Datasets:

Hendrik-a

/

arxiv

like 0

we present and photometry of the resolved stars in the most metal - deficient blue compact dwarf galaxy known , i zw 18 (/50 ) , using _ hubble space telescope_/advanced camera for surveys ( acs ) images , the deepest ones ever obtained for this galaxy . the resulting vs. color - magnitude diagram ( cmd ) reaches limiting magnitudes = = 29 mag . it reveals a young stellar population of blue main - sequence ( ms ) stars ( age 30 myr ) and blue and red supergiants ( 10 myr age 100 myr ) , but also an older evolved population of asymptotic giant branch ( agb ) stars ( 100 myr age 500 myr ) . we derive a distance to i zw 18 in the range 12.6 mpc 15 mpc from the brightness of its agb stars , with preferred values in the higher range . the red giant branch ( rgb ) stars are conspicuous by their absence , although , for a distance of i zw 18 15 mpc , our imaging data go 1 2 mag below the tip of the rgb . thus , the most evolved stars in the galaxy are not older than 500 myr and _ i zw 18 is a bona fide young galaxy_. several star formation episodes can be inferred from the cmds of the main body and the c component . there have been respectively three and two episodes in these two parts , separated by periods of 100 200 myr . in the main body , the younger ms and massive post - ms stars are distributed over a larger area than the older agb stars , suggesting that i zw 18 is still forming from the inside out . in the c component , different star formation episodes are spatially distinct , with stellar population ages decreasing from the northwest to the southeast , also suggesting the ongoing build - up of a young galaxy .

astro-ph0408391

the question of whether there are young galaxies in the local universe forming stars for the first time is of considerable interest for galaxy formation and cosmological studies . there are several reasons for this . first , cold dark matter models predict that low - mass dwarf galaxies could still be forming at the present epoch because they originate from density fluctuations considerably smaller than those giving rise to the giant ones . thus the existence of young dwarf galaxies in the local universe would put strong constraints on the primordial density fluctuation spectrum . second , while much progress has been made in finding large populations of galaxies at high ( @xmath7 @xmath8 3 ) redshifts ( e.g. * ? ? ? * ) , truly young galaxies in the process of forming remain elusive in the distant universe . the spectra of those far - away galaxies generally indicate the presence of a substantial amount of heavy elements , implying previous star formation and metal enrichment . thus it is important to have examples of bona fide young galaxies in the local universe because they can be used as laboratories to study star formation and chemical enrichment processes in environments that are sometimes much more pristine than those in known high - redshift galaxies . moreover , their proximity allows studies of their structure , metal content , and stellar populations with a sensitivity , precision , and spatial resolution that faint distant high - redshift galaxies do not allow . finally , in the hierarchical model of galaxy formation , large galaxies result from the merging of smaller structures . these building - block galaxies are too faint and small to be studied at high redshifts , while we stand a much better chance of understanding them if we can find local examples . the blue compact dwarf ( bcd ) galaxy i zw 18 is one of the best candidates for being a truly young galaxy . @xcite described it as a double system of compact galaxies , which are in fact two bright knots of star formation with an angular separation of 58 . these two star - forming regions are referred to as the brighter northwest ( nw ) and fainter southeast ( se ) components . they form what we will refer to as the main body ( figure [ fig1 ] ) . the presence of ionized gas emission ( e.g. * ? ? ? * ) and of wolf - rayet stars in the nw component @xcite suggest active ongoing star formation in the main body . later studies by @xcite and @xcite have revealed a more complex optical morphology . they pointed out the existence of a prominent diffuse feature @xmath6 22 northwest of the main body , hereafter called the c component ( figure [ fig1 ] ) , which contains a blue irregular star - forming region . @xcite , @xcite and @xcite have shown the c component to have a systemic radial velocity equal to that of the ionized gas in the main body , thus establishing its physical association to i zw 18 . furthermore , the interferometric h i map of i zw 18 by @xcite has shown that the c component is embedded in a common h i envelope with the main body . i zw 18 remains the most metal - poor bcd and the lowest metallicity star - forming galaxy known since its discovery by @xcite . later spectroscopic observations by @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and @xcite have confirmed its low metallicity , with an oxygen abundance of only @xmath6 1/50 the solar value . many studies have focused on the evolutionary state of i zw 18 . @xcite and @xcite have suggested that it may be a young galaxy , recently undergoing its first burst of star formation . the latter authors concluded from _ hubble space telescope _ ( _ hst_)/wide field and planetary camera 2 ( wfpc2 ) images that the colors of the diffuse unresolved component surrounding the se and nw regions are consistent with a population of b and early a stars , with no evidence for older stars . from color - magnitude diagram ( cmd ) studies of the main body based on other _ hst_/wfpc2 images , @xcite have found that star formation in i zw 18 began at least 30 50 myr ago and is still continuing to the present . @xcite have discussed the star formation history in i zw 18 using the same _ hst_/wfpc2 data as @xcite and @xcite , but going deeper thanks to a more sophisticated treatment of the data . by comparing the observed cmds and luminosity functions with the synthetic ones , they concluded that there were two episodes of star formation in i zw 18 . the first star formation episode occurred @xmath6 0.5 1 gyr ago , as evidenced by the presence of asymptotic giant branch ( agb ) stars . this age is more than 10 times larger than the one obtained by @xcite . the more intense second star formation episode took place between 15 and 20 myr ago . subsequently , @xcite has carried out a near - infrared ( nir ) cmd study based on _ hst_/near infrared camera and multi object spectrometer ( nicmos ) @xmath9 ( f110w ) and @xmath10 ( f160w ) images . he concluded that the main body of i zw 18 is dominated by two populations , a 10 20 myr population of red supergiants ( rsg ) and a considerably older ( 0.1 5 gyr ) population of agb stars . because of its faintness , the c component has not been studied in as much detail . its optical spectrum @xcite reveals a blue continuum with weak balmer absorption features and faint h@xmath11 and h@xmath12 in emission , suggesting that the h ii region is ionized by a population of early b stars . in a @xmath13 vs. @xmath0 cmd analysis of the c component , @xcite have found a well - defined stellar upper main - sequence ( ms ) of blue stars with ages @xmath6 40 myr . however , numerous faint red stars are also present , with ages @xmath6 100 300 myr . those authors concluded that the c component consists of an older stellar population with an age of several hundred myr , but that it has experienced more recently a modest starburst in its southeastern half , as evidenced by the presence of blue stars and h@xmath11 emission . @xcite have estimated an age for the c component not exceeding 200 myr . using integrated optical and nir photometric colors derived from ground - based and archival _ hst _ images , @xcite and @xcite have shown that the age of i zw 18 as a whole , including both the main body and the c component , does not exceed 500 myr . this is in agreement with the upper age limits obtained by @xcite , @xcite and @xcite , but in contradiction with the large one of 5 gyr obtained by @xcite . @xcite and @xcite have proposed an extreme scenario in which star formation in i zw 18 proceeds , not only in recent short episodic bursts , but also in a continuous fashion at a low level over a long period of many gyr . this would result in an extended underlying low - surface - brightness ( lsb ) red stellar component in i zw 18 . however , @xcite and @xcite have shown that the existence of such a component is supported neither by spectroscopic nor photometric measurements of the extended emission around i zw 18 . it is evident that despite considerable work , the debate on the age of i zw 18 is still not settled . the evolutionary status of i zw 18 remains elusive because of several limitations in the preceding work . previous studies of stellar populations in i zw 18 based on integrated photometric colors or on spectral synthesis do not provide a unique star formation history . cmd studies constrain better the star formation history , but until now the available _ images on which these studies are based did not go deep enough below the tip of the red giant branch ( trgb ) to reveal whether a faint old ( @xmath14 1 2 gyr ) stellar population of red giant branch ( rgb ) stars exists in i zw 18 or not . evidently deeper _ hst _ observations were needed . the installation of the advanced camera for surveys ( acs ) on _ hst _ in 2003 made those deeper observations possible . we use here new deep _ hst_/acs imaging of i zw 18 in @xmath0 and @xmath1 to go @xmath6 1 2 mag below the trgb and put more stringent constraints on the state of evolution of the bcd . the observations and data reduction are described in sect . 2 . the lack of a rgb population in the cmd of i zw 18 is examined in sect . the distance determination to i zw 18 is discussed in sect . cmds of the stellar populations are studied in sect . we find that i zw 18 does not contain stars older than @xmath6 500 myr , making it a truly young galaxy . our results are summarized in sect . we have obtained _ hst _ images of i zw 18 during the 2003 may 26 june 6 period in the course of eight visits totaling 25 orbits , with the acs / wide field camera ( wfc ) detector through filters f555w and f814w , which we will refer to hereafter as @xmath0 and @xmath1 . the @xmath0 exposure was obtained during five visits totaling 16 orbits and the @xmath1 exposure during three visits with a total of 9 orbits . all observations were obtained with the same orientation of the field of view , with one exposure per orbit , split into two subexposures to permit cosmic ray removal . drizzling was applied so that separate exposures obtained during one visit were slightly offset with respect to one another to permit better spatial sampling . the total exposure time was 43520s ( @xmath6 12.1 hours ) in @xmath0 and 24300s ( @xmath6 6.8 hours ) in @xmath1 . the galaxy was positioned on the wfc1 frame , with the scale being 005 per pixel . preliminary processing of the raw images was done at the space telescope science institute through the standard pipeline . this resulted in distortion - corrected and drizzled images , six in the @xmath0 band and three in the @xmath1 band , which we used in our subsequent reductions with iraf . the separate @xmath0 and @xmath1 images were then combined with a coregistration better than 0.1 pixel and cosmic rays removed . the acs @xmath0 and @xmath1 images of i zw 18 are shown in figures [ fig1]a [ fig1]b , rotated so that north is up and east is to the left . the main body and the c component are labeled in figure [ fig1]b . their enlarged @xmath0 and @xmath3 views are displayed respectively in figures [ fig2]a [ fig2]b and figures [ fig2]c [ fig2]d . the dark regions in the @xmath3 images correspond to blue colors and the white regions to red colors . the extended and filamentary dark regions in the main body ( figure [ fig2]b ) represent ionized gas emission from supernova shells , while extended white sources are background red galaxies . note the presence of several blue stars at large distances from the main body , suggesting recent or ongoing star formation in the halo . however , the extended halo of red stars , which is a common feature of many dwarf galaxies observed with the _ hst _ ( e.g. in ngc 2366 , * ? ? ? * ) , is conspicuous by its absence in i zw 18 . no appreciable ionized gas emission is seen in the c component ( figures [ fig2]c [ fig2]d ) despite the presence of many blue stars . the superior spatial resolution of the _ hst_/acs images permits to resolve individual stars and study stellar populations in i zw 18 by means of cmds . thanks to the higher sensitivity of the acs and longer exposure times , our data is considerably deeper than previous _ hst _ observations of i zw 18 with the wfpc2 . we used the daophot package in iraf to perform point - spread - function ( psf ) fitting photometry in the sky area shown in figure [ fig1 ] . thanks to the drizzling procedure , the several bright and relatively isolated stars around the main body and the c component ( figures [ fig2]a and [ fig2]c ) allow us to derive reliable and well sampled stellar psfs . we obtain full widths at half maximum ( fwhm ) of the stellar profiles of 2.5 pixels ( 0125 ) and 2.8 pixels ( 0140 ) respectively for the @xmath0 and @xmath1 images . as for the aperture radius to use for stellar photometry , we experimented with radii between 2 and 4 pixels to search for the best value giving at the same time a high recovery rate of point sources and small aperture corrections to the derived magnitudes . we find that the best compromise is an aperture radius of 3 pixel ( 015 ) . with smaller radii the aperture corrections are too large ( the calibrating aperture radius is 05 ) and introduce unwanted uncertainties in the photometry . with larger radii the number of recovered stars is sharply reduced , because of the crowding effect which is especially important in the main body . the background level was measured in an annulus with radii 4 and 6 pixels ( 02 and 03 ) around each source and subtracted . the zero - points of the photometry are defined as the magnitude of a star which produces a count rate of one electron per second in a given filter and are set so that the star vega has magnitude 0 at all wavelengths . this gives zero - points of 25.711 mag and 25.487 mag for @xmath0 ( f555w filter ) and @xmath1 ( f814w filter ) respectively . photometric aperture corrections were obtained by a comparison of psf - fitted magnitudes of 12 bright isolated stars with the magnitudes of the same stars measured with the aperture photometry technique within a 10 pixel ( 05 ) aperture . we obtained the corrections @xmath15(05 ) @xmath16 = 0.18 mag and @xmath17(05 ) @xmath18 = 0.23 mag . correction for charge - transfer efficiency loss has been carried out according to the prescriptions of @xcite . we find however , that this effect is negligible ( less than 0.01 mag for sources brighter than 29 mag ) given the high level of the night sky in our long - exposure images . to check our photometry , we compare the magnitudes of several bright isolated stars in the main body , measured through a large aperture of 10 pixel radius , with those measured using the same aperture in the wfpc2 images of @xcite . we find that our magnitudes are fainter by 0.05 mag in @xmath0 and @xmath1 . to put our results on the same scale as the wfpc2 data , we have applied that small correction to our data . since no photometric correction to infinite aperture is yet available for the acs data , we have not considered it . in total , taking into account the aperture corrections and the slight shift between the acs and wfpc2 magnitude scales , the corrections to our psf photometry amount to 0.23 mag in @xmath0 and 0.28 mag in @xmath1 . to minimize the number of false detections , we have adopted the minimum detection level to be 3@xmath19 above the sky noise and a sharpness in the 1.0 + 1.0 range . figure [ fig3 ] shows the distribution of photometric errors as a function of @xmath0 and @xmath1 magnitudes as determined by daophot . it is seen that errors are about 0.2 mag at @xmath0 = 29 mag and @xmath1 = 28.5 mag . they increase to about 0.4 mag at @xmath0 = 30 mag and @xmath1 = 29.5 mag . note that at bright magnitudes ( @xmath20 @xmath4 27 mag ) , the error distribution is very broad at a given brightness because of the effects of crowding , ionized gas emission and a high background level of the unresolved stellar component in the brightest regions of the main body . the total numbers of recovered stars are respectively 1599 , 3555 and 671 in the @xmath0 image , the @xmath1 image and in both of these images at the same time . we have adopted a matching radius of 1 pixel , although changing that radius within the range 0.5 2 pixels does not change appreciably the number of stars in both @xmath0 and @xmath1 frames . that more than half the stars are not matched is likely caused by incompleteness effects and an increasing number of false detections at faint magnitudes . indeed , the number of the stars with @xmath0 = 26 27 mag recovered in both @xmath0 and @xmath1 images is @xmath6 74% that recovered in the @xmath0 image alone , while the number of the stars with @xmath0 = 29 30 mag recovered in both @xmath0 and @xmath1 images is only @xmath6 5% that recovered in the @xmath0 image alone . the transformation of instrumental magnitudes to the johnson - cousins @xmath21 photometric system as defined by @xcite was performed according to the prescriptions of @xcite . the magnitudes and colors of point sources were corrected for galactic interstellar extinction adopting @xmath22 = 0.106 mag @xcite . we have carried out a completeness analysis using the daophot routine addstar . since the crowding level is more important in the main body than in the c component , the analysis was done for each component separately , in the regions determined by the rectangles in figure [ fig4 ] . for each frame we have added artificial stars amounting to @xmath6 5% the number of real stars detected inside each rectangle . we then performed a new photometric reduction using the same procedure as the one applied to the original frame , and checked how many added stars were recovered . this operation was repeated 10 times for each frame and each magnitude bin and the results were averaged . the completeness factor in each magnitude bin defined as the percentage of recovered artificial stars is shown in table [ tab1 ] . it can be seen that the completeness is better for the c component because of its lesser crowding : it is about 69% in @xmath0 and 44% in @xmath1 in the 28 29 mag range , but drops to @xmath6 17% and @xmath6 10% in the 29 30 mag range . for the main body , the completeness is about 55% in @xmath0 and 39% in @xmath1 in the 28 29 mag range , but drops to @xmath6 10% and @xmath6 5% in the 29 30 mag range . figure [ fig5 ] displays the combined @xmath1 vs. @xmath3 cmd for the main body and the c component . the 50% detection limit in both @xmath0 and @xmath1 filters is shown by a dashed straight line . our cmd is more than two magnitudes deeper than the cmds obtained earlier for i zw 18 @xcite . a well populated ms can be seen at @xmath3 @xmath6 0 mag , which begins with the brightest stars ( @xmath1 @xmath6 23.5 mag ) and goes all the way down to 28.5 mag . many bright blue loop ( bl ) and rsg stars with @xmath3 @xmath14 0 mag are present in the upper part of the diagram . two clumps of agb stars with @xmath1 @xmath6 25 mag and @xmath6 26 mag , and with @xmath3 redder than 1.0 mag are detected . there is , however , a conspicuous lack of a rgb stellar population , despite the fact that our data , depending on the adopted distance to i zw 18 , goes @xmath6 1 2 mag fainter than the trgb . only a few faint red ( @xmath3 @xmath23 1.3 mag ) sources close to the detectability limit ( @xmath1 @xmath23 26.6 mag ) are present in the combined cmd . they may be false detections because of their faintness in @xmath0 ( @xmath14 28.5 mag ) . they may also be real stars with erroneous colors because of the large photometric uncertainties ( see their error bars in figure [ fig5 ] ) and/or with colors reddened by dust . their possible nature will be discussed in sect . we have indicated the expected location of the rgb in the cmd by superposing on the data points the isochrone for the very metal - deficient and old globular cluster m15 . its metallicity [ fe / h ] is 2.17 @xcite , lower than the metallicity of the ionized gas in i zw 18 . we have used m15 because it has the lowest metallicity in the @xcite globular cluster sample , and because its isochrone fits well the rgb of the very metal - deficient bcd ugc 4483 ( @xmath2/23 ) @xcite . isochrones of globular clusters with higher metallicities are too red as compared to the data . the dotted , dashed and solid lines show respectively , from left to right , the location of the m15 isochrone for three distances of i zw 18 : 10 mpc , 12.6 mpc and 15 mpc , corresponding to distance moduli @xmath24 = 30 mag , 30.5 mag and 30.88 mag . the upper extremities of each curve are the locations of the trgb at these respective distances . it is clear that there is a dearth of sources in the expected region of rgb stars . this absence is due to a sensitivity effect since it manifests itself at @xmath6 1 2 mag above the 50% completeness line . only stars detected at levels greater than 3@xmath19 above the sky noise in both @xmath0 and @xmath1 are shown in fig . we have also produced a cmd for i zw 18 where the threshould level is only 1@xmath19 above the sky noise . examination of that cmd still reveals a lack of rgb stars in i zw 18 , despite the fact that the cmd goes down now 1 mag deeper ( limiting @xmath0 and @xmath1 magnitudes of above @xmath6 30 mag ) . the cmd of i zw 18 is thus unique among all cmds of galaxies with resolved stellar populations studied so far with the _ hst_. in those galaxies , the rgb stellar population is invariably present and is many times more populated than the agb stellar population ( see e.g. figure 14 of * ? ? ? * where the cmds of 11 dwarf irregular and bcd galaxies are presented ) . to quantify our conclusions about the absence of a rgb stellar population in i zw 18 , we compare the cmd of i zw 18 with that of a nearby galaxy with a well - populated rgb , moved from its actual distance to the one of i zw 18 , so that the data for the comparison galaxy have a resolution and a sensitivity comparable to those of the i zw 18 observations . the cometary blue compact dwarf galaxy ugc 4483 located at the distance of [email protected] mpc is an ideal galaxy for such a comparison . with a metallicity of the ionized gas equal to 1/23 that of the sun , it is one of the bcds with a heavy element content closest to that of i zw 18 . it has been resolved into individual stars with _ hst_/wfpc2 observations by @xcite . the rgb stellar population in ugc 4483 with @xmath26(trgb ) @xmath6 1.3 mag is one of the bluest known among bcds . its cmd has been analyzed by @xcite and is given in fig . we discuss in the following how to perform the resampling of the data for ugc 4483 when moving it to the larger distance of i zw 18 . we need to take into account the fading of stars , the increasing errors of the stellar apparent magnitudes and the increasing crowding effects . first , we consider how the fading of a star affects the probability of its detection . assuming that ugc 4483 is at the distance of i zw 18 and that it is observed with the acs with the same exposure times in @xmath0 and @xmath1 as for i zw 18 , then the detection probability of a star in the cmd of fig . [ fig9]c with magnitudes @xmath0 and @xmath1 is @xmath27 where @xmath28(@xmath29 ) and @xmath28(@xmath30 ) denote the incompleteness factors of the acs observations ( table [ tab1 ] ) , and @xmath29 and @xmath30 are the magnitudes of a star when moved to the distance of i zw 18 . although the incompleteness factors are different for the main body and the c component , we have used only the incompleteness factors for the latter in our modeling . this is because the contamination by ionized gas emission in ugc 4483 is more similar to that in the c component than in the main body of i zw 18 , where it is considerably higher . next we consider the crowding effect : not all stars that are resolved in ugc 4483 at its distance of 3.4 mpc will be resolved at the distance of i zw 18 . we assume that this effect is independent of apparent magnitude . in fig . [ fig6 ] we show the distribution of distances to the nearest companion of each star in i zw 18 ( a ) and in ugc 4483 ( b and c for the parts of the galaxy imaged respectively by the pc and by the wf2+wf3+wf4 frames , hereafter wf frames ) . all distances are expressed in acs pixels . it is seen from fig . [ fig6]a and [ fig6]b that both stars in a pair are resolved by the acs and the pc only if their separation is more than 3 acs pixels . the lower separation limits of the acs and the pc are similar because their resolutions are nearly the same . on the other hand , in the wfpc2/wf frames , because of their twice as worse angular resolution , both stars in a pair are recovered only if their separation is more than @xmath6 6 acs pixels . because the linear resolution at a distance of 15 mpc as compared to one of 3.4 mpc is 15/3.4 = 4.4 times worse , of the detected stars in ugc 4483 , only the ones in pairs with separations @xmath14 [email protected] @xmath32 13 acs will be resolved at the distance of 15 mpc . the parts of the distributions occupied by the resolved pairs of stars are shown by shaded regions in fig . [ fig6]b and [ fig6]c . they correspond to a fraction of stars of @xmath6 0.06 for the pc frame and of @xmath6 0.51 for the wf frames . the remaining pairs with smaller separations are not resolved . we assume that for those , only one of the two stars in a pair is recovered . this corresponds to a fraction of stars in unresolved pairs of 0.94/2 = 0.47 for the pc frame and of 0.49/2 = 0.25 for the wf frames . however , these fractions can be even smaller if there are stars close to the unresolved pairs . to estimate this effect , we consider the distribution of separations between the unresolved pairs of stars and the next nearest star . we find that only @xmath32 60% of unresolved pairs have a nearest star with a separation @xmath14 13 acs pixels . we calculate the fraction of the stars @xmath33 which is recovered at the distance of 15 mpc to be : 0.06 + 0.94/[email protected] @xmath32 0.34 for the pc frame , and 0.51 + 0.49/[email protected] @xmath32 0.66 for the wf frames . in similar manner , we derive @xmath33 = 0.39 ( pc ) and 0.76 ( wf ) for the distance of 12.6 mpc and 0.49 ( pc ) and 0.87 ( wf ) for the distance of 10 mpc . finally , the total probability to recover a detected star in ugc 4483 at the distance of i zw 18 is @xmath34 = @xmath35@xmath31@xmath33 . we show in fig . [ fig7 ] representative simulated cmds of ugc 4483 for three possible distances of i zw 18 , 10 mpc , 12.6 mpc and 15 mpc . they are generated from the observed cmd shown in fig . [ fig9]c according to the detection probability @xmath34 for each star . in addition , we have also recalculated the magnitude of each star by taking into account the errors in the magnitude measurements . for a given stellar magnitude , we have adopted a 1@xmath19 deviation from a fit to the data in fig . [ fig3 ] and distributed the errors in a gaussian fashion . the apparent magnitude scale for each panel ( left ordinate ) has been adjusted so that the absolute magnitude scale ( right ordinate ) is the same in all panels . in all three panels , we also show the isochrone for the globular cluster m15 by a thick line . the two parallel lines in panels ( b ) and ( c ) delineate the location of rgb stars . we conclude from fig . [ fig7 ] that the rgb of ugc 4483 is clearly visible at all three distances . even if the rgb is sparser at the distance of 15 mpc , it is still clearly there . in all likelihood , the absence of a rgb in i zw 18 is not the result of a selection effect . we can quantify the above statement by comparing the observed distribution of the numbers of stars between the two parallel lines ( fig . [ fig7]b , c and fig . [ fig9]a , b ) as a function of @xmath1 magnitude for i zw 18 and for ugc 4483 at distances of 12.6 and 15 mpc . the distributions are shown in fig . we mark by short vertical lines the locations of the trgb at the corresponding distances . it seen that in both galaxies the numbers of agb stars brighter than the trgb are comparable . however , the numbers of stars fainter than the trgb are much larger in the recalculated cmds of ugc 4483 than in the cmd of i zw 18 . a kolmogorov- smirnov test shows that , in the 26 28.5 mag magnitude range , the observed distribution of stars in i zw 18 is different at the @xmath23 99% and @xmath6 96% levels from those in ugc 4483 , for distances of 12.6 mpc and 15 mpc respectively . for a distance of 10 mpc , the difference in the distributions of stars is at the 99.9% level . we have chosen a broad region between the two parallel lines for the construction of the histograms shown in fig . the differences between the histograms of ugc 4483 and that of i zw 18 would be considerably larger if we make that region narrower , and in particular if we exclude from consideration the stars in i zw 18 that are concentrated at the blue edge of the region 0.2 0.3 mag blueward of the m15 isochrone . there are reasons to think that these blue stars may not be rgb stars . since the metallicity of the stars in i zw 18 is unlikely to be much lower than the metallicity [ fe / h ] = 2.17 of m15 , their blue colors are likely due to an age effect . hence these stars are likely considerably younger than the @xmath6 10 gyr old rgb stars in m15 . furthermore , the distributions of stars in this region are very different in i zw 18 and in ugc 4483 . the rgb stars in the cmds of ugc 4483 are distributed more uniformly and not in a narrow blue edge as seen in i zw 18 . there has been some debate about the exact distance of i zw 18 . values that have been proposed range from @xmath6 10 mpc @xcite assuming the observed heliocentric radial velocity of the galaxy of @xmath6 750 km s@xmath36 is a pure hubble flow velocity and a hubble constant of 75 km s@xmath36 mpc@xmath36 , to @xmath6 12.6 mpc @xcite based on a virgocentric flow model , to @xmath6 15 mpc @xcite based on photometric and spectroscopic properties of i zw 18 . with the absence of a rgb stellar population , we can not use the trgb as a distance indicator . other arguments need to be invoked . the redshift distance of i zw 18 without correction for virgocentric infall is 10 mpc . if this distance is correct , then the @xmath1 magnitude of the trgb would be @xmath6 26 mag , corresponding to the upper extremity of the dotted line in figure [ fig5 ] , i.e. it would appear brighter than some agb stars . this seems unlikely as agb stars are usually brighter than the trgb . for example , the agb stars in the bcd vii zw 403 which contains numerous rgb stars are @xmath6 0.5 mag brighter than the trgb @xcite . in another bcd , ugc 4483 , the agb stars are brighter than the trgb by @xmath14 0.7 mag @xcite . thus , we conclude that a distance of 10 mpc to i zw 18 is too small . there is other evidence that goes in the same sense . @xcite , by correcting the radial velocity of i zw 18 for virgocentric infall , obtained @xmath37 = 12.6 mpc . @xcite and @xcite suggested that i zw 18 should be as distant as 15 mpc . thus , @xcite found that the ionized gas emission seen in the southeastern and central parts of the c component can only be produced by stars with ages less than @xmath6 15 myr or with masses @xmath14 10 15 @xmath38 . at a distance of 10 mpc , the absolute luminosities of the brightest stars in the cmd of the c component would not be large enough for such massive stars ( e.g. * ? ? ? furthermore , @xcite have compared the cmds of 11 galaxies ( five bcd and irregular galaxies outside the local group and six local group irregular galaxies ) observed with the _ hst _ and found that the absolute magnitudes of the brightest stars in the main body of i zw 18 would be systematically fainter than those in these galaxies if i zw 18 is at 10 mpc , which is not plausible . increasing the distance of i zw 18 to 12.6 mpc still does not match the observations . the youngest bright stars in the cmd of the c component would still have too low absolute luminosities and too large ages ( @xmath14 25 30 myr ) to account for the ionized gas in the c component . we estimate the distance to i zw 18 by comparing the absolute magnitudes of the agb stars in i zw 18 with those in other bcds . in figure [ fig9 ] we compare the cmds of i zw 18 ( figures [ fig9]a and [ fig9]b ) and the bcd ugc 4483 ( figure [ fig9]c ) . we consider two distances for i zw 18 , 12.6 mpc ( figure [ fig9]a ) as proposed by @xcite and 15 mpc as proposed by @xcite and @xcite ( figure [ fig9]b ) . as in fig . [ fig7 ] , the scales of apparent magnitudes ( left ordinate in each panel ) are adjusted so that the absolute magnitude scale ( right ordinate in each panel ) is the same in all panels . the thick solid line is the isochrone of the globular cluster m15 , the dashed line in figures [ fig9]a [ fig9]b shows the 50% completeness limit . two clumps of agb stars with @xmath1 @xmath6 25 mag and 26 mag are seen in the cmd of i zw 18 . for each cmd , we have derived the mean magnitudes of the agb stars in the shaded regions . in the case of i zw 18 , this corresponds to the fainter clump . the short horizontal lines in fig . [ fig9 ] indicate the mean absolute magnitude of agb stars thus obtained . the absence of rgb stars implies that the agb stars of the fainter clump in i zw 18 should be at least as bright in absolute magnitude as those in ugc 4483 , since they are descendants of more massive stars . however , at the distance of 12.6 mpc ( figure [ fig9]a ) the agb stars in i zw 18 are fainter by 0.25 mag than those in ugc 4483 ( figure [ fig9]c ) . their absolute magnitudes are similar to those of older agb stars in the bcd vii zw 403 @xcite . if we increase the distance of i zw 18 to 15 mpc , then the agb stars in i zw 18 are 0.13 mag brighter , a much more satisfactory state of affair ( figures [ fig9]b and [ fig9]c ) . in fact , the agb stars in i zw 18 and in ugc 4483 have the same absolute magnitudes if the distance to i zw 18 is 14.1 mpc , corresponding to a distance modulus of 30.75 mag . poor statistics of the agb stars in the cmd of i zw 18 and uncertainties in their real absolute magnitudes preclude a more precise determination of the distance to the bcd . all we can say is that it is somewhere between 12.6 mpc and 15 mpc , with the most likely value being in the upper range . our acs data are considerably deeper than all previous imaging of i zw 18 , obtained with the wfpc2 or nicmos . for comparison , the faintest red stars with @xmath1 @xmath6 25 mag and @xmath3 @xmath14 1 mag in the cmd of @xcite ( their figure 7b ) correspond to the brightest agb stars in our cmd ( figure [ fig5 ] ) . likewise , the faintest red stars with @xmath39 @xmath6 7.5 mag ( at the distance of 12.6 mpc ) detected by @xcite correspond to the same brightest agb stars in our data . these agb stars are @xmath6 2.5 mag brighter than the trgb absolute magnitude of @xmath6 5 mag in the @xmath10 band @xcite . furthermore , the acs observations of the c component have two times better spatial resolution as compared to the previous wfpc2 observations ( 005 instead 01 ) . thus , there is no doubt that our deep images allow to study much fainter stellar populations and put more stringent constraints on the evolutionary state of i zw 18 . we have already discussed in section 3 the absence of a rgb stellar population in i zw 18 . we now examine the stellar populations that are present . the cmds for the main body and c component are shown in figures [ fig10]a and [ fig10]b , respectively . the distance of 15 mpc to i zw 18 is adopted hereafter . overplotted on the data are geneva theoretical isochrones ( solid lines ) of single stellar populations for a heavy element mass fraction @xmath40 = 0.0004 @xcite , which corresponds to the metallicity of the ionized gas in i zw 18 . each isochrone is labeled by the logarithm of the age in years . we choose to use geneva instead of padova isochrones @xcite because the latter are not able to reproduce the isochrone of the globular cluster m2 with a similar metallicity , [ fe / h ] = 1.58 @xcite , while the agreement is better with geneva isochrones . several star formation episodes in the main body can be inferred from the cmd in figure [ fig10]a and from other observational data . the ongoing star formation with age @xmath6 4 myr is evidenced by the ionized gas emission ( figures [ fig2]a [ fig2]b ) and the presence of wr stars in the nw component @xcite , located in two compact clusters @xcite . the brightest post - ms star in the cmd with an absolute magnitude @xmath41 @xmath6 9.7 mag and an age of @xmath6 5 6 myr is also part of the present burst . numerous supergiants with @xmath41 @xmath4 8 mag indicate that intense star formation occured @xmath6 10 15 myr ago . there is evidence for two older star formation events in the cmd . a first star formation episode occurred @xmath6 200 myr ago as indicated by the bright agb stars with @xmath41 @xmath6 6 mag . a second star formation happened @xmath6 300 500 myr ago , which is responsible for the oldest stars in the main body , the agb stars with @xmath41 @xmath6 5 mag . these age estimates are in agreement with the previous cmd analyses by @xcite , @xcite and @xcite , although those authors discussed only one previous star formation episode , because their data did not go as deep as ours and the 300 500 myr old agb stars were not seen . most importantly , no old agb stars with ages @xmath14 1 gyr are seen , contrary to the assertion of @xcite on the basis of his _ hst_/nicmos cmd . the presence of such old agb stars would have required numerous rgb stars which are not present in our cmd . we turn next to the stellar populations in the c component . here , there is no substantial ongoing massive star formation as intense ionized gas emission is not seen . this is corroborated by the upper part of the cmd , which is nearly devoid of bright stars ( figure [ fig10]b ) . the lone bright source and the brightest one in the cmd , at @xmath1 @xmath6 22 mag ( shown by a large filled circle in figure [ fig10]b and labeled `` c '' in figure [ fig2]c and figure [ fig10]b ) , is in fact not a single star but represents the central cluster discussed by @xcite . it is partially resolved in our images with a fwhm @xmath6 015 , corresponding to a linear size of @xmath6 10 pc . there is a second resolved cluster ( also shown in figure [ fig10]b by a large filled circle and labeled `` nw '' in figure [ fig2]c and figure [ fig10]b ) with a fwhm @xmath6 021 , corresponding to a linear size of @xmath6 15 pc . it was also noted by @xcite . two star formation episodes can be deduced from the cmd . the most recent one happened @xmath6 15 20 myr ago while the older one took place @xmath6 200 300 myr ago . there is an indication that the c component is slightly younger than the main body because its oldest agb stars are slightly brighter on average ( by @xmath6 0.2 @xmath1 mag ) . our age estimates for the c component are also in agreement with those of @xcite and @xcite based on the _ hst_/wfpc2 data . in figure [ fig11 ] we show cmds for the different regions in the main body and in the c component as delimited in figure [ fig4 ] . ongoing and recent past star formation in the main body is mainly localized in the mii region associated with the se component , and in the miii region associated with the nw component . this is evidenced by the presence of numerous bright ms ( age @xmath4 10 myr ) and bl+rsg ( 10 myr @xmath4 age @xmath4 100 myr ) stars in both cmds ( figures [ fig11]b [ fig11]c ) . some agb stars are also present in region mii . on the other hand , only a few rsg stars are seen in the cmd of the southernmost region mi . the population of agb stars is more numerous there , indicating older star formation ( age @xmath14 100 myr ) ( figure [ fig11]a ) . in the cmd of the southernmost region ci of the c component ( figure [ fig11]d ) only stellar populations with ages @xmath4 100 myr are present . the age of the youngest stars in this region is @xmath6 15 20 myr as evidenced by the brightest post - ms stars ( @xmath41 @xmath6 8 mag ) in the cmd and the presence of weak h@xmath12 and h@xmath11 emission lines in its spectrum ( e.g. , * ? ? ? * ) . in the central region cii ( figure [ fig11]e ) , older stars with ages @xmath6 200 300 myr are present . there is a population of younger stars , with ages @xmath6 30 myr as deduced from the @xmath41 @xmath6 6.4 mag of the brightest post - ms stars . although they are not resolved , the youngest stars ( age @xmath6 15 myr ) are likely present in the central cluster , as evidenced by the presence of ionized gas emission there @xcite . region ciii is the oldest ( age @xmath14 300 myr ) region in the c component ( figure [ fig11]f ) . there is no recent star formation here since no bright ms star is seen . the agb stars are fainter than those in region cii , also suggesting a larger age . probably , the youngest object in region ciii is the nw stellar cluster . however , the absence of ionized gas emission around this cluster @xcite and its relatively red @xmath3 @xmath6 0.35 mag ( figure [ fig11]f ) implies that its age is @xmath14 100 myr . all previous cmd studies of i zw 18 did not go deep enough to allow setting an upper limit to its age . our new deep acs images permit us to do so . for a distance of i zw 18 in the range between 12.6 mpc and 15 mpc , no rgb stars are seen in our cmds , and the age upper limit can be set to 1 2 gyr . however , the real age for i zw 18 is likely smaller as only stars with ages @xmath4 500 myr are present ( figures [ fig10 ] and [ fig11 ] ) . this age upper limit of @xmath6 500 myr is in excellent agreement with the age estimates obtained by @xcite and @xcite by examining the integrated optical and near - infrared colors of i zw 18 . while we favor a distance of i zw 18 in the upper range of the 12.6 mpc 15 mpc interval , our conclusions will be reenforced if the smaller distance of 12.6 mpc is adopted for the bcd . in this case , it would be easier to detect faint rgb stars since the trgb s apparent magnitude would be up to 0.38 mag brighter ( compare figures [ fig9]a and [ fig9]b with the simulated cmds for ugc 4483 in figures [ fig7]b and [ fig7]c ) . the ages of the oldest stellar populations will increase at most by a few hundred myr . however , we have argued in section 4 that such a small distance would not be in agreement with photometric and spectroscopic observations of i zw 18 . the spatial distributions of stars with different ages in i zw 18 can give useful information on its star formation history . to carry out the study , we divide the stars in the cmd in figure [ fig5 ] into three categories : 1 ) ms stars with @xmath3 @xmath42 0.05 mag , 2 ) bl stars with 0.05 mag @xmath5 @xmath3 @xmath42 0.4 mag and rsg stars with @xmath1 @xmath42 24 mag and @xmath3 @xmath8 0.4 mag , and 3 ) agb stars with 26.4 mag @xmath5 @xmath1 @xmath5 24 mag and @xmath3 @xmath14 0.8 mag . we have also included in the last category the progenitors of the agb stars located in the lower right corner of the cmd with 26.4 mag @xmath5 @xmath1 @xmath5 27.5 mag and 0.6 mag @xmath5 @xmath3 @xmath5 1.2 mag . the boundaries of the regions in the cmd of i zw 18 where these stars are located are shown in figure [ fig5 ] . the spatial distributions of different types of stars are displayed by in figures [ fig12]a [ fig12]d . the open circles indicate the locations of the nw and se components in the main body and the central star cluster in the c component . the spatial distribution of all stars irrespective of type is shown in figure [ fig12]a . it is seen that the vast majority of the stars are located in the main body and in the c component . the few scattered points outside i zw 18 are probably false detections . in the main body , the ms and bl+rsg stars ( figures [ fig12]b [ fig12]c ) are distributed uniformly , suggesting that ongoing and recent past star formation took place over the whole body . on the other hand , the agb stars are mainly located in the southeastern part ( figure [ fig12]d ) . this was noted earlier by @xcite and @xcite based on wfpc2 images . the absence of agb stars in the northwestern region may be caused in part by a more severe crowding and a larger contribution of ionized gas emission in the @xmath0 image ( figures [ fig2]a [ fig2]b ) . this may make daophot miss more of the agb stars since they are fainter in @xmath0 than the ms and bl+rsg stars . however , we believe that some of the effect is real . the most striking feature is that the older agb stars are not more spread out spatially than the younger stars , a fact already noted by @xcite . such a distribution is drastically different from the situation in other galaxies where the old agb and rgb stars are distributed over a considerably larger area as compared to younger stars , due to diffusion and relaxation processes of stellar ensembles . if anything , the reverse appears to be true here : the ms and bl+rsg stars are distributed over a larger area around the main body as compared to the agb stars . this suggests that star formation in the past responsible for the agb stars was more concentrated in the main body , while recent star formation responsible for the ms and bl+rsg stars is more spread out . evidently , i zw 18 is a galaxy in the process of forming from the inside out . the absence of a halo of agb stars and of their progenitors around i zw 18 , despite an easier detectability at large distances because of less crowding , is in direct contradiction with the scenario of @xcite and @xcite . these authors have proposed that continuous low - mass star formation has proceeded in i zw 18 on cosmological time scales in a large area around the main body . they predict the existence of a red extended underlying stellar component which @xcite claim to have detected from their optical and near - infrared surface photometry , but which was not confirmed by independent photometry by @xcite and @xcite . in fact , the spatial distributions in figure [ fig12 ] suggest that the agb stars in the main body are relatively young and have had no time to migrate to large distances ( cf . * ) . in the c component , the spatial distributions of the different types of stars are very different from those in the main body . the ms stars are seen primarily in the southeastern part and are nearly absent in the northwestern part ( figure [ fig12]b ) . the bl+rsg stars are also mainly distributed in the southeastern part , however their location is offset as compared to the ms stars : the ms stars are mostly aligned in the east - west direction while the bl+rsg stars are mainly distributed in the southeast - northwest direction . on the other hand , agb stars are located principally in the northwestern part . some of these stars are also present to the south of the central cluster . it appears that the two major episodes of star formation in the c component have occurred in spatially different regions . this is characteristic of the mode of star formation in bcds where the centers of star formation move about in a stochastic manner . this also implies that , as for the main body , the formation of the c component is still proceeding . it started in the northwestern part @xmath6 200 300 myr ago , lasting for a relatively short period , and continues now in the southeastern part . that ms stars are not seen in the northwestern part does not mean that such stars are not present there . they are simply too faint to be detected because the ms turn - off for a 200 300 myr old stellar population is @xmath41 @xmath6 1 mag ( e.g. * ? ? ? * ) or @xmath1 @xmath6 30 mag , below the detectability limit . the absence of ms stars also means that no detectable star formation has occurred during the last @xmath6 200 myr in the northwestern part . one possible exception is the nw cluster with the age @xmath14 100 myr . on the other hand , the absence of agb stars in the southeastern region , where ms stars reside , suggests that this region is younger , with an age @xmath4 100 myr . a similar age estimate was made by @xcite from a spectroscopic study . finally , we consider the spatial distribution and nature of the few faintest ( @xmath1 @xmath23 26.6 mag ) and reddest ( @xmath3 @xmath23 1.3 mag ) sources in the cmd of figure [ fig5 ] . they are shown by crosses in figure [ fig12]d . a few of these sources are scattered in the field and are likely false detections . two other are in the c component , while the remaining six are in the southeastern part of the main body . their @xmath3 colors are redder than those of globular cluster stars in m15 . the location of some of these sources in the cmd of the main body to the right of the 10 gyr geneva theoretical isochrone with @xmath40 = 0.0004 ( figure [ fig10]a ) implies that they are older than the universe , which is absurd . the situation is worse when padova isochrones are used . thus , we conclude that the red colors of the faint sources are not due to their large ages , but to some other reason . their very red @xmath3 colors may be explained in part by large photometric uncertainties , which at these faint magnitudes , can reach @xmath4 0.5 mag ( see the error bars in figure [ fig5 ] ) . furthermore , these sources may be reddened by dust known to be present in the main body , with a maximum extinction @xmath22 @xmath6 0.5 mag in the southeastern part @xcite . however , this value is derived over large areas , and the extinction can be clumpy and significantly higher on small scales . it is seen from figure [ fig5 ] that correction for interstellar reddening with @xmath22 @xmath14 1 mag will move all red faint sources in the cmd region where stars with ages @xmath4 500 myr reside . in any case , the faintness of these sources and the unknown distribution of extinction on small scales in the main body and in the c component precludes a more reliable explanation of their nature . another way to study the properties of stellar populations is to consider their integrated characteristics as given by the surface brightness and color profiles in different regions of the galaxy . the advantage of this approach is that it includes both resolved and unresolved stars . the disadvantage is that populations with different ages contribute to the integrated light and assumptions have to be made on the star formation history to derive the age distribution of stars . in figure [ fig13 ] , we show the @xmath0 ( a ) and @xmath1 ( b ) surface - brightness and @xmath3 ( c ) color distributions averaged over a 7 wide strip along the position angle 47 connecting the main body and the c component . the origin is taken to be at the center of the nw component of the main body . surface brightnesses and colors have been transformed to the standard @xmath43 photometric system according to the prescriptions of @xcite and have been corrected for galactic extinction with @xmath22 = 0.106 mag @xcite . the nw and se components of the main body and the central cluster in the c component are labeled in each panel . the bluest color @xmath6 0.5 mag is in a region @xmath6 3 northwest of the nw component . the equivalent width of the h@xmath11 emission line in this region exceeds 1000 @xcite and hence the blue color is due to ionized gas emission . other regions in the main body with @xmath3 as blue as @xmath6 0.3 - 0.4 mag are also strongly contaminated by ionized gas emission . the only region in the main body free of ionized gas emission is the southernmost one ( region mi ) which , with @xmath3 @xmath6 0.3 0.4 mag , is the reddest part in i zw 18 . in the c component , the contribution of ionized gas emission is negligible @xcite and the bluest color @xmath3 @xmath6 0 mag in the southeastern part ( region ci ) really reflects that of a young stellar population . the reddest color @xmath3 @xmath6 0.2 0.3 mag is in the northwestern part of the c component . it is slightly bluer than the color of the reddest region in the main body , suggesting that the c component may be slightly younger than the main body . the surface brightness profiles also show the absence of an extended lsb component , with a precipitous drop of 6 mag from the nw component to the edge of the galaxy over a distance @xmath6 10 or 750 pc . to model the colors of the reddest regions in i zw 18 , we use spectral energy distributions of single stellar populations with a heavy element mass fraction @xmath40 = 0.0004 , calculated with the pegase.2 code of @xcite . because geneva stellar evolutionary models @xcite do not include agb stars , the padova models @xcite are used . the salpeter initial mass function ( imf ) with a slope @xmath11 = 2.35 , and lower and upper stellar mass limits of 0.1 @xmath38 and 120 @xmath38 respectively are adopted . the models predict that the @xmath3 color of an instantaneous burst with age between 200 and 500 myr is @xmath6 0.5 mag , or @xmath6 0.1 0.2 mag redder than the observed reddest color . only for an instantaneous burst with age @xmath4 150 myr , several 100 myr smaller than the cmd - derived ages , are the modeled colors consistent with the observed ones . we have also considered the case of continuous star formation . if star formation has occurred at a constant rate continuously during the period 50 160 myr ago , then the @xmath3 color is @xmath6 0.35 mag , in agreement with the observed value . but the upper age limit is again significantly lower than the cmd - derived ages . we can increase the upper age limit for continuous star formation to make it consistent with the cmd - derived value of @xmath6 500 myr , but to reproduce @xmath3 @xmath6 0.35 mag , we must at the same time decrease the lower age limit to @xmath6 10 myr . in this case young massive stars should be present in the reddest regions of i zw 18 , and they are not seen . a possible cause for the color discrepancy may be that the metallicity of stars in i zw 18 is not equal to the metallicity of the ionized gas as assumed , but lower . we have checked into this possibility with models that are four times as metal - poor ( @xmath40 = 0.0001 ) . @xmath3 does become bluer , but the blueing is not sufficient . for instantaneous bursts with ages in the range 200 500 myr , @xmath3 @xmath6 0.46 mag , still too red as compared to the observed value . we have also varied the parameters of the imf for a single stellar population with a heavy element mass fraction @xmath40 = 0.0004 . making the slope flatter or steeper with @xmath11 in the range 2 - 3 and increasing the lower mass limit from 0.1 @xmath38 to 1 @xmath38 does not result in a significantly bluer @xmath3 color . the @xmath3 colors are in the range 0.45 0.50 mag for instantaneous bursts with ages in the range 200 500 myr . we conclude from the previous considerations that , despite the uncertainties of the models and the fact that they can not reproduce precisely the observed reddest colors in i zw 18 , they do predict young ages which are in the ballpark of cmd - derived ages . we present a photometric study of the resolved stellar populations in i zw 18 , the most metal - deficient blue compact dwarf ( bcd ) galaxy known . the analysis of the color - magnitude diagram ( cmd ) of i zw 18 , based on _ hubble space telescope_/advanced camera for surveys @xmath0 and @xmath1 images , the deepest ever obtained for the bcd , have led us to the following conclusions : \1 . the cmd of i zw 18 is populated by stars with different ages including the youngest hydrogen core burning main - sequence ( ms ) stars ( age @xmath4 30 myr ) , evolved massive stars with helium core burning [ blue loop ( bl ) stars and red supergiants ( rsg ) ] with ages between 10 myr and 100 myr , and asymptotic giant branch ( agb ) helium shell burning stars with ages between 100 myr and 500 myr . however , _ i zw 18 is the first galaxy with resolved stellar populations where no red giant branch ( rgb ) stars are seen _ , although our data go 1 2 mag deeper than the tip of the rgb for a distance of i zw 18 in the range 12.6 mpc 15 mpc ( see conclusion 2 ) . the oldest stars located mainly in the southeastern part of the main body and the northwestern part of the c component have an age not exceeding @xmath6 500 myr . thus _ i zw 18 is a bona fide young galaxy_. \2 . since no rgb stars are seen , we can not use the brightness of the tip of the rgb to derive the distance to i zw 18 . instead , we compare the brightness of the agb stars in i zw 18 with those in the very metal - deficient bcd ugc 4483 with a heavy element abundance of @xmath2/23 to derive a distance in the range 12.6 mpc 15 mpc , with the most likely value in the upper range . \3 . several star formation episodes in i zw 18 can be inferred from its cmd . however star formation proceeds differently in the main body and in the c component . in the main body , three star formation episodes are indicated , separated by periods of @xmath6 100 200 myr . examination of the spatial distribution of the stellar populations suggests that the star formation process is still gradually building up the main body from the inside out as the young ms and bl+rsg stars occupy larger areas as compared to the older agb stars . in the c component , two star formation episodes are inferred separated by a period of @xmath6 200 myr . these separate star formation episodes occurred in spatially different regions , reflecting the stochastic mode of star formation in bcds . the southeastern region of the c component is a few hundred myr younger than the northwestern region . both in the main body and in the c component , the spatial distributions of the stellar populations in i zw 18 strongly suggest that the galaxy is still in the process of forming . t.x.t . has been partially supported by grant hst - go-08769.01-a . the research described in this publication was made possible in part by award no . up1 - 2551-kv-03 of the u.s . civilian research & development foundation for the independent states of the former soviet union ( crdf ) and a grant no . m/85 - 2004 of the ministry of education and science of ukraine . we are also grateful for the partial financial support of nsf grant ast-02 - 05785 . thanks the hospitality of the astronomy department of the university of virginia . dufour , r. j. , garnett , d. r. , skillman , e. d. , & shields , g. a. 1996b , in asp conference series 98 , from stars to galaxies , ed . c. leitherer , u. fritze-v.alvensleben , & j. huchra ( san francisco : asp ) , 358 crrcrr 22 23&100.0&100.0 & & 100.0&100.0 + 23 24 & 97.0 @xmath25 4.9 & 99.9 @xmath25 1.7 & & 100.0&100.0 + 24 25 & 91.7 @xmath25 5.3 & 96.1 @xmath25 3.1 & & 100.0 & 98.2 @xmath25 1.7 + 25 26 & 90.6 @xmath25 7.7 & 89.7 @xmath25 2.3 & & 97.4 @xmath25 3.0 & 98.2 @xmath25 1.7 + 26 27 & 85.1 @xmath25 8.7 & 85.7 @xmath25 6.5 & & 95.4 @xmath25 4.4 & 94.5 @xmath25 2.8 + 27 28 & 78.9 @xmath25 8.5 & 70.7 @xmath25 10.4 & & 92.9 @xmath25 9.2 & 80.4 @xmath25 4.2 + 28 29 & 55.3 @xmath25 15.7 & 39.0 @xmath25 6.3 & & 69.4 @xmath25 9.9 & 44.2 @xmath25 10.1 + 29 30 & 10.4 @xmath25 7.7 & 5.4 @xmath25 7.1 & & 16.5 @xmath25 9.4 & 9.8 @xmath25 8.4 + 30 31 & 0.6 @xmath25 2.5 & & & 0.6 @xmath25 3.0 & +

we describe algorithms that detect 21 cm line self - absorption ( hisa ) in large data sets and extract it for analysis . our search method identifies hisa as spatially and spectrally confined dark features that appear as negative residuals after removing larger - scale emission components with a modified clean algorithm . adjacent hisa volume - pixels ( voxels ) are grouped into features in space , and the brightness of voxels outside the 3-d feature boundaries is smoothly interpolated to estimate the absorption amplitude and the unabsorbed emission brightness . the reliability and completeness of our hisa detection scheme have been tested extensively with model data . we detect most features over a wide range of sizes , linewidths , amplitudes , and background levels , with poor detection only where the absorption brightness temperature amplitude is weak , the absorption scale approaches that of the correlated noise , or the background level is too faint for hisa to be distinguished reliably from emission gaps . false detection rates are very low in all parts of the parameter space except at sizes and amplitudes approaching those of noise fluctuations . absorption measurement biases introduced by the method are generally small and appear to arise from cases of incomplete hisa detection . this paper is the third in a series examining hisa at high angular resolution . a companion paper ( paper ii ) uses our hisa search and extraction method to investigate the cold atomic gas distribution in the canadian galactic plane survey .

astro-ph0503119

the 21 cm line of neutral atomic hydrogen ( ) is a key probe of the galactic interstellar medium . although the cold ( @xmath1 k ) gas distribution is difficult to map in emission , self - absorption ( hisa ) allows cold foreground gas to be distinguished from warmer background gas at the same radial velocity @xcite . until recently , hisa has been studied in limited low - resolution maps ( e.g. , @xcite ) , or in a few isolated objects at higher resolution ( e.g. , @xcite ) , but no detailed , systematic surveys have been made . high resolution allows feature structure to be studied and unabsorbed background brightness to be estimated accurately . coverage of a wide area enables an unbiased look at the hisa population , e.g. , without the a priori expectation that hisa is found only in molecular clouds . high - resolution , wide - area hisa surveys have now become possible with the advent of several major synthesis surveys : the canadian galactic plane survey ( cgps ; @xcite ) , the southern galactic plane survey ( sgps ; @xcite ) , and the vla galactic plane survey ( vgps ; @xcite ) . past hisa studies have identified absorption features by eye , but this approach is no longer adequate . the very richness of the synthesis survey data sets requires that they be analyzed in a rigorous , repeatable manner . we have therefore designed automated algorithms to identify and extract hisa features from longitude - latitude - velocity @xmath0 data cubes . in this paper , we describe our hisa search and extraction algorithms . we also explain how we have tested our software with model data to determine its reliability under a range of different conditions . large surveys are playing an increasingly significant role in modern astrophysics , and it is essential that their underlying methods are understood so their results can be interpreted properly . following criteria established in @xcite ( @xcite ; hereafter paper i ) , our hisa search software seeks finely - structured dark features against bright backgrounds that can not be confused with simple gaps in emission . although its parameters are optimized to identify hisa in the cgps , the software is easily adapted to work with other surveys ( e.g. , the vgps : @xcite ) . the cgps uses a hexagonal grid of full - synthesis fields with single - dish observations to enable the detection of all scales of structure down to the synthesized beam . the cgps data have a @xmath2 beam , 0.824 velocity sampling , and a field - center noise of @xmath3 k in empty channels ; @xmath4 doubles when the 107 primary beam is filled with 100 k emission , and it can be up to 60% greater between field centers . the initial phase of the survey mapped a @xmath5 region along the galactic plane with longitudes @xmath6 and latitudes @xmath7 ( @xmath8 ) , and extensions in both @xmath9 and @xmath10 have followed . below , we describe our method of hisa identification and extraction at some length ( [ sec : ext ] ) and evaluate the method s performance with models ( [ sec : eval ] ) . a companion paper presents the results of our hisa search of the @xmath5 phase i cgps ( @xcite ; hereafter paper ii ) . subsequent papers in this series will apply the hisa search to other data sets . many hisa features are apparent to the eye ( e.g. , see figs . [ fig : real_hisa_id_maps_overview ] - [ fig : real_hisa_id_stage_spectra ] ) , but a complete visual search is unlikely to be uniform , repeatable , or thorough , and it is also impractical given the sheer volume and complexity of the cgps data . thus , an automated search is needed . the search algorithm should find features meeting simple criteria that can be confirmed by eye , but it should also be tested with model data to evaluate its performance quantitatively ( [ sec : eval ] ) . the nature and appearance of hisa dictate how it can be identified . first , while the cold from which it arises can have any extent , no hisa feature can exceed the @xmath0 boundaries of its bright background emission , or it ceases to be absorption . second , hisa must have different @xmath0 structure than the background for it to be distinguishable from background fluctuations . we choose to search for hisa that is more finely - structured than the background , since this is consistent with the first constraint , most cgps hisa that can be visually identified is of this nature , and the exceptions ( e.g. , @xcite ) are difficult to identify algorithmically . we seek features that can only be explained as hisa . we prefer this conservative approach over the alternative of including significant false detections in our survey sample . as given in paper i , our conditions for distinguishing hisa from simple gaps in emission are : ( 1 ) narrower line widths than most observed emission features ; ( 2 ) steeper line wings ; ( 3 ) more small - scale angular structure ; and ( 4 ) a minimum emission background level . the first two conditions are related for gaussian line profiles , since these have line wing slopes proportional to amplitude over width , but real hisa need not be gaussian . the last condition excludes the finely - structured emission gaps that are common at interarm velocities in the outer galaxy , where smooth , bright backgrounds are often absent . these four criteria exclude hisa on larger angular and velocity scales or against weaker backgrounds , but they are adequate for capturing most visible features . we do not require the extra condition of molecular line emission to confirm hisa features ( e.g. , @xcite ) , since many hisa features are visible without emission in the cgps ( paper i ; @xcite ) and in other surveys ( e.g. , @xcite ) . we tried and rejected many different methods before selecting the algorithm described in this paper . discarded techniques include various derivative measures to detect sharp edges , spatial and spectral curvature tests to look for dips , fourier and wavelet filtering methods , and unsharp masking . most of these were successful in locating the strongest features , but few were robust against noise , and many also produced large numbers of artifacts and false detections . the latter were especially frequent in methods that used only spectral or spatial searches rather than both combined . our chosen method is based on a variant of the clean algorithm @xcite developed by @xcite ( hereafter the sdi clean ) . we remove large - scale spectral and spatial emission structures from the data iteratively and flag the small - scale negative residuals as self - absorption features . for computational efficiency , these operations are carried out separately on the spectrum at each spatial position @xmath11 and on the channel map at each radial velocity @xmath12 in the data cube , and the results are combined afterward . the identified hisa is then subtracted from the cube , and the whole process is repeated until significant hisa can no longer be found ; such iteration allows features larger than the chosen clean scales to be mapped . the two spectral and spatial search algorithms are described below . each has been tuned to find as much visually identifiable hisa as possible while minimizing false detections . the latter are further reduced by subsequently requiring the hisa at any 3-d position @xmath0 to be detected by both searches ( [ sec : ext_dt ] ) . the two algorithms were tuned by visually comparing the search output against the observed for many different parameter value combinations , using a range of different hisa features with different emission backgrounds in the cgps data . the parameters that yielded the most complete hisa detections with the fewest false detections were used in the model - based search performance evaluations ( [ sec : eval ] ) and in the cgps hisa survey ( paper ii ) . in the following discussions , the galactic coordinate variables @xmath0 are replaced by their pixel coordinate analogs @xmath13 . adopted values for the search parameters are given in square brackets [ ] . these give the best performance for a hisa search of cgps data , but they may not be universal . in particular , the best filter scales and minimum background level may differ for hisa searches elsewhere in the galaxy . at each spatial position @xmath14 , the spectral search algorithm builds an approximation of the `` unabsorbed spectrum '' @xmath15 that would be observed if no hisa were present . the algorithm assumes that @xmath15 can be constructed from gaussian functions of a characteristic width @xmath16 that is narrower than the dominant emission features but broader than the width of any expected hisa feature . any channels in which the observed spectrum @xmath17 deviates significantly negatively from @xmath15 are flagged as possible hisa . the iterative procedure used to derive @xmath15 is a modification of the sdi clean . @xmath15 is initially set to zero , and the `` residual spectrum '' @xmath18 is set equal to @xmath19 , a smoothed version of @xmath17 . smoothed data are used to improve the signal - to - noise for the cleaning process . @xmath19 is a _ spatial _ average of @xmath20 pixels centered at @xmath14 , i.e. , the average of @xmath21 intensities where @xmath22 and @xmath23 [ and @xmath24 pixels @xmath25 . independent of this spatial averaging , the _ spectral _ _ rms _ noise @xmath26 in @xmath17 is computed as the lowest of three _ rms _ noise measures over equal thirds of @xmath17 . in the clean loop , the following steps are performed : 1 . if @xmath27 , the peak value of @xmath18 , is less than a preset fraction [ 3% ] of the peak value of @xmath19 , the iteration ceases . 2 . for any channel @xmath28 where @xmath18 exceeds a given clip level [ 0.8 ] @xmath29 , a `` correction spectrum '' @xmath30 is set to a preset gain [ 0.25 ] @xmath31 ; elsewhere , @xmath30 is set to zero . @xmath30 is convolved with a gaussian whose full width at half maximum ( fwhm ) is @xmath16 [ 8 ] , and the resulting spectrum is added to @xmath15 . the new residual spectrum is set to @xmath32 , and @xmath33 , the _ rms _ of all _ positive _ values of @xmath18 is computed over all channels . as @xmath15 approaches @xmath19 , @xmath33 decreases ; if @xmath34 , the iteration is terminated . if the iteration has not terminated due to one of the above convergence criteria , and a maximum number of loops [ 1000 ] has not been reached , steps 1 - 5 are repeated . after the clean loop is completed , adjacent channels where @xmath35 a factor @xmath36 [ @xmath37 are grouped into `` segments '' of suspected hisa . in each segment , @xmath19 and @xmath15 are evaluated at the channel @xmath38 where @xmath18 has a local minimum . if @xmath39 , the segment is rejected as a likely noise feature . if @xmath40 , where @xmath41 is a preset brightness level [ 30 k ] , the segment is rejected as having an insufficiently bright background to identify hisa clearly . a second , more conservative @xmath42 [ 70 k ] is applied later when the spatial and spectral search results are combined ( [ sec : ext_dt ] ) . a gaussian profile is fitted to the absorption magnitude spectrum @xmath18 of each remaining channel segment . real hisa line profiles may not be gaussian , but this shape is assumed for simplicity . for computational speed , the central channel of the gaussian is fixed at @xmath38 , and the fwhm is fixed at one of two values , @xmath43 [ 2.0 ] or @xmath44 [ 4.0 ] , to capture hisa with a range of widths . model tests ( [ sec : eval ] ) show that many features outside this range are also detected . in each fit , a sloping linear base level is derived along with the gaussian amplitude @xmath45 and the standard deviation @xmath46 of the fit . a fit is rejected as statistically unreliable if @xmath47 [ 2.0 ] or @xmath48 . it can also be rejected if @xmath17 lacks a morphological `` dip '' at @xmath38 . this is determined with a filter function that returns a value of 1.0 for a dip between two equal peaks , 0.5 for a `` dip '' that drops only to the level of the adjacent spectral data on one side of @xmath38 , and 0.0 on a linearly rising or falling spectrum . fits that return a value below the chosen threshhold [ 0.6 ] are rejected . this filter inhibits the detection of hisa on the edges of emission features that are steeper than @xmath16 would allow ; otherwise , significant false hisa detections result . for accepted fits , the channels in a _ narrow _ or _ broad _ hisa spectrum ( both initially zero ) are set equal to the fitted gaussian if the amplitude exceeds a given fraction [ 5% ] of @xmath45 . then , a `` detected '' hisa spectrum is created that consists of the maximum value in each channel from the _ narrow _ and _ broad _ gaussian fits to the hisa line profile , to ensure full detection of the feature . in the case where the _ broad _ line wings do not correspond to real hisa in a narrow feature , these will not be detected by the spatial search and will be removed at a later stage of analysis . finally , when the hisa amplitudes have all been computed , these are spatially smoothed [ with a 1.5 beam ] to join together groups of flagged `` flecks '' into more coherent features , and those that are sufficiently weak and isolated are culled if their amplitude falls below a specified threshhold [ 2 k ] . an additional cosmetic improvement is made by excluding strong hica from the set of spectrally - identified hisa . this is done by dropping any sight lines from the search that contain channels whose continuum - subtracted line brightness is significantly negative , i.e. , if @xmath49 . weaker hica will survive this filter to contaminate the set of detected hisa . such contamination is difficult to remove in a way that leaves the `` pure '' hisa in the same sight lines intact . to illustrate the algorithm , we plot the spectrum of paper i s perseus hisa globule ( @xmath50 ) in figure [ fig : globule_spectrum_raw ] and several stages of the algorithm s analysis in figure [ fig : globule_spectrum_hi_dips ] . these data supersede those used in paper i , which contained a flaw that had no serious impact on the results . appendix [ app : erratum ] gives further details . in figure [ fig : globule_spectrum_hi_dips ] , the estimation of @xmath15 converged when the peak residual became less than 3% of the initial spectral peak , giving @xmath51 k ( the dashed line indicates a negative deviation of twice this value ) . suspected hisa was identified in eight channel segments . one of these ( 4 ) had no counterpart dip in @xmath17 . six others were rejected because @xmath52 k. for simplicity , we used @xmath53 k here to show what would survive the ultimate @xmath54 k filter . in the remaining segment ( 5 ) , channels in which the `` merged '' hisa spectrum ( maximum of the narrow and broad hisa spectra ) is non - zero were then flagged as having `` detected '' hisa . note that , because of the initial smoothing , the detected hisa has a smaller amplitude than in figure [ fig : globule_spectrum_raw ] . the full hisa amplitude is recovered when the spectral and spatial search results are merged ( [ sec : ext_dt ] ) . the spatial search algorithm is similar in principle to the spectral search , although it does not attempt to fit absorption features with gaussian shapes , nor does it require that they satisfy a morphological `` dip '' filter . it begins by estimating the unabsorbed brightness distribution @xmath55 in a given spectral channel @xmath28 . the algorithm assumes that @xmath55 can be constructed from two - dimensional circular gaussian components of a characteristic width @xmath56 that is narrow enough to represent most emission structure but broader than any expected hisa features . clearly the choice of @xmath56 limits the angular size of hisa features that will be detected , although this can be alleviated with repeated searches . the iterative procedure used to derive @xmath55 is again a modification of the sdi clean . @xmath55 is initially set to zero , and the residual map @xmath57 is set equal to @xmath58 , a spatially smoothed copy of the observed channel map @xmath59 . use of @xmath58 , computed as an @xmath20 pixel average of @xmath59 [ with @xmath60 pixels @xmath61 , improves the clean convergence . on a larger angular scale [ 20 ] , an estimate of the typical _ rms _ noise @xmath26 in @xmath59 and its gross variation across the channel map are derived in a manner similar to the spectral _ rms _ noise in [ sec : ext_spectral ] . from this , the average _ rms _ noise in @xmath58 , termed @xmath62 , and its variation over the map are deduced . in the clean loop , the following steps are performed : 1 . @xmath27 , the mth [ 10th ] highest value of @xmath57 , is found . this is chosen rather than the peak value so that the iteration process is not dominated by one noisy pixel . if @xmath27 is less than a preset fraction [ 3% ] of the peak value of @xmath58 , the iteration ceases . 2 . for any pixel @xmath14 where @xmath57 exceeds a given clip level [ 0.5 ] @xmath29 , a correction map @xmath63 is set to a preset gain [ 0.25 ] @xmath64 ; elsewhere , @xmath63 is set to zero . @xmath63 is convolved with a 2-d gaussian whose fwhm is @xmath56 [ 20 ] , and the resulting image is added to @xmath55 . the new residual map is set to @xmath65 , and the _ rms_value @xmath33 of the _ positive _ values of @xmath57 is computed . in deriving @xmath33 , allowance is made for the fact that the noise may vary across the image by applying suitable weights to the values of @xmath57 . if @xmath66 , the iteration is terminated . if the iteration has not terminated due to one of the above convergence criteria , and a preset maximum number of loops [ 1000 ] has not been reached , steps 1 - 5 are repeated . after the clean loop is completed , all pixels where @xmath67 a factor @xmath36 [ @xmath37 are noted , as are those where @xmath68 . a map of `` suspected hisa '' is set equal to @xmath69 for all pixels @xmath14 where either condition is met and zero elsewhere . this map is then filtered to remove pixels with amplitudes less than a specified cutoff [ 4 k ] , as well as those for which @xmath70 [ 30 k ] . lastly , as in the spectral algorithm , the suspected hisa map is smoothed [ with a 1.5 beam ] to improve feature coherence , and a final cull is made of smoothed amplitudes below a lower threshhold [ 2 k ] ; surviving features are deemed `` detected hisa '' . the spatial search is illustrated in figure [ fig : globule_xcut_hi_abs ] , with longitude profiles taken through the perseus hisa globule position ( @xmath71 ) from one channel map at different stages of processing . the determination of @xmath55 took 114 iterations , ending when @xmath33 became less than @xmath72 k. the physical properties of the absorbing gas can not be understood without knowing the hisa brightness temperature amplitude @xmath73 , where @xmath74 is the observed brightness on the hisa feature , and @xmath75 is the unabsorbed emission that would be measured if no hisa were present . since only @xmath76 is directly observed at the hisa position , @xmath77 must be estimated from @xmath78 , the brightness off the hisa feature in space and/or velocity . for clarity , we note that @xmath77 represents _ all _ of the unabsorbed emission along the line of sight at radial velocity @xmath79 . the emission from behind the hisa feature that is subject to absorption is @xmath80 , where @xmath81 ; the exact value of @xmath82 depends upon the sight - line geometry ( see paper i ) . several means of estimating @xmath77 have been used in past studies . in the spectral domain , the velocity edges of a hisa feature can be fitted with straight lines @xcite or more complex functions @xcite to estimate @xmath77 at intervening velocities . in the spatial domain , the brightness at positions adjacent to the hisa feature can be used directly as @xmath77 ( paper i ; @xcite ) or as anchor points for spatial fits across the feature @xcite . a variant on this approach assumes hisa is sufficiently diluted in the broad beam of a single dish telescope to use the single dish spectrum at the feature position as @xmath77 @xcite . our approach is more general . we group the hisa volume - pixels ( voxels ) identified in [ sec : ext_spectral]-[sec : ext_spatial ] into contiguous 3-d features in the spectral line cube . for each feature , we estimate @xmath77 by interpolating the @xmath78 values of the non - hisa voxels that border the feature in @xmath0 space . the interpolation uses a 3-d gaussian weighting function to ensure smoothness on the scale of the feature . specifically , at each position @xmath0 within the hisa feature , @xmath83 where @xmath84 indexes the list of @xmath85 off - hisa voxels with coordinates @xmath86 , and the weight @xmath87 is given by @xmath88 \;\ ; .\ ] ] the gaussian dispersions @xmath89 are set so that each fwhm ( @xmath90 ) is half the maximum length of any contiguous row of hisa voxels in that dimension , with fwhm lower limits of 1.2 and 3.3 and upper limits of 20and 8 , the hisa search clean scales . the result is somewhat similar to that of 1-d spectral fitting methods , but its structure is constrained by all three dimensions of the data . this method yields @xmath77 and @xmath91 estimates superior to those of our separate spectral and spatial searches . the @xmath77 estimation algorithm considers two confidence levels of hisa . first , a _ union _ filter requiring a hisa identification from either the spectral or spatial search is applied . this filter includes nearly every hisa feature that the eye can detect , as well as many non - hisa features that are discarded later . all accepted voxels are grouped into tentative hisa features and interpolated over to obtain @xmath77 , which is subtracted from the unsmoothed data to get @xmath91 with the full cgps angular resolution . then , an _ intersection _ filter requiring hisa identification in _ both _ spectral and spatial searches is applied . voxels not satisfying this filter are unflagged as hisa , and their @xmath77 is reset to the observed brightness . computing @xmath77 and @xmath91 for a union voxel set and applying an intersection filter afterward ensures that ( 1 ) only the most likely hisa features survive , and ( 2 ) any `` penumbral '' contamination from undetected hisa in their @xmath78 voxels is minimized ; otherwise , @xmath77 and @xmath92 could be significantly underestimated . an alternative is to interpolate only over hisa satisfying the intersection filter with all union - filter voxels dropped from the @xmath78 ensemble , but this frequently leaves too few edge voxels for a robust @xmath77 estimate . three additional filters are applied with the intersection filter . voxels with @xmath93 are discarded , as are those in the noisy peripheries of the survey and those for which @xmath94 [ 70 k ] , a stricter value than the previous @xmath41 [ 30 k ] of [ sec : ext_spectral]-[sec : ext_spatial ] . the peripheral culling rejects hisa voxels with cgps field mosaic weights @xmath95 , the lowest weight that occurs between synthesis field centers . since @xmath96 , this allows a maximum noise of 1.618 times the field center value ( see @xcite ) , which is typically @xmath97 k for the @xmath98 k levels of our hisa features . the choice of @xmath54 k is empirically based . as noted in paper i and [ sec : ext_strategy ] , finely - structured emission is common in the cgps data where the total amount of emission is low , i.e. , off the plane and at interarm velocities . without some sort of @xmath99 filtering , the hisa identification software is easily fooled in these regions , flagging many false hisa features adjacent to and between sharp - edged emission . we are confident these are false hisa features , since the absorbing would have to be unrealistically cold to absorb against such faint backgrounds , and such apparently strong features are far less abundant in brighter emission fields where they should be easier to detect and where more gas should be found generally . there is no single @xmath42 value that excludes all such false hisa while retaining all real hisa . we chose @xmath54 k to balance these two needs , with greater priority placed on the first . for the cgps , 70 k rejects essentially all false hisa arising from sharp emission edges while keeping most real hisa . the model tests of [ sec : eval ] show that the false hisa rejection is quite successful . a few cases of some real hisa being missed or truncated are discussed below and in paper ii . figures [ fig : real_hisa_id_maps_overview]-[fig : real_hisa_id_spectra ] illustrate the hisa amplitude extraction process with sample channel maps and spectra that include the same features shown in figures [ fig : globule_spectrum_raw]-[fig : globule_xcut_hi_abs ] . these give the initial data , the @xmath91 amplitudes computed in different stages of the analysis , and the final @xmath77 . as the figures show , most of the visually apparent hisa is readily extracted . some residual hisa remains in @xmath77 , but its amplitude is a few k at most ; the @xmath91 value of @xmath100 k extracted for the perseus globule is only 2 k weaker than that obtained with the more conservatively - chosen paper i spatial @xmath78 boxes ( table [ tab : cg_prop ] ) . the hisa identification software can not detect features larger than its clean scales [ 8 and 20 ] and instead flags only their darkest parts . to overcome this limitation , we feed the @xmath77 cubes back into our search algorithms to identify hisa missed on the previous pass ( [ sec : ext_strategy ] ) . subsequent @xmath91 extractions are made with unions of hisa flags from all prior search passes but always use the original data for @xmath78 . three such passes are adequate for the cgps data set . of the hisa voxels extracted in all three passes , 86.5% were found in the first pass , 11.2% in the second , and only 2.3% in the third . for brevity , figures [ fig : real_hisa_id_maps_overview]-[fig : real_hisa_id_spectra ] display only third - pass results , which differ little from the first pass for this example . a case with more dramatic differences between passes is illustrated in figures [ fig : real_hisa_id_stage_maps ] & [ fig : real_hisa_id_stage_spectra ] . features this big require multiple passes to capture . hisa flagging is significantly improved after the first pass in both the spatial and spectral domains . we show only first- and third - pass results here , since the second pass closely resembles the third . after three passes , hisa flagging is incomplete in only a few places due to @xmath101 k truncation , mostly near the northern edge of the map ( fig . [ fig : real_hisa_id_stage_maps ] ) . aside from minor losses from @xmath77 changes , the flagged hisa generally increases , with the fraction of the total flagged per pass being 72.9% , 22.5% , and 4.6% for passes 1 , 2 , and 3 . the smooth @xmath102 structure in figures [ fig : real_hisa_id_stage_maps ] & [ fig : real_hisa_id_stage_spectra ] shows that our @xmath77 estimation method follows the large - scale emission brightness reasonably well . we have chosen to attribute fine - scale structure in @xmath76 to @xmath91 , leaving @xmath77 smooth on the scale of the hisa feature . this approach presumes that the absorbing gas is finely structured and the background is not , consistent with our adopted hisa identification strategy ( [ sec : ext_strategy ] ) . such consistency allows the identified hisa structure to be removed so that subsequent search passes do not flag it again . if however some @xmath76 structure arises from @xmath77 ( e.g. , @xcite ) , the true @xmath91 is smoother than we have found . we feel that our choice of method is reasonable for most circumstances . small - scale emission structure is common in the general ism but appears minimized in the bright , smooth fields where we see most cgps hisa . the eye is the first means of identifying hisa features that meet the appropriate criteria , and the search algorithms of [ sec : ext ] were designed primarily to mimic visual detection . however , the eye can be fooled ; for example , it often finds false patterns in noise , perhaps due to evolutionary pressures to spot predators @xcite . we made our hisa search and extraction algorithms as rigorous as possible , but they remain limited by a number of factors , including : 1 . * @xmath77 faintness : * to avoid confusion with emission gaps at low column densities , hisa detection is blocked if @xmath101 k. where this occurs , small features or parts of large ones may be missed . * @xmath77 underestimation : * we assume @xmath77 is not finely structured and estimate it from @xmath78 voxels surrounding the hisa feature in 3-d . however , many hisa features occur near spectral emission peaks . if @xmath103 , then we underestimate @xmath77 and @xmath92 . both can also be underestimated if hisa flagging is incomplete and an unidentified `` penumbra '' of faint hisa contaminates @xmath78 . * noise degradation : * despite smoothing , some low - amplitude hisa will be lost to noise . whole features may be missed , or just their cores may be detected , making them appear smaller , clumpier , and more fragmented than they really are . in addition , false hisa detections will be introduced by noise fluctuations at low @xmath92 . * overlarge structure : * by design , the hisa search algorithms can not flag whole features larger than the adopted @xmath104 and @xmath105 clean filter scales . the use of multiple search passes eases this limitation but may not remove it entirely . * unresolved structure : * small - scale hisa structure may be diluted or missed entirely if undersampled . angular structure down to the @xmath106 cgps beam is seen ( paper i ) , so smaller - scale structure seems likely . hisa linewidths narrower than the cgps nyquist limit of @xmath107 also exist @xcite . such linewidths are rare in random hica sight lines ( e.g. , @xcite ) , but since continuum backgrounds can be brighter than backgrounds , hica can include warmer absorbing gas than hisa , and it s possible that hisa lines may be narrower on average . hisa velocity dilution is thus a real concern in synthesis surveys , while angular dilution will be more severe for single - dish telescopes ; no present instrumentation can adequately sample both the angular and velocity structure of hisa . to evaluate such limitations objectively and quantitatively , we tested our software s ability to extract hisa features from model data . our goal was to understand ( 1 ) what fraction of hisa the software detects , ( 2 ) how many detections are false positives , and ( 3 ) how much the detected features differ in size and amplitude from their input versions . the model 21 cm spectral line cubes were sums of noisy , positive - amplitude emission backgrounds and noise - free , negative - amplitude absorption features . gas properties and radiative transfer effects were not considered , as these are irrelevant to the detection software s performance . to test this under varying conditions , 64 randomly - configured model cubes were made . each cube used the standard cgps pixel and channel sizes , with dimensions half those of a standard cgps mosaic cube for computational efficiency : @xmath108 ( @xmath109 voxels ) . sample model data are shown in figure [ fig : hockey_pucks ] . each model hisa feature was given a cylindrical shape in the line cube , with a gaussian velocity profile and a flat - disk spatial profile convolved with a 60 circular beam . although simple , these angular and velocity profiles are similar enough to typical hisa for testing purposes . the features , known as `` hockey pucks '' for their usually oblate aspects in the cgps voxel grid , are parameterized by their unconvolved angular fwhm @xmath110 , velocity fwhm @xmath111 , and ( negative ) central amplitude @xmath112 . 2048 hockey pucks were inserted into each model cube with random sizes , amplitudes , and positions . the @xmath0 and @xmath112 distributions were uniformly random , except that puck overlaps in @xmath0 were prevented , with a minimum separation of 1 voxel enforced between pucks at an absorption threshhold of 0.005 k. the @xmath110 and @xmath111 distributions were skewed toward small features , with relative probabilities of @xmath113 and @xmath114 . this was done to counter the fact that larger pucks have more voxels . we measured angular and velocity widths locally from each voxel in the performance analysis ( [ sec : eval_anal ] ) , and @xmath115 and @xmath116 made our voxel - based size distributions more evenly sampled . the puck parameter ranges used were @xmath117 , @xmath118 , and @xmath119 k @xmath120 @xmath100 k. @xmath121 results in `` unresolved '' structures that get diluted in the cgps beam similarly , @xmath122 , which would occur for purely thermal linewidths at 2.73 k , would be unresolved by the cgps in velocity . both cases test the detection limits for fine - scale hisa structure . at the other extreme , the software s sensitivity to structures larger than the 20 and 8clean filter scales is also tested . the background emission fields were similarly constructed of random ensembles of cylindrically - symmetric components . these differed from the hockey puck absorption features in that they had positive amplitudes , simple gaussian angular profiles , and minimum sizes equal to the clean scales . they were also allowed to overlap and fill the entire cube , so we refer to them as emission components rather than discrete features . size ranges were @xmath123 and @xmath124 , with @xmath125 and @xmath126 the same as for the hisa pucks . the amplitude range was @xmath127 k @xmath128 @xmath129 $ ] k , where the component volume @xmath130 , @xmath131 , and the @xmath132 distribution was further skewed as @xmath133 . these adjustments placed most of the power at large scales , as is seen in real emission @xcite . 4096 components were summed to make each model cube s emission field . this was subsequently rescaled to give a median brightness temperature of 70 k , so that half the cube on average would allow hisa detections , and @xmath134 k effects could be easily studied . for greater realism , noise was added to the model . a 3-d field of uncorrelated gaussian random voxel noise was convolved with a 60 fwhm gaussian beam and a 1.319 fwhm gaussian velocity point spread function ( psf ) to mimic the structure of correlated noise in the cgps data , and the _ rms _ noise amplitude was scaled to match the 6 k level found in cgps field centers filled with 100 k emission . unlike the cgps noise , the model noise does not vary with distance from field centers , its beam is declination - independent , and its velocity psf is not the true cgps velocity psf , which is the fourier transform of a gaussian truncated at 20% of peak amplitude , with an effective fwhm of @xmath135 channels . however , none of these differences should seriously affect the performance analysis . the same procedures used to search for hisa in the cgps data ( [ sec : ext ] ) were applied to the model data . the software performance was then evaluated by comparing the input and extracted absorption . each of the 64 model cubes was analyzed separately , and the results were merged afterward to maximize coverage of the model parameter space . four observables were extracted from the hisa data : the absorption amplitude @xmath91 , unabsorbed brightness @xmath77 , angular width @xmath136 , and velocity width @xmath137 of the absorption . all four were measured at each voxel @xmath0 position rather than on a per - feature basis , because the cgps hisa has complex structure , and properties can vary within one feature . however , @xmath136 and @xmath137 are still aggregate properties that depend on the local distribution of hisa around them . the velocity width @xmath137 measures the line fwhm . for each hisa voxel , all hisa contiguous in @xmath79 at the same @xmath11 position is examined to find the channel with maximum @xmath92 . on either side of this channel , the closest channels for which @xmath138 are identified ; non - hisa channels with @xmath139 are included if necessary . the half - maximum velocities are refined to sub - channel accuracy by linear interpolation . the difference between them is @xmath137 . this @xmath137 is assigned to all hisa voxels in the same velocity grouping at the same @xmath11 . we make no attempt to correct for instrumental broadening ( e.g. , @xcite ) , since this is nontrivial for the complex line structure of some hisa , and only the narrowest features will be broadened significantly in the cgps . figure [ fig : da_dv_examples ] shows a sample map of @xmath137 . on average , the broader linewidths occur in larger hisa features . the angular width @xmath136 measures the diameter of the largest circle containing the @xmath11 position and zero non - hisa voxels at the same velocity . this scheme measures the edge - to - edge feature width on a local scale . unlike the fwhm - based @xmath137 , @xmath136 uses the full hisa extent . experiments with an angular fwhm proved too sensitive to complex internal structure in the @xmath140 distribution to be interpreted easily . the resulting @xmath136 measures can be a bit larger than the fwhm - based hockey puck @xmath110 , especially for large puck amplitudes @xmath112 , but since the same @xmath136 measure is taken of the hisa model inputs and outputs , the method is internally consistent . figure [ fig : da_dv_examples ] illustrates how @xmath136 is measured . from each hisa voxel , the angular offset @xmath141 to the nearest non - hisa voxel with the same velocity is found . @xmath142 maps hisa `` skeletons '' whose ridge - like maxima equal half the local width of the feature . to build a @xmath136 map , we step over all @xmath11 positions and write @xmath143 to all points @xmath144 in a new map for which @xmath145 , where the largest imposed @xmath146 value is always retained . this yields the angular width of hisa filamentary structure at that velocity . the hisa extraction software s performance was measured in three ways : the `` throughput '' fraction @xmath147 of model hisa detected ; the `` true fraction '' @xmath148 of hisa detections corresponding with model input ; and the `` drifts '' @xmath149 , @xmath150 , @xmath151 , and @xmath152 between input and output properties . all were measured as functions of the hisa observables @xmath153 , which define a 4-d parameter space in which the software performance is evaluated . the @xmath0 positions and @xmath153 properties were first tabulated for all hisa voxels in both the input and output model @xmath0 cubes . the 4-d voxel count histograms @xmath154 and @xmath155 were constructed from these voxel tables , using bin dimensions of 2.5 k @xmath156 2.5 k @xmath156 0.5 @xmath156 0.5 . in parallel , the voxel count histogram @xmath157 was made from all input voxels with output at the same @xmath0 , and @xmath158 was made from all output voxels with input at the same @xmath0 . throughputs and true fractions were then derived as @xmath159 and @xmath160 . from the subset of voxels appearing in both the input and output hisa cubes , four 4-d drift histograms of the average changes undergone by @xmath161 , @xmath162 , @xmath163 , and @xmath164 as functions of @xmath165 were assembled , e.g. , as @xmath166 , with the average taken over all hisa voxels in @xmath0 with the same @xmath165 properties . the 4-d performance histograms were computed for all 64 hisa models , which were identical apart from different random number inputs . the results were merged together into a single set of histograms and smoothed with 4-d gaussians to improve the performance measure reliability and coverage of the parameter space . variable smoothing scales were used , because the parameter space coverage was sparser in some areas than in others . the smoothing fwhm were @xmath167 , @xmath168 , @xmath169 , and @xmath170 , with minimum values of 6.0 k , 6.0 k , 1.0 , and 1.319 to match the model @xmath171 and cgps resolution . this scheme preserved structure in the well - sampled parts of the parameter space and interpolated it smoothly elsewhere . @xmath155 histograms were also computed for all 36 mosaic cubes of real cgps hisa and summed together to assess the distribution of observed hisa in the survey . as with the hisa feature catalog in paper ii , sight lines with @xmath172 k were excluded . to examine general @xmath153 parameter distributions and trends , we made 2-d projections of the unsmoothed model @xmath173 and @xmath174 and the real cgps @xmath174 by summing the counts along the 2 other axes in the 4-d parameter space . a number of these 2-d projections are shown in figure [ fig:4hist_counts ] . the input models fill ranges of @xmath175 k , @xmath176 k , @xmath177 , and @xmath178 , with peaks at @xmath179 k and small @xmath92 and @xmath136 ; the @xmath137 distribution was relatively flat . the peaked @xmath91 and @xmath136 distributions occurred despite the shapes chosen for the puck property distributions ( [ sec : eval_mod ] ) . the low-@xmath92 peak is due to faint hisa in feature line wings and spatial envelopes . the low-@xmath136 peak results from @xmath136 being measured from hisa voxels above a minimum @xmath92 , which makes pucks appear smaller off the line center . in a similar way , pucks with the same @xmath110 have greater @xmath136 if @xmath180 is larger , and no voxels with @xmath181 and @xmath182 k are found . the extracted model @xmath91 peak is shifted to @xmath183 k. a tail of strong absorption extends to @xmath184 k. although they account for only 2% of the total hisa voxels , these @xmath185 k points demonstrate that some @xmath91 drift occurs . the extracted @xmath77 is truncated at @xmath186 k but otherwise appears unchanged from the input model . large @xmath136 and @xmath137 values are both truncated as predicted in [ sec : eval_mot ] , although less severely for @xmath136 , since the non - gaussian angular profiles better survive the clean process . iterative extraction ( [ sec : ext ] ) allows much of the puck structure @xmath187 to be recovered here , but the purely gaussian puck velocity profiles with @xmath188 are cleaned out of the data with great efficiency . peaks at the same location as the input data but is more concentrated . @xmath137 is also now concentrated toward low values . the cgps hisa @xmath91 has a larger range than in the extracted models , due to a few very strong features like ghisa [email protected]@xmath18902 and ghisa [email protected]@xmath19003 ( see paper ii for feature details ) . its @xmath91 peak is similar to the models. the cgps hisa is truncated for @xmath101 k as well as @xmath191 k , where the maximum brightness is reached . @xmath136 peaks at the same scale as the model output but has a lower maximum scale , perhaps because real hisa is more porous . the model output @xmath137 range is slightly exceeded . the @xmath137 peak and maximum value are both a little broader than for the extracted model hisa . the cgps hisa fills almost the same parameter space as the extracted model hisa . since some model parameter ranges are larger on input than output , the ranges of real hisa properties may exceed those observed in the cgps . hisa with @xmath101 k is already known ( e.g. , @xcite ) , and hisa with @xmath192 , @xmath193 , or @xmath194 k is possible , although @xmath195 is required by the 4-component radiative transfer equation ( paper ii , eqn . 1 ) , where @xmath196 is the spin ( excitation ) temperature of the absorbing gas . also , @xmath136 is limited by feature porosity . ghisa [email protected]@xmath19003 exceeds the @xmath136 limit in gross extent but is not completely solid . larger features are known ( e.g. , the @xcite `` cold cloud '' toward the galactic center ) , but their porosity at 1 resolution has not been reported . the input models had no built - in correlations of feature properties , and the same is largely true for the real hisa . certainly @xmath91 and @xmath77 are not related in the cgps , except that the strongest @xmath91 s prefer some @xmath77 values over others . however , the peak cgps @xmath136 and @xmath137 both increase gradually with @xmath92 out to @xmath197 k , @xmath198 , and @xmath199 . these trends have considerable scatter , and there are weaker versions in the extracted model hisa . but if they reflect real hisa behavior , then stronger absorption is more likely to have larger contiguous angular structure or broader linewidths , although @xmath136 and @xmath137 do not correlate as well with each other as they do with @xmath91 . figure [ fig:4hist_fid_fot ] presents selected 2-d slices through the 4-d parameter space to illustrate the behavior of the throughput @xmath147 and true fraction @xmath148 . these have been smoothed as described in [ sec : eval_anal_meas ] . we find that most hisa is detected if it is significantly stronger than the noise , larger than a few beams , narrower than a few , and has @xmath200 k. furthermore , the vast majority of detected hisa is reliable , except for hisa that can be mimicked by beam - scale noise fluctuations . the throughput is high for much of the parameter space : @xmath201 where @xmath202 k , @xmath200 k , @xmath203 , and @xmath204 , reaching a maximum of @xmath205 where all of these criteria are well - met . where one or more of them is not met , @xmath147 drops rapidly , with @xmath206 for @xmath207 k , @xmath208 k , @xmath209 , or @xmath210 . most of this behavior can be explained as losing features in the noise , underbright @xmath77 , or overbroad linewidths poorly fitted by the spectral search method s @xmath44 [ 4 ] gaussians ( [ sec : ext_spectral ] ) . however , low @xmath147 seems to occur for low @xmath136 even when @xmath92 is large . this suggests some beam - scale hisa features may be missed by our search if they are isolated from larger structures . the perseus hisa globule of paper i is detected easily , but it is also attached to the complex ghisa [email protected]@xmath19040 . by contrast , narrow - line hisa is detected with great efficiency : high @xmath147 is found for @xmath137 as low as the 0.8 cgps channel width , so long as @xmath211 k. to allay the concerns of @xcite , lines narrower than this should also be detectable if their intrinsic amplitudes are larger to compensate for spectral dilution . a thermally broadened @xmath212 k hisa line would be diluted by a factor of 2.4 but easily detected if @xmath213 k. hisa of this strength or greater is common in the cgps . the true fraction has much simpler behavior : @xmath214 almost everywhere that @xmath202 k , @xmath215 , or @xmath216 . if none of these holds , noise fluctuations in the data produce significant false positive detections , with @xmath217 in the worst cases . figure [ fig:4hist_drift ] illustrates trends in the parameter drifts @xmath149 , @xmath151 , and @xmath152 . since @xmath218 ( [ sec : ext_dt ] ) and @xmath76 is fixed , @xmath219 . with minor exceptions , the behavior of all the parameter drifts is fairly simple : @xmath92 and @xmath77 are often underestimated by a few k in well - detected features , while incomplete detections of large features ( e.g. , fig . [ fig : hockey_pucks ] ) cause @xmath136 and @xmath137 to be underestimated as well . the drift in @xmath91 is positive , i.e. , toward reduced amplitudes , if these three conditions are met : @xmath220 the 6 k noise level , @xmath221 80 k , and @xmath222 2 . if one of them is not met , @xmath223 . there is no strong dependence on @xmath137 . the amount of drift is typically a few k , with a range of @xmath224 k in most areas but more negative for @xmath225 k. the @xmath149 behavior has a similar shape to @xmath147 above , suggesting that detection sensitivity governs @xmath91 drift . features with intrinsically low @xmath92 , low @xmath77 , or very small @xmath136 appear to have larger @xmath92 ( and @xmath77 ) if they are detected . but @xmath147 shows most are _ not _ detected ; those that are represent a biased sample in which @xmath92 and @xmath77 happened to be boosted in the right direction to make them detectable . by contrast , easily detectable features appear to have lower @xmath92 than they should . since @xmath219 , their @xmath77 is underestimated as well ; one of the mechanisms noted in [ sec : eval_mot ] may be to blame . but whether positive or negative , the magnitude of @xmath149 is usually only a few k. the large drifts that produced the @xmath91 outliers in the model @xmath174 results ( [ sec : eval_res_dist ] ; fig . [ fig:4hist_counts ] ) are exceptional cases . the drift in @xmath136 is negative everywhere that @xmath226 1 . it covers a range of @xmath227 , becoming steadily more negative for larger @xmath136 , with only minor dependencies on other parameters . the small positive drifts occur when @xmath228 hisa is augmented by beam - scale noise fluctuations ; @xmath228 can occur in the line wings of 1features , since pucks appear smaller off the line center ( [ sec : eval_res_dist ] ) . the much larger negative drifts are from those large features that are incompletely detected , as confirmed by visual inspection of the @xmath0 data ( e.g. , fig . [ fig : hockey_pucks ] ) . some of these partial detections are caused by @xmath134 k boundaries , but most are from noise . 2- and 3-@xmath229 noise fluctuations frequently poke holes in fairly strong features , reducing their @xmath136 measures . this is especially common in the line wings where @xmath92 is less . the drift in @xmath137 is positive only where both @xmath230 and @xmath231 and is negative everywhere else . it covers a range of @xmath232 , becoming steadily more negative for larger @xmath137 , with only minor dependencies on other parameters . as with @xmath151 , noise degradation is a major cause of this @xmath152 trend . but in addition , the spectral hisa search itself is optimized for the detection of hisa linewidths @xmath233 , and gaussian features broader than 8 are cleaned out almost entirely ( [ sec : eval_res_fid_fot ] ) . lastly , features near emission peaks may have @xmath77 underestimated ( [ sec : eval_mot ] ) , leading to @xmath92 underestimates in line wings , and thus @xmath137 underestimation . since our @xmath77 estimation is not a simple linear interpolation ( [ sec : ext_dt ] ) , this effect may not be as severe as that noted by @xcite , but the @xmath234 results above for well - detected features suggest it is not zero either . @xcite also note that @xmath77 gradients will cause @xmath137 to be narrower than the hisa optical depth profile fwhm , and the line center in @xmath91 will appear shifted from @xmath235 , the maximum optical depth . however , the performance of our hisa software is only concerned with @xmath91 , so these biases do not apply here . and while our voxel - based evaluation method is not able to track changes of position , visual inspection shows that filaments and other structures within features do not shift between input and output ; only the centroids of whole features may shift if the features are not completely detected . these results have many uses . in addition to statistically describing how well hisa is detected , they can be applied directly to particular features to assess their reliability and completeness . the former is done by measuring @xmath153 at each hisa voxel @xmath0 position and interpolating @xmath148 from the 4-d histogram described in [ sec : eval_res_fid_fot ] . @xmath148 is the likelihood that a hisa detection represents real absorption . we have determined @xmath148 for each cgps hisa voxel . figure [ fig : feature_with_fot ] shows @xmath148 contours on a sample hisa feature . the hisa detection reliability in this case is quite high . in paper ii , we use @xmath236 in analyses of total cgps hisa coverage and the distributions of weak and strong absorption . we also assess the completeness of our hisa detections . the actual detection fraction @xmath147 is not recoverable from observed data , but we consider the fraction of detections with @xmath237 k , since such hisa has @xmath238 if its size and strength are appreciable ( [ sec : eval_res_fid_fot ] ) . we have described algorithms that identify and extract self - absorption ( hisa ) features in high - resolution 21 cm line data cubes . these algorithms were designed to carry out a hisa survey of cold in the initial @xmath239 phase of the arcminute - resolution canadian galactic plane survey ( cgps ) , but they should have more general applicability . our search algorithms use clean - based spatial and spectral filtering to remove large - scale emission structure and identify hisa as significant negative residuals . features identified in both spectral and spatial domains are flagged as hisa , and the unabsorbed brightness @xmath77 along the feature sightline is estimated from a 3-d interpolation of the off - feature brightness temperature @xmath78 . hisa detections in overly noisy regions are rejected , as are those for which @xmath101 k , lest significant false detections result from gaps between sharply - structured emission features with faint backgrounds . in order to capture features larger than the clean filter scale , identified hisa is removed and the search process is repeated ; a total of three such passes suffices for the cgps data . we performed detailed tests of our hisa - finding software with model data to determine its detection limits , false positive rates , and measurement biases as functions of feature size , amplitude , and background field brightness . the tests show that hisa is well detected within the software design criteria , with high detection rates for hisa significantly stronger than the noise level , larger than a few beams , narrower than a few , and with @xmath200 k. at the same time , the bulk of hisa detections are reliable , with very low false positive rates in most parts of the parameter space except those occupied by beam - scale noise fluctuations . measurement drifts are small in well - detected features , with @xmath77 underestimated by a few k due to contamination of @xmath78 by faint , undetected hisa near the feature . where detections are truncated by noise fluctuations or faint @xmath77 , the bias may be somewhat larger . incomplete detections also make features appear smaller in angular size and linewidth than in reality due to truncation . this paper is the third in an ongoing series investigating hisa at high resolution in the galactic plane . a companion paper ( paper ii ) presents hisa survey results for the cgps . subsequent papers will further analyze the cgps hisa and also examine hisa in cgps extensions and in the vla galactic plane survey . we thank w. mccutcheon , t. landecker , and j. stil for a number of useful discussions on this project , and the anonymous referee for constructive comments on the manuscript . j. stil helped with multipanel figure layout . we are very grateful to r. gooch for tireless computing support , including continued expansion of the capabilities of the karma visualization software package @xcite , which was used extensively for this work . the dominion radio astrophysical observatory is operated as a national facility by the national research council of canada . the canadian galactic plane survey ( cgps ) is a canadian project with international partners . the cgps is described in @xcite . the main cgps data set is available at the canadian astronomy data centre . the cgps is supported by a grant from the natural sciences and engineering research council of canada . the data in the vicinity of @xmath240 have been revised from those used in paper i. a single synthesis field was assigned the wrong flux scale in the data used in that paper , and this error was not discovered until after publication . as a result , the hisa amplitudes presented in paper i for the perseus hisa complex and globule features were in error , and the correct hisa amplitudes are smaller than those found in paper i. with revised data , these features have warmer spin temperatures and lower optical depths than those derived in paper i , but the column densities and masses are only mildly affected . table [ tab : cg_prop ] lists the corrected results for both features . the correct globule spectrum is plotted in figure [ fig : globule_spectrum_raw ] , and the positions of both features are marked in figure [ fig : real_hisa_id_maps_zoom ] . the local hisa filament presented in paper i was unaffected by this problem . baker , p. l. , & burton , w. b. 1979 , , 35 , 129 bania , t. m. , & lockman , f. j. 1984 , , 54 , 513 colgan , s. w. j. , salpeter , e. e. , & terzian , y. 1988 , , 328 , 275 feldt , c. 1993 , , 276 , 531 gibson , s. j. 2002 , asp conf . 276 , _ seeing through the dust : the detection of and the exploration of the ism in galaxies _ , eds . a. r. taylor , t. l. landecker , & a. g. willis , 235 gibson , s. j. , taylor , a. r. , dewdney , p. e. , & higgs , l. a. 2000 , , 540 , 851 ( paper i ) gibson , s. j. , taylor , a. r. , higgs , l. a. , brunt , c. m. , & dewdney , p. e. 2005 , , submitted ( paper ii ) gibson , s. j. , taylor , a. r. , stil , j. m. , higgs , l. a. , dewdney , p. e. , & brunt , c. m. 2004 , in _ how does the galaxy work ? a galactic tertulia with don cox and ron reynolds _ , eds . e. j. alfaro , e. prez , & j. franco ( dordrecht : kluwer academic publishers ) , 47 gooch , r. 1996 , asp conf . ser . 101 , _ astronomical data analysis software and systems v _ , eds . g. h. jacoby & j. barnes , 80 green , d. a. 1993 , , 262 , 327 hasegawa , t. , sato , f. , & fukui , y. 1983 , , 88 , 658 hgbom , j. a. 1974 , , 15 , 417 kavars , d. w. , dickey , j. m. , mcclure - griffiths , n. m. , gaensler , b. m. , & green , a. j. 2003 , , 598 , 1048 kerton , c. r. 2005 , , accepted knapp , g. r. 1974 , , 79 , 527 knee , l. b. g. , & brunt , c. m. 2001 , , 412 , 308 levinson , f. h. , & brown , r. l. 1980 , , 242 , 416 li , d. , & goldsmith , p. f. 2003 , , 585 , 823 mcclure - griffiths , n. m. , green , a. j. , dickey , j. m. , gaensler , b. m. , haynes , r. f. , & wieringa , m. h. 2001 , , 551 , 394 . mccutcheon , w. h. , shuter , w. l. h. , & booth , r. s. 1978 , , 185 , 755 minter , a. h. , lockman , f. j. , langston , g. i. , & lockman , j. a. 2001 , , 555 , 868 montgomery , a. s. , bates , b. , & davies , r. d. 1995 , , 273 , 449 peebles , p. j. e. 1993 , _ principles of physical cosmology _ ( princeton : princeton university press ) , 35 peters , w. l. , & bash , f. n. 1987 , , 317 , 646 riegel , k. w. , & crutcher , r. m. 1972 , , 18 , 55 steer , d. g. , dewdney , p. e. , & ito , m. r. 1984 , , 137 , 159 taylor , a. r. , et al . 2003 , , 125 , 3145 taylor , a. r. , stil , j. m. , dickey , j. m. , mcclure - griffiths , n. m. , martin , p. g. , rothwell , t. , & lockman , f. j. 2002 , asp conf . 276 , _ seeing through the dust : the detection of and the exploration of the ism in galaxies _ , eds . a. r. taylor , t. l. landecker , & a. g. willis , 68 van der werf , p. p. , goss , w. m. , & vanden bout , p. a. 1988 , , 201 , 311 ll|cc|cc + & @xmath76 [ k ] & 69 & 71 & 47 & 62 + & @xmath78 [ k ] & 107 & 99 & 112 & 104 + & @xmath91 [ k ] & @xmath241 & @xmath242 & @xmath243 & @xmath244 + & + & @xmath196 [ k ] & @xmath245 & @xmath246 & @xmath247 & @xmath248 + & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 + & @xmath254 [ 10@xmath255 @xmath256 & @xmath257 & @xmath258 & @xmath259 & @xmath260 + & @xmath261 [ @xmath262 & @xmath263 & @xmath264 & @xmath265 & @xmath266 + & @xmath267 [ ] & @xmath268 & @xmath269 & @xmath270 & @xmath271 + + & @xmath196 [ k ] & @xmath272 & @xmath272 & @xmath272 & @xmath272 + & @xmath249 & @xmath273 & @xmath274 & @xmath275 & @xmath276 + & @xmath254 [ 10@xmath255 @xmath256 & @xmath277 & @xmath278 & @xmath279 & @xmath280 + & @xmath261 [ @xmath262 & @xmath281 & @xmath281 & @xmath281 & @xmath281 + & @xmath267 [ ] & @xmath282 & @xmath283 & @xmath284 & @xmath285 + & @xmath286 [ 10@xmath255 @xmath256 & @xmath287 & @xmath288 & @xmath289 & @xmath290 + & @xmath291 [ @xmath262 & @xmath292 & @xmath292 & @xmath292 & @xmath292 + & @xmath293 [ ] & @xmath294 & @xmath295 & @xmath296 & @xmath283 + [ tab : cg_prop ]

we study the automorphism group of graphons ( graph limits ) . we prove that after an appropriate `` standardization '' of the graphon , the automorphism group is compact . furthermore , we characterize the orbits of the automorphism group on-tuples of points . among applications we study the graph algebras defined by finite rank graphons and the space of node - transitive graphons .

1406.4958

graphons have been introduced as limit objects of convergent sequences of dense simple graphs , and many aspects of graphs can be extended to graphons . the goal of this paper is to describe a natural way to extend the notion of graph automorphisms to graphons . our notion of automorphism group satisfies the natural requirement that it is invariant under weak isomorphism of graphons . ( weakly isomorphic graphons represent the limit objects of the same convergent graph sequences . ) thus our study of the automorphisms of graphons fits well into graph limit theory . in this paper we heavily use the topological aspects of graph limit theory developed in @xcite . it was shown in @xcite that every graphon has two `` canonical '' representations on metric spaces , which we call , informally , the _ neighborhood metric _ and the _ @xmath1-neighborhood metric_. these metric spaces depend only on the weak isomorphism class of the graphon . ( in @xcite , these are called the `` neighborhood metric '' and the `` similarity metric '' . ) the neighborhood metric space is simpler to define and work with , but it is not compact in general ; the @xmath1-neighborhood metric space is compact . the automorphism group acts on each of these as a subgroup of isometries . it is a rather straightforward consequence of the compactness of the @xmath1-neighborhood metric that the automorphism group is always a compact topological group ( theorem [ thm : aut - comp ] ) . this fact is also closely related to ( and could be derived from ) a theorem of vershik and habck @xcite on the compactness of isometry groups of multivariate functions . as a consequence we prove that for node - transitive graphons the neighborhood metric is also compact . the space of graphons ( with weakly isomorphic graphons identified ) is compact in a natural topology ( defined by the `` cut distance '' ) . another result of this paper is that the set of node - transitive graphons is closed , and hence compact , in this topology . as we will see , graph limit theory restricted to this closed set gives rise to a rather interesting limit theory for functions on groups . such a theory was initiated in @xcite , and it was a crucial component of the limit approach to higher order fourier analysis ( see @xcite ) . we give a characterization of the orbits of the automorphism group on @xmath0-tuples of points . this generalizes results in @xcite from finite graphs to graphons , as well as the characterization of weak isomorphism of graphons by borgs , chayes and lovsz @xcite . we use this characterization to connect the topic of graph algebras with group theory . as an application , we give a group theoretic description of the graph algebras defined by finite rank graphons . it follows from our results that the limit of a convergent sequence of finite graphs , each having a node - transitive automorphism group , is a node - transitive graphon . however , the relationship between the automorphism groups of the finite graphs and that of the limit graphon is more involved . a _ @xmath0-labeled graph _ is a graph ( simple or multi ) with @xmath0 of its nodes labeled @xmath2 ( @xmath3 ) . we denote by @xmath4 the set of @xmath0-labeled multigraphs , by @xmath5 the set of @xmath0-labeled simple graphs , and by @xmath6 the set of @xmath0-labeled simple graphs with nonadjacent labeled nodes . in particular , @xmath7 is the set of unlabeled simple graphs . we will need some special @xmath0-labeled graphs and multigraphs . we denote by @xmath8 the ( unlabeled ) graph with two nodes and one edge , and by @xmath9 the multigraph consisting of two nodes connected by two edges . we denote by @xmath10 and @xmath11 the graph @xmath8 with one and two nodes labeled , respectively ; @xmath12 and @xmath13 are defined analogously . we denote by @xmath14 the path with @xmath15 nodes , with its two endpoints labeled . for two simple graphs @xmath16 and @xmath17 , let @xmath18 denote the number of homomorphisms ( adjacency - preserving maps ) @xmath19 . we define the _ homomorphism density _ @xmath20 a _ graphon _ consists of a standard probability space @xmath21 and a symmetric measurable function @xmath22 $ ] . to simplify notation , we will omit some letters that may be understood . for the standard probability space @xmath21 , we let @xmath23 denote the underlying sigma - algebra and let @xmath24 denote the probability measure . also , we write @xmath25 instead of @xmath26 in integrals if there is only one probability measure considered . every graphon @xmath27 defines an integral operator @xmath28 on the hilbert space @xmath29 by @xmath30 we say that @xmath31 has finite rank if this operator has finite rank ( i.e. , its range is a finite dimensional subspace of @xmath29 . for every graphon @xmath27 and every graph @xmath32 , we define @xmath33 we note that the formula makes sense for multigraphs @xmath16 , but we exclude loops . we write @xmath34 instead of @xmath35 if the underlying probability space is clear . graphons were introduced to describe limit objects of convergent sequences of dense graphs . a sequence of simple graphs @xmath36 is called _ convergent _ , if the numerical sequence @xmath37 converges for every simple graph @xmath16 . in this case , there is a graphon @xmath27 such that @xmath37 converges to @xmath34 for every simple graph @xmath16 @xcite . the limiting graphon is not strictly uniquely determined . quite often one assumes that @xmath38 $ ] ( with the lebesgue measure ) . in this paper , different underlying spaces will be more useful . we say that two graphons @xmath27 and @xmath39 are _ weakly isomorphic _ , if @xmath40 for every simple graph @xmath16 . every graphon is weakly isomorphic to a graphon on @xmath41 $ ] , but this is not always the most convenient representative of a weak isomorphism class . weakly isomorphic pairs of graphons were characterized in @xcite . let @xmath21 and @xmath42 be standard probability spaces and let @xmath43 be a measure preserving map . for any function @xmath44 , we define the function @xmath45 by @xmath46 it is clear that if @xmath47 is a graphon , then so is @xmath48 , which we call the _ pullback _ of @xmath47 along @xmath49 . it is easy to see that the graphons @xmath47 and @xmath48 are weakly isomorphic . it follows that all pullbacks of the same graphon are weakly isomorphic . the main result of @xcite asserts that two graphons are weakly isomorphic if and only if they are pullbacks of the same graphon . we can define a sequence of graphons @xmath50 to be _ convergent _ with limit graphon @xmath31 if @xmath51 converges to @xmath34 for every simple graph @xmath16 . ( there is a semimetric , called the `` cut distance '' , on the set of graphons that makes this space compact , and which defines this same notion of convergence . we do nt need the cut distance in this paper , however . ) we can define homomorphism densities of @xmath0-labeled graphs in graphons , but these will be @xmath0-variable functions @xmath52 rather than numbers . these _ restricted homomorphism densities _ are defined by not integrating the variables corresponding to labeled nodes : @xmath53 } } \prod_{ij\in e } w(x_i , x_j)\,\prod_{i\in v\setminus[k ] } d\pi(x_i).\ ] ] graph algebras are important algebraic structures associated with graph parameters . we give a quick introduction to the subject . for more details see @xcite . for two simple graphs @xmath54 , the product @xmath55 is defined as the graph obtained from @xmath17 and @xmath56 by identifying vertices with the same label and by reducing multiple edges . this product defines a commutative semigroup structure on @xmath57 . let @xmath58 denote the set of formal @xmath59-linear combinations of elements from @xmath57 . such linear combinations are usually called _ quantum graphs_. the multiplication extends to @xmath58 from @xmath57 using the distributive law and thus @xmath58 becomes a commutative algebra . in other words , @xmath58 is the semigroup algebra of @xmath57 . let @xmath60 be the graph obtained by removing the labels in the graph @xmath17 . we extend this notation to quantum graphs by linearity . an arbitrary graph parameter @xmath61 can be extended to quantum graphs by linearity . similarly , restricted homomorphism densities can be extended to @xmath0-labeled quantum graphs by linearity : if @xmath62 , then @xmath63 every graph parameter gives rise to a symmetric bilinear form on @xmath64 defined by @xmath65 . let @xmath66 be the set of elements @xmath67 in @xmath58 such that @xmath68 for every @xmath69 . then @xmath66 is an ideal and @xmath70 is the _ graph algebra _ corresponding to @xmath71 . the infinite matrix @xmath72 defined by @xmath73 is called the _ @xmath0-th connection matrix _ of @xmath71 . it is easy to see that the rank of @xmath74 is the dimension of @xmath75 . we will be interested in the special case when @xmath71 is defined by @xmath76 for some fixed graphon @xmath31 . in this case , @xmath75 depends only on the weak isomorphism class of @xmath31 . it was shown in @xcite that in this case the inner product @xmath77 is positive semidefinite , and hence so are the connection matrices . there is a concrete representation of @xmath75 that will be convenient to use and that will create a connection between automorphisms of @xmath31 and the graph algebras . let @xmath27 be an arbitrary graphon . we define a map @xmath78 by letting @xmath79 be the @xmath0-variable function @xmath80 . we extend this map linearly to general quantum graphs . note that @xmath81 is a commutative algebra with pointwise multiplication and addition , and @xmath82 is an algebra homomorphism . the kernel of @xmath82 is equal to @xmath66 and thus the range of @xmath82 is isomorphic to the @xmath0-th graph algebra of @xmath31 . we denote by @xmath83 this subalgebra of @xmath81 . we will need a subalgebra of @xmath84 : let @xmath85 denote the linear span of functions @xmath79 , where in @xmath86 ( so its labeled points are non - adjacent ) . by definition @xmath87 . for two points @xmath88 of a graphon @xmath27 , we define their _ neighborhood distance _ by @xmath89 it may happen that @xmath90 is not measurable for some @xmath91 ; however , we can always change @xmath31 on a set of measure @xmath92 to make these one - variable sections of it measurable . we will assume in the sequel that these functions are measurable . the distance function @xmath93 is not necessarily a metric , only a semimetric , meaning that @xmath94 may be @xmath92 for distinct points @xmath91 and @xmath95 . such points are called _ twins_. how to merge twins to get a weakly isomorphic graphon for which @xmath93 is a metric , was described in @xcite ( see also @xcite ) . as a further step of `` purifying '' a graphon , we can replace the metric space @xmath96 by its completion . furthermore , in this new topology the underlying probability measure may not have full support ; we may restrict the graphon to the support of the measure ( which is a closed and therefore complete subspace ) . this procedure is described in @xcite . we call a graphon @xmath27 _ pure _ , if @xmath96 is a complete metric space , and @xmath24 has full support ( i.e. , every open set has positive measure ) . the procedure described above implies that every graphon is weakly isomorphic to a pure graphon . pure graphons will be crucial in this paper , even in order to define automorphisms . it will be sometimes convenient to use the @xmath97-distance instead of the @xmath98-distance : we consider @xmath99 since trivially @xmath100 , these two metrics define the same topology . in particular , the metric space @xmath101 associated with a pure graphon @xmath27 is also complete and the measure @xmath24 has full support . one advantage of @xmath102 is that it can be expressed in terms of restricted homomorphism densities . we consider the @xmath1-labeled quantum graph @xmath103 in figure [ fig : h ] . then it is easy to check that @xmath104 in the representation of the metric @xmath102 . ] while the difference between the metrics @xmath93 and @xmath102 is not essential , the @xmath1-neighborhood metric ( called the _ similarity metric _ in @xcite ) is more substantially different @xcite . one way to define it is to introduce the `` operator square '' of a graphon : @xmath105 and then consider the neighborhood distance of the graphon @xmath106 : @xmath107 this definition looks artificial , but in fact it has many nice properties . it is easy to see that @xmath108 . if @xmath27 is a pure graphon , then @xmath109 is a metric space ( in particular , the distance between distinct points is positive ) , which is not necessarily complete , but we can consider its completion @xmath110 . we can extend the probability measure to @xmath111 by defining it to be @xmath92 on the set of new points . we can also extend the function @xmath31 to @xmath112 $ ] so that @xmath113 is a graphon , and the metric @xmath114 is equal to the completion of the metric @xmath115 ( this takes some care ) . we will not distinguish @xmath114 and @xmath115 in the sequel . on the other hand the metric @xmath116 is quite different : in terms of the @xmath115 metric , all open sets have positive measure , while the set @xmath117 of new points is closed and has measure @xmath92 . on the other hand , in terms of the @xmath116 metric , the set @xmath117 is open ( of measure zero ) . the main property of this completion , which we will need , is that the space @xmath110 is compact ( @xcite ; see also @xcite , corollary 13.28 ) . the metric @xmath115 has another important property ( @xcite , theorem 13.27 ) : [ prop : weakww ] if @xmath27 is pure graphon , then the metric @xmath118 defines exactly the weak topology on @xmath111 . in other words , @xmath119 @xmath120 if and only if @xmath121 for every bounded measurable function @xmath122 . a further important property of the metric @xmath115 , which we do nt use in this paper but is worth mentioning , is that a decomposition of @xmath111 into sets with small @xmath115-diameter corresponds to a ( weak ) regularity partition . we refer to @xcite for the exact statement of this correspondence . we start with citing lemma 13.19 from @xcite : [ lem : contin ] let @xmath27 be a pure graphon and let @xmath32 be a @xmath0-labeled multigraph with nonadjacent labeled nodes . then @xmath123 for all @xmath124 . we need a version of this lemma for the @xmath115-distance instead of the @xmath93-distance . some special cases of this were proved in @xcite , section 13.4 . [ lem : hom - cont ] let @xmath27 be a pure graphon , and let @xmath32 be a @xmath0-labeled simple graph with nonadjacent labeled nodes . then the restricted homomorphism function @xmath125 is continuous in each of its variables @xmath126 on the metric space @xmath110 . consider any point @xmath127 , and let @xmath128 be such that @xmath129 if @xmath130 . we want to show that @xmath131 let @xmath132 , and let @xmath133 be obtained from @xmath16 by deleting node @xmath134 and labeling nodes @xmath135 . then @xmath136 the condition @xmath129 implies that @xmath137 weakly as @xmath130 ( @xcite , theorem 13.7 ) . it is easy to see that this implies that @xmath138 ( weakly as a function of @xmath139 ) , which in turn implies that @xmath140 as claimed . let us discuss the restrictions in these lemmas . it is obvious that these lemmas do not remain valid if we allow edges between labeled nodes : for example , @xmath141 itself is not necessarily continuous . lemma [ lem : contin ] implies that @xmath142 is continuous ( even lipschitz ) in the neighborhood distance , simultaneously in all variables . lemma [ lem : hom - cont ] , however , fails to hold in this stronger sense ; see example [ exa : sym ] below ( adapted from @xcite , example 13.30 ) . this example also shows that in lemma [ lem : hom - cont ] we have to restrict @xmath16 to simple graphs . inequality also shows that , for a fixed @xmath16 , the difference @xmath143 can be estimated by @xmath144 , independently of @xmath31 . example [ exa : non - lip ] below shows that , even in the case @xmath145 , no such estimate can be given in terms of @xmath146 . [ exa : sym ] for @xmath147 , let @xmath148 be the binary expansion of @xmath95 . define @xmath149 for @xmath150 and @xmath151 . define @xmath152 for all @xmath95 . this function is not symmetric , so we put it together with a reflected copy to get a graphon : @xmath153 let @xmath154 , then ( as noted in @xcite ) the sequence @xmath155 converges to the point @xmath92 in the metric @xmath115 . on the other hand , for the @xmath156-node path labeled at both endpoints @xmath157 but @xmath158 showing that @xmath159 is not continuous at @xmath160 , and that @xmath161 , as a function of @xmath91 and @xmath95 , is not continuous at @xmath162 . [ exa : non - lip ] consider the weighted graph @xmath56 given by the matrix of edgeweights @xmath163 and the vector of nodeweights @xmath164 this weighted graph @xmath56 can be considered as a pure graphon with a @xmath165-point underlying space . then @xmath166 is the weighted graph given by the matrix of edgeweights @xmath167 and the same nodeweights as before . let @xmath168 and @xmath169 be the last two nodes , then @xmath170 , but @xmath171 and @xmath172 . we have seen that excluding edges between the labeled nodes is essential in both previous lemmas . the next lemma expresses @xmath31 in terms of restricted homomorphism numbers for graphs with nonadjacent labeled nodes , and gives a ( rather weak , but still useful ) remedy for this restriction . we consider two sequences of quantum graphs @xmath173 and @xmath174 ( figure [ fig : gg ] ) , and define @xmath175 , @xmath173 and @xmath174 . here @xmath176 is obtained by gluing together @xmath15 copies of @xmath71 along the labeled nodes ; one of these nodes is then unlabeled ( shown in white ) . ] [ lem : formula ] let @xmath27 be a pure graphon , and let @xmath177 $ ] be defined by . then @xmath178 @xmath179 . we have @xmath180 here the numerator can be expressed as @xmath181 while the denominator is @xmath182 integrating over @xmath95 , and using cauchy schwarz , we get @xmath183 it is not hard to see that the right hand side tends to @xmath92 as @xmath130 , which implies the lemma ( in fact , a little more : @xmath184 in the @xmath98 metric for every @xmath91 ) . it only makes sense to define automorphisms of pure graphons . of course , one could define an `` automorphism '' of any graphon @xmath27 as an invertible measure preserving map @xmath185 such that @xmath186 for almost all @xmath187 . however , there is a lot of trouble with this notion : weakly isomorphic graphons will have wildly different automorphism groups . an example with many automorphisms is a stepfunction @xmath31 : here @xmath188 contains the group of all invertible measure preserving transformations that leave the steps invariant ( in addition to all the automorphisms of the corresponding weighted graph ) . note , however , that if we purify a stepfunction , then we get a finite weighted graph , and so the large and `` ugly '' subgroups consisting of measure preserving transformations of the steps disappear . another problem would be that any permutation of points of a zero - measure set should be considered an automorphism , so every graphon would have a transitive automorphism group . [ def : auto ] let @xmath27 be a pure graphon . a measure preserving bijection @xmath185 is called an _ automorphism _ of @xmath27 if , for every @xmath189 , the equality @xmath186 holds for almost all @xmath190 . note the change in the phrasing of the last condition : it is stronger that requiring that @xmath186 for almost all @xmath187 . this modification will exclude `` automorphisms '' like interchanging two points . ( the simpler but inadequate definition is given in @xcite ; the results announced there hold true with the definition given here . ) it is clear that every automorphism preserves the distances @xmath93 and @xmath115 , and hence it extends to an automorphism of @xmath113 . the points of @xmath191 can be identified in the graphon @xmath113 by the property that every @xmath116-neighborhood of them has positive measure . so the automorphism groups of a pure graphon @xmath27 and its completion @xmath113 are essentially the same . in this section , we will mostly work with @xmath113 . we can endow @xmath188 with a metric ( and through this , with a topology ) by @xmath192 not every isometry of the metric space @xmath96 ( or of the metric space @xmath110 ) is an automorphism . [ exa : isom - not ] let @xmath193,w)$ ] be the pure graphon @xmath194 , and consider the direct sum @xmath193,w)\oplus ( [ 0,1],1-w)$ ] . this is pure as well , and interchanging the two components is an isometry but not an automorphism in general . the following technical lemma shows that a slight apparent weakening of the second condition in the definition of an automorphism leads to the same concept . we will formulate it for the @xmath115-metric ; for the @xmath93-metric the proof is similar ( in fact , much simpler ) . [ lem : iso - null ] let @xmath27 be a pure graphon , and let @xmath195 be a bijective measure preserving map that is an isometry of @xmath110 and satisfies @xmath196 almost everywhere . then @xmath49 is an automorphism . let us call a point @xmath197 _ nice _ , if @xmath198 for almost all @xmath199 . the condition that @xmath196 almost everywhere implies that almost all points are nice , but we want to show that _ all _ points are nice . to this end , let us fix @xmath189 . since the measure has full support in @xmath110 , every neighborhood of @xmath91 has positive measure , and hence there is a sequence of nice points @xmath200 such that @xmath201 . this means that @xmath202 also , since @xmath49 is an isometry , @xmath203 since @xmath49 is measure preserving , we can replace @xmath95 by @xmath204 and @xmath205 by @xmath206 in this equation : @xmath207 since the points @xmath200 are nice , @xmath208 for almost all @xmath95 , and similarly @xmath209 for almost all pairs @xmath210 . this implies that @xmath211 comparing with , we get @xmath212 the left hand side does not depend on @xmath15 , and hence it follows that @xmath213 for almost all @xmath205 . we can choose a sequence @xmath214 for which this holds and for which @xmath215 it is easy to see that this implies that @xmath216 a similar argument gives @xmath217 subtracting , we get @xmath218 which implies that @xmath219 for almost all @xmath95 . this proves the lemma . the following fact is stated ( without proof ) in section 13.5 of @xcite . [ thm : aut - comp ] the automorphism group of a pure graphon is compact . this theorem is an immediate consequence of the following fact . [ lem : closed ] the automorphisms of a pure graphon @xmath27 form a closed subgroup of the isometry group of @xmath220 . clearly every automorphism of @xmath27 is an isometry of @xmath110 , and these isometries form a subgroup . we want to prove that this subgroup is closed in the topology of pointwise convergence . let @xmath221 be a sequence of automorphisms of @xmath27 , and assume that they converge to an isometry @xmath49 . we want to prove that @xmath49 is not only an isometry , but an automorphism . by lemma [ lem : iso - null ] , it suffices to prove the following claims . [ claim : open1 ] for every open set x , @xmath222 as @xmath130 . indeed , since @xmath223 for every @xmath189 , it follows that for every @xmath224 , @xmath225 if @xmath15 is large enough . this means that every point belongs to a finite number of sets @xmath226 only , which implies that @xmath227 . by a similar argument , @xmath228 this implies the claim . [ claim : closed1 ] the map @xmath49 is measure preserving . it suffices to show that @xmath49 preserves the measure of any open set @xmath229 . by claim [ claim : open1 ] , @xmath230 as @xmath130 . since @xmath231 is measure preserving , this implies that @xmath232 . [ claim : closed2 ] @xmath196 almost everywhere . it suffices to prove that for any two open sets @xmath233 and @xmath234 , @xmath235 for every @xmath190 , @xmath236 using that the maps @xmath231 are automorphisms , @xmath237 the first term on the right side tends to @xmath238 by proposition [ prop : weakww ] , and the difference in the last line tends to @xmath92 as @xmath130 by and claim [ claim : open1 ] . this proves claim [ claim : closed2 ] , and thereby the lemma . since @xmath31 is bounded , the operator @xmath28 is hilbert - schmidt and hence it has a spectral decomposition @xmath239 where the @xmath240 are its nonzero eigenvalues and the functions @xmath241 are the corresponding eigenfunctions , forming an orthonormal system . here by definition @xmath243 almost everywhere . we assume that @xmath90 is measurable for every @xmath91 , and we can change @xmath244 on a set of measure @xmath92 so that holds for every @xmath189 . we note that implies that @xmath244 is bounded : @xmath245 we need the following simple observation : for every @xmath189 , @xmath246 indeed , using and the fact that @xmath247 is an orthonormal system , we get @xmath248 ( the last equality follows because even though @xmath247 may not be a complete orthogonal system , it can be extended by functions in the nullspace of @xmath28 to such a system , and these additional functions contribute @xmath92 terms ) . the second equality in is trivial by definition . in turn implies that @xmath249 expansion may not hold pointwise , only in @xmath97 ; but it follows from basic results on hilbert - schmidt operators that if we take the inner product with any function @xmath250 , then we get an equation : @xmath251 where the sum on the right side is absolutely convergent . we need the following stronger fact : [ lem : more - conv ] let @xmath27 be a graphon , and let be its spectral decomposition . \(a ) for @xmath252 and @xmath190 , the sum @xmath253 is absolutely convergent . \(b ) for every bounded measurable function @xmath254 and for almost all @xmath190 , @xmath255 \(a ) we have @xmath256 here the first factor is the tail of a convergent sum by , and hence it tends to @xmath92 as @xmath130 . furthermore , @xmath247 is an orthonormal system , and hence @xmath257 proving ( a ) . let @xmath258 and @xmath259 be the functions on the left and right sides of equation . then for any bounded measurable function @xmath260 , @xmath261 ( where we use and the fact that the sum in the third line is absolutely convergent ) . this proves that @xmath262 almost everywhere . in this chapter we use the topological properties of pure graphons to formulate finer statements about spectral decompositions . first of all , note that if @xmath27 is a pure graphon then eigenfunctions of @xmath31 are continuous functions on @xmath111 in the metric @xmath115 ( @xcite , corollary 13.29 ) . furthermore , the eigenfunctions separate the points of @xmath21 : [ specsepar ] if @xmath27 is a pure graphon , then for every pair of distinct points @xmath187 there is an eigenfunction @xmath71 of @xmath31 such that @xmath263 . by way of contradiction , assume that @xmath91 and @xmath95 can not be separated this way . from @xmath264 we obtain that @xmath265 , and thus the functions @xmath266 and @xmath267 have a positive distance in @xmath29 . on the other hand , @xmath268 holds for every fixed @xmath269 where the sum is @xmath97-convergent . applying this formula for @xmath270 and @xmath271 together with our assumption that @xmath272 , we get a contradiction . [ lem : unif - eigen ] if @xmath27 is a pure graphon , then the sum on the left side of converges uniformly for @xmath197 . using continuity of the eigenfunctions we obtain that every term on the left side of is continuous in @xmath115 , and so is the right side by lemma [ lem : hom - cont ] . since every term on the left side is nonnegative , it follows by dini s theorem that the convergence is uniform in @xmath91 . this allows us to get the following stronger version of lemma [ lem : more - conv ] for pure graphons : [ lem : unif - conv ] ( a ) if @xmath27 is a pure graphon , then the sum is uniformly absolute convergent for @xmath273 . \(b ) if , in addition , @xmath274 is a continuous function of @xmath95 for every @xmath189 in the neighborhood distance , then the expansion holds for every @xmath190 . \(a ) by lemma [ lem : unif - eigen ] , @xmath275 hence the computation in gives an estimate of the tail uniformly for all @xmath273 . \(b ) the left side of defines a continuous function of @xmath190 in the metric @xmath93 . every term on the right side is also continuous , and the convergence is uniform by the estimate , using . hence the limit is a continuous function of @xmath190 . the space @xmath21 has the property that every nonempty open set has positive measure . if two continuous functions are equal almost everywhere on such a space , then they are equal everywhere . [ cor : trans - compact ] if the automorphism group of a pure graphon @xmath27 is transitive on @xmath21 , then @xmath27 is compact and @xmath276 . let @xmath189 , then the orbit of @xmath91 is a continuous image of @xmath277 , and so it is compact in the metric @xmath93 . if the automorphism group is transitive on @xmath21 , then this orbit is @xmath21 , and hence @xmath96 is compact . since @xmath278 , this implies that @xmath279 is compact , and since @xmath21 is dense in @xmath110 , it follows that @xmath280 . we use our results above about spectra to describe a way , more explicit than convergence in @xmath97 , of the convergence of the expansion . for a graphon @xmath27 and @xmath281 , we define the graphon @xmath282_\lambda)$ ] by the following partial sum of ( [ eq : spec1 ] ) : @xmath283_\lambda(x , y)=\sum_{|\lambda_r|\geq\lambda } \lambda_r f_r(x)f_r(y).\ ] ] note that this sum is finite . if @xmath31 has multiple eigenvalues , then the terms @xmath284 depend on the basis chosen in the eigenspaces , but @xmath285_\lambda$ ] does not depend on this basis . let @xmath286 where @xmath287 is the eigenspace of @xmath31 corresponding to @xmath240 . let @xmath288 denote the orthogonal projection of @xmath29 onto @xmath289 . then @xmath290_\lambda}=t_w\pi_\lambda$ ] . from the inequality @xmath291 , it follows that the rank of @xmath285_\lambda$ ] ( the dimension of @xmath289 ) is at most @xmath292 . assume that the eigenvalues are ordered so that @xmath293 . let @xmath294 denote the probability distribution of the vector @xmath295 , where @xmath296 and @xmath189 is chosen randomly , and let @xmath297 be the support of @xmath294 . then the purification of @xmath282_\lambda)$ ] can be defined as @xmath298 , where @xmath299 a coordinate - independent way of describing @xmath294 is to consider the dual space of @xmath289 . for each @xmath189 , we consider the linear functional @xmath300 ( @xmath301 ) . if @xmath189 is chosen randomly we obtain the probability distribution @xmath294 on @xmath302 , and we can define @xmath303 as its support . we will need the next lemma , which is a direct consequence of the results in the paper @xcite . [ meascon ] let @xmath304 be a convergent sequence of graphons with limit @xmath31 . assume that @xmath281 is not an eigenvalue of @xmath31 . then there is subsequence @xmath305 in @xmath304 and choices of orthonormal eigenvectors for @xmath306_\lambda$ ] and @xmath307 $ ] such that the measures @xmath308 constructed above for @xmath309 converge to @xmath294 weakly . if @xmath310 , then the projection @xmath311 ( by forgetting the last @xmath312 coordinates ) transforms @xmath313 into @xmath314 . the map @xmath315 is surjective from @xmath316 to @xmath317 . let @xmath318 be a decreasing sequence tending to @xmath92 . let @xmath319 be the inverse limit of the system @xmath320 . this means that @xmath321 the limit of @xmath322 defines a probability measure @xmath323 on the compact set @xmath319 . let @xmath324 be the graphon defined on @xmath319 using the formula for the @xmath325-th coordinate . [ specpure ] for every graphon @xmath27 there is a measure preserving homeomorphism @xmath326 such that @xmath327_{\alpha_i}$ ] holds for every @xmath325 . notice that the construction of @xmath324 depends only on the weak isomorphism class of @xmath31 , and so we can assume that @xmath27 is pure . the maps @xmath328 from @xmath111 to @xmath329 ( where @xmath330 ) are continuous in the @xmath115 metric . hence the map @xmath331 is also continuous . since @xmath332 separates elements in @xmath111 ( to see this , apply lemma [ specsepar ] for @xmath333 ) , it is a bijection between @xmath111 and @xmath319 . the desired property is clear from the definition of @xmath332 . as an application of spectral decomposition , we prove the following generalization of lemma 5.1 in @xcite ( which will be needed later on ) . [ lem : orbits2 ] let @xmath334 and @xmath335 be two pure graphons and let @xmath336 , @xmath337 . let @xmath103 be a @xmath0-labeled quantum multigraph . in every constituent of @xmath103 , select an edge such that at least one endpoint of it is unlabeled , and let @xmath338 denote the @xmath0-labeled quantum multigraph obtained from @xmath103 by subdividing the selected edge by @xmath339 new nodes in every constituent . suppose there exists an @xmath340 such that @xmath341 for every @xmath342 . then @xmath343 . let @xmath344 be obtained from @xmath103 by keeping only those terms in which one endpoint of the selected edge is labeled @xmath325 ( @xmath345 ) . let @xmath346 be the sum of the remaining terms , where the selected edge has no labeled endpoint . let @xmath347 be the @xmath348-labeled quantum multigraph obtained from @xmath344 by deleting the selected edge from each constituent and labeling its unlabeled endpoint by @xmath349 . let @xmath350 be the @xmath351-labeled quantum multigraph obtained from @xmath346 by deleting the selected edge from each constituent and labeling its endpoints by @xmath349 and @xmath352 . then @xmath353 we use the spectral decomposition @xmath354 this decomposition holds almost everywhere for @xmath355 , but for @xmath356 , we can only claim that the sums on the right sides converge to the function on the left in @xmath97 . since the graphon is pure , lemma [ lem : unif - conv ] implies that the expansion @xmath357 holds for all @xmath358 . we have an analogous expansion for @xmath359 . if these two expressions are equal for every integer @xmath342 , then they are also equal for @xmath356 ( see e.g. @xcite , proposition a.21 ) . [ cor : orbits2 ] let @xmath334 and @xmath335 be two pure graphons and let @xmath336 , @xmath337 . \(a ) if @xmath360 for every @xmath0-labeled simple graph @xmath16 , then holds for every @xmath0-labeled multigraph @xmath16 . \(b ) if holds for every @xmath0-labeled simple graph @xmath16 with nonadjacent labeled nodes , then holds for every @xmath0-labeled multigraph @xmath16 with nonadjacent labeled nodes . \(b ) follows from lemma [ lem : orbits2 ] by induction on the number of parallel edges . to prove ( a ) , it suffices to note that @xmath361 follows by considering the simple graph @xmath16 with a single edge connecting @xmath325 and @xmath362 . let @xmath363 be an automorphism of a graphon @xmath27 . notice that if @xmath71 is an eigenfunction of length @xmath134 of @xmath31 then @xmath364 is also an eigenfunction of length @xmath134 corresponding to the same eigenvalue . as a consequence every automorphism of @xmath31 acts on the space @xmath289 defined in as an element in @xmath365 where @xmath366 is the orthogonal group on @xmath287 . the corresponding action on the dual space @xmath367 leaves the measure @xmath294 invariant . we will denote by @xmath368 the finite dimensional compact group formed by all elements @xmath369 that preserve @xmath294 . ( note that @xmath368 is the automorphism group of @xmath285_\lambda$ ] . ) the group @xmath370 acts on both @xmath371 and @xmath372 . since @xmath373 is an invariant subspace of @xmath370 , the group @xmath370 acts on @xmath374 as well . in particular , there is a homomorphism @xmath375 . we denote by @xmath376 the inverse limit of the system @xmath377 . we can describe the automorphism group of a compact graphon using representation of a graphon above . [ specaut ] for every graphon @xmath27 the action of @xmath188 on @xmath111 can be obtained as @xmath378 , where @xmath332 is the function in lemma [ specpure ] . we may assume that the graphon @xmath27 is pure . first we show that @xmath379 . every automorphism of @xmath31 , restricted to @xmath380 @xmath381 , induces a consistent sequence of elements in @xmath382 . it follows that @xmath383 . the other containment is a direct consequence of lemma [ specpure ] : elements of @xmath378 act on @xmath111 continuously and leave @xmath384_{\alpha_i}$ ] invariant for every @xmath325 . this means that they also fix @xmath385 . the following theorem characterizes the orbits of the automorphism group of a graphon . [ thm : orbits ] let @xmath27 be a pure graphon , and let @xmath386 . then there exists an automorphism @xmath387 such that @xmath388 if and only if @xmath389 for every @xmath0-labeled simple graph @xmath16 in which the labeled nodes are independent . the following version is more general ( at least formally ) . [ thm : orbits2 ] let @xmath334 and @xmath335 be two pure graphons and let @xmath390 . then there exists a measure preserving bijection @xmath391 such that @xmath392 almost everywhere and @xmath393 if and only if @xmath394 for every @xmath0-labeled simple graph @xmath16 . the proof of this theorem is a modification of the proof of the main result of @xcite , combined with more recent methods involving pure graphons . first , we note that the condition in the theorem is self - sharpening : by corollary [ cor : orbits2 ] , the condition holds for every @xmath0-labeled multigraph @xmath16 . the following lemma is the main step in the proof . [ lem : couple ] let @xmath334 and @xmath335 be two graphons and let @xmath395 , @xmath337 such that @xmath396 for every @xmath0-labeled multigraph @xmath16 . let @xmath397 denote the probability measure of @xmath398 . then we can couple @xmath399 with @xmath400 so that if @xmath401 is a pair from the coupling distribution , then @xmath402 almost surely for every @xmath348-labeled multigraph @xmath16 . consider two random points @xmath403 from @xmath24 and @xmath404 from @xmath405 , and the random variables @xmath406 with values in @xmath41^{\ff_{k+1}}$ ] . we claim that the variables @xmath233 and @xmath234 have the same distribution . it suffices to show that @xmath233 and @xmath234 have the same mixed moments . if @xmath407 , and @xmath408 are nonnegative integers , then the corresponding moment of @xmath233 is @xmath409 where the multigraph @xmath16 is obtained by unlabeling the node labeled @xmath349 in the multigraph @xmath410 . expressing the moments of @xmath234 in a similar way , we see that they are equal by hypothesis . this proves that @xmath233 and @xmath234 have the same distribution . using lemma 6.2 of @xcite it follows that we can couple the variables @xmath403 and @xmath404 so that @xmath411 with probability 1 . in other words , @xmath412 for every @xmath413 with probability 1 . for an infinite sequence @xmath414 , let @xmath415 $ ] denote its prefix of length @xmath15 . [ lem : couple2 ] under the conditions of the previous lemma , we can couple @xmath416 with @xmath417 so that if @xmath401 is a pair from the coupling distribution , then for every @xmath418 and every @xmath419-labeled graph @xmath16 , @xmath420}(f , w_1)=t_{b_1\dots b_ky[n]}(f , w_2)\ ] ] almost surely . by lemma [ lem : couple ] , we can define recursively a coupling @xmath421 of @xmath422 with @xmath423 so that @xmath424 almost surely for every @xmath425 , and @xmath426 , projected to the first @xmath15 coordinates in both spaces , gives @xmath421 . the distributions @xmath421 give a distribution @xmath427 on @xmath428 , which clearly has the desired properties . the following lemma can be considered as a version of the theorem for infinite sequences . [ lem : mapping ] let @xmath334 and @xmath335 be two pure graphons , and let @xmath429 be a sequence whose elements are dense in @xmath398 . suppose that @xmath430 for every partially labeled multigraph @xmath16 . then there is a measure preserving bijection @xmath391 such that @xmath392 almost everywhere and @xmath431 for all @xmath432 . the notation @xmath433 , where @xmath168 is an infinite sequence , means that only those elements of @xmath168 are considered whose subscript occurs in @xmath16 as a label . we start with noticing that @xmath434 this follows by and the hypothesis of the lemma . for @xmath435 , take a subsequence @xmath436 such that @xmath437 . then @xmath436 is a cauchy sequence , and hence , by , so is the sequence @xmath438 , and since @xmath439 is complete , it has a limit @xmath440 . it is easy to see that this map is well - defined ( i.e. , it does not depend on the choice of the sequence @xmath436 ) , and that @xmath49 is bijective . next , we claim that for every sequence @xmath441 and every multigraph @xmath16 with nonadjacent labeled nodes @xmath442 indeed , this holds if every @xmath126 is an element of the sequence @xmath443 by hypothesis , and then it follows for all @xmath126 by the continuity of @xmath444 ( lemma [ lem : contin ] ) . finally , consider the function @xmath445 by lemma [ lem : formula ] , @xmath446 as @xmath130 . also , by , @xmath447 and applying lemma [ lem : formula ] again , @xmath448 as @xmath130 . this implies that @xmath449 almost everywhere . now we are ready to prove the main theorem of this section . theorem [ thm : orbits2 ] let @xmath450 be independent random points of @xmath451 , and let @xmath452 be independent random points of @xmath453 . applying lemma [ lem : couple2 ] repeatedly , we can couple @xmath450 with @xmath452 so that , for any @xmath454-labeled graph @xmath16 , @xmath455 with probability @xmath134 , the elements of both sequences @xmath456 and @xmath457 are dense in @xmath451 and @xmath453 , respectively . let us fix such a choice , then by lemma [ lem : mapping ] there is a measure preserving bijection @xmath391 such that @xmath392 almost everywhere and @xmath458 for all @xmath459 . this proves the theorem . [ cor : alginv ] let @xmath27 be a pure graphon . then the closure of @xmath85 in @xmath460 consists of all continuous @xmath277-invariant functions on @xmath461 . lemma [ lem : contin ] implies that all functions in @xmath85 are continuous and clearly they are invariant under automorphisms . the other containment follows from the stone - weierstrass theorem , since @xmath461 is compact , and by theorem [ thm : orbits ] the elements in @xmath85 separate the orbits of @xmath277 . let @xmath462 be the automorphism group of the pure graphon @xmath27 . we consider the natural action of @xmath462 on functions on @xmath21 defined by @xmath463 . similarly @xmath464 acts diagonally on functions on @xmath465 . for a subset @xmath466 we denote by @xmath467 the set of @xmath462-invariant elements in @xmath319 . it is clear that restricted homomorphism functions are invariant under the action of @xmath462 and thus all the algebras @xmath84 are @xmath462-invariant . a graphon is called _ node - transitive _ if the automorphism group of its pure representation @xmath27 acts transitively on @xmath21 . the next theorem gives an algebraic characterization of node - transitive graphons . [ thm : algtr ] let @xmath27 be a graphon . the following statements are equivalent . \(i ) @xmath27 is node - transitive . \(ii ) the functions @xmath468 are essentially constant on @xmath21 for all @xmath469 . \(iii ) @xmath470 . \(iv ) the first connection matrix @xmath471 of @xmath31 has rank @xmath134 . \(v ) @xmath472 for all @xmath473 . we may assume that @xmath27 is pure . if ( i ) holds , then @xmath462 is transitive on @xmath21 , and so every function @xmath468 is constant on @xmath21 , which implies ( ii ) . conversely , ( ii ) implies by theorem [ thm : orbits ] that @xmath464 is transitive on @xmath21 , so ( i ) holds . thus ( i ) and ( ii ) are equivalent . every constant function is in @xmath474 , hence @xmath475 , and so ( ii ) is equivalent to ( iii ) . we know that @xmath476 , so ( iv ) is just a re - statement of ( iii ) . finally , ( v ) is a re - statement of ( iv ) , since @xmath74 is positive semidefinite . examples for node - transitive graphons are finite node - transitive graphs . other examples are graphons defined on compact topological groups . [ def : cayley ] let @xmath462 be a second countable compact topological group , which , together with its haar measure , defines a standard probability space . let @xmath477 $ ] be a measurable function such that @xmath478 . then the graphon @xmath479 $ ] defined by @xmath480 is called a _ cayley graphon_. note that the condition @xmath478 is needed to guarantee that @xmath31 is symmetric . by omitting this condition we get `` directed cayley graphons '' . [ thm : trcayley ] cayley graphons are node - transitive . conversely , every node - transitive graphon is weakly isomorphic to a cayley graphon . note that a finite node - transitive graph @xmath17 is not necessarily a cayley graph ( for example , the petersen graph ) . however one can obtain a cayley graph @xmath481 from @xmath17 by replacing every vertex by @xmath482 vertices and every edge by a complete bipartite graph @xmath483 . the value @xmath482 is the size of the stabilizer of a vertex in @xmath17 in the automorphism group . the graph @xmath481 is weakly isomorphic to @xmath17 as a graphon . let @xmath484 be a cayley graphon on the compact topological group @xmath464 . it is clear that @xmath462 acts transitively ( with multiplication from the right ) on this graphon . ( however , @xmath31 might not be pure . ) it follows that the restricted homomorphism functions @xmath468 are all constant on @xmath462 . the third condition in theorem [ thm : algtr ] shows that @xmath31 is node - transitive . to prove the second assertion , let @xmath485 be a node - transitive graphon ; we may assume that it is pure . let @xmath464 be its automorphism group . we know that @xmath464 is compact , and so it has a normalized haar measure @xmath323 . let us fix an element @xmath486 , and define the function @xmath487 $ ] by @xmath488 . we claim that @xmath489 is a cayley graphon weakly isomorphic to @xmath485 . [ claim : mp ] the map @xmath490 defined on @xmath462 is measure preserving . the definition of the metric on @xmath188 implies that @xmath491 for every @xmath492 . this shows that @xmath493 is continuous and hence , measurable . let @xmath494 denote the sigma - algebra on @xmath462 formed by the sets @xmath495 , where @xmath403 is a borel set in @xmath21 , and let @xmath496 denote the measure on @xmath494 that is the pullback of @xmath24 . it is clear that @xmath496 is @xmath462-invariant . standard topological group theory shows that @xmath496 extends to the borel sigma - algebra on @xmath462 as the normalized haar measure @xmath323 . this proves the claim . since by definition @xmath497 , the claim implies that @xmath498 is weakly isomorphic to @xmath27 . let @xmath499 $ ] be defined by @xmath500 . then @xmath501 , so @xmath498 is a cayley graphon . [ rem : groups ] theorem [ thm : trcayley ] creates a connection between graph limit theory and an interesting and rich limit theory for functions on groups ( see @xcite , @xcite ) . the idea is the following . let @xmath502\}_{i=1}^\infty$ ] be a sequence of measurable functions on compact groups . we say that the sequence @xmath503 is convergent if the corresponding cayley graphons @xmath504 converge . by proposition [ prop : trans - lim ] and theorem [ thm : trcayley ] , we prove that the limit of @xmath504 is weakly isomorphic to a cayley graphon defined by a measurable function @xmath505 $ ] on a compact group . we say that @xmath71 is the _ limit object _ of the sequence @xmath506 . it turns out that one can define this limit concept without passing to graphons . this point of view was heavily used in the second author s approach @xcite to higher order fourier analysis . we start with the observation that , if a convergent graph sequence consists of node - transitive graphs , then their limit graphon is node - transitive as well . more generally , we have the following consequence of the fifth condition in theorem [ thm : algtr ] . [ prop : trans - lim ] if a sequence of node - transitive graphons is convergent , then their limit graphon is also node - transitive . what makes this simple assertion interesting is the fact that the automorphism group of the limit graphon is not determined by the automorphism groups of graphs or graphons in the convergent sequence . [ exa : cyclic ] fix any @xmath507 , and define the graph @xmath36 by @xmath508 $ ] , where every @xmath509 $ ] is connected to the next and previous @xmath510 nodes ( modulo @xmath15 ) . the automorphism group of @xmath36 is the dihedral group @xmath511 . this sequence tends to the pure graphon on @xmath512 , with @xmath513 , whose automorphism group is @xmath514 , the continuous version of the dihedral groups . no surprise so far . but let us consider the graphs @xmath515 . add edges connecting every node @xmath516 to @xmath517 , where @xmath518 . let @xmath519 denote the resulting graph . identifying node @xmath516 with @xmath520 $ ] , it is not hard to see that @xmath521 . the limit of this graph sequence is the pure graphon @xmath27 , where @xmath21 is the torus @xmath522 , and @xmath523 , whose automorphism group is the wreath product of @xmath514 with @xmath524 , a @xmath1-dimensional group different from @xmath514 . the next example ( in a slightly different form ) is from the papers @xcite and @xcite . it shows that even if the underlying group @xmath462 is the same for a convergent sequence of cayley graphons , a transitive action on the limit graphon may need a different , bigger group . let @xmath525 be the circle group and let @xmath526 be the character defined by @xmath527 . let @xmath173 be the function @xmath528 where @xmath529 denotes the imaginary part . it is not hard to see that the limit of the cayley graphons corresponding to @xmath173 is the cayley graphon corresponding to the function @xmath530 on the torus @xmath531 . in the light of the previous example the next theorem ( which we quote from @xcite ) is somewhat surprising . we need a definition . let @xmath462 be a compact group with haar measure @xmath323 . let @xmath532 denote the subspace of @xmath533 spanned by the @xmath462-invariant subspaces of dimension at most @xmath15 . we say that @xmath462 is _ weakly random _ if @xmath532 is finite dimensional for every @xmath15 . [ thm : cayley ] let @xmath462 be a weakly random compact group . let @xmath534\}_{n=1}^\infty$ ] be a sequence of measurable functions such that the corresponding cayley graphons converge . then the limit graphon is again a cayley graphon on @xmath462 . the best known example for a weakly random group is the orthogonal group @xmath535 . this shows that cayley graphons on @xmath535 behave very differently from cayley graphons on @xmath514 . cayley graphons on @xmath535 are closed with respect to graphon convergence , however cayley graphons on @xmath514 are not closed . we cite a related result , which is a consequence of a theorem of gowers @xcite , indicating further , more subtle , relations between the automorphism groups of graphs and their limits . [ thm : gowers ] let @xmath36 be a cayley graph of a group @xmath536 @xmath537 , where the edge - density of @xmath36 tends to a limit @xmath538 , and the minimum dimension in which @xmath536 has a nontrivial representation tends to infinity . then the sequence @xmath539 is quasirandom , i.e. , it tends to a pure graphon @xmath27 where @xmath21 has a single point . our goal is to determine the automorphism group of the limit of a sequence of node transitive graphs . using lemma [ specaut ] one can reduce the problem of computing the automorphism group of @xmath31 to the same problem about bounded rank graphons . to demonstrate this principle we show the next theorem . recall that a compact group @xmath540 is _ abelian by pro - finite _ if it has a closed abelian normal subgroup @xmath233 such that @xmath541 is the inverse limit of finite groups . let @xmath542 be a sequence of node - transitive graphs converging to a graphon @xmath27 . then @xmath27 is weakly isomorphic to a cayley graphon on an abelian by pro - finite group . we want to show that @xmath543 has a closed , abelian by pro - finite subgroup that acts transitively on @xmath21 . let @xmath318 be a decreasing sequence of real numbers with @xmath544 that contains no eigenvalue of @xmath31 . we can assume that @xmath27 is pure . since @xmath31 is node transitive , theorem [ thm : aut - comp ] implies that @xmath21 is compact and @xmath276 . we will use the notation from chapter [ sec : specpure ] . for every @xmath545 let @xmath546 denote the measure defined above for @xmath545 in the explicit coordinate system @xmath547 where @xmath548 . for finitely many values of @xmath362 the measure @xmath546 may exist in a different dimension but we ignore those values . by choosing a subsequence we can assume without loss of generality that the conditions of the lemma [ meascon ] hold for every @xmath325 . let @xmath549 denote the automorphism group of @xmath546 and let @xmath550 denote the closed subgroup in @xmath551 whose elements are ultra - limits ( for some fixed ultrafilter @xmath552 ) of sequences @xmath553 where @xmath554 . it is clear that elements of @xmath550 preserve @xmath555 and it acts transitively on @xmath329 . we claim that @xmath550 is abelian by finite . a classical theorem by camille jordan @xcite states that there is a function @xmath556 such that any finite subgroup of @xmath557 contains an abelian group of index at most @xmath556 . using this theorem , we see that each @xmath558 has an abelian subgroup of index at most @xmath559 . it is a standard technique to show that this property is inherited by the ultralimit @xmath550 . if the groups @xmath558 are all abelian , then the continuity of the commutator word shows that @xmath550 is abelian . for the general case , choose @xmath559 coset representatives @xmath560 in each group @xmath558 for the abelian subgroup where @xmath561 . their limits as @xmath562 will be coset representatives for the limiting abelian group . to finish the proof , let @xmath56 be the inverse limit of the groups @xmath550 with respect to the homomorphisms @xmath563 . then @xmath564 and @xmath56 acts transitively on @xmath319 . by lemma [ specaut ] we obtain that @xmath565 is transitive on @xmath21 . we conclude with an application of our results on automorphisms of graphons to characterize graph algebras of graphons that have finite rank as integral kernel operators . let @xmath27 be a pure graphon with finite rank . the spectral decomposition takes the simpler form @xmath566 for any sufficiently small @xmath281 , we have @xmath285_\lambda = w$ ] , and so the considerations in section [ sec : specpure ] imply that @xmath96 is compact . let @xmath567 , and let @xmath319 be the function algebra generated by the eigenfunctions of @xmath31 . we denote by @xmath568 the space of homogeneous polynomials of degree @xmath15 in the eigenfunctions of @xmath31 , so that @xmath569 . substituting in the definition of restricted homomorphism numbers , we see that @xmath570 . since the functions in @xmath474 are @xmath464-invariant , it follows that @xmath571 . our main goal is to prove that equality holds here . assertion ( a ) follows by substituting formula ( [ eq : specdec ] ) in ( [ eq : higmult ] ) . to prove ( b ) , let @xmath580 , and let @xmath581 denote the one - labeled quantum graph obtained from @xmath582 by connecting a new node with label @xmath134 to all the labeled nodes and then we removing the original labels . then @xmath583 . finally , ( c ) follows by replacing @xmath584 by @xmath585 in the formula for @xmath586 . since the action of @xmath462 is measure preserving , the integration over @xmath587 is equivalent to the integration over @xmath588 . if @xmath591 , then @xmath592 every function @xmath589 can be expressed as a linear combination of functions such as that on the right side of the previous formula . since @xmath593 is linear in @xmath103 , this completes the proof . by lemma [ lem : fr2 ] , @xmath595 for a suitable @xmath596 . let @xmath597 . it is clear that @xmath103 is @xmath462-invariant . by lemma [ lem : fr1 ] and the linearity of @xmath598 in the first variable it follows that @xmath599 . trivially @xmath601 . to prove the reverse , let @xmath594 . trivially @xmath589 , so it suffices to prove that @xmath602 . lemma [ lem : fr3 ] shows that @xmath599 for some @xmath462-invariant function @xmath603 . using corollary [ cor : alginv ] , there is a sequence of functions @xmath604 such that @xmath605 ( @xmath606 ) uniformly in @xmath91 . by lemma [ lem : fr1 ] , @xmath607 ( as a function of @xmath189 ) , and clearly @xmath608 uniformly in @xmath91 . this implies that @xmath602 . we have seen that @xmath610 . to prove the reverse , we note that every function @xmath611 is a finite sum of functions @xmath612 , where @xmath613 , and if @xmath71 is @xmath464-invariant , then so are the terms @xmath173 . hence @xmath467 is the linear span of the spaces @xmath614 . by the previous lemma we get that @xmath615 . the algebra @xmath319 is a finitely generated commutative algebra and the compact group @xmath462 acts on @xmath319 via automorphisms . hilbert s theorem on @xmath464-invariant rings implies that @xmath616 is finitely generated . vershik and u. habck : compactness of the congruence group of measurable functions in several variables , _ zapiski nauchnykh sem . pomi _ * 334 * ( 2006 ) , 5767 ; english translation : _ j. math . ( n.y . ) * 141 * ( 2007 ) , 16011607 .

stars form in cold molecular clouds . however , molecular gas is difficult to observe because the most abundant molecule ( ) lacks a permanent dipole moment . rotational transitions of co are often used as a tracer of , but co is much less abundant and the conversion from co intensity to mass is often highly uncertain . here we present a new method for estimating the column density of cold molecular gas ( ) using optical spectroscopy . we utilise the spatially resolved h maps of flux and velocity dispersion from the sydney - aao multi - object integral - field spectrograph ( sami ) galaxy survey . we derive maps of by inverting the multi - freefall star formation relation , which connects the star formation rate surface density ( ) with and the turbulent mach number ( ) . based on the measured range of and , we predict in the star - forming regions of our sample of 260 sami galaxies . these values are close to previously measured obtained directly with unresolved co observations of similar galaxies at low redshift . we classify each galaxy in our sample as ` star - forming ' ( 219 ) or ` composite / agn / shock ' ( 41 ) , and find that in ` composite / agn / shock ' galaxies the average , , and are enhanced by factors of , , and , respectively , compared to star - forming galaxies . we compare our predictions of with those obtained by inverting the kennicutt - schmidt relation and find that our new method is a factor of two more accurate in predicting , with an average deviation of 32% from the actual . [ firstpage ] galaxies : ism galaxies : star formation galaxies : structure stars : formation techniques : spectroscopic turbulence

1703.09224

the coalescence of gases by turbulence and gravity intricately controls star formation within giant molecular clouds @xcite . on one hand , turbulence has the ability to hinder star formation by providing kinetic energy that can oppose gravity . on the other , the supersonic turbulence ubiquitously observed in the molecular phase of the interstellar medium ( ism ) produces local shocks and compressions , which lead to enhanced gas densities that are key for triggering star formation @xcite . understanding the complex effects of turbulence in the ism is therefore crucial to understanding the process of galaxy evolution . the cold turbulent gas that provides the fuel for star formation is only visible in the millimetre / sub - millimetre to radio wavelengths , and is often faint , making it difficult to detect at high spatial resolutions . a standard method to measure the mean column density of molecular gas ( @xmath1 ) is to use rotational lines of co. a severe problem with this method is that , because co is about @xmath8 times less abundant than the main mass carrier , @xmath0 , one requires a co - to-@xmath0 conversion factor , which is typically calibrated based on measurements in our own galaxy . however , the co - to-@xmath0 conversion factor may depend on metallicity , environment and redshift , introducing high uncertainties in the reconstruction of the total gas surface densities from measurements of co @xcite . another method is to measure dust emission or dust extinction and assuming a gas - to - dust ratio to infer the molecular gas masses and surface densities . these methods can suffer from uncertainties in the gas - to - dust ratio , especially for low - metallicity galaxies where this ratio becomes increasingly uncertain . both co and dust observations require telescopes and instruments that work at millimetre / sub - millimetre wavelengths , which may not always be available and/or may have relatively low spatial resolution . here we present a new method to estimate @xmath1 based on the star formation rate ( sfr ) , which can be obtained with optical spectroscopy . large optical integral field spectroscopy ( ifs ) surveys have started to provide us with details regarding the chemical distribution and kinematics of extragalactic sources at a size and uniformity unprecedented until recent times . large galaxy surveys such as the sloan digital sky survey ( sdss ; * ? ? ? * ; * ? ? ? * ) , 2-degreefield galaxy redshift survey ( 2dfgrs ; * ? ? ? * ) , the cosmic evolution survey ( cosmos ; * ? ? ? * ) , the vimos vlt deep survey ( vvds ; * ? ? ? * ) , and the galaxy and mass assembly survey ( gama ; * ? ? ? * ; * ? ? ? * ) have contributed more than 3.5 million spectra that have been of extraordinary aid to our understanding of galaxy evolution . however , those spectra have been taken with a single fibre or slit , and provide only a single , global spectrum per galaxy @xcite . these spectra are therefore susceptible to aperture effects because differing parts or fractions of the galaxies are recorded for each source , thus making each observation dependent on the size and distance of the galaxy , as well as the positioning of the fibre @xcite . conversely , ifs can spatially resolve each galaxy observed , thus assigning individual spectra at many locations across the galaxy . here we utilise data from the sami galaxy survey , an ifs survey with the aim to observe 3400 galaxies over a broad range of environments and stellar masses . we use the sfrs measured in sami in order to provide a tool for estimating @xmath1 . the basis of our @xmath1 reconstruction method is a recent star formation relation developed in the multi - freefall framework of turbulent gas @xcite . there have been many ongoing efforts to find an intrinsic relation between the amount of gas and the rate at which stars form in a molecular cloud . initiated by @xcite , hereafter k98 , @xmath3 correlates with @xmath1 @xcite , which can be approximated by an empirical power law with exponent @xmath9 , @xmath10 for a sample of low - redshift disc and starburst galaxies , k98 found an exponent of @xmath11 . however , significant scatter and discrepancies between different sets of data exist within this framework , commonly referred to as the kennicutt - schmidt relation . these discrepancies suggest that @xmath3 does not only depend on @xmath1 , but also on factors such as the turbulence and the freefall time of the dense gas on small scales . motivated by the fact that dense gas forms stars at a higher rate , a new star formation correlator was derived in @xcite , hereafter sfk15 . this descriptor , denoted by @xmath12 and called the ` maximum or multi - freefall gas consumption rate ' ( mgcr ) , is dependent on the probability density function ( pdf ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) of molecular gas , @xmath13 where @xmath4 is the mach number of the turbulence , @xmath14 is the turbulence driving parameter @xcite and @xmath15 is the ratio of thermal to magnetic pressure @xcite in the molecular gas . the sfk15 model for @xmath3 given by eq . ( [ eq : newsflaw ] ) is built upon foundational concepts laid out by @xcite , hereafter kdm12 , which had parameterised @xmath3 by the ratio between @xmath1 and the average ( single ) freefall time @xmath16 , a correlator hereon denoted by @xmath17 @xcite . our new correlator instead uses the concept of a multi - freefall time , which was pioneered by @xcite , tested with numerical simulations in @xcite , and used in sfk15 as a stepping stone to expand upon the kdm12 model . sfk15 found that @xmath3 is equal to @xmath18 of the mgcr by placing observations of milky way clouds and the small magellanic cloud ( smc ) in the k98 , kdm12 and sfk15 frameworks , confirming the measured low efficiency of star formation @xcite . statistical tests in sfk15 showed that a significantly better correlation between @xmath3 and @xmath12 was achieved than that which could be attained between either the @xmath1 or @xmath17 parameterisations of the previous star formation relations by k98 and kdm12 , respectively . the scatter in the sfk15 relation was found to be a factor of 34 lower than in the k98 and kdm12 relations , suggesting that it provides a better physical model for @xmath3 compared to the empirical relation by k98 and compared to the single - freefall relation by kdm12 . the aim of the current work is to formulate a method to predict the distribution of @xmath1 by inverting eq . ( [ eq : newsflaw ] ) and using optical observations , which will be plentiful in the coming few years . here we use the h@xmath2 luminosities and velocity dispersions provided by the sami galaxy survey to estimate @xmath1 from measurements of @xmath3 and @xmath4 . in sec . [ sec : obs ] we describe the observations of our sami galaxy sample . [ sec : theory ] introduces our new method to derive @xmath1 by inverting the sfk15 relation . in sec . [ sec : results ] we present our results and compare purely star - forming with composite / agn / shock galaxies in our sample . in sec . [ sec : comp ] we compare our own and other observations and predictions to previous star formation relations within the kennicutt - schmidt framework . in sec . [ sec : k98vssfk15 ] we demonstrate that our new method for predicting @xmath1 is superior to inverting the k98 relation . our conclusions are summarised in sec . [ sec : conc ] . the new data products for each sami galaxy in our sample derived here ( average turbulent mach number , cold gas density , freefall time , etc . , and finally @xmath1 ) are listed in tab . [ tab : galaxies ] in appendix [ app : online ] and are available for download in the online version of the journal or by contacting the authors . we selected a sample of 260 galaxies from the sami galaxy survey internal data release version 0.9 . the sydney - aao multi - object integral field spectrograph ( sami ; * ? ? ? * ) is a front - end fibre feed system for the aaomega spectrograph @xcite , consisting of 13 bundles of 61 fibres each ( hexabundles ; * ? ? ? * ; * ? ? ? * ) that can be deployed over a 1 degree diameter field of view . sami therefore enables simultaneous spatially - resolved spectroscopy of twelve galaxies and one calibration star with a 15 diameter field - of - view on each object . the aaomega spectrograph can be configured to provide different resolutions and wavelength ranges ; the sami galaxy survey employs the 570v grating to obtain a resolution of @xmath19 ( @xmath20 ) at and the 1000r grating to obtain @xmath21 ( @xmath22 ) at . sami datacubes are reduced and re - gridded to a spatial scale of @xmath23 @xcite and the spatial resolution is about @xmath24 ( green et al . , in prep ) . the sami galaxy survey plans to include more than 3000 galaxies at redshift @xmath25 covering a wide range of stellar masses and environments . the sample is drawn from gama @xcite with additional entries from eight nearby clusters to cover denser environments ( * ? ? ? * owers et al . , in prep ) . reduced datacubes and a variety of emission line based higher - level data products are included in the first public data release ( * ? ? ? * green et al . , in prep ) . the emission lines of sami galaxies have been analysed using the spectral fitting pipeline lzifu @xcite to extract emission line fluxes and kinematics for each spectrum . the spectrum associated with each spectral / spatial pixel ( ` spaxel ' ) is first fit with a stellar template using the ` penalized pixel - fitting ' ( ppxf ) routine @xcite before fitting up to three gaussian line profiles to each of eleven strong emission lines simultaneously . for this paper , we choose to use the single gaussian fits , and make use of the emission line flux maps , gas velocity maps , and gas velocity dispersion maps below . also available in the sami galaxy survey database are maps of sfr and @xmath3 ( in units of @xmath26 ) . these maps are made using extinction - corrected h@xmath2 fluxes converted to sfrs following the relation derived in @xcite . the sfr maps are fully described in medling et al . ( in prep ) . from the pool of sami galaxies , we select a sub - sample of galaxies according to the criteria described below . we only consider spaxels with a sufficiently high signal - to - noise ( s / n ) ratio . the s / n was defined to be the ratio of the total emission line flux to the statistical one - sigma error in the line flux . this error was inferred using the levenberg - marquardt technique of chi - squared minimisation @xcite . in the following , we list the selection criteria : 1 . source extractor ( sextractor ) ellipticity values are available . these values were obtained from the gama database @xcite . we require the ellipticity for each galaxy to estimate the physical volume of gas within each spaxel ( explained in detail in sec . [ sec : tff ] below ) . 2 . the s / n ratio must be @xmath27 in the h@xmath2 , h@xmath15 , [ n`ii ` ] , [ s`ii ` ] , [ o`i ` ] and [ o`iii ` ] emission lines . this allows reliable classification of the emission mechanism . however , in order to measure velocity dispersions down to about @xmath28 , we require and impose an s / n ratio of @xmath29 in the measured velocity dispersion ( explained in detail in sec . [ subsec : veldisp ] below ) . we also require that beam - smearing ( see sec . [ subsec : veldisp ] ) did not have a significant effect on the measured velocity dispersion . 3 . after removing spaxels that have low s / n and/or are affected by beam - smearing , the galaxy must have more than ten star - forming spaxels remaining . the star - forming spaxels were filtered using the optical classification criteria given in @xcite , an example of which is shown in fig . [ fig : bpt ] . this classification scheme uses optical emission line ratios ( bpt / vo diagrams ; * ? ? * ; * ? ? ? * ) , in order to distinguish between star - forming galaxies and galaxies that are dominated by an active galactic nucleus ( agn ) or by shocks . the h@xmath2-to - sfr conversion factor used in this work is only valid for star - forming regions , because agn / shock - dominated spaxels are contaminated with emission from agn / shock regions @xcite . emission line fluxes of each spaxel were corrected for extinction using the balmer decrement and the @xcite reddening curve . standard extinction for the diffuse ism was assumed , with an @xmath30 value of 3.1 being utilised throughout the analysis @xcite . each galaxy was classified as either a ` star - forming ' or ` composite / agn / shock ' galaxy . to be classified as star - forming , the galaxy had to have at least @xmath31 of all valid spaxels lying below and to the left - hand side of the @xcite classification line in the [ o`iii`]/h@xmath15 versus [ n`ii`]/h@xmath2 diagram , and below and to the left - hand side of the @xcite line in the [ s`ii`]/h@xmath2 and [ o`i`]/h@xmath2 diagrams , as described in @xcite ( see fig . [ fig : bpt ] ) . a galaxy was classified as composite / agn / shock , if at least @xmath32 of all valid spaxels lie above the @xcite classification line on the [ o`iii`]/h@xmath15 versus [ n`ii`]/h@xmath2 diagram and above the @xcite classification line on the [ s`ii`]/h@xmath2 and [ o`i`]/h@xmath2 diagnostic diagrams . thus , composite / agn / shock galaxies may include composite , agn , or shock @xcite galaxies according to the classification in @xcite . these classifications resulted in a sample of 219 star - forming and 41 composite / agn / shock classified galaxies . here we exploit the spatially resolved sami h@xmath2 flux and @xmath3 maps in combination with the h@xmath2 velocity dispersion maps to derive predictions of @xmath1 across each galaxy in our sample . examples of @xmath3 maps are shown in the left - hand panels of fig . [ fig : maps ] for the star - forming and composite / agn / shock classified galaxies from fig . [ fig : bpt ] . we further derive spaxel - averaged values of the physical parameters for each galaxy in our subsample . the sfk15 model relies on the availability of the sonic mach number @xmath33 , the ratio between the gas velocity dispersion and the local speed of sound . this was prompted by the findings of @xcite , showing that the observed scatter within the k98 and kdm12 relations may be primarily attributed to the physical variations in @xmath33 . as direct measurements of @xmath33 are unavailable for the sami sample , for every pixel we estimate a value using the method described in the following sections . the result of these procedures is shown in the middle panels of fig . [ fig : maps ] , for the two example galaxies , gama 492414 ( top ) and gama 69740 ( bottom ) . molecular clouds in galactic spiral arms exhibit a gas temperature range of , while those in the galactic centre can have temperatures up to @xmath34 @xcite . all @xmath0 gas should lie within this temperature range , otherwise it will cease to be molecular under typical conditions in the ism @xcite . the local sound speed ( @xmath35 ) of the gas is given by @xmath36^{1/2},\ ] ] with the boltzmann constant @xmath37 , the mass of a hydrogen atom @xmath38 , and the mean particle weight @xmath39 . the latter is @xmath40 for molecular gas and @xmath41 for ionised gas , assuming standard cosmic abundances @xcite . therefore temperatures of @xmath42 and @xmath43 correspond to molecular sound speeds of @xmath44 and @xmath45 , respectively . we hence estimate the mach number of the gas in each spaxel by dividing the velocity dispersion by the molecular sound speed of @xmath46 , which is appropriate for the dense , cold star - forming phase of the ism in the temperature range . in order to apply our star formation relation , eq . ( [ eq : newsflaw ] ) , we need an estimate of the turbulent velocity dispersion of the molecular gas in order to construct the turbulent mach number . here we use the h@xmath2 velocity dispersion to approximate the velocity dispersion of the cold gas . the h@xmath2 velocity dispersion is similar ( to within a factor of ) to the molecular gas velocity dispersion , because of the coupling of turbulent gas flows between the hot , warm and cold phases of the ism . for instance , it has been found that for m33 , the second - most luminous spiral galaxy in our local group , the atomic h`i ` dispersions are a fair estimator of the co dispersions @xcite . in m33 , h@xmath2 velocities have been found to trace h`i ` velocities reasonably well @xcite . however , the h@xmath2 velocity dispersion is expected to be somewhat higher than the h@xmath47 velocity dispersion , because the ionised emission comes from h`ii ` regions close to massive stars , which directly contribute to driving turbulence . we thus expect the velocity dispersion in the direct vicinity of massive stars to be overestimated . in order to take this effect into account , we crudely approximate the h@xmath47 velocity dispersion with half the h@xmath2 velocity dispersion , @xmath48 . while this provides only a rough estimate of the molecular velocity dispersion ( trustworthy only to within a factor of 23 ) , we show below that the uncertainties that this introduces into our @xmath1 estimate are only of the order of @xmath49 . this is because of the relatively weak dependence of @xmath1 on @xmath4 , as we will derive in sec . [ sec : uncertainties ] below . to demonstrate this , we investigate a case below , where we assume that the molecular gas velocity dispersion is equal to the h@xmath2 velocity dispersion , @xmath50 , which yields @xmath51 lower @xmath1 . thus , even though our velocity dispersion estimate is uncertain by factors of , the final uncertainty in @xmath1 is @xmath52 . [ [ sn - requirements ] ] s / n requirements : + + + + + + + + + + + + + + + + + the sami / aaomega spectrograph setup has an instrumental velocity resolution of @xmath53 at the wavelength of h@xmath2 ( see sec . [ sec : sami ] ) . velocity dispersions below this resolution limit can still be reliably measured if the s / n in the observed ( instrument - convolved ) velocity dispersion is sufficiently high . in the following , we estimate the required s / n in order to reconstruct intrinsic velocity dispersions down to @xmath54 . we choose this cutoff of @xmath28 , because it is the sound speed of the ionised gas , eq . ( [ eq : cs ] ) with @xmath55 and @xmath41 , and thus represents a physical lower limit for @xmath56 . the intrinsic ( true ) velocity dispersion ( @xmath57 ) can be obtained by subtracting the instrumental velocity resolution ( @xmath58 ) from the observed ( instrument - convolved ) velocity dispersion ( @xmath59 ) in quadrature , with @xmath60 the same relation holds for the uncertainties ( noise ) in the velocity dispersion , @xmath61 assuming that the instrumental velocity resolution is fixed , we can use @xmath62 and simplify the last equation to @xmath63 dividing both sides by @xmath57 and substituting eq . ( [ eq : s_n_cuts ] ) yields @xmath64 since @xmath65 and @xmath66 are the observed ( instrument - convolved ) and intrinsic s / n ratios , respectively , we can estimate the required @xmath67 for the target intrinsic @xmath68 and the target intrinsic velocity dispersion that we want to resolve , @xmath54 , by evaluating @xmath69 = 34.\end{aligned}\ ] ] thus , for spaxels with observed ( instrument - convolved ) velocity dispersion s / n ratios greater or equal to 34 , we can reliably reconstruct the intrinsic ( instrument - corrected ) velocity dispersion down to @xmath28 , with an intrinsic s / n ratio of at least 5 . we note that the sami database provides the instrument - subtracted velocity dispersion @xmath70 ( `` vdisp ` ' ) and its error @xmath71 ( `` vdisp_err ` ' ) based on the lzifu fits @xcite . thus , in order to apply the s / n cut of 34 derived in eq . ( [ eq : sncutobs ] ) , we first reconstruct @xmath72 and its error @xmath73 , using error propagation . this criterion is functionally equivalent to setting a s / n cut on the instrument - subtracted velocity dispersion , @xmath74 after applying our s / n cuts of 34 to the observed ( instrument - convolved ) velocity dispersion , any spaxels with velocity dispersions less than @xmath28 are disregarded . we note that this final cut only removes 1% of the spaxels with @xmath75 . [ [ beam - smearing ] ] beam smearing : + + + + + + + + + + + + + + we also have to account for ` beam smearing ' , a phenomenon that occurs because of the limitation in spatial resolution of the instrument . beam smearing occurs for a physical velocity field that changes on spatial scales smaller than the spatial resolution of the observation . if there is a steep velocity gradient across neighbouring pixels , such as near the centre of a galaxy , beam smearing leads to an artificial increase in the measured velocity dispersion at such spatial locations . to account for beam smearing , we follow the method in @xcite and estimate the local velocity gradient @xmath76 for a given spaxel with coordinate indices @xmath77 as the magnitude of the vector sum of the difference in the velocities in the adjacent pixels , @xmath78 ^ 2 + [ v(i , j+1)-v(i , j-1)]^2}. \end{aligned}\ ] ] note that the differencing to compute @xmath76 occurs over a linear scale of three sami pixels along @xmath79 and @xmath80 and thus covers roughly the spatial resolution of the seeing - limited sami observations with @xmath81 ( see sec . [ sec : obs ] ) . if a pixel has a neighbour that is undefined ( e.g. , because of low s / n ) , the gradient in that direction is not taken into account . as our standard criterion to account for beam smearing , we cut any pixels in which the velocity dispersion is less than twice that of the velocity gradient ( @xmath82 ) and disregard such pixels in further analyses , leaving only spaxels that are largely unaffected by beam smearing . in addition to our fiducial beam - smearing criterion ( @xmath82 ) , we test a case with a relaxed beam - smearing cut of @xmath83 , and find nearly identical results ( see tab . [ tab : sum ] below ) . we note that our standard beam - smearing cut with @xmath82 tends to remove spaxels near the centre of some of the galaxies ( see e.g. , fig . [ fig : maps ] ) . however , using the relaxed beam - smearing cut with @xmath83 yields global ( galaxy - averaged ) mach numbers and global @xmath1 estimates that agree to within 4% with our standard beam - smearing cut ( see tab . [ tab : sum ] ) , demonstrating that our results are largely unaffected by beam smearing . [ [ turbulent - velocity - dispersion - versus - systematic - motions ] ] turbulent velocity dispersion versus systematic motions : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + beam - smearing is the result of un - resolved velocity gradients in the plane - of - the - sky . however , systematic velocity gradients ( such as resulting from rotation or large - scale shear ) along the line - of - sight ( los ) also increase the velocity dispersion ( even for arbitrarily high spatial resolution ) by los - blending . these large - scale systematic motions do not represent turbulent gas flows ( see e.g. , the recent study of turbulent motions in the galactic - centre cloud brick , which is subject to large - scale shear * ? ? ? * ) . as we have not subtracted or accounted for these factors , our values of the turbulent velocity dispersion may be overestimated . in summary , we emphasise that the turbulent velocity dispersion has large uncertainties and is only accurate to within a factor of @xmath84@xmath85 . however , the uncertainties that this introduces into our final product ( @xmath1 ) are @xmath52 , because of the relatively weak dependence of @xmath1 on @xmath4 ( derived in detail in sec . [ sec : uncertainties ] below ) . to find the mgcr @xmath12 , we divide @xmath3 ( left - hand panels of fig . [ fig : maps ] ) by the sfr efficiency of 0.45@xmath86 found in sfk15 . that is , we invert eq . ( [ eq : newsflaw ] ) , @xmath87 = \frac{{\sigma_\mathrm{sfr}}}{0.0045 } .\end{aligned}\ ] ] in order to find the ratio between the gas column density and the freefall time at the average gas density , @xmath17 , we take the mach number calculated in sec . [ sec : mach ] and convert @xmath12 to @xmath17 , @xmath88 = \frac{{({\sigma_\mathrm{gas}}/t)_\mathrm{multi - ff}}}{\left(1 + b^2 \mathcal{m}^2 \frac{\beta}{\beta+1}\right)^{3/8}}.\end{aligned}\ ] ] in the following , we will assume a fixed turbulence driving parameter @xmath89 , representing a natural mixture @xcite , and assume an absence of magnetic fields such that @xmath90 . although both of these are strong assumptions , we emphasise that , in the absence of constraints on @xmath14 or @xmath15 in these galaxies , we have to assume fixed , typical values for them and allow that these assumptions contribute to the uncertainties of the @xmath1 estimation . however , if these parameters will be measured in the future , they can be used in eqs . ( [ eq : newsflaw ] ) and ( [ eq : gastff ] ) to obtain a more accurate prediction of @xmath1 . for simplicity , here we fix @xmath14 and @xmath15 , and only consider the remaining dependence on @xmath4 . now that we have @xmath92 from eq . ( [ eq : gastff ] ) , we need an estimate of the average freefall time @xmath93 to obtain @xmath1 from @xmath17 . thus , we need an estimate of the local gas density @xmath91 , which requires some geometrical considerations and assumptions similar to the ones outlined in kdm12 . ) . in this diagram the green rectangle represents a section of a galaxy with scale height @xmath94 , which is inclined an angle @xmath95 away from the vertical . @xmath96 is the column length used to estimate the local gas density and freefall time . ] first , we make the assumption that the galaxy has a uniform gas disc geometry with a scale height of @xcite . however , depending on the viewing angle with respect to the orientation of the galactic disc in the plane of the sky , the los length through the gas may be greater than the scale height ( see fig . [ fig : geometry ] ) . this angle can be estimated from the observed ellipticity of the galaxy . to correct for the viewing angle , we obtain sextractor ellipticity values , @xmath97 , for each galaxy from the gama database , from which we obtain the inclination angle , @xmath95 , of the galaxy : @xmath98 the column length , @xmath96 , can then be inferred by dividing the scale height , @xmath94 , by the cosine of the inclination angle , as pictured in fig . [ fig : geometry ] , @xmath99 = \frac{h}{\cos(\theta ) } = \frac{100}{\cos(\theta ) } = \frac{100}{1 - \varepsilon}. \label{eq : l}\end{aligned}\ ] ] we caution that the assumed cylindrical geometry is a drastic simplification , as there has been much evidence to suggest that the scale height of a galaxy follows a relation dependent upon its radius from the galactic centre @xcite . this may cause our predicted gas maps to underestimate the gas density towards the centre and overestimate the gas density towards the outskirts of the galaxy . however , the general shape of the predicted distribution of gas and especially the galaxy - averaged gas surface density should not be affected significantly by this geometrical simplification . a refinement in the geometry is relatively straightforward to implement , if one requires more accurate maps . we estimate that the relative uncertainties in @xmath96 may be up to @xmath100 . however , our final result ( @xmath1 ) , does not depend significantly on @xmath96 ( see detailed discussion in sec . [ sec : uncertainties ] ) . given the column length @xmath96 , we can write the gas density @xmath101 since we do not have @xmath1 because it is our final product , we now substitute a rearrangement of the definition of @xmath17 , @xmath102 as well as the definition of the freefall time in terms of @xmath91 , @xmath103 where @xmath104 is the gravitational constant . we combine the three previous equations and solve for the gas density , @xmath105 we substitute @xmath91 back into eq . ( [ eq : tff ] ) to obtain the freefall time @xmath106 for the average gas density @xmath91 . finally , we obtain our prediction for @xmath1 either by multiplying the freefall time from sec . [ sec : tff ] by @xmath17 calculated in sec . [ sec : gastff ] , i.e. , using eq . ( [ eq : gas_single_tff ] ) , or by multiplying the volume density @xmath91 from eq . ( [ eq : rho ] ) by the column length @xmath96 from eq . ( [ eq : l ] ) . in terms of the principle observables , @xmath3 and @xmath107 , as well as our assumptions for the parameters @xmath108 , @xmath14 and @xmath15 , this corresponds to the final expression for @xmath1 given by @xmath109^{2/3}.\end{aligned}\ ] ] two examples of the spatially resolved maps of estimated @xmath1 based on the new method provided by eq . ( [ eq : gas ] ) are shown in the right - hand panels of fig . [ fig : maps ] . here we estimate the uncertainties in our @xmath1 prediction based on eq . ( [ eq : gas ] ) . we derive the uncertainties by error propagation of all variables in eq . ( [ eq : gas ] ) . first , we note that the dependence of @xmath1 on @xmath96 is weak ( @xmath110 ) and the dependence on @xmath4 is also relatively weak ( @xmath111 ) , which means that the uncertainties in @xmath96 and @xmath4 enter the final uncertainty in @xmath1 with a weight of 1/3 and 1/2 , respectively . the strongest dependence of @xmath1 is on the sfr , i.e. , @xmath112 , so the uncertainties in @xmath3 are weighted by 2/3 , and we thus expect these to dominate the final uncertainties . rigorously , the relative uncertainty @xmath113 from eq . ( [ eq : gas ] ) is given by @xmath114^{1/2},\end{aligned}\ ] ] where we approximated the denominator @xmath115 in eq . ( [ eq : gas ] ) as @xmath116 for the uncertainty propagation ( recall that we also assumed @xmath117 ) , because @xmath118 , based on our velocity dispersion cut and sound speed ( see sec . [ subsec : veldisp ] ) . with typical relative uncertainties of 70% in @xmath96 , @xmath100 in @xmath4 ( see sec . [ sec : mach ] ) and @xmath119 in @xmath3 ( based on our s / n cuts of 5 on the h@xmath2 flux ; see sec . [ sec : oursubsample ] ) , we find a relative uncertainty of @xmath120 , which is dominated by the uncertainty in @xmath3 . even if the uncertainties in both @xmath96 and @xmath4 were @xmath100 and @xmath121 respectively , we would still be able to estimate @xmath1 with an uncertainty of @xmath122 . in summary , despite the large uncertainties in @xmath4 and @xmath96 ( see sec . [ sec : mach ] and [ sec : tff ] ) , our final uncertainties in @xmath1 are less than a factor of 2 . our main objective is to estimate @xmath1 from @xmath3 and the turbulence properties ( @xmath4 ) in our sami galaxy sample . we do this by applying the new method introduced in the previous section ( sec . [ sec : theory ] ) , going step - by - step from @xmath3 to @xmath1 . [ fig : derivation ] shows each of the @xmath3 parameterisations explored in sfk15 , presented in the same order as the computations of our @xmath1 derivations ( sec . [ sec : theory ] ) . the framework of the first panel assumes a direct correlation between @xmath3 and @xmath12 . that is , it assumes the star formation relation of eq . ( [ eq : newsflaw ] ) to hold , thus by construction the sami data points in this framework lie along the same line . the data points from sfk15 which were used to obtain this relation are also shown . we note that in the sfk15 derivation of eq . ( [ eq : newsflaw ] ) , the k98 galaxies were omitted because they did not have @xmath12 values assigned to them due to their lack of @xmath4 measurements . they are thus similarly excluded in this panel . compared to the observational data published in sfk15 , we updated and corrected some of the previous data , and added new observations in figure [ fig : derivation ] . first , we replace the @xcite data for the smc by the most recent @xmath123 resolution data provided in @xcite ( j16 ) . we also add the large magellanic cloud ( lmc ) data from @xcite and assume that the smc and lmc data have mach numbers in the range , i.e. , we basically treat the mach number as unconstrained , i.e. , varying in a plausible range , but we currently do not have direct measurements of @xmath4 in the smc or lmc . resolution data from @xcite and @xcite are more consistent with velocity dispersions that correspond to for the smc and lmc . however , without a direct measurement of the velocity dispersion and gas temperature , the mach number remains rather unconstrained for the smc and lmc . ] second , we replace the global cmz data from @xcite by the local cmz cloud g0.253 + 0.016 ` brick ' @xcite for which significantly more information is available . we take the values of @xmath1 , @xmath4 , @xmath14 , and @xmath15 measured in @xcite ( f16 ) and use the sfr per freefall time estimate of 2% from @xcite to obtain @xmath3 for the ` brick ' . the other cloud data are identical to those published in kdm12 , f13 and sfk15 , which were taken from @xcite ( h10 ) , @xcite ( g11 ) , @xcite ( w10 ) , and @xcite ( h10 ) . however , we corrected the error bar on the l10 clouds , which showed the standard deviation instead of the standard deviation of the mean in sfk15 . we further propagated the uncertainties in @xmath3 , @xmath4 and @xmath16 between @xmath12 , @xmath17 and @xmath1 . finally , we note that the observational data included in fig . [ fig : derivation ] cover a wide range in spatial and spectral resolution ( for details we refer the reader to the source publications of these data ) , which allowed us to test the universality of the sfk15 relation . in the future , when turbulence estimates become available for high - redshift data , those need to be included as well , to revisit the question of universality of the star formation relation derived in sfk15 . the second panel of fig . [ fig : derivation ] depicts the kdm12 parameterisation , @xmath3 versus @xmath17 . the derivation of this value for the sami galaxies required inputs from both the h@xmath2 flux and velocity dispersion , with @xmath17 computed from eq . ( [ eq : gastff ] ) . in addition to the observational data shown in the left - hand panel , we added the individual k98 disc and starburst galaxies from kdm12 ( with corrections based on * ? ? ? * ; * ? ? ? * ) . the third panel of fig . [ fig : derivation ] shows the final product of our @xmath1 predictions ; the average gas column density estimate for each of the sami galaxies in our sample . these predictions span a range of . we note that the estimated @xmath1 values for the sami galaxies are close to the @xmath1 values of the k98 low - redshift disc galaxies . this is encouraging , because they are the most similar in type to our sample of sami galaxies . the offset in @xmath1 and @xmath3 by @xmath124 between the sami and k98 galaxies can be understood as a consequence of spatial resolution . in contrast to our spaxel - resolved analysis of the sami galaxies ( with spatial resolution of @xmath125 ; see sec . [ sec : sami ] ) , the k98 galaxies are unresolved , which reduces the inferred @xmath3 @xcite . the reason is that , although the total h@xmath2 flux ( @xmath126 ) remains similar even at lower resolutions , the area @xmath127 over which h@xmath2 is emitted tends to be overestimated and hence the @xmath3 tends to be underestimated ( @xmath128 ) for the global k98 data . similar holds for @xmath1 , because it depends on @xmath127 in the same way as @xmath3 , and indeed , we find that the resolved sami galaxies tend to lie at somewhat higher @xmath1 compared to the unresolved k98 disc galaxy sample . for sami galaxies classified as pure star - forming galaxies ( blue ) and composite / agn / shock galaxies ( orange ) . ] lcc physical parameter & average ( standard deviation ) & average ( standard deviation ) + & for star - forming galaxies & for composite / agn / shock galaxies + @xmath129 $ ] & @xmath130 & @xmath131 + + mach number ( @xmath33 ) & @xmath132 & @xmath133 + @xmath134 $ ] & @xmath135 & @xmath136 + + mach number ( @xmath33 ) & @xmath137 & @xmath138 + @xmath134 $ ] & @xmath139 & @xmath140 + + mach number ( @xmath33 ) & @xmath141 & @xmath142 + @xmath134 $ ] & @xmath143 & @xmath144 + [ tab : sum ] * notes . * note that the values in brackets denote the standard deviation ( galaxy - to - galaxy variations ) of each physical parameter ; not the uncertainty in the parameter . uncertainties are discussed in sec . [ sec : uncertainties ] . [ fig : derivation ] suggests that the distributions of @xmath1 and @xmath3 are similar between the star - forming and composite / agn / shock galaxies . to quantify any statistical differences in @xmath1 between these two sub - samples , we investigate the distribution functions of @xmath1 . [ fig : gas_histograms ] shows the histograms of @xmath1 . we see that @xmath1 is enhanced in composite / agn / shock galaxies compared to star - forming galaxies . this difference in @xmath1 between star - forming and composite / agn / shock type galaxies is primarily a consequence of the differences in @xmath3 , and secondarily a consequence of the differences in @xmath4 between the two classes , i.e. , the dependences of @xmath1 on @xmath3 and @xmath4 ( see eqs . [ eq : gas ] and [ eq : gas_uncertainty ] ) . other dependences are realtively insignificant , such as the dependence on the assumed scale height of the galaxies . in this context , we have checked that any differences in the ellipticity distributions between the star - forming and composite / agn / shock galaxies are statistically insignificant . the measured mean and standard deviation of @xmath3 , @xmath4 and @xmath1 in the star - forming and composite / agn / shock galaxy samples are listed in table [ tab : sum ] ( the full list of physical parameters derived for each galaxy is provided in table [ tab : galaxies ] ) . the sfr surface densities and mach numbers of the star - forming and composite / agn / shock sample are @xmath145 and @xmath146 , and @xmath147 and @xmath148 , respectively . the resulting average @xmath1 values are @xmath149 and @xmath150 for star - forming and composite / agn / shock galaxies , respectively . table [ tab : sum ] further shows that changing the beam - smearing cutoff from the fiducial @xmath82 to a less strict cutoff ( @xmath83 ) yields nearly identical results . finally , the last two rows of table [ tab : sum ] show that using the velocity dispersion of the ionized gas ( @xmath50 ) instead of the approximate velocity dispersion of the molecular gas ( @xmath48 ) , reduces the derived @xmath1 by @xmath151 . thus , even with the large uncertainties in @xmath152 and hence in @xmath4 ( see sec . [ sec : mach ] ) , our final estimates in @xmath1 can be considered accurate to within a factor 2 . ) as a function of the derived @xmath1 for star - forming ( blue ) and composite / agn / shock ( orange ) galaxies in our sami sample . the large open circles show the same data , but binned in steps of 0.1 dex in @xmath1 . the error bars show the standard deviation of the mach number in each bin . despite the significant variations in mach number per individual @xmath1 bin , composite / agn / shock galaxies tend to have on average @xmath153 higher @xmath4 than star - forming galaxies . ] here we investigate global differences in the gas kinematics between star - forming and composite / agn / shock galaxies in our sami sample . [ fig : mach_gas ] shows our measurements of the mach number ( c.f . [ sec : mach ] ) as a function of derived @xmath1 for the two galaxy classes . we see that overall and also for fixed @xmath1 , composite / agn / shock galaxies have higher mach number by a factor of @xmath154 compared to star - forming galaxies . this may be a consequence of agn and/or shocks raising the velocity dispersion over turbulence driven by pure star - formation feedback . [ fig : mach_gas ] further reveals a significant scatter in mach number for fixed @xmath1 , which is somewhat more pronounced in the case of composite / agn / shock galaxies compared to purely star - forming ones . this may indicate that different driving sources of the turbulence act together and possibly dominate at different times in different galaxies . such driving sources can be divided into two main categories : i ) stellar feedback ( such as supernova explosions , stellar jets , and/or radiation pressure ) and ii ) galaxy dynamics ( such as galactic shear , magneto - rotational instability , gravitational instabilities , and/or accretion onto the galaxy ) @xcite . our results here suggest that agn feedback may be another important , potentially highly variable source of the turbulent gas velocity dispersion in galaxies . many studies in the literature have attempted to measure the correlation between @xmath3 and @xmath1 within different sets of data . however , there is no clear consensus on the coefficients and scaling exponents , due to the intrinsic scatter in the kennicutt - schmidt relation ( kdm12 , * ? ? ? * sfk15 ) . some studies find breaks in the power - law relations , which can be interpreted as thresholds @xcite , while other studies do not find evidence for such thresholds @xcite . here we explore how our @xmath1 predictions for the sami galaxies compare to the relations published in the literature . in fig . [ fig : gas_comp ] , we show an enhanced version of the right - hand panel of fig . [ fig : derivation ] , in order to compare our @xmath1 estimates to previously derived star formation relations in the @xmath3 versus @xmath1 framework . the relations we investigate are described in k98 , @xcite ( b08 ) , @xcite ( w10 ) , @xcite ( h10 ) and @xcite ( b11 ) . these relations are shown as lines in fig . [ fig : gas_comp ] . we see that different sets of data follow different relations , which often show significant deviations from one another . the sami galaxies have higher @xmath3 than described by the k98 , b08 , h10 , or b11 relations , but lower @xmath3 than described by the w10 relation . a power - law fit to all the sami galaxies yields a power - law exponent of @xmath155 instead of @xmath156 ( k98 ) ; see eq . ( [ eq : k98 ] ) . in summary , we find that none of the previously proposed scaling relations of @xmath3 as a function of @xmath1 describes the entirety of the data well . the reason for this is that the sfr ( @xmath3 ) depends on more than just gas density ( @xmath1 ) . instead , star formation also strongly depends on the turbulence of the gas ( mach number and driving mode ) , the magnetic field , and on the virial parameter ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * sfk15 ) . a complete understanding and prediction of star formation requires taking into account the dependences on these variables , in addition to gas ( surface ) density . here we compare the prediction of @xmath1 based on inverting the k98 relation with the @xmath1 prediction based on the sfk15 framework developed here . first , we note that a popular way of obtaining @xmath1 estimates in the absence of a direct measurement of it , is to invert the k98 relation , i.e. , to invert eq . ( [ eq : k98 ] ) , which yields @xmath157 = \left(\frac{{\sigma_\mathrm{sfr}}\,\left[{\mathrm{{\mathrm{m_{\odot}}}\,{\mathrm{yr}}^{-1}\,{\mathrm{kpc}}^{-2}}}\right]}{a}\right)^{1/n},\ ] ] with @xmath158 and @xmath11 ( k98 ) . here we derived an alternative way to estimate @xmath1 from @xmath3 , which is given by eq . ( [ eq : gas ] ) , and contains additional dependences on the geometry ( @xmath96 ) , turbulence ( @xmath4 and @xmath14 ) , and on the magnetic field ( @xmath15 ) . for the @xmath1 estimates of the sami galaxies here , we fixed @xmath96 , @xmath14 and @xmath15 for simplicity , and only included the mach number based on the measured velocity dispersion as an additional parameter to @xmath3 ( compared to the k98 relation , which depends on @xmath3 only ) . we now want to see how the @xmath1 estimates based on our new relation ( eq . [ eq : gas ] ) compare to inverting the k98 relation ( eq . [ eq : k98inverse ] ) . [ fig : sf_law_comp ] shows the direct comparison of the two as a function of @xmath3 . we plot the logarithmic difference of @xmath1(predicted ) and @xmath1(measured ) on the ordinate of fig . [ fig : sf_law_comp ] for @xmath1(predicted ) based on k98 ( eq . [ eq : k98inverse ] ) in grey and @xmath1(predicted ) based on eq . ( [ eq : gas ] ) in colour . in addition to the observational data already shown in figs . [ fig : derivation ] and [ fig : gas_comp ] , we add direct estimates of @xmath1 for a subset of 56 star - forming sami galaxies for which _ herschel _ @xmath159 dust measurements were available , using the methods in @xcite . a detailed description of how the dust emission was converted to @xmath1 is provided in appendix [ app : herschel_dust_to_gas ] . in fig . [ fig : sf_law_comp ] we see that our new method of estimating @xmath1 from @xmath3 given by eq . ( [ eq : gas ] ) is significantly better than simply inverting the k98 relation , eq . ( [ eq : k98inverse ] ) . we find that our new relation provides @xmath1 estimates with an average deviation of 0.12 dex ( 32% ) , while inverting the k98 relation yields an average deviation from the true ( measured ) @xmath1 by 0.42 dex ( 160% ) . this shows that our method provides a significantly better @xmath1 prediction from @xmath3 than inverting the k98 relation . our improved @xmath1 estimate comes at the cost of requiring an estimate of the mach number ( velocity dispersion ) as an additional parameter for the reconstruction ( prediction ) of @xmath1 . however , if @xmath3 is obtained from h@xmath2 ( as for the sami galaxies analysed here ) , then we have shown that the velocity dispersion of h@xmath2 can be used to estimate the mach number ( sec . [ sec : mach ] ) . even better @xmath1 predictions based on eq . ( [ eq : gas ] ) are expected if the exact scale height @xmath94 , the turbulence driving parameter @xmath14 , and the magnetic field plasma @xmath15 are available from future observations and/or by combining different observational datasets . we presented a new method to estimate the molecular gas column density ( @xmath1 ) of a galaxy using only optical ifs data , by inverting the star formation relation derived in sfk15 . our method utilises observed values of @xmath3 and velocity dispersion ( here from h@xmath2 ) as inputs and returns an estimate of the molecular @xmath1 . the derivation of our method is explained in detail in sec . [ sec : theory ] , with the final result given by eq . ( [ eq : gas ] ) . we apply our new method to estimate @xmath1 for star - forming and composite / agn / shock galaxies classified and observed in the sami galaxy survey . our main findings from this study are the following : * from the range in and mach number measured for the sami galaxies , we predict in the star - forming regions of our sami galaxy sample , consisting of 260 galaxies in total . the predicted values of @xmath1 are similar to those of unresolved low - redshift disc galaxies observed in k98 . while the k98 galaxies required co detections , here we estimate @xmath1 solely based on h@xmath2 emission lines . * we classify each galaxy in our sample as star - forming or composite / agn / shock . based on the sample - averaged @xmath145 and @xmath146 , and @xmath147 and @xmath148 for star - forming and composite / agn / shock galaxies , respectively , we estimate @xmath160 and @xmath150 , respectively ( see table [ tab : sum ] ) . we therefore find that on average , the composite / agn / shock galaxies have enhanced @xmath3 , @xmath4 , and @xmath1 by factors of @xmath5 , @xmath6 , and @xmath7 , respectively , compared to the star - forming sami galaxies ( see table [ tab : sum ] ; for each individual sami galaxy , see table [ tab : galaxies ] ) . * we discussed methods to account for finite spectral resolution and beam - smearing in sec . [ subsec : veldisp ] . while the uncertainties are large in the velocity dispersion used to estimate the turbulent mach number of the molecular gas ( sec . [ sec : mach ] ) , we show that the final estimate of @xmath1 is accurate to within a factor of 2 ( see sec . [ sec : uncertainties ] ) . * we compare our new method of estimating @xmath1 from @xmath3 with a simple inversion of the k98 relation ( fig . [ fig : sf_law_comp ] ) . we find that our new method yields a significantly better estimate of @xmath1 than inverting the k98 relation , with average deviations from the intrinsic @xmath1 by 32% for our new method , compared to average deviations of 160% from inverting the k98 relation . we thank mark krumholz and the anonymous referee for their useful comments , which helped to improve this work . cf acknowledges funding provided by the australian research council s ( arc ) discovery projects ( grants dp150104329 and dp170100603 ) . dms is supported by an australian government s new colombo plan scholarship . support for amm is provided by nasa through hubble fellowship grant # hst - hf2 - 51377 awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas5 - 26555 . bag gratefully acknowledges the support of the arc as the recipient of a future fellowship ( ft140101202 ) . ljk gratefully acknowledges the support of an arc laureate fellowship . sb acknowledges the funding support from the arc through a future fellowship ( ft140101166 ) . smc acknowledges the support of an arc future fellowship ( ft100100457 ) . ns acknowledges support of a university of sydney postdoctoral research fellowship . the sami galaxy survey is based on observations made at the anglo - australian telescope . the sydney - aao multi - object integral field spectrograph ( sami ) was developed jointly by the university of sydney and the australian astronomical observatory . the sami input catalogue is based on data taken from the sloan digital sky survey , the gama survey and the vst atlas survey . the sami galaxy survey is funded by the arc centre of excellence for all - sky astrophysics ( caastro ) , through project number ce110001020 , and other participating institutions . the sami galaxy survey website is http://sami - survey.org/. table [ tab : galaxies ] lists the derived physical parameters of all sami galaxies analysed here . listed are the first ten galaxies in each of our two galaxy classes ( star - forming and composite / agn / shock ) . the complete table is available in the online version of the journal or upon request . lccccccccccc gama & redshift & @xmath97 & @xmath161 & @xmath162 & @xmath3 & @xmath33 & @xmath91 & @xmath163 & @xmath12 & @xmath17 & @xmath1 + i d & & & & [ @xmath164 & @xmath165}$ ] & & @xmath166 $ ] & [ @xmath167 & @xmath165}$ ] & @xmath165}$ ] & @xmath168}$ ] + + @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 & @xmath179 & @xmath180 + @xmath181 & @xmath182 & @xmath183 & @xmath184 & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 & @xmath192 + @xmath193 & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 & @xmath203 & @xmath204 + @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath209 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath214 & @xmath215 & @xmath216 + @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 & @xmath176 & @xmath177 & @xmath224 & @xmath225 & @xmath226 + @xmath227 & @xmath228 & @xmath171 & @xmath229 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 + @xmath238 & @xmath239 & @xmath195 & @xmath240 & @xmath241 & @xmath198 & @xmath242 & @xmath200 & @xmath201 & @xmath202 & @xmath243 & @xmath244 + @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & @xmath231 & @xmath250 & @xmath251 & @xmath252 & @xmath235 & @xmath253 & @xmath254 + @xmath255 & @xmath246 & @xmath247 & @xmath256 & @xmath249 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 + @xmath264 & @xmath265 & @xmath266 & @xmath248 & @xmath209 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 + & & & & & & & & & & & + + @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 + @xmath286 & @xmath287 & @xmath288 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath282 & @xmath294 & @xmath295 & @xmath244 + @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath173 & @xmath300 & @xmath301 & @xmath302 & @xmath201 & @xmath303 & @xmath304 & @xmath305 + @xmath306 & @xmath170 & @xmath307 & @xmath308 & @xmath173 & @xmath309 & @xmath199 & @xmath176 & @xmath177 & @xmath310 & @xmath311 & @xmath312 + @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath222 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath323 + @xmath324 & @xmath314 & @xmath325 & @xmath326 & @xmath317 & @xmath327 & @xmath328 & @xmath329 & @xmath330 & @xmath310 & @xmath295 & @xmath331 + @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 & @xmath343 + @xmath344 & @xmath345 & @xmath288 & @xmath196 & @xmath346 & @xmath347 & @xmath301 & @xmath348 & @xmath310 & @xmath271 & @xmath349 & @xmath350 + @xmath351 & @xmath352 & @xmath353 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & @xmath358 & @xmath202 & @xmath359 & @xmath360 & @xmath361 + @xmath362 & @xmath352 & @xmath363 & @xmath354 & @xmath364 & @xmath365 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & @xmath179 & @xmath370 + & & & & & & & & & & & + * notes . * all galaxy parameters are based on a straight average over all valid spaxels and the uncertainties were propagated based on the average values . column 1 : gama i d . column 2 : redshift . column 3 : ellipticity . column 4 : number of valid spaxels for gas column density estimate . column 5 : linear size of each spaxel . column 6 : spaxel - averaged @xmath371 . column 7 : spaxel - averaged turbulent mach number . column 8 : spaxel - averaged gas volume density , estimated based on eq . ( [ eq : rho ] ) . column 9 : spaxel - averaged freefall time based on eq . ( [ eq : tff ] ) . column 10 : spaxel - averaged multi - freefall gas consumption rate , @xmath12 ; eq . ( [ eq : gastmff ] ) . column 11 : spaxel - averaged single - freefall gas consumption rate , @xmath17 ; eq . ( [ eq : gastff ] ) . column 12 : spaxel - averaged molecular gas surface density @xmath1 , estimated with eq . ( [ eq : gas ] ) . [ tab : galaxies ] to provide an independent measure of @xmath1 for our sami galaxy sample , we used the empirical relation determined by @xcite , correlating the total gas mass of galaxies with their sub - mm dust luminosities . using a sample of nearby galaxies , @xcite found that the total ( atomic@xmath372molecular ) gas mass of galaxies ( @xmath373 ) could be determined within 0.12 dex using the monochromatic @xmath159 luminosity ( @xmath374 ) , with @xmath375 to determine the sub - mm luminosity of the sami galaxies , we made use of the _ herschel_-atlas survey @xcite , a wide 550 square degrees infrared survey of the sky by the _ herschel _ space observatory , that covers the gama regions from which the sami galaxy survey sample arise . in particular , we cross - matched the 219 star - forming sami galaxies classified here against the single - entry source catalog from _ herschel_-atlas data release 1 @xcite . of the 219 sami star - forming galaxies , 128 have _ herschel _ detections . of these , 56 have significant detections ( signal - to - noise @xmath376 ) in the spire @xmath159 band . to convert the total gas mass to a gas surface density we required a surface area over which the infrared flux is emitted . given the large beam size of the spire @xmath159 observations , the sami galaxies are unresolved . however , as can be seen in the radial profiles of the nearby galaxy sample used in @xcite ( in particular their figure 7 and online figures ) , the highest surface brightness regions occur within half an optical radius ( @xmath377 or @xmath378 based on * ? ? ? * ) , with most of the infrared luminosity ( and molecular gas mass ) occurring within this radius . @xcite further find that at this radius , the atomic and molecular gas surface densities are about the same ( the ratio of the total atomic and molecular gas masses inside @xmath379 is also about unity ) . based on those findings , we approximated the molecular gas mass within @xmath379 with @xmath380 . therefore , we derive the molecular @xmath1 through @xmath381 where the effective radius @xmath382 , and ellipticity @xmath97 , of the sami galaxies are as derived in the gama survey @xcite . , l. j. , & dopita , m. a. 2003 , in revista mexicana de astronomia y astrofisica conference series , vol . 17 , revista mexicana de astronomia y astrofisica conference series , ed . v. avila - reese , c. firmani , c. s. frenk , & c. allen , 8384

gamma - ray bursts ( grbs ) are among the most violent occurrences in the universe . they are powerful explosions , visible to high redshift , and thought to be the signature of black hole birth . they are highly luminous events and provide excellent probes of the distant universe . grb research has greatly advanced over the past 10 years with the results from swift , fermi and an active follow - up community . in this review we survey the interplay between these recent observations and the theoretical models of the prompt grb emission and the subsequent afterglows . = 1

1301.0840

grbs are the most extreme explosive events in the universe . the initial prompt phase lasts typically less than 100 s and has an energy content of @xmath0 ergs , giving a luminosity that is a million times larger than the peak electromagnetic luminosity of a supernova . the grb name is a good one because their spectra peak in the gamma - ray band between @xmath1 kev and @xmath2 mev . the source of the energy powering the bursts is thought to be the gravitational collapse of matter to form a black hole or other compact object . grbs were discovered in the late 1960s by the vela satellites monitoring the nuclear test ban treaty between the us and the soviet union @xcite . it was slow progress for 20 years learning the origin of these brilliant flashes . gamma - ray instruments of the time had poor positioning capability so only wide - field , insensitive telescopes could follow - up the bursts to look for counterparts at other wavelengths . nothing associated with the grbs was seen in searches hours to days after their occurrence . in the 1990 s the burst and transient source experiment ( batse ) onboard the compton gamma ray observation ( cgro ) obtained the positions of @xmath3 grbs and showed that they were uniformly distributed on the sky @xcite , indicating either an extragalactic or galactic - halo origin . batse also found that grbs separate into two duration classes , short and long grbs , with dividing line at @xmath4 s @xcite . the detection of x - ray afterglows and more accurate localizations delivered by the bepposax mission @xcite enabled optical redshifts of grbs to be measured and their extragalactic origin to be confirmed . the long bursts were found to be associated with far - away galaxies at typical redshifts @xmath5 , implying energy releases in excess of @xmath6 ergs , after taking into account the jet beaming factor . the afterglow time decay often steepened after a day indicating a geometrical beaming of the radiation into a jet of opening angle @xmath7 @xcite . key open issues prior to the swift and fermi era included : ( 1 ) the origin of short grbs , ( 2 ) the nature of the high energy radiation from grbs , ( 3 ) the redshift distribution of bursts and their usage for early universe studies , and ( 4 ) the physics of the jetted outflows . swift and fermi , launched in 2004 and 2008 respectively , have opened a new era in grb research . swift is a nasa mission and fermi a nasa collaboration with the us department of energy . they are both major international partnerships and have different and complementary capabilities . swift has a wide - field imaging camera in the hard x - ray band that detects the bursts at a rate of @xmath8 per year , providing positions with arcminute accuracy . the spacecraft then autonomously and rapidly ( 100 s ) reorients itself for sensitive x - ray and uv / optical observations of the afterglow . fermi has two wide - field instruments . one detects bursts in the gamma - ray band at a rate of @xmath9 per year , providing spectroscopy and positions with 10-degree accuracy . the other observes bursts in the largely - unexplored high - energy gamma - ray band at a rate of @xmath10 per year . combined , the two missions are advancing our understanding of all aspects of grbs , including the origin of short bursts , the nature of bursts coming from the explosion of early stars in the universe and the physics of the fireball outflows that produce the gamma - ray emission . the swift mission @xcite has three instruments : the burst alert telescope ( bat ; @xcite ) , the x - ray telescope ( xrt ; @xcite ) and the uv optical telescope ( uvot ; @xcite ) . the bat detects bursts and locates them to @xmath4 arcminute accuracy . the position is then sent to the spacecraft to repoint the xrt and uvot at the event . positions are also rapidly sent to the ground so that ground telescopes can follow the afterglows . there are more than 50 such telescopes of all sizes that participate in these world - wide follow - up campaigns . measurements of the redshift and studies of host galaxies are typically done with large ground - based telescopes , which receive immediate alerts from the spacecraft when grbs are detected . swift has , by far , detected the largest number of well - localized bursts with afterglow observations and redshift determinations . as of 1 september 2012 , bat has detected 709 grbs ( annual average rate of @xmath8 per year ) . approximately @xmath11 of the bat - detected grbs have rapid repointings ( the remaining @xmath12 have spacecraft constraints that prevent rapid slewing ) . of those , virtually all long bursts observed promptly have detected x - ray afterglow . short bursts are more likely to have negligible x - ray afterglow , fading rapidly below the xrt sensitivity limit . the fraction of rapid - pointing grbs that have uvot detection is @xmath13 . combined with ground - based optical observations , about @xmath14 of swift grb have optical afterglow detection . there are 225 swift grbs with redshifts compared with 41 pre - swift . swift was specifically designed to investigate grb afterglows by filling the temporal gap between observations of the prompt emission and the afterglow @xcite . figure [ fig : swiftxray ] shows example x - ray afterglows from xrt from long and short grbs @xcite . the combined power of the bat and xrt has revealed that in long grbs the non - thermal prompt x - ray emission smoothly transitions into the decaying afterglow . often , a steep - to - shallow transition is found suggesting that prompt emission and the afterglow are distinct emission components @xcite . the early steep - decay phase seen in the majority of grbs is a surprise . the current best explanation is that we are seeing high - latitude emission due to termination of central engine activity @xcite . swift has discovered erratic flaring behavior ( figure [ fig : swiftxray ] ) , lasting long after the prompt phase in @xmath15 of x - ray afterglows . the most extreme examples are flares with integrated power similar to or exceeding the initial burst @xcite . the rapid rise and decay , multiple flares in the same burst , and cases of fluence comparable with the prompt emission , suggest that these flares are due to the central engine . long grbs ( lgrbs ) are associated with the brightest regions of galaxies where the most massive stars occur @xcite . lgrbs occur over a large redshift range from @xmath16 ( grb 980425 ) to @xmath17 ( grb 090423 & grb 090429b ) . with few exceptions , lgrbs that occur near enough for supernova detection have accompanying type ib or ic supernovae . the exceptions are cases of bursts that may be misclassified as long or that may have exceptionally weak supernovae . these facts support the growing evidence that long bursts are caused by `` collapsars '' where the central core of a massive star collapses to a compact object such as a black hole @xcite or possibly a magnetar @xcite . theories for the origin of lgrbs predate swift , but are supported by the new data . in them , a solar rest mass worth of gravitational energy is released in a very short time ( seconds or less ) in a small region of the order of tens of kilometers by a cataclysmic stellar event . the energy source is the collapse of the core of a massive star . only a small fraction of this energy is converted into electromagnetic radiation , through the dissipation of the kinetic energy of a collimated relativistic outflow , a fireball with bulk lorentz factors of @xmath18 , expanding out from the central engine powered by the gravitational accretion of surrounding matter into the collapsed core or black hole . mission & statistics the fermi instruments are the gamma - ray burst monitor ( gbm , @xcite ) and the large area telescope ( lat , @xcite ) . the gbm has scintillation detectors and covers the energy range from 8 kev to 40 mev . it measures spectra of grbs and determines their position to @xmath19 accuracy . the lat is a pair conversion telescope covering the energy range from 20 mev to 300 gev . it measures spectra of sources and positions them to an accuracy of @xmath20 . the gbm detects grbs at a rate of @xmath21 per year , of which on average @xmath12 are short bursts . the lat detects bursts at a rate of @xmath10 per year . lat bursts have shown two common and interesting features : ( 1 ) delayed emission compared to lower energy bands and ( 2 ) longer lasting prompt emission compared to lower every bands . the four brightest lat bursts are grb 080916c @xcite , grb 090510 @xcite , grb 090902b @xcite , and grb 090926a @xcite . they have yielded hundreds of @xmath22 mev photons each , and together with the lower energy gbm observations , have given unprecedented broad - band spectra . in grb 080916c , the gev emission appears only in a second pulse , delayed by @xmath23 s relative to the first pulse ( figure [ fig : lc080916c ] ) . such a delay is present also in short bursts , such as grb 090510 , where it is a fraction of a second . this soft - to - hard spectral evolution is clearly seen in all four of these bright lat bursts , and to various degrees a similar behavior is seen in other weaker lat bursts . in some bursts , such as grb 080916c and several others , the broad - band gamma - ray spectra consist of a simple band - type broken power - law function in all time bins . in grb 080916c the first pulse has a soft high energy index disappearing at gev energies , while the second and subsequent pulses have harder high energy indices reaching into the multi - gev range . in some other bursts , such as grb 090510 @xcite and grb 090902b @xcite , a second hard spectral component extending above 10 gev without any obvious break appears in addition to common band spectral component dominant in the lower 8 kev-10 mev band . an exciting discovery , unanticipated by results from the energetic gamma - ray experiment telescope ( egret ) instrument on cgro , was the detection of high - energy emission from two short bursts ( grb 081024b @xcite and grb 090510 @xcite ) . their general behavior ( including a gev delay ) is qualitatively similar to that of long bursts . the ratio of detection rates of short to long grbs by lat is @xmath24 ( 2 out of 27 ) , which is significantly smaller than the @xmath25 by gbm . while the statistics on short grbs are too small to draw firm conclusions , so far , the ratio of the lat fluence to the gbm fluence is @xmath26 for the short bursts as compared to @xmath27@xmath28 for the long bursts . it is also noteworthy that , for both long and short grbs , the @xmath29 mev emission lasts longer than the gbm emission in the @xmath30 mev range . the flux of the long - lived lat emission decays as a power law with time , which is more reminiscent of the smooth temporal decay of the afterglow x - ray and optical fluxes rather than the variable temporal structure in the prompt kev mev flux . this similarity in the smooth temporal evolution of the fluxes in different wave bands has been detected most clearly in grb 090510 @xcite although this burst also requires a separate prompt component to the lat emission @xcite . this short burst was at @xmath31 , and was jointly observed by the lat , gbm , bat , xrt and uvot . the lat detects only @xmath32 of the bursts triggered by the gbm which were in the common gbm - lat field of view . this may be related to the fact that the lat - detected grbs , both long and short , are generally among the highest fluence bursts , as well as being among the intrinsically most energetic grbs . for instance , grb 080916c was at @xmath33 and had an isotropic - equivalent energy of @xmath34 ergs in gamma rays , the largest ever measured from any burst @xcite . the long lat bursts grb 090902b @xcite at @xmath35 had @xmath36 ergs , while grb 090926a @xcite at @xmath37 had @xmath38 ergs . even the short burst grb 090510 at @xmath39 produced , within the first 2 s , an @xmath40 ergs @xcite . swift is detecting grbs at higher redshift than previous missions due to its higher sensitivity and rapid afterglow observations . the average redshift for the swift grbs is 2.3 compared to 1.2 for previous observations . although statistics are poor , the highest redshift grbs are seen to have high luminosity , resulting in fluxes well above the detection threshold . such bursts are also strong at other wavelengths . table 1 presents optical data for the highest redshift grbs observed to date . the optical brightness is given in magnitudes in @xmath41 , @xmath42 @xmath43 and @xmath44 wavelength bands . the bursts are 34 orders of magnitude brighter than typical galaxies at these redshift . to date , swift has seen seven grbs with @xmath45 . the highest @xmath46 grbs so far are @xmath47 ( grb 050904 @xcite ) , @xmath48 ( grb 080913 @xcite ) , @xmath49 ( grb 090423 @xcite ) , and @xmath50 ( grb 090429b @xcite ) . figure [ fig:090423 ] shows a spectrum of the current record holder with a spectroscopic redshift , grb 090423 @xcite . because grbs are so bright across the electromagnetic spectrum , they are excellent tools for studying the high redshift universe . the information that can be provided by grbs is complementary to what can be learned by studies of galaxies and quasars . for example , at @xmath51 it is expected that the bulk of the star formation activity is taking place in small galaxies below the detection limits of hst . the chemical evolution of the universe can be studied with grbs as illustrated in figure [ fig : metallicity ] @xcite . grbs provide data to higher redshift than active galaxies . the metallicity bias of grbs is currently contradictory . plots like figure [ fig : metallicity ] show grbs in higher metallicity star forming regions of galaxies . however , comparing star forming regions , grbs actually tend to favor ones with lower metallicity @xcite . another way that grbs are contributing to our understanding of the high - redshift universe is in the determination of the star formation history ( figure [ fig : starformation ] ) . lgrbs are the endpoints of the lives of massive stars and their rate is therefore approximately proportional to the star formation rate . they give information at high redshift where the rate is highly uncertain . there may be evolutionary biases , such as a dependence of lgrbs on the metallicity of host galaxies , so studies relating them to the star formation rate must include these factors into account @xcite . .optical observations of high redshift grbs [ cols="<,<,<,<",options="header " , ] at the time of swift s launch , the greatest mystery of grb astronomy was the nature of short - duration , hard - spectrum bursts ( sgrbs ) . although more than 50 long grbs had afterglow detections , no afterglow had been found for any short burst . in summer 2005 swift and hete-2 , precisely located three short bursts for which afterglow observations were obtained leading to a breakthrough in our understanding of short bursts @xcite . as of 2012 , bat has detected 65 sgrbs , @xmath11 of which have xrt detections , and 15 of which have redshifts . in contrast to long bursts , the evidence is that sgrbs typically originate in host galaxies with a wide range of star formation properties , including low formation rate . their host properties are substantially different than those of long bursts @xcite indicating a different origin . also , nearby sgrbs show no evidence for simultaneous supernovae ( @xcite ; and references therein ) , very different than long bursts . taken together , these results support the interpretation that sgrbs arise from an old populations of stars and are due to mergers of compact binaries ( i.e. , double neutron star or neutron star - black hole ) @xcite . it is possible that there is a subpopulation of high - redshift , high - luminosity short grbs with a different origin @xcite . measurements or constraining limits on beaming from light curve break searches have been hard to come by with the typically weak afterglow of sgrbs . with large uncertainties associated with small number statistics , the distribution of beaming angles for sgrbs appears to range from @xmath19 to @xmath52 @xcite , roughly consistent but perhaps somewhat larger than that of lgrbs . swift observations also revealed long ( @xmath1 s ) `` tails '' with softer spectra than the first episode following the prompt emission for about @xmath53 of short bursts @xcite . swift localization of a short grb helps narrow the search window for gravitational waves from that grb @xcite . detection of gravitational waves from a swift grb would lead to great scientific payoff for merger physics , progenitor types , and ns equations of state . here is a list of unusual transients detected by swift @xcite . _ short grb 050925 : _ this unusual burst triggered the bat with a single peaked outburst of duration @xmath54 ms @xcite . it occurred near the galactic plane , and nothing was seen in the uvot . however , the @xmath55-band extinction toward the source was @xmath56 mag . the xrt spectra and lightcurve show no significant x - ray emission in the field , suggesting that any x - ray counterpart to this burst was faint . markwardt et al . ( 2005 ) @xcite were able to fit a power law spectrum to the bat data , with a photon index @xmath57 ; they found a better fit could be obtained with a blackbody spectrum @xmath58 kev , a value consistent with small - flare events from sgrs . the low galactic latitude and soft spectrum indicate a possible galactic source or sgr . _ grbs 060505 & 060614 : _ grb 060505 and grb 060614 were nearby grbs ( @xmath59 and 0.125 , resp . ) with no coincident sn ( table 2 ) , to deep limits @xcite . grb 060614 was bright ( 15150 kev fluence of @xmath60 erg @xmath61 ) , and , with a @xmath62 of 102 s , seemed to be a secure long grb . host galaxies were found @xcite and deep searches were made for coincident sne . all other well - observed nearby grbs have had sne , but grb 060614 did not to limits @xmath63 times fainter than previous detections @xcite . grb 060614 shares some characteristics with sgrbs @xcite . the bat light curve shows an initial short hard flare lasting @xmath27 s , followed by an extended softer episode , @xmath1 s. the light curve is similar to some swift sgrbs and a subclass of batse sgrbs @xcite . grb 060614 also falls in the same region of the lag - luminosity diagram as sgrbs ( figure [ fig : lag ] ) . thus grb 060614 is problematic to classify . it is a lgrb by the traditional definition , but lacks an associated sn . it shares some similarities with sgrbs , but the soft episode is brighter , which would be difficult to account for in the ns - ns merger scenario . _ hostless grb 070125 : _ there is no obvious host galaxy for grb 070125 . deep ground - based imaging reveals no host to @xmath64 mag . cenko et al . ( 2008 ) @xcite present an analysis of spectroscopic data which reveals weak mg ii lines indicative of halo gas . in the field are two blue galaxies offset by @xmath65 kpc at @xmath66 . if there is an association with one of them , it would imply a velocity @xmath67 km s@xmath68 over a @xmath69 myr lifetime of the massive progenitor . the only known way of achieving this would have been a prior close interaction with a massive bh . however , this interpretation was muddied by chandra et al . ( 2008 ) @xcite , who inferred a dense environment , based on bright , self - absorbed radio afterglow . they proposed a scenario in which the high density material lies close to the explosion site , and the lower density material further away . they note grb 070125 was one of the brightest grbs ever detected , with an isotropic release of @xmath70 erg ( by comparison , @xmath71 erg ) . the prompt emission from grb 070125 was also seen by suzaku / wam @xcite . _ galactic grb 070610 : _ discovered initially as grb 070610 , this object , now dubbed swift j195509.6 + 261406 ( sw1955 + 26 ) , is thought to represent a member of a relatively new class of fast x - ray nova containing a bh . it had a duration of @xmath27 s , and also shows large variability . kasliwal et al . ( 2008 ) @xcite discuss several possibilities for this source , and propose an analogy with v4641 sgr , an unusual bh binary which had a major outburst in 1999 @xcite . v4641 sgr is a binary with a b9 iii star orbiting a @xmath72 bh @xcite and also exhibited strong and fast x - ray and optical variability . the analog is imperfect in that the normal star in sw1955 + 26 is a cool dwarf rather than a b9 giant , suggesting a physical origin for the bursting behavior in the accretion disk and/or jet rather than the mass donor star . a @xmath27 s burst is certainly distinct from the @xmath73 month long fast - rise exponential decay seen in systems like a0620- 00 in 1975 @xcite which are thought to be due to a large scale storage and dumping of material in an accretion disk @xcite , and may be more in line with either a disk - corona @xcite or disk - jet instability @xcite . rea et al . ( 2011 ) @xcite derive stringent upper limits on the quiescent x - ray emission from sw1955 + 26 using a @xmath74 ks chandra observation , and use this to argue against a magnetar interpretation . simon et al . ( 2012 ) @xcite present an analysis of optical emission during the 2007 outburst and find the optical emission manifests as spikes which decrease in duration as the burst progresses . they show that the emission can be explained by pure synchrotron emission . the main part of the optical outburst following the swift / bat trigger lasted @xmath75 d , with subsequent echo outbursts between 1 and 3 day , as in x - ray novae . the main part of the outburst has an exponential decay reminiscent of the 1975 outburst in a0620 - 00 , but with a much shorter duration . if the global accretion disk limit cycle @xcite the mechanism for this outburst , it suggests a much shorter orbital period than the 7.75 hr for a0620 - 00 @xcite . _ grb 101225 christmas burst : _ grb 101225 was quite unusual : it had a @xmath76 s and exhibited a curving decay when plotted in the traditional @xmath77-@xmath78 coordinates . the total bat fluence was @xmath79 erg @xmath61 . the xrt and uvot found a bright , long - lasting counterpart . ground based telescopes followed the event , mainly in @xmath41 and @xmath80 , and failed to detect any spectral features . at later times a color change from blue to red was seen ; hst observations at 20 d found a very red object with no apparent host . observations from the spanish gran telescopio canarias at 180 d detected grb 101225 at @xmath81 mag and @xmath82 @xcite . considered together , these characteristics are unique to this burst ( figure [ fig:101225 ] ) , and led campana et al . ( 2011 ) @xcite to propose that it was caused by a minor body like an asteroid or comet becoming disrupted and accreted by a ns . depending on its composition , the tidal disruption radius would be @xmath83@xmath84 km . campana et al . find an adequate fit to the light curve by positing a @xmath85 g asteroid with a periastron radius @xmath86 km . if half the asteroid mass is accreted they derive a total fluence @xmath87 erg @xmath61 and distance @xmath88 kpc . thne et al . ( 2011 ) @xcite offer a different explanation for grb 101225 the merger of a he star and a ns leading to a concomitant sn . they derive a pseudo - redshift @xmath89 by fitting the spectral - energy distribution and light curve of the optical emission with a grb - sn template . thus in their interpretation the event was much more distant and energetic . they argue for the presence of a faint , unresolved galaxy in deep optical observations , and fit the long - term light curve with a template of the broad - line type ic sn 1998bw associated with grb 980425 . if their distance is correct as well as their interpretation of a component emerging at 10 d as being a sn , then its absolute peak magnitude @xmath90 mag would make it the faintest sn associated with a long grb . the isotropic - equivalent energy release at @xmath91 would be @xmath92 erg which is typical of other long grbs but greater than most other low - redshift grbs associated with sne . band , red ) and 7700 angstroms ( @xmath80 band , orange ) @xcite . the inset ( with t in seconds ) shows the swift / xrt light curve . shaded regions highlight the periastron passages calculated using a tidal disruption model @xcite . , width=307 ] gamma - ray bursts are thought to be the endpoints of evolution of massive , @xmath93 , stars ( favorite for long grbs ) @xcite or neutron star - neutron star / black hole close binaries ( favorite for short grbs ) @xcite . in either case a black hole and a short - lived accretion disc are thought to be formed after the core collapse ( long grbs ) or binary merger ( short grbs ) . in some cases a highly magnetized neutron star , called a magnetar , may form as an intermediate state which eventually collapses to a black hole @xcite . gravitational binding energy of the system is released within a short time scale which eventually produces the grb phenomenon when some fraction of the gravitational energy converts to gamma rays . this is done via accretion of material / gas onto the black hole . the compact core of a few solar mass with the schwarzschild radius @xmath94 cm forms the central engine of a grb . infalling gas onto the newly - formed black hole is accreted with high efficiency . most of the gravitational energy , @xmath95 ergs , is radiated in @xmath10 mev thermal neutrinos as in core - collapse sne and a lesser amount in gravitational wave . only a small fraction , @xmath0 erg , is converted to a grb fireball , which eventually produces a highly - relativistic bi - polar jet and is responsible for the observed @xmath96 rays @xcite . as discussed previously , observations of supernovae associated with long grbs provide strong support for the core - collapse schenario for this class of grbs . the binary merger scenario of short grbs has not been confirmed , but remains the most popular model . the grb fireball is produced when large amount of energy is released within a short time scale . the isotropic - equivalent energy is @xmath97 , where @xmath98 is the jet solid angle . the average solid angle @xmath99 or the jet half - opening angle @xmath100 @xcite . the luminosity of the outflowing material is @xmath101 , where @xmath102 is the typical duration of a long grb . the temperature of the outflowing material at the base of the jet , with an inner radius of an accretion disc @xmath103 cm , is @xmath104 few mev . here @xmath105 ergs @xmath106 k@xmath107 is the radiation density constant . the fireball contains gamma rays , electron - positron pairs and relatively fewer baryons ( mostly protons ) . due to the radiation pressure , the optically thick hot plasma expands and overcome the gravitational pressure and forms the relativistic jet @xcite . an initially thermal energy dominated grb jet becomes kinetic energy dominated when the baryons carry the bulk of the initial fireball energy at a saturation radius @xmath108 @xcite . the bulk lorentz factor of the jet initially increases as @xmath109 and becomes constant at a value @xmath110@xmath111 , where @xmath112 is defined as the ratio beteen the total energy to mass flow and is called the baryon loading of the fireball . the radius at which the fireball attains this coasting lorentz factor is @xmath113 @xcite . the prompt @xmath96-rays from a grb are thought to be emitted when a significant fraction of the fireball kinetic energy is converted back to radiation energy . the internal shocks model is currently the most popular model for prompt @xmath96-ray emission @xcite . collisions between shells of plasma or fireball produce shocks and a fraction of the bulk kinetic energy is converted into random particle energy which is then radiated , via synchrotron mechanism e.g. , as the observed non - thermal @xmath96 rays . given a timescale @xmath114 from observations of highly variable prompt @xmath96-ray emission , internal shock collisions take place , due to a difference in lorentz factor @xmath115 between the shells , at a radius @xmath116 cm . this is typically much larger than the photospheric radius of a baryon - dominated grb jet , given by @xmath117 cm ( see , however , @xcite for different scaling of the photosphere ) . the observed pulse structure can be produced in colliding shells model @xcite . note that the variability time scale can be as small as @xmath2 ms , corresponding to the size scale of the jet base @xmath118 . the internal shocks are semi - relativistic , with lorentz factor @xmath119 few . a fraction of the shock energy is thought to amplify magnetic field , via some kind of instabilities , up to an energy that is at some equipartition level @xmath120 with the jet luminosity @xmath121 @xcite . another fraction of the shock energy is thought to accelerate electrons , via a fermi mechanism , to relativistic energies . in a fast - cooling scenario , where electrons lose most of their energy to the observed radiation , e.g. via synchrotron / compton mechanism , within the dynamic time scale , the prompt @xmath96-ray luminosity @xmath122 can be used to infer @xmath121 ( and eventually @xmath123 from the duration of the burst ) as @xmath124 . the magnetic field in the jet frame can also be inferred from the prompt @xmath96-ray observations as @xmath125 gauss @xcite . with a minimum lorentz factor of the electrons , @xmath126 , characteristic synchrotron radiation peak can explain the observed @xmath127 few hundred kev typical @xmath128 peak of the @xmath96-ray spectrum . the power - law part of the observed band spectra above @xmath129 can be explained as due to synchrotron radiation from the power - law distributed electron lorentz factor , @xmath130 , as @xmath131 . the synchrotron shock model as described above has difficulties to explain the observed @xmath96-ray spectrum below @xmath129 in a large fraction of the bursts @xcite . the fast - cooling synchrotron model is expected to produce a spectrum @xmath132 below @xmath129 as contrary to @xmath133 , typical observed value . a harder , @xmath134 , synchrotron spectrum from electrons in random megnetic field is often called the `` synchrotron death line '' @xcite which too has been observed violated in a sizeable fraction of grbs . there have been many attempts to solve this band spectrum crisis within the synchro - compton mechanism . explanations include synchrotron self - absorption in the kev range @xcite , compton scattering of optical photons to kev energies @xcite , low - pitch angle scattering or jitter radiation @xcite , time - dependent acceleration @xcite as well as mixing of kev photons with axion - like particles @xcite . photospheric thermal emission , as we decribe next , holds promise to explain the hard band spectrum below @xmath129 . in case of a baryon - loaded grb fireball , as we discussed before , most of the thermal energy converts to kinetic energy at a radius @xmath135 . the photospheric radius @xmath136 is also typically much larger than @xmath137 . in case the fireball is highly baryon dominated @xmath136 is smaller and could become comparable to @xmath137 . a significant fraction of the thermal energy can be released from the fireball in this case @xcite . moreover dissipation of kinetic energy at or below the photosphere , by shocks or other mechanisms such as neutron - proton decoupling and interactions @xcite , could also enhance thermal radiation for moderately baryon - loaded fireballs as well as explain highly - efficient @xmath96-ray emission from grbs , which is difficult to explain with internal shocks . simple photospheric models should produce thermal emission peaking at @xmath2 mev , the temperature of the fireball at the base of the jet , which is in the range of observed band peaks @xcite . this leads to a plausible physical explanations for the amati @xcite or ghirlanda @xcite relations between the @xmath128 peak energy and the total @xmath96-ray energy @xcite . the temperature of the photosphere can be much below @xmath2 mev in more complicated models , see e.g. @xcite . neutron - proton decoupling , their interactions and resulting radiation from cascade particles at the photosphere or compton scattering of thermal photons by relativistic electrons in sub - photospheric shocks have also been explored to produce the phenomenological band spectrum . these processes explain the power - law spectrum above @xmath129 . the `` synchrotron death line '' issue @xcite below @xmath129 can be avoided in photospheric models . recent modeling of fermi data from grb 090902b provides some evidence of photospheric emission being dominant in this burst @xcite . in some other fermi bursts a photospheric component in addition to the band or comptonized spectrum seems to improve fits , although thermal emission is not dominant in these bursts @xcite . fermi observations of grbs over a broad energy range and non - detection of a strong photospheric component in almost all bursts , and specially after grb 080916c @xcite , have renewed interests in magnetic or poynting - flux dominated grb jets , . a large dissipation radius , @xmath138 cm , yet high efficiency inferred from modeling fermi data can also be explained more naturally with magnetic models . particles are accelerated in the reconnection region of the magnetic field lines and produce the observed radiation . magnetic models can be generalized into two categories . in the first category the baryons are completely absent in the jet or dynamically negligible , at least initially @xcite , and in the second category the baryon load is significant but dynamically sub - dominant relative to the magnetic stresses @xcite . a hybrid model , dubbed icmart @xcite involves an initially poynting - flux dominated outflow with few baryons that lead to internal shocks when the magnetic energy subsides at a large radius . the origin of @xmath29 mev radiation from grbs and associated additional spectral component ( figure [ fig : spectra090926a ] ) has been intensely debated in recent years . in the context of the internal shock scenario , synchrotron - self - compton ( ssc ) scattering of soft target photons by relativistic electrons is the first choice to produce the hard power - law component in a simple one - zone model @xcite . the observed time delay ( figure [ fig : lc080916c ] ) between the kev mev radiation and @xmath29 mev radiation , however , is longer than the expected delay for the ssc emission , which can be of the order of the variability time scale ( @xmath139 s for long bursts ) at best . compton radiation from late internal shocks , with much longer variability time at gev than in kev mev , can explain delayed lat onset in some cases @xcite . an alternate mechanism to produce @xmath29 mev radiation in the prompt grb phase is through hadronic emission , either proton - synchrotron radiation and radiation from @xmath140 pair cascades @xcite ( figure [ fig : psynchrotron ] ) or photohadronic interactions and associated cascade radiation @xcite ( figure [ fig : pgamma ] ) . in the one - zone radiating shells , the hadronic models can account for the delayed detection of the hard power - law component through required proton acceleration and cooling time . contribute to the observed spectra . high - energy neutrinos are also expected in this model . from @xcite . ] the total energy of the burst , however , needs to be rather large in hadronic models because of a high jet lorentz factor , @xmath141 , required to avoid @xmath142 pair production _ in - situ _ by @xmath143 gev photons . in particular , to generate the observed @xmath144 s delay in lat for grb 110731a the comoving frame jet magnetic field needs to be @xmath145 @xmath146 g , in case of proton - synchrotron model . the corresponding isotropic - equivalent jet luminosity is @xmath147 @xmath148 erg s@xmath68 @xcite . while the hadronic models require a larger total energy than the leptonic models , the energy budget can be brought down for @xmath149 , see also @xcite . a high @xmath150 is not a prerequisite to avoid @xmath142 absorption in case of two - zone radiation , in either leptonic or hadronic models , where lat emission comes from late internal shocks @xcite . the late internal shocks , however , can not explain temporally extended lat emission which is apparently non - variabile . the delayed onset , longer duration and smooth temporal decay of the @xmath29 mev flux ( figure [ fig : latflux ] ) led to the development of an early afterglow interpretation of high - energy radiation from the external forward shock @xcite . the time delay between kev mev radiation and @xmath29 mev radiation is explained as the time required for deceleration of the grb jet with high @xmath150 . the hard power - law component and temporally - extended lat emission are interpretated as synchrotron radiation from the external shock while kev mev emission from the internal shocks in these two - zone emission models . detailed modeling , however , shows that an extrapolation of the external shock emission at earlier time fails to reproduce all of the @xmath29 mev flux in the prompt phase , requiring a significant contribution from the internal dissipation regions @xcite . mev radiation from a few lat - detected grbs , which show decaying lat flux as a common feature . the grey band with slope @xmath151is the expected decay behavior for afterglow emission from a radiative blast wave . from @xcite ] both in the internal- and external - shock scenario , modeling of high - energy radiation by the leptonic and/or hadronic processes depends crucially on @xmath150 . currently @xmath152 pair - production constraint in internal shocks is the most widely used estimate of a lower limit on @xmath150 @xcite . in the case of grb 090926a and grb 110731a a cut off in the additional power - law component allowed to estimate @xmath150 rather than its lower limit , assuming that the cut off at gev energy is due to attenuation of flux by @xmath152 pair absorption . figure [ fig : g0 ] shows the estimates of @xmath150 from lat - detected bright grbs . there are , however , large ( a factor 23 ) uncertainty due to theoretical modeling of target radiation fields and environment @xcite . pair absorption effect of gev photons assuming that they are produced in the same physical region as the lower , @xmath153 mev , photons . the lower limits , @xmath154 , are calculated in cases where there were no cutoff in the spectra in the gev range . in the `` pl only '' lower limits on @xmath154 were calculated using only the additional polwer - law components of the spectra . estimates of @xmath150 rather than @xmath154 were made when the observed @xmath143 gev cutoff in the spectra were asumed due to @xmath152 pair absorption.,width=307 ] regardless of the details of the dissipation mechanism of the grb jet to produce prompt @xmath96-ray emission , the accumulated surrounding material by the jet drives a blast wave and external forward shock @xcite . a reverse shock is also expected in the baryon - loaded fireballs . the deceleration takes place with the jet energy compoarable to the kinetic energy of the swept - up material in the blast wave , @xmath155 . the deceleration radius is @xmath156 cm for a constant density external medium . after deceleration , the blast wave evolves in a self - similar fashion with the bulk lorentz factor decreasing as @xmath157 for an adiabatic fieball and @xmath158 for a radiative fireball , both in constant density medium @xcite . particles are expected to be accelerated in the forward shock , which is highly relativistic . magnetic field also assumed to be generated in the forward shock , which leads to synchrotron radiation by relativistic particles @xcite . as the jet continues to be decelerated in the course of sweeping up more and more external matter , the peak of the synchrotron flux shifts from high - frequency to low - frequency bands . smooth decay of the flux across different energy bands is a characteristic feature of the afterglow synchrotron model @xcite , whose predicted detection allowed the first measurements of host galaxies and redshift distances @xcite . broadband x - ray to radio emission from grbs has been quite successfully described by synchrotron radiation from a decelerating blast wave with or without variation of the straightforward models . prompt optical emission @xcite , at the time the deceleration has been detected , with robotic ground - based telescopes such as rotse and others . the external shock interpretation of the late afterglow has proven robust overall , although some detailed features such as the x - ray steep decays followed by flat plateaus and occasional large flares are not well - understood yet . as discussed in section [ subsec : gev ] , temporally extended @xmath159 mev emission detected by fermi - lat from many grbs well after the prompt kev mev emission has been advocated as afterglow emission by a number of authors @xcite . to test an early afterglow interpretation of the lat data from grbs one needs to confront models with simultaneous multiwavelength data sets . most recently , modeling of multiwavelength light curves ( figure [ fig : grb110731a_lc ] ) and broadband spectral energy distribution ( sed ) from grb 110731a ( figure [ fig : grb110731a_sed ] ) strongly indicate that @xcite ( i ) lat extended emission as early as @xmath160 s , from broadband sed , is compatible with afterglow synchrotron emission in other bands . ( ii ) a wind afterglow model @xcite is favored over an ism model , from the behavior of the light curves and seds . detailed modeling of grb 110731a data also resulted in @xmath161 just before deceleration of the blast wave @xcite . an internal shock contribution to lat data before @xmath160 s can not be ruled out , however , and requires detailed investigation with future swift - fermi bursts . s suggests their origin , including lat emission , as from a decelerating blast wave in stellar wind environment . from @xcite . , width=316 ] imply that lat emission is afterglow . from @xcite . ] neutrinos with energies @xmath162 100 tev are expected from photohadronic ( @xmath163 ) interactions of relativistic protons accelerated in the fireball internal or external shocks . in the case of internal shocks , the interaction is expected to take place between mev photons produced by radiation from accelerated electrons and protons which are co - accelerated in the same shocks @xcite . neutrinos are expected to be produced more efficiently at an energy @xmath164 tev . neutrino production in internal shocks is expected to suppress at an energy @xmath165 pev due to synchrotron and adiabatic losses of the high - energy pions and muon @xcite . in the case of external shocks , optical / uv photons predominantly take part in @xmath163 interactions to produce @xmath166 eev neutrinos @xcite . the most intense neutrino emission at @xmath167 tev energies from grbs or collapsars might be due to @xmath168 and @xmath169 interactions occurring _ inside _ the stellar envelope while the jet is still burrowing its way out of the star @xcite and before it successfully break through the stellar envelope to produce a grb or choke inside . figure [ fig : nuflux ] shows different neutrino flux models @xcite from a nearby grb 030329 at @xmath170 which was also connected with a supernova . the icecube collaboration has recently constrained high - energy neutrino emission from the internal shocks by stacking electromagnetic data from grbs @xcite . this constraint , in the @xmath220 pev energy range , calculated from 40 and 59 string data severely excludes efficient neutrino production in the internal shocks . this could be due to the high bulk lorentz factor , @xmath171 , of the grb jet which reduces photopion production efficiency , e.g. @xcite , or due to inefficient aceleration of protons in the internal shocks . to accommodate icecube non - detection , revisions to the neutrino flux from internal shocks have also been proposed @xcite . new calculations in the models alternate to internal shocks are being done as well , e.g. @xcite . data from the full icecube detector will either constrain the neutrino production efficiency in the internal shocks or rule out these models . note that tev neutrinos from the buried grb jet @xcite and eev neutrinos from the afterglow @xcite remain viable . . an accompanying supernova was detected from this long grb as well . from @xcite.,width=316 ] binary mergers of ns - ns and/or ns - bh , the most popular model for short grbs , are the strongest candidate sources of gravitational waves ( gws ) with a total energy reaching @xmath95 ergs @xcite . the luminosity of gw from long grbs , originating from core - collapse of massive stars , is highly model - dependent @xcite . recent numerical calculations of gw emission from long grbs range from pessimistic @xcite to modest @xcite predictions . the expected detection rates of gw from compact binary mergers in coincidences with short grbs have been estimated to be several per year @xcite for the advanced ligo and virgo . coordinated detection of gw with electromagnetic signal helps to reduce background significantly , if short grbs are the sources . combining sub - threshold signal from electromagnetic , neutrino and gw observatories can also lead to significant detection @xcite . both swift and fermi are healthy and adequately - funded by nasa and foreign partners . neither has lifetime limits from expendable propulsion gasses or cryogens , and both have orbital lifetimes beyond 2025 . they are joined by integral ( e.g. , mereghetti et al . 2003 @xcite ) and maxi @xcite , which also detect grbs and the interplanetary network of grb detectors @xcite that localize events by triangulating their arrival times . there are several future missions with grb capabilities that are being developed or proposed . they include : 1 . svom a chinese - french mission to detect grbs and perform rapid optical and x - ray follow - up @xcite . 2 . uffo a korean payload flying on the russian lomonosov mission @xcite . the payload includes a wide - field hard x - ray burst detector and a rapidly - orienting optical telescope . the pathfinder version of the mission will fly in 2012 . lobster a wide - field focusing x - ray instrument operating in the 0.44 kev band . there are several versions that have been proposed including a uk instrument on the international space station @xcite , a nasa explorer mission including an afterglow ir telescope @xcite , and an explorer mission of opportunity on the international space station @xcite . 4 . loft a large - area x - ray timing mission selected for a phase a study for esa s m2 opportunity @xcite . astrosat an indian mission with hard x - ray imager , scanning sky monitoring , x - ray and uv telescopes @xcite . fermi lat detection of @xmath172 gev photons indicate that the prospects for detection of vhe radiation from grbs by ground - based telescopes are promising . highly sensitive imaging atmospheric cherenkov teleescopes ( iacts ) such as magic , veritas and hess with a threshold energy as low as 25 gev ( special grb mode for magic ) are currently slewing to the grb directions . water cherenkov detector hawc is also under construction . simultaneous observations by fermi lat and cherenkov telescopes in the overlapping energy range will be extremely helpful to understand the nature of high - energy radiation from grbs . such observations can also pose strong constraints on both the leptonic and hadronic models , especially independent handles on the magnetic field strength are possible . long - sought ssc radiation from the external forward shock or hadronic radiation in the prompt phase can also be searched for and/or constrained . detection of very - high energy radiation from grbs can be modeled to estimate the total explosion energy , which has implications for graviational wave detection by ligo / virgo . there is a great opportunity for new science coming from future observations coordinated with grbs of gravitational waves , neutrinos and across the electromagnetic spectrum . there are major new facilities coming on line and planned such as ligo / virgo , icecube , panstarrs , lsst and the 30 m telescopes . we are in an era of rapid progress in understanding grbs . swift , fermi , integral , maxi , and ipn are all detecting grbs . a vibrant community of observers is following up the grb detections with new and capable instruments across the electromagnetic spectrum . key findings from the past several years has included the following : 1 . x - ray afterglow features , steep and shallow decays , flares , 2 . afterglows and accurate localizations of short bursts supporting a possible merger origin , 3 . confirmation of long grb - supernova connection , 4 . detection of high - redshift , @xmath173 , grbs , 5 . delayed onset of gev emission , 6 . high bulk lorentz factor of the jet . with all this progress , there are many objectives that we hope to accomplish in the next 10 years . these include : 1 . confirmation of short burst origin , 2 . jet opening angle and measurements of lighcurve breaks 3 . metallicity dependence 4 . relation to star formation history 5 . nature and interpretation of high energy spectral components . we look forward to many years more of grb discovery . we thanks john cannizzo and charles dermer for valuable consultation and comments on this paper . all data from the swift and fermi observatories are publicly available through nasa s high energy astrophysics science archive research center ( heasarc ) at http://heasarc.gsfc.nasa.gov/. work of s.r . was supported by nasa fermi guest investigator program and was performed at the naval research lab while under contract . b. zhang , b. -b . zhang , f. j. virgili , e. -w . liang , d. a. kann , x. -f . wu , d. proga and h. -j . lv _ et al . _ , astrophys . j. * 703 * , 1696 ( 2009 ) o. bromberg , e. nakar , t. piran and r. e . sari , arxiv:1210.0068 [ astro-ph.he ] . n. gehrels & p. mszros , science , 337 , 932 ( 2012 ) d. a. frail , s. r. kulkarni , r. sari , s. g. djorgovski , j. s. bloom , t. j. galama , d. e. reichart and e. berger _ et al . _ , . j. * 562 * , l55 ( 2001 ) s. b. cenko , d. a. frail , f. a. harrison , s. r. kulkarni , e. nakar , p. chandra , n. r. butler and d. b. fox _ et al . _ , astrophys . j. * 711 * , 641 ( 2010 ) a. shemi and t. piran , astrophys . j. * 365 * , l55 ( 1990 ) . p. mszros , p. laguna and m. j. rees , astrophys . j. * 415 * , 181 ( 1993 ) p. mszros and m. j. rees , astrophys . j. * 530 * , 292 ( 2000 ) [ astro - ph/9908126 ] . e. m. rossi , a. m. beloborodov and m. j. rees , mon . not . soc . * 369 * , 1797 ( 2006 ) m. j. rees and p. mszros , astrophys . j. * 430 * , l93 ( 1994 ) p. veres and p. mszros , astrophys . j. * 755 * , 12 ( 2012 ) c. d. dermer , astrophys . j. * 614 * , 284 ( 2004 ) s. kobayashi , t. piran and r. e . sari , astrophys . j. * 490 * , 92 ( 1997 ) m. v. medvedev and a. loeb , astrophys . j. * 526 * , 697 ( 1999 ) t. inoue , k. asano and k. ioka , astrophys . j. * 734 * , 77 ( 2011 ) s. razzaque , p. mszros and b. zhang , astrophys . j. * 613 * , 1072 ( 2004 ) r. d. preece , m. s. briggs , r. s. mallozzi , g. n. pendleton , w. s. paciesas and d. l. band , astrophys . j. lett . 506 , l23 ( 1998 ) y. kaneko , r. d. preece , m. s. briggs , w. s. paciesas , c. a. meegan and d. l. band , astrophys . j. suppl . 166 , 298 ( 2006 ) a. goldstein , j. m. burgess , r. d. preece , m. s. briggs , s. guiriec , a. j. van der horst , v. connaughton and c. a. wilson - hodge _ et al . _ , astrophys . j. suppl . * 199 * , 19 ( 2012 ) j. granot , t. piran , and r. sari , astrophys . j. lett . 534 , l163 ( 2000 ) a. panaitescu and p. mszros , astrophys . j. * 544 * , l17 ( 2000 ) m. v. medvedev , astrophys . j. * 540 * , 704 ( 2000 ) m. v. medvedev , astrophys . j. * 637 * , 869 ( 2006 ) n. m. lloyd - ronning and v. petrosian , astrophys . j. 565 , 182 ( 2002 ) o. mena , s. razzaque and f. villaescusa - navarro , jcap * 1102 * , 030 ( 2011 ) e. v. derishev , v. v. kocharovsky & v. v. kocharovsky astrophys . j. 521 , 640 ( 1999 ) m. j. rees and p. mszros , astrophys . j. * 628 * , 847 ( 2005 ) a. peer , p. mszros and m. j. rees , astrophys . j. * 642 * , 995 ( 2006 ) l. amati , f. frontera , m. tavani , j. j. m. in t zand , a. antonelli , e. costa , m. feroci and c. guidorzi _ et al . _ , astron . astrophys . * 390 * , 81 ( 2002 ) g. ghirlanda , g. ghisellini and d. lazzati , astrophys . j. * 616 * , 331 ( 2004 ) c. thompson , p. mszros and m. j. rees , astrophys . j. * 666 * , 1012 ( 2007 ) a. peer , b. -b . zhang , f. ryde , s. mcglynn , b. zhang , r. d. preece and c. kouveliotou , mnras 420 , 468 ( 2012 ) s. guiriec , m. s. briggs , v. connaugthon , e. kara , f. daigne , c. kouveliotou , a. j. van der horst and w. paciesas _ et al . _ , astrophys . j. * 725 * , 225 ( 2010 ) s. guiriec , v. connaughton , m. s. briggs , m. burgess , f. ryde , f. daigne , p. mszros and a. goldstein _ et al . _ , astron . j. * 727 * , l33 ( 2011 ) b. zhang and a. peer , astrophys . j. * 700 * , l65 ( 2009 ) v. v. usov , m.n.r.a.s 267 , 1035 ( 1994 ) p. mszros and m. j. rees , astrophys . j. lett . * 482 * , l29 ( 1997 ) m. lyutikov and r. d. blandford , astro - ph/0312347 . c. thompson , mon . not . astron . soc . * 270 * , 480 ( 1994 ) . g. drenkhahn , astron . astrophys . 387 , 714 ( 2002 ) g. drenkhahn and h. c. spruit , astron . astrophys . 391 , 1141 ( 2002 ) c. thompson , astrophys . j. * 651 * , 333 ( 2006 ) b. zhang and h. yan , astrophys . j. * 726 * , 90 ( 2011 ) m. ackermann , et al . , astrophys . j. in press arxiv:1212.0973 [ astro-ph.he ] x. -y . wang , z. li , z. -g . dai and p. mszros , astrophys . j. * 698 * , l98 ( 2009 ) z. bosnjak , f. daigne and g. dubus , astron astrophys . 498 , 677 ( 2009 ) k. toma , x. -f . wu and p. mszros , astrophys . j. * 707 * , 1404 ( 2009 ) k. toma , x. -f . wu and p. mszros , mnras 415 , 1663 ( 2011 ) s. razzaque , c. d. dermer and j. d. finke , open astron . j. * 3 * , 150 ( 2010 ) k. asano , s. guiriec and p. mszros , astrophys . j. * 705 * , l191 ( 2009 ) k. murase , k. asano , t. terasawa and p. mszros , astrophys . j. * 746 * , 164 ( 2012 ) k. asano , s. inoue and p. mszros , astrophys . j. * 725 * , l121 ( 2010 ) y. -c . zou , y. -z . fan and t. piran , astrophys . j. lett . 726 , l2 ( 2011 ) arxiv:1008.2253 [ astro-ph.he ] . kumar and r. b. duran , mon . not . 409 * , 226 ( 2010 ) g. ghisellini , g. ghirlanda , l. nava & celotti , a. mnras 403 , 926 ( 2010 ) a. corsi , d. guetta and l. piro , astrophys . j. * 720 * , 1008 ( 2010 ) b. -b . zhang , b. zhang , e. -w . liang , y. -z . fan , x. -f . wu , a. peer , a. maxham and h. gao _ et al . _ , astrophys . j. * 730 * , 141 ( 2011 ) r. -y . liu and x. -y . wang , astrophys . j. * 730 * , 1 ( 2011 ) a. maxham , b .- b zhang and b. zhang , mnras , * 415 * , 77 ( 2011 ) p. mszros and m. j. rees , astrophys . j. * 476 * , 232 ( 1997 ) r. a. m. j. wijers , m. j. rees and p. mszros , mon . not . * 288 * , l51 ( 1997 ) r. a. chevalier & z .- y . li , astrophys . j. 536 , 195 ( 2000 ) e. waxman and j. n. bahcall , phys . lett . * 78 * , 2292 ( 1997 ) j. p. rachen and p. mszros , phys . d * 58 * , 123005 ( 1998 ) c. d. dermer and a. atoyan , phys . lett . * 91 * , 071102 ( 2003 ) d. guetta , d. hooper , j. alvarez - muniz , f. halzen and e. reuveni , astropart . phys . * 20 * , 429 ( 2004 ) t. kashti and e. waxman , phys . lett . * 95 * , 181101 ( 2005 ) p. lipari , m. lusignoli and d. meloni , phys . d * 75 * , 123005 ( 2007 ) m. vietri , phys . * 80 * , 3690 ( 1998 ) e. waxman and j. n. bahcall , astrophys . j. * 541 * , 707 ( 2000 ) z. g. dai & t. lu 2001 , astrophys . j. 551 , 249 ( 2001 ) p. mszros and e. waxman , phys . lett . * 87 * , 171102 ( 2001 ) s. razzaque , p. mszros and e. waxman , phys . d * 68 * , 083001 ( 2003 ) s. razzaque , p. mszros and e. waxman , phys . d * 69 * , 023001 ( 2004 ) r. abbasi _ et al . _ [ icecube collaboration ] , phys . lett . * 106 * , 141101 ( 2011 ) m. ahlers , m. c. gonzalez - garcia and f. halzen , astropart . phys . * 35 * , 87 ( 2011 ) r. abbasi _ et al . _ [ icecube collaboration ] , nature * 484 * , 351 ( 2012 ) s. razzaque , c. d. dermer , j. d. finke and a. atoyan , aip conf . proc . * 1133 * , 328 ( 2009 ) s. hummer , p. baerwald and w. winter , phys . lett . * 108 * , 231101 ( 2012 ) [ arxiv:1112.1076 [ astro-ph.he ] ] . h. -n . he , r. -y . liu , x. -y . wang , s. nagataki , k. murase and z. -g . dai , astrophys . j. * 752 * , 29 ( 2012 ) b. zhang and p. kumar , arxiv:1210.0647 [ astro-ph.he ] . c. s. kochanek and t. piran , astrophys j. * 417 * , l17 ( 1993 ) m. ruffert and h. t. janka , astron . astrophys . * 344 * , 573 ( 1999 ) c. l. fryer , s. e. woosley and a. heger , astrophys . j. * 550 * , 372 ( 2001 ) c. d. ott , c. reisswig , e. schnetter , e. oconnor , u. sperhake , f. loffler , p. diener and e. abdikamalov _ et al . _ , phys . lett . * 106 * , 161103 ( 2011 ) k. kiuchi , m. shibata , p. j. montero and j. a. font , phys . rev . * 106 * , 251102 ( 2011 ) j. abadie _ et al . _ [ ligo scientific and virgo collaborations ] , * 27 * , 173001 ( 2010 ) m. w. e. smith , d. b. fox , d. f. cowen , p. meszaros , g. tesic , j. fixelle , i. bartos and p. sommers _ et al . arxiv:1211.5602 [ astro-ph.he ] . s. mereghetti , et al . , a & a 411 , 291 ( 2003 )

we present feynman type diagrams for calculating the-point function of the primordial curvature perturbation in terms of scalar field perturbations during inflation . the diagrams can be used to evaluate the corresponding terms in the-point function at tree level or any required loop level . rules are presented for drawing the diagrams and writing down the corresponding terms in real space and fourier space . we show that vertices can be renormalised to automatically account for diagrams with dressed vertices . we apply these rules to calculate the primordial power spectrum up to two loops , the bispectrum including loop corrections , and the trispectrum .

0705.4096

there is currently a great deal of interest in the statistical properties of primordial perturbations from inflation , because measurements of any non - gaussianity will improve by about an order of magnitude over the next few years , for example with the atacama cosmology telescope @xcite and planck @xcite . this will provide a key way to discriminate between the many models of inflation . although single field models of slow - roll inflation typically generate a small level of non - gaussianity @xcite , there may be an observable level generated in multiple field inflation @xcite , for example the curvaton scenario @xcite , or some single field models such as dirac - born - infeld inflation ( dbi ) @xcite . currently observations of the cmb have concentrated on constraining the 3-point function ( bispectrum ) @xcite , but the 4-point function ( trispectrum ) has also been considered @xcite , and in principle higher order @xmath0-point functions may also be observable if they are sufficiently large . in some models the first signal of non - gaussianity may come through the trispectrum , for example some special cases of the curvaton scenario @xcite . furthermore higher order statistics in principle carry more information and therefore could distinguish between different sources of non - gaussianity , for example in many cases the bispectrum can be paremetrised by a single non - linearity parameter , @xmath1 , while the trispectrum depends on two wavenumber independent parameters , @xmath2 and @xmath3 @xcite . we will calculate the primordial curvature perturbation on uniform density hypersurfaces , @xmath4 , on large scales employing the @xmath5 formalism @xcite . we use the separate universe approach @xcite . this considers each super - hubble scale patch to be evolving like a separate friedmann - robertson - walker universe which are locally homogeneous . by patching these regions together we can track the evolution of the perturbations on large scales just by using background quantities which greatly simplifies our calculations . we also need to know the perturbations of the scalar fields at hubble - exit , but in many cases these are extremely close to gaussian in which case their statistical properties are given purely in terms of the power spectrum of the fields . the number of @xmath6-foldings , @xmath7 , given by n=^t_fin_t_inih(t ) dt , is evaluated from an initial flat hypersurface to a final uniform - density hypersurface . the perturbation in the number of @xmath6-foldings , @xmath5 , is the difference between the curvature perturbations on the initial and final hypersurfaces . we wish to calculate primordial perturbations , hence we pick a final uniform density hypersurface to be at a fixed time during the standard radiation dominated era , for example during primordial nucleosynthesis . the initial time is arbitrary provided it is after the hubble exit time of all relevant scales . it is often convenient to pick this time to be shortly after hubble - exit . we introduce the method for calculating the primordial curvature perturbation in terms of the @xmath5 formalism in more detail below . here we present a feynman diagram type approach to calculating the primordial non - gaussianity , where every term in the @xmath0-point function of @xmath4 has a diagrammatic representation . similar diagrams were introduced in the context of non - linear perturbation theory of large scale structure @xcite and further developed in @xcite ( see also @xcite ) . it is possible to read off the corresponding mathematical term in the @xmath0-point function for every diagram and it is possible to draw all the diagrams in a systematic way so that all terms are included . this approach extends to loop corrections of any required order . although the loop corrections are expected to be small in general compared to the tree level terms there is no proof of this , except for a particular loop correction to the bispectrum in some specific models of inflation @xcite . we will see that in general the loop corrections depend on a large scale cut off ; depending on this cut off the loop correction may contribute significantly to observable quantities , see for example @xcite . in fact there is a specific model of inflation where the one loop term gives the dominant and potentially observable contribution to the bispectrum @xcite . a diagrammatic method to calculate field fluctuations during inflation was presented in @xcite . we first introduce the primordial curvature perturbation . in the next two sections of this paper we assume the initial field perturbations are gaussian , which in many cases is a good approximation to make , for example slow - roll inflation . in sec . [ sec : realspacediagrams ] the rules for drawing diagrams in real space are given , along with a discussion of why we want to calculate connected rather than disconnected diagrams . in sec . [ sec : fourierspacediagrams ] the equivalent rules in fourier space are given . an application of the fourier space rules for the power spectrum including second order loop corrections is given in sec . [ sec : powerspectrum ] . in sec . [ sec : renormalisation ] we discuss a way to renormalise the vertices of the diagrams such that all diagrams with dressed vertices can be absorbed , with a proof given in the appendix . in sec . [ sec : ng ] the assumption of gaussian initial fields is dropped and the complete rules presented and applied to several examples . finally we conclude in sec . [ sec : conclusion ] . we can write the primordial curvature perturbation , @xmath4 , in terms of derivatives of @xmath7 with respect to the fields , @xmath8 and the field perturbation , @xmath9 , at the initial time @xmath10 . by neglecting any additional dependence on @xmath11 at hubble - exit we are assuming the field perturbations on large scales are overdamped and thus we can neglect the decaying mode . this is valid during slow roll , but also more generally holds for light fields during inflation . the assumption that @xmath7 is a function of field values only is also valid in dbi inflation @xcite . we use the notation that the fields are @xmath12 , where the perturbed field @xmath13 satisfies @xmath14 . the curvature perturbation is given by @xmath15 the dummy index @xmath16 labels the light scalar fields relevant during inflation , and summation convention is used throughout this paper . it is more convenient to work with the @xmath4 defined above since it satisfies @xmath17 , even though often @xmath18 is used in the literature . this equation can hence be used to calculate the primordial @xmath0-point function of @xmath4 , although the result will depend on the @xmath0-point function of the fields at hubble exit . this has so far been explicitly calculated for the 2- , 3- and 4-point functions , @xcite . the connected 2- , 3- and 4-point functions of the fields are defined by @xmath19 note that only the 4-point function ( and higher ) depend on the direction of the @xmath20 vectors , the 2- and 3-point functions of the fields just depend on the magnitude of the vectors , @xmath21 . homogeneity of the random fields implies that the sum of the @xmath20 vectors must be zero , while isotropy implies that the @xmath0-point functions are invariant under reorientations of this closed configuration . at lowest order in slow roll different fields are uncorrelated at hubble exit and all light fields have the same power spectrum @xcite , so @xmath22 where @xmath23 is the kronecker delta - function , and the variance per logarithmic interval in @xmath24-space is given by @xmath25 where the hubble parameter @xmath26 is evaluated at hubble - exit , @xmath27 . at zeroth order in slow - roll parameters , @xmath28 is independent of wavenumber , i.e. we have a scale invariant spectrum for the field fluctuations . the bispectrum satisfies @xmath29 @xcite , where @xmath30 is a slow - roll parameter , so it is zero at lowest order in slow roll . however the trispectrum is not zero at lowest order in slow roll @xcite , @xmath31 , but is suppressed by @xmath32 compared to the power spectrum . for a discussion of the slow - roll order of higher order @xmath0-point functions of the fields see @xcite . initially we will assume that all of the fields have a purely gaussian distribution at an initial time , e.g. shortly after hubble exit . then every odd @xmath0-point function of the fields is zero and every even @xmath0-point function of the fields can be reduced to a product of 2-point functions . this assumption is not required and the extension to non - gaussian field perturbations will be presented in sec . [ sec : ng ] . however the approximation of initially gaussian field perturbations is frequently made in the literature and is often a good approximation , for example @xcite shows that the 3-point function of the fields adds an unobservably small contribution to the bispectrum around hubble exit assuming slow roll inflation with a standard kinetic term of the scalar fields , while @xcite show that the 4-point function of the fields adds an unobservably small contribution to the trispectrum around hubble exit under the same assumptions . working to zeroth order in slow roll implies that ( [ cfreefields ] ) holds and that the 3-point function of the fields is zero . so although working to lowest order in slow roll is not the same as working with gaussian field perturbations , in practice the conditions are related . hence when assuming that the field perturbations have a gaussian distribution at hubble exit we will also work at zeroth order in slow roll . it is likely that any observable amount of non - gaussianity can be calculated to sufficiently high accuracy when making these two approximations , assuming a standard kinetic term . assuming that the fields are uncorrelated gaussian variables , with identical distributions , as discussed above , the two point function of the fields is given by [ varphicorrelation ] _ x_i^a_x_j^b = \ { l ^a bg(|x_i - x_j| ) ij , + ^a b^2 i = j . . all of the higher @xmath0-point functions of the fields can be written in terms of @xmath33 and the variance of the fields @xmath34 . the two point function @xmath33 only depends on @xmath35 because we are assuming the background is homogenous and isotropic . the diagrammatic rules for the connected @xmath0-point function , @xmath36 , are : 1 . draw @xmath0 points representing the @xmath0 spatial points @xmath37 and connect them with @xmath38 propagators ( dashed lines ) which attach two of the positions @xmath39 ( it can be two distinct positions or it can have both ends attached to the same position ) . the diagram should be connected in order to calculate the connected @xmath0 point function , see sec . [ sec : disconnecteddiagrams ] . we require @xmath40 in order to draw a connected diagram and diagrams with @xmath41 are tree level , while those with @xmath42 include loop corrections . 2 . label each end of each propagator with the field indices @xmath43 . 3 . assign a factor @xmath44 to each spatial point , @xmath39 , where the number of derivatives of @xmath7 is the number of propagators attached to that @xmath39 . 4 . assign a factor of @xmath45 to each propagator , where @xmath46 are the appropriate field indices attached to the propagator and @xmath47 are the positions at either end of the propagator . if both ends of the propagator are attached to the same @xmath39 instead assign a factor @xmath48 . divide by the appropriate numerical factor . whenever @xmath49 propagators attach the same @xmath39 and @xmath50 at both ends this gives a factor of @xmath51 . whenever @xmath49 propagators dress an @xmath39 this gives an additional factor of @xmath52 ( as well as the factor of @xmath51 due to the previous rule ) . a propagator dresses an @xmath39 if both ends of the propagator are attached to the same @xmath39 . add all permutations of the diagrams which is all of the distinct ways to relabel the spatial points . the number of permutations depends on the symmetries of the diagram , a diagram with complete symmetry between all of the spatial points has only one term , while for a diagram with no symmetries between the spatial points there are @xmath53 permutations . in fig . [ fig : realspace234pt ] the diagrams for the 2-point function at tree level and one loop level , plus the tree level terms for the 3- and 4-point functions are shown . note that for the 4-point function there are two tree level terms . the terms corresponding to the diagrams are given below , in the same order as the diagrams , [ 2ptfnreal ] _ x_1_x_2&= & ( n_a n^a + n_a n^ab_b)g(|x_1-x_2| ) + 12n_abn^abg(|x_1-x_2|)^2 , + _ x_1_x_2_x_3&= & n_a n_b n^ab ( g(|x_1-x_2| ) g(|x_1-x_3|)+2 ) , + [ 4ptfnreal ] _ x_1_x_2_x_3_x_4_c & = & n_a n_b n_c n^abc ( g(|x_1-x_2| ) g(|x_1-x_3| ) g(|x_1-x_4| ) + 3 ) + & & + n_a n_b n^a_c n^bc ( g(|x_1-x_2|)g(|x_2-x_3|)g(|x_3-x_4|)+11 ) . although the second diagram for the two point function , fig . [ fig : realspace234pt ] , has a dressed vertex with an associated numerical factor of @xmath54 , there is another permutation of this diagram with the @xmath55 dressed which gives an equal contribution . the third diagram has two propagators attaching @xmath56 and @xmath55 and there are no other permutations of this diagram so there is a numerical factor of @xmath54 , as shown in the third term of ( [ 2ptfnreal ] ) . the diagrams for the 3- and 4-point functions are all at tree level so they all have a numerical factor of 1 . the first diagram for the 4-point function has 4 permutations since we have 4 choices of which of the @xmath57 has 3 propagators attached to it as given in the first term of ( [ 4ptfnreal ] ) , the second diagram of the 4-point function has 12 permutations because there are @xmath58 choices of which 2 of the @xmath57 should have 2 propagators attached to them in general any @xmath0-point function has a connected and a disconnected part . throughout this paper we will only consider the connected @xmath0-point function because the disconnected contribution contains no new information over the lower @xmath0-point functions that it is split into . there is no disconnected contribution to the 2- or 3-point function of the fields or zeta , because we assume @xmath59 . the disconnected contribution is only non - zero if a subset of the @xmath20 vectors sums to zero . for example the disconnected part of the primordial 4-point function , @xmath60 , is the product of two 2-point functions such as @xmath61 and in total there are three such terms . this is non - zero even if @xmath4 is purely gaussian , but it contains no new information compared to the primordial 2-point function . this term only contributes if @xmath62 for some @xmath63 , so provided the sum of any two @xmath20 vectors is not zero any contribution is from the connected part . the connected part of the primordial 4-point function is only non - zero if @xmath4 is not purely gaussian and it contains information that is not observable from the primordial 2- or 3-point functions . we present here the rules for drawing the connected diagrams ( see sec . [ sec : disconnecteddiagrams ] ) of the @xmath0-point function of @xmath4 at @xmath38-th order ( i.e. @xmath64 ) , for @xmath40 . the tree level terms correspond to @xmath41 . 1 . draw all distinct connected diagrams with @xmath0-external ( solid ) lines and @xmath38 ( dashed ) propagators . every vertex must consist of 1 external line and at least 1 propagator . 2 . label the external legs with incoming momenta @xmath65 for @xmath66 and label the propagators with internal momenta @xmath67 for @xmath68 . label each end of each propagator with a field index @xmath69 . 3 . assign a factor @xmath70 to each vertex . the number of derivatives of @xmath7 corresponds to the number of propagators attached to each vertex . we use the convention that incoming momentum is positive . the @xmath71 function ensures momentum is conserved at each vertex . see fig . [ fig : fourierspacerulesg ] . 4 . assign a factor @xmath72 to each propagator , where @xmath46 are the appropriate field indices that the propagator is labelled with at either vertex and @xmath73 is the momentum attached to the propagator . see fig . [ fig : fourierspacerulesg ] . integrate over the propagator momenta , @xmath74 . the first @xmath75 integrations are trivial because of the @xmath71 functions but any further integrations ( in the case of a diagram with loop corrections ) can not in general be evaluated analytically . divide by the appropriate numerical factor . whenever @xmath49 propagators attach the same vertices at both ends this gives a factor of @xmath51 . whenever @xmath49 propagators dress a vertex this gives an additional factor of @xmath52 ( as well as the factor of @xmath51 due to the previous rule ) . a propagator dresses a vertex if both ends of the propagator are attached to the same vertex . add all permutations of the diagrams which is all of the distinct ways to relabel the @xmath65 attached to the external lines . the number of permutations depends on the symmetries of the diagram , a diagram with complete symmetry between all of the external lines has only one term , while for a diagram with no symmetries between the external lines there are @xmath53 permutations . going between the unlabelled diagrams in fourier or real space is straightforward , to go from fourier space to real space one simply removes the external lines and places an @xmath39 at each vertex . to go from real space to fourier space one attaches an external line in place of every @xmath39 . in fourier space rule 1 for the diagram is to draw every possible diagram at the appropriate order . rule 2 gives the rules for labelling the diagrams . rules 35 give the associated mathematical expression for each diagram , up to a numerical factor given in rule 6 . a derivation of the numerical factors is given in sec . [ sec : renormalisation ] and app . [ sec : app : nptfn ] . the numerical factor for all tree level diagrams is 1 . finally rule 7 tells one to include every permutation of the labelling of the @xmath65 on the external legs that are distinct . the primordial power spectrum in fourier space is defined by @xmath76 the tree level and one loop correction terms to the 2-point function is diagrammatically given by fig . [ fig:2pt1loopg ] . after carrying out the trivial integration over the propagator momentum the tree level term is given by @xmath77 the one loop terms are given by [ 2ptfn1loop ] p_^1(k)=d^3q ( 12n_abn^abp(q)p(|-|)+ n_an^ab_bp(k)p(q ) ) , where we have already carried out one integration . both of the diagrams associated with these terms have a numerical factor of 2 , however there are also 2 permutations of the second term because it has a non symmetric diagram . because of momentum conservation , @xmath78 , it follows that @xmath79 . after enforcing @xmath80 the two permutations of the second diagram give an equal contribution which cancels the numerical factor of 2 . going to 2 loops , terms of order @xmath81 there are four terms . diagrammatically the terms are given in fig . [ fig:2pt2loopg ] which corresponds to [ 2ptfn2loop ] p_^2&=&d^3q_1d^3q_2 ( 14n_abc^abn^cp(q_1)p(q_2)p(k)+14n_ab^an^bc_cp(k)p(q_1)p(q_2 ) . + 12n_abn^abc_cp(q_1)p(q_2)p(|-| ) + 16n_abcn^abcp(q_1)p(|-|)p(|-| ) ) . only the fourth diagram of fig . [ fig:2pt2loopg ] does not have a dressed vertex . in fact at every level there is only one diagram without dressed vertices , as has already been shown explicitly at tree and one loop level in fig . [ fig:2pt1loopg ] . in the case where a single field @xmath82 generates the primordial curvature perturbation , so @xmath83 , we have p_^tree(k)=(n)^2 p(k ) , where @xmath84 . in the case of a scale invariant spectrum we can also evaluate the 1 loop integrals if we apply a large scale cut off @xmath85 . for a discussion of how to evaluate the integral over the loop momenta and the dependence of this term on the cut off @xmath85 see for example @xcite . we also apply a small scale cut off @xmath24 . the result is p_^1loop(k)= p(k)((n)^2+nn ) ( kl ) , where the variance @xmath86 was defined by eq . ( [ p ] ) . in analogy with eq . ( [ p ] ) we define @xmath87 which from observations of the cmb is of the order @xmath88 @xcite . we hence have [ p1loop ] = _^tree ( + ) ( kl ) . we can write the derivatives of @xmath7 in terms of the ( in principle ) observable non - linearity parameters @xmath1 and @xmath3 which in this case can be defined by @xcite [ fandgdefn ] = _ 1 + 35f_nl_1 ^ 2+g_nl_1 ^ 3 + , where @xmath89 is gaussian because it is directly proportional to the initial gaussian field perturbation , @xmath90 , and the dimensionless non - linearity parameters , @xmath1 and @xmath3 , are given by [ fnl1field ] f_nl&=&56 , + [ gnl1field ] g_nl&=&. for the extension of these two formula to the case of multiple field inflation see for example @xcite . substituting ( [ fnl1field ] ) and ( [ gnl1field ] ) into ( [ p1loop ] ) we find [ p1loopfg ] = _^tree ( f_nl^2+g_nl ) ( kl ) . if @xmath91 and we take the large scale cut off to be comparable to the present day hubble scale then we can take @xmath92 @xcite , and the observational bound on the bispectrum , @xmath93 @xcite bounds the first term of the ratio above to be less than @xmath94 . there is not yet any equivalent observational bound on @xmath3 but it would have to be extremely large in order to give a significant contribution to eq . ( [ p1loopfg ] ) . it therefore seems that the only way for the 1 loop contribution to be significant compared to the tree level term of the power spectrum is to take an exponentially large cut off @xmath85 . for a discussion of this possibility see for example @xcite there is a way to reduce the number of diagrams significantly , by renormalising the vertices such that the diagrams with dressed vertices are automatically accounted for . we do this by renormalising the factors @xmath44 , that are attached to each vertex . the derivative of the number of @xmath6-foldings for the given background @xmath95 is @xmath96 , and we can relate this to the number of @xmath6-foldings at a general point @xmath97 by [ ntdefn ] n ( ( ) ) = n|__0+n_a^a+n_ab^a^b + n_abc^a^b^c+ . the expectation of @xmath98 is given by [ nt ] _ abc = n_abc+ n_abcd^d^2 + 18 n_abcde^de^2 ^ 2 + , where we have used wick s theorem to decompose the expectation of powers of @xmath99 according to [ wick ] ^2n=(2n-1)!!^2^n , ^2n-1=0 . the double factorial is defined by @xmath100 . the coefficient of the general term in ( [ nt ] ) , @xmath101 is given by ( [ ntdefn ] ) and ( [ wick ] ) and is @xmath102 . this explains the numerical factor associated with every dressed vertex , see rule 6 , sec . [ sec : fourierspacediagrams ] . in fourier space the ensemble average of @xmath98 is [ ntfourier ] _ abc = n_abc+ n_abcd^d d^3qp(q ) + 18 n_abcde^de d^3q_1 d^3q_2 p(q_1)p(|-| ) + , which in real space is given by eq . ( [ nt ] ) . in fig . [ fig : ntprimed ] the graphical representation of @xmath103 is given up to 2 loops . so far the derivatives of @xmath7 had been calculated at @xmath95 , however if we instead calculate the expectation of the derivative of @xmath7 at a general point , and attach this factor @xmath104 to each vertex ( which we call a renormalised vertex ) then all diagrams with dressed vertices are automatically included in the expansion ( [ nt ] ) and do nt need to be drawn . we provide a proof of this in appendix [ sec : app : renn ] . the new diagrammatic rules for the connected @xmath0-point function with renormalised vertices are : 1 . draw @xmath0 points representing the @xmath0 spatial points @xmath37 and connect them with @xmath38 propagators ( dashed lines ) which attach two distinct positions @xmath39 and @xmath50 . the vertices are drawn with a solid dot to show that they are renormalised , see fig . [ fig : ren2ptfn2loop ] for examples of such diagrams . the diagram should be connected in order to calculate the connected @xmath0 point function , see sec . [ sec : disconnecteddiagrams ] . we require @xmath40 in order to draw a connected diagram and diagrams with @xmath41 are tree level , while those with @xmath42 include loop corrections . 2 . label each end of each propagator with the field indices @xmath43 . 3 . assign a factor @xmath104 to each spatial point , @xmath39 , where the number of derivatives of @xmath105 is the number of propagators attached to that @xmath39 . 4 . assign a factor of @xmath45 to each propagator , where @xmath46 are the appropriate field indices attached to the propagator and @xmath47 are the positions at either end of the propagator . divide by the appropriate numerical factor . whenever @xmath49 propagators attach the same @xmath39 and @xmath50 at both ends this gives a factor of @xmath51 . add all permutations of the diagrams which is all of the distinct ways to relabel the spatial points . the number of permutations depends on the symmetries of the diagram , a diagram with complete symmetry between all of the spatial points has only one term , while for a diagram with no symmetries between the spatial points there are @xmath53 permutations . in fig . [ fig : ren2ptfn2loop ] the diagrams for the 2-point function up to two loop level is presented . the terms corresponding to the diagrams are given below , in the same order as the diagrams , _ x_1_x_2&= & _ a^ag(|x_1-x_2| ) + 12_ab^abg(|x_1-x_2|)^2 + & & + _ abc^abcg(|x_1-x_2|)^3 + . note that with the renormalised diagrams there is only one diagram at every order in loops . this result is also derived to all orders in app . [ sec : app : renn ] . rules for drawing the connected diagrams ( see sec . [ sec : disconnecteddiagrams ] ) of the @xmath0-point function of @xmath4 at @xmath38-th order ( i.e. @xmath64 ) , for @xmath106 . the tree level terms correspond to @xmath41 . 1 . draw all distinct connected diagrams with @xmath0 external ( solid ) lines and @xmath38 ( dashed ) propagators . every renormalised vertex must consist of 1 external line and at least 1 propagator , and drawn with a solid dot to show that the vertex is renormalised . the propagators can not have both ends attached to the same vertex . 2 . label the external legs with incoming momenta @xmath65 for @xmath66 and label the propagators with internal momenta @xmath67 for @xmath68 . label each end of each propagator with a field index @xmath69 . 3 . assign a factor @xmath107 to each vertex . the number of derivatives of @xmath105 corresponds to the number of propagators attached to each vertex . we use the convention that incoming momentum is positive . the @xmath71 function ensures momentum is conserved at each vertex . 4 . assign a factor @xmath72 to each propagator , where @xmath46 are the appropriate field indices that the propagator is labelled with at either vertex and @xmath73 is the momentum attached to the propagator . 5 . integrate over the propagator momenta , @xmath74 . the first @xmath75 integrations are trivial because of the @xmath71 functions but any further integrations ( in the case of a diagram with loop corrections ) can not in general be evaluated analytically . divide by the appropriate numerical factor . whenever @xmath49 propagators attach the same vertices at both ends this gives a factor of @xmath51 . add all permutations of the diagrams which is all of the distinct ways to relabel the @xmath65 attached to the external lines . the number of permutations depends on the symmetries of the diagram , a diagram with complete symmetry between all of the external lines has only one term , while for a diagram with no symmetries between the external lines there are @xmath53 permutations . we again calculate the power spectrum up to 2 loop level , this time with renormalised vertices . there is only one diagram at every loop level , we present the three lowest order diagrams in fig . [ fig : ren2ptfn ] , which correspond to p^_&= & _ a^ap(k ) + d^3q _ ab^abp(q)p(|-| ) + & & + d^3q_1 d^3q_2 _ abc^abcp(q_1)p(|-|)p(|-| ) . we calculate the bispectrum including the leading order loop corrections . the bispectrum is defined by [ bispectrum ] _ _ k_3b _ ( k_1,k_2,k_3 ) ^3 ( ) . at tree level there is one term and at 1 loop level there are two terms , which are shown diagrammatically by fig . [ fig : ren3ptfn1loop ] and these give the contribution @xmath108 if we do nt assume that the scalar fields are gaussian at the initial time ( e.g. hubble exit ) we have to take account of the higher @xmath0-point functions of the fields . this is likely to be important in models with a non - standard kinetic term where the non - gaussianity of the fields may be large at hubble exit @xcite , for example dbi inflation inspired from brane world models , @xcite . the connected 2- , 3- and 4-point function of the fields are defined in ( [ fieldcorrelators])([4ptfnfields ] ) . our analysis relies on the @xmath5-formalism which assumes that @xmath7 is a function of the field values @xmath109 but not the field derivatives independently as well . however we can still go beyond leading order in a slow - roll expansion and derive expressions which are valid to arbitrary order in slow roll , for example the different fields are in general correlated at hubble exit , at first order in slow roll , see @xcite . the diagrams are more complicated since while the 2-point function is still drawn as a propagator the higher @xmath0-point functions are drawn by an @xmath0-point vertex of dashed lines . the rules for the diagrams are now : 1 . draw all distinct connected diagrams with @xmath0-external ( solid ) lines and the appropriate number of internal correlators , which can be propagators ( 2-point functions ) or higher @xmath0-point correlators of the fields . the order of the connected @xmath0-point function of the field perturbations is @xmath75 , i.e. @xmath110 . there can not be more than one external line attached to each vertex . every dashed line must be attached to an external line on at least one end , if it is attached to an external line on both ends it is a propagator ( a correlator of order 2 ) . 2 . label the external lines with incoming momenta @xmath65 for @xmath66 and label each dashed line with internal momenta @xmath67 for @xmath68 . label each dashed line at a vertex containing an external line with a field index @xmath111 . 3 . assign a factor @xmath112 to each vertex which includes an external line , where the number of derivatives of @xmath7 corresponds to the number of dashed lines attached to that vertex . the @xmath71 function ensures momentum is conserved at each vertex . see fig . [ fig : fourierspacerulesng ] . 4 . assign a factor @xmath113 to each propagator , where @xmath46 are the appropriate field indices that the propagator is labelled with at either vertex and @xmath73 is the momentum attached to the propagator . see fig . [ fig : fourierspacerulesng ] . 5 . assign a factor @xmath114 to each vertex of three dashed lines and no external lines ( which corresponds to a 3-point function of the fields ) , where @xmath115 and @xmath116 are the field indices attached to the dashed lines where they meet a vertex with an external line . see fig . [ fig : fourierspacerulesng ] again . similar rules hold for vertices of dashed lines at higher order . integrate over all internal momenta , @xmath74 . for a tree diagram all of the integrations are trivial , but in the case of a diagram with loop corrections there will be some integrals which in general can not be evaluated analytically . divide by the appropriate numerical factor . whenever @xmath49 propagators or higher order correlators of the fields attach the same vertices ( and with the same number of lines to each vertex ) at both ends this gives a factor of @xmath51 . if @xmath117 legs from a higher order correlator are attached to the same vertex this gives an extra factor of @xmath118 . it follows that if @xmath49 correlators , each of order @xmath119 , all dress the same vertex then this gives a numerical factor of @xmath120 . see app . [ sec : app : ng ] for a proof of these rules and see fig . [ fig : ntng ] for two example of dressed vertices , the associated terms with the numerical factors are given in ( [ ntngeg1 ] ) , ( [ ntngeg2 ] ) . add all permutations of the diagrams which is all of the distinct ways to relabel the @xmath65 attached to the external lines . the number of permutations depends on the symmetries of the diagram , a diagram with complete symmetry between all of the external lines has only one term , while for a diagram with no symmetries between the external lines there are @xmath53 permutations . we give again the power spectrum defined by ( [ powerspectrumdefn ] ) , but this time for non - gaussian initial fields and valid to all orders in slow roll . there is one extra diagram at one loop level , and six extra diagrams at two loop level . because of momentum conservation , @xmath78 , it follows that @xmath121 . therefore even the non - symmetric diagrams which would have two permutations only correspond to a single term after enforcing this symmetry . the tree level term is given by @xmath122 notice that at leading order in slow roll using ( [ cfreefields ] ) this reduces to ( [ 2ptfntree ] ) . the one loop terms are given by p_^1(k)&=&d^3q ( 12n_abn_cdc^ac(q)c^bd(|-|)+ n_an_bcdc^ab(k)c^cd(q ) + & & + n_an_bcb^abc(k , q,|-| ) ) , where the extra one loop term which does not appear for gaussian fields is shown in fig . [ fig:2pt1loopng ] , and the first two terms were shown diagrammatically in fig . [ fig:2pt2loopg ] . going to 2 loops , terms of order @xmath81 there are ten terms , only the last four of which are non - zero for a gaussian field . diagrammatically the terms involving a 3-point or 4-point function of the fields are given in fig . [ fig:2pt2looptbng ] , which corresponds to the first six terms of [ 2ptfn2loopng ] p_^2&=&d^3q_1d^3q_2 ( 13n_an_bcdt^abcd(,-,- , ) . + & & + 14n_abn_cdt^abcd(,-,,- ) + & & + n_abn_cdeb^bde(|+|,|-|,|-|)c^ac(q_1 ) + 12n_abn_cdeb^abc(k , q_1,|+|)c^de(q_2 ) + & & + 13n_an_bcdeb^cde(q_1,|+|,q_2)c^ab(k ) + 12n_an_bcdeb^abc(q_1,|-|,k)c^de(q_2 ) + & & + 14n_an_bcdefc^ab(k)c^cd(q_1)c^ef(q_2 ) + 14n_abcn_defc^ad(k)c^bc(q_1)c^ef(q_2 ) + & & + 12n_abcdn_efc^ab(q_1)c^ce(|-|)c^df(q_2 ) + & & . + 16n_abcn_defc^ad(q_1)c^be(|-|)c^cf(|-| ) ) . the final four terms were already shown diagrammatically in fig . [ fig:2pt2loopg ] , they correspond to the last four terms of ( [ 2ptfn2loopng ] ) . however the mathematical expressions for these four diagrams given in ( [ 2ptfn2loop ] ) are correct at leading order in slow roll , while the expression above is valid to all orders in slow roll . note that 1 of the 1 loop diagrams and 4 of the 2 loop diagrams have dressed vertices , these can be removed by renormalising the vertices in a similar way to the case with gaussian fields , see app . [ sec : app : ng ] . the connected part of the primordial trispectrum is defined by @xmath123 diagrammatically all of the tree level terms for the 4-point function are given by fig . [ 4ptfntreenew ] . writing down these terms in order we get [ trispectrumfull ] t_^ ( , , , ) & = & n_an_bn_cn_d t^abcd ( , , , ) + & & + n_abn_c n_d n_e ( c^ac(k_1)b^bde(|+|,k_3,k_4)+ ( 11 perms ) ) + & & + n_abn_cdn_e n_f ( c^ac(|+|)c^bd(k_3)c^df(k_4 ) + ( 11 perms ) ) + & & + n_abcn_d n_e n_f ( c^ad(k_2)c^be(k_3)c^cf(k_4)+(3perms ) ) . this result was first derived in @xcite without using diagrams ( see also @xcite ) . note that all of the numerical coefficients are one , because all of the terms are tree level , and that only the final two terms are non - zero if the field fluctuations are gaussian . the real space diagrams for these two terms were given in fig . [ fig : realspace234pt ] . we have presented a diagrammatic method for calculating any @xmath0-point function of the primordial curvature perturbation , @xmath4 , at tree - level or any required loop level . rules are given for first drawing diagrams at the required order , and then for writing down the corresponding terms in the @xmath0-point function , in either real or fourier space . for example , we have drawn diagrams corresponding to all the 1- and 2-loop terms to the power spectrum and have given the corresponding corrections to the power spectrum . our method is based on using the @xmath5-formalism which identifies the local curvature perturbation with the perturbation of the integrated expansion from an initial hypersurface . the expansion on large scales ( typically larger than the hubble scale ) can be determined as a ( non - linear ) function of the initial local values of scalar fields during inflation using the homogeneous ( frw ) equations of motion in the separate universe approach . in practice we are working with fields that are smoothed on scales @xmath124 . in particular we neglect any ultraviolet divergence of the fields on small scales . we show that it is possible to renormalise vertices by absorbing an infinite sum of terms , corresponding to loop corrections to vertices . these are automatically accounted for if we work in terms of the average local expansion @xmath125 and its derivatives , defined in eq . ( [ nt ] ) for gaussian fields , instead of the background expansion , @xmath126 . this removes terms such as @xmath34 which is divergent for scale - invariant spectra , leaving all @xmath0-point functions finite in real space so long as @xmath125 and its derivatives are finite . however divergent terms remain in fourier space due to integrals over loop momenta between vertices which may diverge in the infrared . a discussion of the effect of the large scale cut off on observables is given in @xcite . results for the primordial @xmath0-point function are given in terms of derivatives of @xmath7 with respect to the fields at hubble exit and the @xmath0-point functions of the fields evaluated at hubble exit . in single - field models of inflation the curvature perturbation is constant on super hubble scales , so it is possible to calculate the derivatives of @xmath7 explicitly at hubble exit during inflation @xcite without needing to know the details of the end of inflation or reheating , etc . in this case both the bispectrum and trispectrum have been shown to be small , regardless of whether the fields are assumed to be gaussian or not , but assuming the field has a standard kinetic term in the action . during inflation the derivatives of @xmath7 can be calculated explicitly even in multiple field models for a separable potential , this been done to second @xcite and third order @xcite . several specific multiple field models have been studied in greater detail , e.g. @xmath127-flation @xcite and a model of multiple field inflation in which @xmath7 can be calculated exactly , @xcite . for all examples considered so far in the slow roll regime the non - gaussianity has proved to be small ( see also @xcite ) , however there are many models which explicitly generate a large non - gaussianity after inflation . examples include the curvaton scenario @xcite and modulated reheating @xcite , also preheating , for example with hybrid inflation @xcite . in general a numerical calculation of the derivatives of @xmath7 is required , the advantage of the @xmath5 formalism is that we only need to solve the homogenous background equations . if one assumes the fields are gaussian then only the 2-point function of the fields is required , but in general one requires the @xmath128-point function of the fields when working to @xmath38th order for any @xmath0-point function of @xmath4 . so far the 3- and 4-point functions have been calculated , both at leading order in slow roll and ignoring any quantum mechanical loop corrections . it is possible but not expected that these corrections may be important @xcite . in the most commonly considered case of slow - roll inflation , the field is extremely close to gaussian at hubble exit so it is a good approximation to assume that the field is gaussian and that any observable non - gaussianity is generated after hubble exit . methods to generate a large non - gaussianity during inflation include having a break in the potential , @xcite , or to have a non - standard kinetic term so that the non - gaussianity of the fields is large at hubble exit @xcite , for example dbi inflation inspired from brane world models , @xcite . the authors are grateful to david lyth and robert crittenden for comments . kk is supported by pparc ( stfc ) . ms is supported in part by jsps grants - in - aid for scientific research ( a ) no . 18204024 and ( b ) no . the authors acknowledge support by the jsps japan - uk collaboration program . here we provide a proof that renormalising vertices removes the diagrams with dressed vertices . in real space this is equivalent to removing terms with @xmath129 . we first provide a proof for the power spectrum where the notation is simpler . we then provide a proof for the general @xmath0-point function , initially assuming that the fields are gaussian at hubble exit , sec . [ sec : app : nptfn ] , and finally extend this proof to allow for arbitrary field correlations at hubble exit , sec . [ sec : app : ng ] . the gauge invariant curvature perturbation is defined by ( [ deltan ] ) as = n - n= _ p=1^ n__p ( ^a^b^c- ^a^b^c ) , where the subscript @xmath119 means that there are @xmath119 field indices above the brace . from the definition of @xmath130 , the 2-point function is given by [ app:2ptfn ] _ p , q=1^n__pn__q ( _ x^a_x^b_x^c-_x^a_x^b_x^c ) ( _ y^d_y^e_y^f- _ y^d_y^f ) . because the fields are gaussian every @xmath0-point function of the fields can be split into products of 2-point functions . we can choose the pairs to be connected pairs , i.e. each @xmath99 is evaluated at a different point or disconnected pairs which are @xmath129 . we consider every possible way of splitting the term @xmath131 into pairs : 1 connected pair : there are @xmath132 different ways to select this term . this gives & & _ p , q n__p n__q ^ad g(|x - y| ) _ x^b_x^c _ y^e_y^f = _ a^ag(|x - y| ) , + & & _ a= _ p=1^n_a_p-1^b^c . 2 connected pairs : there are @xmath133 ways to choose these two connected pairs . the factor of @xmath134 comes because the order in which we pick the pairs does not matter . this gives & & _ p , q n_ab_p-2 n_de_q-2 ^ad^be g^2(|x - y| ) _ y^f + & = & _ ab^abg^2(|x - y| ) . continuing this for every possible number of connected pairs , we find that ( [ app:2ptfn ] ) is equivalent to _ x_y= _ r=1^__r^abcg^r(|x - y| ) . similar to the calculation of the 2-point function , we have to split the term @xmath135 into pairs in every possible way . in the case that some of the @xmath136 do not appear in a connected pair there is no contribution to the @xmath0-point function , for example if @xmath137 terms do not appear in any connected terms then the last bracket of ( [ app : nptfn ] ) is @xmath138 . we are left with [ app : reducednptfn ] _ x_y_z&= & _ _ n n__pn__q n__r + & & _ x^a_x^b_x^c _ z^g_z^h_z^i_cc , where the subscript @xmath139 to the angular brackets means that they must be decomposed into pairs such that every @xmath140 appears in at least one connected pair . we introduce the notation that @xmath141 is the number of connected pairs of @xmath142 and @xmath143 and similarly for every other possible connected pair . furthermore @xmath144 is the number of @xmath145 that form part of a connected pair , so @xmath146 . note that we require @xmath147 . we consider a term in ( [ app : reducednptfn ] ) from an arbitrary choice of @xmath148 . the number of ways to choose the @xmath149 connected pairs of @xmath150 correlators is given by @xmath151 , then the number of ways to choose the @xmath152 connected correlators linking @xmath142 and @xmath153 is given by @xmath154 . continuing this counting for all the pairs , we find the number of ways to split into the required number of connected pairs is . this choice of @xmath155 therefore gives the following term to ( [ app : nptfn ] ) , & & _ p , q, ,r n_abcn_def n_ghi ^ad^be + & & g^l_xy(|x - y|)g^l_yz(|y - z|)_p - n_x_r - n_z + & = & _ _ n_x__n_z^ad^beg^l_xy(|x - y|)g^l_yz(|y - z| ) , where there are @xmath149 contraction of field indices of the first two factors of @xmath104 etc . repeating this for all allowed choices of @xmath156 we get _ x_y_z= _ l_xy,,l_yz _ _ n_x__n_z^ad^beg^l_xy(|x - y|)g^l_yz(|y - z| ) . diagrammatically a term @xmath157 corresponds to @xmath149 propagators which all have both ends attached to the same spatial points @xmath142 and @xmath143 . this explains the numerical factor of @xmath158 , as given by rule 5 in real space ( or equivalently rule 6 in fourier space ) . here we extend the proof that renormalising vertices eliminates the diagrams with dressed vertices also holds for non - gaussian fields . by doing this we also derive the numerical factors that are associated with every diagram . the proof is more complex than before because we must consider every possible correlation of an arbitrary number of fields , rather than reducing every @xmath0-point correlation to products of 2-point functions . we now have [ app : generalisednt ] _ abc&= & n_abc+ n_abcde^d^e+n_abcdef ^d^e^f + & & + n_abcdefg ^d^e^f^g+ . the correlators here includes the connected and disconnected parts . again the @xmath0-point function of @xmath130 is given by ( [ app : nptfn ] ) and we can reduce it to ( [ app : reducednptfn ] ) in a similar way to the case with gaussian fields . however we now have to decompose the term in angle brackets into connected correlators of arbitrary order in every possible way . we extend the notation that @xmath149 is the number of terms like @xmath159 to more general correlators like @xmath160 is the number of terms like @xmath161 . every @xmath49 must have at least 2 distinct @xmath39 in the subscript . we define @xmath144 as the number of @xmath142 s contained in the correlators and again we require that @xmath162 and similarly for @xmath163 , etc . we consider the term in ( [ app : reducednptfn ] ) coming from a particular choice of @xmath164 , which we denote by p^abcefgh & = & _ l_xy _ l_xxz + & = & _ p_w _ c^l_xxzz and the product on the last line is over all chosen correlators of the fields and we have also dropped the field indices on this line . we can repeat the calculation for the number of ways to split the total correlator of ( [ app : reducednptfn ] ) into the appropriate connected pairs . the only new point is that whenever an @xmath39 appears more then once in the same @xmath165 the order in which we pick those @xmath142 s does not matter , so we have to divide by the factorial of the number of @xmath142 s raised to the power @xmath165 . in total we get , where the product in the denominator is over all @xmath49 s and the number of @xmath39 s repeated in each individual @xmath49 . therefore this choice of @xmath49 s gives the following contribution to ( [ app : nptfn ] ) , & & _ p , q,,r n_abcn_def n_ghi + & & p^abgh _ p - n_x_r - n_z + & = & _ _ n_x__n_zp^abgh . repeating this for all allowed choices of the @xmath49 s this gives [ app : nptfnng ] _ x_y_z= _ l_xy,,l_xx z _ _ n_x__n_zp^abi . ( [ app : nptfnng ] ) explains the numerical factor of rule 7 in sec . [ sec : ng ] , except for the case of dressed vertices . the numerical factor for dressed vertices requires a calculation of the number of ways to reduce the @xmath0-point function in ( [ app : generalisednt ] ) into connected pairs of the required order . if @xmath49 correlators of order @xmath0 each dress the same vertex then we need to know the number of ways to split the @xmath166-point function into @xmath49 lots of connected @xmath0-point functions . the result is ^lnc_n^n(l-1)c_n^2nc_n = , where the denominator of @xmath51 arises because the order in which we pick the @xmath0-point functions does not matter . the factor of @xmath167 is cancelled by the numerical factor in ( [ app : generalisednt ] ) . the generalisation to more complicated dressed vertices is straightforward to calculate , for example the number of ways to split an @xmath168-point function into @xmath169 lots of connected @xmath170-point functions and @xmath171 lots of connected @xmath172-point functions is . in fig . [ fig : ntng ] we give two examples of these rules , both of which are 4 loop terms of the dressed vertex of @xmath173 , the first diagram has @xmath174 and @xmath175 while the second diagram has @xmath176 and @xmath177 . the associated mathematical expressions are & & [ ntngeg1 ] n_abcdefg^b^c^d^e^f^g , + & & n_abcdefgh^b^c^d^e^f^g^h . [ ntngeg2 ] v. acquaviva , n. bartolo , s. matarrese and a. riotto , nucl . b * 667 * , 119 ( 2003 ) [ arxiv : astro - ph/0209156 ] . j. m. maldacena , jhep * 0305 * , 013 ( 2003 ) [ arxiv : astro - ph/0210603 ] . b. a. bassett , s. tsujikawa and d. wands , rev . phys . * 78 * , 537 ( 2006 ) [ arxiv : astro - ph/0507632 ] . k. enqvist and m. s. sloth , nucl . b * 626 * , 395 ( 2002 ) [ arxiv : hep - ph/0109214 ] ; d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) [ arxiv : hep - ph/0110002 ] ; t. moroi and t. takahashi , phys . lett . b * 522 * , 215 ( 2001 ) [ erratum - ibid . b * 539 * , 303 ( 2002 ) ] [ arxiv : hep - ph/0110096 ] ; k. enqvist and s. nurmi , jcap * 0510 * , 013 ( 2005 ) [ arxiv : astro - ph/0508573 ] ; k. a. malik and d. h. lyth , jcap * 0609 * , 008 ( 2006 ) [ arxiv : astro - ph/0604387 ] ; m. sasaki , j. valiviita and d. wands , phys . d * 74 * , 103003 ( 2006 ) [ arxiv : astro - ph/0607627 ] . e. silverstein and d. tong , phys . rev . d * 70 * , 103505 ( 2004 ) [ arxiv : hep - th/0310221 ] ; m. alishahiha , e. silverstein and d. tong , phys . d * 70 * , 123505 ( 2004 ) [ arxiv : hep - th/0404084 ] ; m. x. huang and g. shiu , phys . d * 74 * , 121301 ( 2006 ) [ arxiv : hep - th/0610235 ] . l. verde , l. m. wang , a. heavens and m. kamionkowski , mon . not . soc . * 313 * , l141 ( 2000 ) [ arxiv : astro - ph/9906301 ] . d. h. lyth , k. a. malik and m. sasaki , jcap * 0505 * , 004 ( 2005 ) [ arxiv : astro - ph/0411220 ] . d. h. lyth and y. rodriguez , phys . lett . * 95 * , 121302 ( 2005 ) [ arxiv : astro - ph/0504045 ] . d. s. salopek and j. r. bond , phys . d * 42 * , 3936 ( 1990 ) . m. sasaki and t. tanaka , prog . * 99 * , 763 ( 1998 ) [ arxiv : gr - qc/9801017 ] . d. wands , k. a. malik , d. h. lyth and a. r. liddle , phys . d * 62 * , 043527 ( 2000 ) [ arxiv : astro - ph/0003278 ] . r. scoccimarro and j. frieman , astrophys . j. suppl . * 105 * , 37 ( 1996 ) [ arxiv : astro - ph/9509047 ] . m. crocce and r. scoccimarro , phys . d * 73 * , 063519 ( 2006 ) [ arxiv : astro - ph/0509418 ] . i. zaballa , y. rodriguez and d. h. lyth , jcap * 0606 * , 013 ( 2006 ) [ arxiv : astro - ph/0603534 ] . l. boubekeur and d. h. lyth , phys . d * 73 * , 021301 ( 2006 ) [ arxiv : astro - ph/0504046 ] . m. musso , arxiv : hep - th/0611258 . m. x. huang , g. shiu and b. underwood , arxiv:0709.3299 [ hep - th ] . d. seery and j. e. lidsey , jcap * 0509 * , 011 ( 2005 ) [ arxiv : astro - ph/0506056 ] . d. seery , j. e. lidsey and m. s. sloth , jcap * 0701 * , 027 ( 2007 ) [ arxiv : astro - ph/0610210 ] . b. j. w. van tent , class . quant . * 21 * , 349 ( 2004 ) [ arxiv : astro - ph/0307048 ] . c. t. byrnes and d. wands , phys . d * 74 * , 043529 ( 2006 ) [ arxiv : astro - ph/0605679 ] . p. r. jarnhus and m. s. sloth , arxiv:0709.2708 [ hep - th ] . d. h. lyth and i. zaballa , jcap * 0510 * , 005 ( 2005 ) [ arxiv : astro - ph/0507608 ] . d. binosi and l. theussl , comput . commun . * 161 * , 76 ( 2004 ) [ arxiv : hep - ph/0309015 ] . d. h. lyth , arxiv:0707.0361 [ astro - ph ] . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 170 * , 377 ( 2007 ) [ arxiv : astro - ph/0603449 ] . m. sasaki , j. valiviita and d. wands , phys . d * 74 * , 103003 ( 2006 ) [ arxiv : astro - ph/0607627 ] . d. seery , arxiv:0707.3377 [ astro - ph ] . f. vernizzi and d. wands , jcap * 0605 * , 019 ( 2006 ) [ arxiv : astro - ph/0603799 ] . t. battefeld and r. easther , jcap * 0703 * , 020 ( 2007 ) [ arxiv : astro - ph/0610296 ] . k. y. choi , l. m. h. hall and c. van de bruck , jcap * 0702 * , 029 ( 2007 ) [ arxiv : astro - ph/0701247 ] . s. a. kim and a. r. liddle , phys . d * 74 * , 063522 ( 2006 ) [ arxiv : astro - ph/0608186 ] . d. battefeld and t. battefeld , jcap * 0705 * , 012 ( 2007 ) [ arxiv : hep - th/0703012 ] . m. sasaki , class . quant . * 24 * , 2433 ( 2007 ) [ arxiv : astro - ph/0702182 ] . l. alabidi and d. h. lyth , jcap * 0605 * , 016 ( 2006 ) [ arxiv : astro - ph/0510441 ] . g. i. rigopoulos , e. p. s. shellard and b. j. w. van tent , arxiv : astro - ph/0511041 . s. yokoyama , t. suyama and t. tanaka , arxiv:0705.3178 [ astro - ph ] . g. dvali , a. gruzinov and m. zaldarriaga , phys . d * 69 * , 023505 ( 2004 ) [ arxiv : astro - ph/0303591 ] ; l. kofman , arxiv : astro - ph/0303614 . n. barnaby and j. m. cline , phys . d * 73 * , 106012 ( 2006 ) [ arxiv : astro - ph/0601481 ] ; n. barnaby and j. m. cline , phys . d * 75 * , 086004 ( 2007 ) [ arxiv : astro - ph/0611750 ] . s. weinberg , phys . d * 74 * , 023508 ( 2006 ) [ arxiv : hep - th/0605244 ] . x. chen , m. x. huang , s. kachru and g. shiu , jcap * 0701 * , 002 ( 2007 ) [ arxiv : hep - th/0605045 ] .

we present and analyze the orbits of eight nearby planetary nebulae ( pns ) using two different galactic models . the errors of the derived orbital parameters are determined with a monte carlo method . based on the derived orbital parameters , we find that sh 2 - 216 , deht 5 , ngc 7293 , a21 , and ton 320 belong to the thin - disk population , and pg 1034 + 001 and a31 belong to the thick - disk population . puwe 1 probably belongs to the thick - disk population , but its population classification is very uncertain due to the large errors of its derived orbital parameters . the pn - ism interactions are observed for the eight pns in our sample . the position angles of the proper motions of the pns are consistent with the directions of the pn - ism interaction regions . the kinematic ages of pns are much smaller than the time for them to cross the galactic plane . using the models of borkowski et al . and soker et al . , the pn - ism interaction can be used to derive the local density of ism in the vicinity of evolved pns . according to the three - dimensional hydrodynamic simulations of wareing et al . ( wzo ) , sh 2 - 216 , a21 , and ton 320 are in the wzo 3 stage , pg 1034 + 001 and ngc 7293 are in the wzo 1 stage , and puwe 1 is in the wzo 2 stage .

1102.1309

planetary nebulae ( pns ) are the descendants of asymptotic giant branch ( agb ) stars . according to the interacting stellar wind ( isw ) model @xcite , an agb star loses most of its envelope mass with a slow stellar wind ( agb wind ) until the hot core is exposed . after reaching the pn nucleus stage , a new faster wind ( post - agb wind ) sweeps up the circumstellar material originating in the slow wind and forms the high - density and ionized nebular shell . most pns are not spherical and exhibit complex shapes : round , elliptical , butterfly , bilobed , and irregular . stellar rotation , magnetic field , and binarity are important ingredients in shaping pns @xcite . the interaction of a moving pn with the interstellar medium ( ism ) was first discussed by @xcite . * hereafter bss ) examined the morphology of nine pns with a center star ( cs ) whose proper motions exceeded 0.015 arcsec yr@xmath0 and found that seven of the examined pns interact with the ism . they also examined the morphology of large , nearby pns , and found that many were asymmetric . @xcite presented a ccd imaging atlas of large pns ( @xmath1 ) and found that 21 of 27 pns in their sample revealed significant interaction with the ism . using deep narrowband ccd images of the nine largest , nearby pns , @xcite found that eight pns in their sample revealed features indicating an interaction between the pns and the ism . more recently , @xcite presented mid - infrared images for five pns which appear to interact with the ism . using the data of the int photometric h@xmath2 survey ( iphas ) @xcite , @xcite identified 21 objects as possible pns interacting with the ism . on the theoretical side , the first theoretical study of the interaction between pns and ism was made by @xcite . using a thin - shell approximation and simple analytic formulae , @xcite found that the motion of a pn through the ism would be expected to cause a shift of the cs toward the leading surface of the interacting pn and ism . bss used simple analytical estimates to study the interaction between a moving pn and the ism and found that the interaction produces asymmetries in the surface brightness of the pn . using a two - dimensional numerical hydrodynamic code , ( * ? ? ? * hereafter sbs ) simulated the interaction between the pn and the ism . they considered two limiting cases : adiabatic and isothermal models . in their adiabatic model , a pn moves with large relative velocity ( @xmath3 km s@xmath0 ) within the ambient ism with low density so that the cooling time is larger than the flow time . the adiabatic model is relevant for pns moving in the galactic halo . in the isothermal model , which is relevant for pns in the galactic disk , the pn moves through the high - density ism with moderate relative velocity @xmath4 km s@xmath0 and the cooling is very effective . the isothermal model of sbs validates the thin - shell approximation originally developed by @xcite . sbs also found that the simple analytic model breaks down in their adiabatic model in which a rayleigh - taylor ( rt ) instability develops , leading to shell fragmentation , while the isothermal model shows no sign of the rt instability . @xcite studied the interaction of a pn with a magnetized ism and showed that the ism magnetic field , which was not incorporated in the numerical simulations of sbs , can lead to the development of rt instability even in the isothermal environment . according to the simple thin - shell model for the pns moving through the ism , the interaction between pns and ism can be divided into three phases @xcite . first , when the density of the young pn is much larger than that of the ambient ism , the pn will expand freely . second , when the pn density drops below a critical value , the ism pressure becomes comparable to the one in the pn shell and the upstream side of the pn shell will be compressed and the surface brightness in the interaction region will increase . when the density in the pn shell has dropped to a lower value , the shape of the pn shell will be significantly distorted and the cs of the pn will move close to the upstream edge . finally , the cs of the pn will move outside of the pn shell , and the pn shell will be stripped and mixed with the ism . in the pn - ism interaction models of bss and sbs , they have assumed that the pn shell has formed before the interaction with the ism and the interaction only be observable during the late stage of the pn evolution . moreover , high ism density , large relative velocity , or both explain the observed asymmetry in the external shell of a pn . @xcite performed two - dimensional numerical simulations of the evolution of an agb star moving through its surrounding ism . they found that during the early stage of the evolution of an agb star , the slow agb wind can interact with the ism . the pn forms later when the fast post - agb wind sweeps up the stellar material in the slow agb wind . a spherical innermost shell shaped by the fast post - agb wind and an external shell with a bow shock character can form . in simulations @xcite adopted two low ism densities and a very conservative relative velocity of the pn , 20 km s@xmath0 . thus , high ism density , large relative velocity , and a magnetic field are not necessary to explain the observed asymmetries in the external shells of pns . moreover , interaction can occur even during early agb evolution , which is in agreement with the observed common asymmetries relating to the interaction with ism in most of the halos of pns , not just in those of old pns . considering a wide range of pn velocities , mass - loss rates of the cs and ism densities , ( * hereafter wzo ) performed three - dimensional hydrodynamic simulations , from the beginning of the slow agb wind phase to the end of the post - agb / pn phase . they developed a `` triple - wind '' model including a slow agb wind , a fast post - agb wind , and a third wind reflecting the movement through the ism . wzo found that the ism interaction strongly affects the outer structures of pns . the pn - ism interaction process can be divided into four stages ( wzo 1 4 ) . in the first stage ( wzo 1 ) , the pn expands freely within the bubble of the undisturbed slow agb wind material and is unaffected by the agb wind - ism interaction . in the second stage ( wzo 2 ) , the pns interact with the bow shock formed during the agb wind - ism interaction process , and the density and the emission in the upstream side increase accordingly . the third stage ( wzo 3 ) is defined by the geometric center of the pn moving downstream away from the cs . in the fourth and final stage ( wzo 4 ) , the pn is completely disrupted and the cs is outside the pn . the relative velocity of the pn and the ism density are important parameters for us to consider in order to model and understand the pn - ism interaction processes . both of these parameters are relevant to the galactic orbits of pns . the orbital parameters can also be used to distinguish different populations in the galaxy @xcite , but the available orbital parameters for pns in the galaxy are very rare . @xcite calculated the orbit of pg @xmath5 and found that this pn is in an advanced stage of interaction with the ism . @xcite calculated the orbits of four pns and confirmed that the motion of the pn shell through the ism is the origin of the pn - ism interaction process . in this research , using improved observation data , we calculated the galactic orbits for eight nearby pns , which doubles the number of pns whose orbits are available . using the derived orbital parameters of those pns , we also discuss the pn - ism interaction processes in our sample in detail . we present the derived galactic orbits of our pn sample in section 2 . in section 3 , for each pn in our sample , using the derived kinematical data , we discuss the interaction between the pn and the ism . our conclusions are given in section 4 . the position , distance , absolute proper motion , and radial velocity are essential data for deriving the galactic orbit of a pn . first , we searched the literature to find pns whose distance data are available . there are many indirect methods of pn distance determination . on the one hand , statistical distances can be underestimates . on the other hand , spectroscopic distances can be overestimates @xcite . we consider only pns whose distances are derived from direct parallax measurements . @xcite presented trigonometric parallaxes of the css of 16 nearby pns . the median error in their derived parallaxes is 0.42 mas which is smaller than the @xmath6 mas typical error for most _ hipparcos _ parallaxes @xcite . using the _ hubble space telescope _ , the absolute parallaxes of the css of four pns in the sample of @xcite were also derived by @xcite . the derived errors in @xcite are 2 3 times smaller than those in @xcite . based on a large number of multi - wavelength data , @xcite found that the nebula around the cs phl 932 is a small region and not a real pn . rejecting phl 932 from the sample of @xcite , we searched the proper motion and radial velocity data for the other 15 pns in the literature . using available all - sky astrometric catalogs , @xcite collected reliable proper motion information ( with significance better than @xmath7 in at least one component ) for a sample of 234 pns . five pns in the sample of @xcite are not listed in the catalog of @xcite : ngc 6853 , re 1738 + 665 , pg 1034 + 001 , a7 , and a24 . we used the vizier catalog access tool to search the proper motion data for these five pns . no proper motion data are available for ngc 6853 , re 1738 + 665 , and a24 . we found a proper motion for a7 in the ucac 2 catalog @xcite , but the errors are three times larger than the proper motions . we found a proper motion for pg 1034 + 001 in the ucac 3 catalog @xcite , but the proper motions are three times larger than their errors in both components . so , pg 1034 + 001 is included in the following analysis . for the 10 pns included in the catalog of @xcite , proper motions derived from different astrometric catalogs are listed . those catalogs include _ hipparcos _ @xcite , _ tycho 2 _ @xcite , ucac 2 @xcite , gsc 2 @xcite , and usno - b @xcite . the former four catalogs are all based on the international celestial reference system ( icrs ) , but the proper motions of usno - b were derived relative to the ys4.0 catalog and are not absolute . we do not adopt the data taken from the usno - b catalog . the ucac 3 catalog was made available in 2009 . in addition to the early epoch data used for the ucac 2 proper motions , ucac 3 utilizes data from about 5000 unpublished astrographic plates as well as a complete re - reduction of the spm first epoch data to significantly lower the systematic errors of ucac 3 proper motions @xcite . the proper motions in ucac 3 are improved with respect to those in _ hipparcos _ , _ tycho 2 _ , and ucac 2 . thus , for cs with @xmath8 mag in the sample of @xcite , the proper motions from ucac 3 are adopted . for stars with @xmath9 mag , there are larger system errors in the proper motions of ucac 3 . so , for cs with @xmath9 mag in the sample of @xcite , the proper motions derived from the gsc 2 catalog and listed in @xcite are adopted , but , the proper motions of ngc 6270 derived from ucac 2 are very different from those derived from gsc 2 , and no data can be found in the ucac 3 catalog for this object . we rejected this pn in our analysis . the simbad database was used to search the radial velocity data for the 10 pns with known distances and proper motions . in the end , eight pns with reliable distances , proper motions , and radial velocities were included in our final sample . in table [ tb1 ] the positions ( @xmath2 , @xmath10 ) of each pn adopted from @xcite are listed in columns 2 and 3 , the @xmath11 magnitudes of the css adopted from @xcite are listed in column 4 , column 5 lists the derived parallaxes from @xcite and @xcite , proper motions ( @xmath12 , @xmath13 ) taken from the ucac 3 and gsc 2 catalogs are listed in columns 6 and 7 , the position angles ( p.a.s ) of proper motions are listed in column 8 , and radial velocities and their references are listed in columns 9 and 10 . in table [ tb21 ] the angular diameters of the eight pns , except for pg 1034 + 001 , are taken from @xcite and are listed in column 2 . using the distances listed in column 4 , the linear diameters of these pns were derived and are listed in column 3 . the electron densities , @xmath14 , of pns , if available , are listed in column 5 . the volume densities , @xmath15 , of the diffuse warm neutral medium ( wnm ) and warm ionized medium ( wim ) at the observed positions of each pn were calculated according to the model of @xcite and are listed in columns 6 and 7 . the expansion velocities of each pn , if available are listed in column 8 , and the corresponding references are listed in the final column . using the linear diameter and expansion velocity , the kinematic age for each pn was calculated and is listed in column 9 . @xcite used a hydrostatic galactic model , which assume that the interstellar magnetic field , gas , and cosmic rays form a dynamic equilibrium with the galactic gravitational potential . they considered various galactic gas phases in their model ; these gases include cold neutral medium ( cnm ) , wnm , wim , neutral halo medium ( nhm ) , and hot halo medium ( hhm ) . because the probability of a pn - ism interaction occurring for cnm is very small @xcite , and all of the pns in our sample belong to the galactic disk population ( see the next section ) , where the densities of nhm and hhm are much lower , these two gases need not be considered in the pn - ism interaction for disk pns . hence we calculated the density for wnm and wim only with the model of @xcite . table [ tb21 ] indicates that the total densities , @xmath15 , of wnm and wim around the eight pns are close to @xmath16 0.1 @xmath17 . using the data listed in table [ tb1 ] and adopting a solar motion ( @xmath18 , @xmath11 , @xmath19)@xmath20 ( 10.0 , 5.2 , 7.2 ) km s@xmath0 @xcite , the local standard of rest ( lsr ) velocities of pns were determined . the lsr velocities were then corrected to the galactic standard of rest ( gsr ) by adopting the galactocentric distance of the sun @xmath21 kpc and a rotation velocity of the lsr at the position of the sun of 220 km s@xmath0 ( * ? ? ? * and reference therein ) . the observed gsr positions and velocities of the eight nearby pns are listed in table [ tb2 ] . the position coordinates @xmath22 , @xmath23 , @xmath24 and velocity components @xmath18 , @xmath11 , @xmath19 refer to a galactocentric righthanded cartesian coordinate system with the directions pointing to the galactic center , direction of rotation , and the galactic north pole . the errors in the velocity components listed in table [ tb2 ] include the errors in distances , proper motions , and radial velocities of pns listed in table [ tb1 ] . the lsr velocities of the pns are listed in the last column . we employed the axisymmetric galactic gravitational potential models of ( * ? ? ? * hereafter as91 ) and * hereafter fsc96 ) . these two models consist of three components : bulge , disk , and spherical halo . both models admit two conserved quantities : the total energy @xmath25 and the @xmath24-component @xmath26 of the angular momentum vector . the disks are very different between the two models @xcite . using the initial conditions listed in table [ tb2 ] , the orbits of the pns were calculated backward in a time interval of 2 gyr with the two given gravitational potential models . the bulirsch - stoer algorithm of @xcite was used for the integration . the relative change in the total energy over the 2 gyr integration time is of the order of @xmath27 . the orbital parameters @xmath28 , @xmath29 , and @xmath30 calculated with the as91 and fsc96 models are listed in table [ tb3 ] . @xmath31 is the orbital eccentricity , where @xmath32 and @xmath33 are apogalactic and perigalactic distances from the galactic center , respectively , which are determined from the average maximum and minimum galactocentric distances in the calculated orbits within the integration time of 2 gyr . @xmath29 is the average maximum vertical distance above the galactic plane @xcite . @xmath30 is the mean time interval for the pn to cross the galactic plane from one @xmath34 to the other in the opposite direction @xcite . the @xmath24-component @xmath26 of the angular momentum vector of each pns is also listed in table [ tb3 ] . for all eight pns in our sample , tables [ tb21 ] and [ tb3 ] indicate that the mean time @xmath30 for each pn is much larger than its kinematic age . the errors of orbital parameters listed in table [ tb3 ] were calculated based on a monte carlo method . because the intrinsic uncertainties of the orbits within the 2 gyr integration interval are very small , only the errors in the input data were used to calculate the errors of the derived orbital parameters @xcite . for each pn , the initial condition was generated by adding gaussian deviates to the observed proper motion , radial velocity , and distance . the errors for the input data listed in table [ tb2 ] are adopted as the standard deviations . for each pn , the errors for its orbital parameters were calculated based on 1000 separate integrations . the average difference for the orbital parameter @xmath28 , @xmath29 , and @xmath30 derived from the as91 and fsc96 models are @xmath35 , @xmath36 pc , and @xmath37 myr , respectively . table [ tb3 ] shows that , the relative differences in the derived orbital parameters due to the adopted different potential models are smaller than those derived from observational errors . the relative large differences in @xmath29 and @xmath30 for the two models are due to the fact that the mass of the disk in the fsc96 model is smaller than that in the as91 model @xcite . the orbits calculated with the as91 and fsc96 models in the time interval of 2 gyr are presented in figure [ fg1 ] , where @xmath38 . for all pns in our sample , their orbits projected on the galactic plane in figure [ fg1 ] clearly indicate the periodic motions of pns in the plane . for most of the pns , the meridional orbits in figure [ fg1 ] are of a boxy - like type and pns move in the meridional plane within the limited regions almost filling the boxes . however , the meridional orbits of sh 2 - 216 and puwe 1 in the as91 model and that of pg1034 + 001 in the fsc96 model indicate that they only move with special paths within the meridional plane . thus , the weird orbits of sh 2 - 216 and puwe 1 derived from the as91 model and that of pg1034 + 001 derived from the fsc96 model are not used in the following analysis . these weird orbits are caused by the galactic models adopted . thus , we need to check each model for every object in detail , which is beyond the scope of this paper . this problem is under investigation and will be discussed in a subsequent paper . @xcite re - examined the velocity of the sun with respect to the lsr . they found that the classical determination of its component @xmath39 in the direction of galactic rotation is undermined by the metallicity gradient in the disk . using a chemodynamical model which accounts for these effects , they obtained ( @xmath18 , @xmath11 , @xmath19)@xmath20 ( 11.1 , 12.24 , 7.25 ) km s@xmath0 . in particular , @xmath39 is 7 km s@xmath0 larger than that of @xcite adopted in this study . using the new parameters of the sun derived by @xcite , we re - calculated the orbits of the eight pns in our sample with the as91 and fsc96 models . using the as91 model , the mean differences for the orbital parameters @xmath28 , @xmath29 , and @xmath30 derived with the different parameters of the sun @xcite are @xmath40 , @xmath41 pc , and @xmath42 myr , respectively . the corresponding mean differences for @xmath28 , @xmath29 , and @xmath30 with the fsc96 model are @xmath43 , @xmath44 pc , and @xmath45 myr , respectively . the effects of adopting different parameters of the sun are very small for the derived orbital parameters with different galactic models . @xcite analyzed the orbital eccentricity distributions for different galactic populations and found that no thin - disk star has an eccentricity greater than 0.3 and no thick - disk star has an eccentricity less than 0.1 . the orbital eccentricities of the thin - disk and the thick - disk stars overlap between 0.1 and 0.3 . the mean @xmath28 for the thin - disk stars is @xmath46 and the mean for thick - disk stars is @xmath47 . in our sample , only sh 2 - 216 has @xmath28 less than 0.1 which indicates it is a thin - disk pn . it is difficult to assign the population membership for the other seven pns based only on their orbital parameter @xmath28 . @xcite presented kinematic criteria for distinguishing thin - disk , thick - disk , and halo populations in their da white dwarf sample . their kinematic criteria were deduced from a sample of f and g stars chosen from the literature , for which chemical criteria can be used to distinguish different populations . their kinematic population classification scheme is based on the position in the @xmath48 velocity diagram , the position in the @xmath49 diagram , and the galactic orbits themselves . among these three kinematic criteria , the @xmath49 diagram can distinguish different populations more clearly ( comparing figures 2 and 3 of @xcite ) . in the @xmath49 diagram of @xcite , the thin - disk stars lie in a v - shaped area with @xmath50 and @xmath26 around 1800 kpc km s@xmath0 . the thick - disk stars possess @xmath51 and @xmath52 kpc km s@xmath0 . because of our adopted right - handed orientation of the coordinate system , the direction of @xmath26 vector is opposite to that of @xcite . we present the @xmath53 diagram in figure [ fg2 ] , the adopted orbital parameters @xmath28 are derived from the as91 model . because of the weird orbits of sh 2 - 216 and puwe 1 in the as91 model , their parameters @xmath28 derived from the fsc96 model are adopted in figure [ fg2 ] . the error of @xmath28 for each pn in figure [ fg2 ] includes the error due to the uncertainty in the input data and that derived with different galactic models . due to the weird orbits of sh 2 - 216 , puwe 1 , and pg1034 + 001 in different models , their errors of @xmath28 in different models can not be confirmed . thus , only errors derived from the input data are considered for the three above - mentioned pns . in figure [ fg2 ] , dashed lines divide this figure into two regions for different populations . the top left region enclosed by these two dashed lines is the one for the thin - disk population . the other region is occupied by the thick - disk population . figure [ fg2 ] shows that pg 1034 + 001 and a31 lie in the region of the thick - disk population and sh 2 - 216 , deht 5 , ngc 7293 , a21 , and ton 320 belong to the thin - disk population . puwe 1 has low @xmath54 and @xmath55 which indicates it likely belongs to the thick - disk population , but the large errors in its orbital parameters make its population classification more uncertain . the orbit of sh 2 - 216 was first calculated by @xcite with the as91 model . using the method of @xcite , @xcite identified sh 2 - 216 as a thin - disk pn . they used the trigonometrical distance derived by @xcite which is consistent with that derived by @xcite . the proper motions ( @xmath56 , @xmath57 ) @xmath58 ( 21.7 , @xmath59 ) mas yr@xmath0 used by @xcite were taken from the ucac 2 catalog which are also consistent with those taken from the ucac 3 catalog . @xcite used the radial velocity of 11.9 km s @xmath0 derived by @xcite which is half of that derived by @xcite . @xcite have pointed out that the result of @xcite may be less accurate because the cs of this pn was possibly not well centered in the observation aperture . using the radial velocity of @xcite , we re - calculated the orbit of sh 2 - 216 in the as91 model , the orbital parameters @xmath28 , @xmath29 , and @xmath30 are derived as 0.03 , 142 pc , and 33.3 myr , respectively . the adopted different radial velocities of sh 2 - 216 do not affect the derived orbital parameters with the same galactic model . using the orbital parameters derived with the radial velocity of @xcite , based on its position in figure [ fg2 ] , sh 2 - 216 is still identified as a thin - disk pn , which is also consistent with that of @xcite . therefore , using the radial velocity of @xcite will not affect the population membership determination of sh 2 - 216 . the interaction between sh 2 - 216 and the ism was studied by @xcite in detail . a region of enhanced emission in the east of the pn and the displacement of the cs with respect to the center of the pn clearly indicate the pn - ism interaction . the p.a . of the proper motions of sh 2 - 216 also indicates that the direction of the movement of the pn is consistent with the position of the pn - ism interaction region . using the kinematic age 0.45 myr and the @xmath19 velocity of sh 2 - 216 listed in table [ tb2 ] , this pn only moved @xmath60 pc away from its birthplace in the vertical direction . we have pointed out in the previous section that the kinematic age of this pn is much smaller than the mean time @xmath30 for it to cross the galactic plane . assuming that the ism in the vicinity of the sun has the same rotation velocity of the lsr , the lsr velocities of the pns listed in table [ tb2 ] can be considered as the approximately relative velocities of those pns with respect to their nearby ism . using a relative velocity of @xmath16 20 km s@xmath0 , with an estimated electron density of @xmath61 8 @xmath17 @xcite and an expansion velocity of @xmath16 4 km s@xmath0 @xcite , the models of bss and sbs should need a local ism density of @xmath62 1.4 @xmath17 to explain the observed pn - ism interaction of sh 2 - 216 . according to the model of @xcite , the density of the diffuse ism at the position of sh 2 - 216 is 0.12 @xmath17 is much lower than that estimated from the models of bss and sbs . @xcite estimated a relative velocity of only @xmath16 5 km s@xmath0 ; an unusually high ism density of @xmath62 4 @xmath17 can be estimated with this much lower relative velocity . if we use the radial velocity of @xcite , based on the models of bss and sbs , the local ism density can be derived as 2.2 @xmath17 , which is still much larger than that derived from the model of @xcite . based on the above analysis , we find that the local density of ism near sh 2 - 216 derived from the models of bss and sbs is much larger than that derived from the model of @xcite . the model of @xcite represents the large - scale ism distribution in the galaxy , the values listed in table [ tb21 ] are the average and smoothed density of ism at the position of sh 2 - 216 . in fact , large density perturbations on small scales of ism are observed @xcite , hence the model of @xcite may underestimate the local density of ism interacting with sh 2 - 216 . on the other hand , if the density of ism derived from the model of @xcite is correct at the position of sh 2 - 216 , other factors must be considered in the models of bss and sbs to explain the observed pn - ism interaction for sh 2 - 216 . @xcite pointed out that the ism magnetic field is needed to mold the interaction between sh 2 - 216 and its nearby ism . @xcite presented 1420 mhz polarization images of sh 2 - 216 . an arc of low polarized intensity in the northeast portion was observed and was coincident with the optically identified pn - ism interaction region . several polarization - angle `` knots '' appeared along the arc . @xcite estimated the rotation measure through the knots and derived a line - of - sight magnetic field in the interaction region as @xmath63 @xmath64 g . this result is consistent with that estimated by @xcite . in the simulations of @xcite and wzo , with a low relative velocity @xmath16 20 km s@xmath0 , the pn - ism interaction can be observed for a lower ism density of @xmath65 @xmath17 . the magnetic field is also not considered in their simulations . the cs has been displaced from the center of the pn , so sh 2 - 216 is in the wzo 3 stage . figure 2 of @xcite shows that puwe 1 has a spherical enhanced emission shell interacting with the ism , but the northwest sector of the shell is dimmer than the other portions . the p.a . of the proper motions of puwe 1 indicates that this pn is moving to the southeast which is consistent with the direction of the pn - ism interaction region . the orbit of puwe 1 shows that it is moving toward its maximum distance from the plane of the galaxy . using the kinematic age 0.04 myr and the @xmath19 velocity of puwe 1 listed in table [ tb2 ] , this pn only moved @xmath66 pc away from its birthplace in the vertical direction . the relative velocity of puwe 1 is @xmath16 50 km s@xmath0 which is the typical velocity in the models of bss and sbs for disk pns . using the average volume filling factor of galactic disk pns @xmath67 @xcite , the mean electron density of puwe 1 can be estimated as 3.1 cm @xmath68 @xcite . according to bss and sbs , with an expansion velocity of 27 km s@xmath0 , the ism density can be calculated as @xmath69 cm @xmath68 which is close to the average diffuse ism density @xmath70 cm @xmath68 derived from the model of @xcite at the position of puwe 1 . because the cs is still in the center of the pn , puwe 1 is in the wzo 2 stage . both the [ ] @xcite and h@xmath2 @xcite images of a21 show that this pn is a one - side nebula , whose bright filaments are in the southeast and a faint rim is on the opposite side . the bright filamentary structure in the southeast indicates the pn - ism interaction . the p.a . of the proper motions of this pn is consistent with the direction of the pn - ism interaction regions . @xcite explained the one - side structure of a21 as the result of a highly inhomogeneous ism in the vicinity of this pn . the orbits of a21 indicate that it is a typical thin - disk object . using the derived kinematic age 0.02 myr and a @xmath19 velocity of a21 , this pn only moved @xmath71 pc away from its birthplace in the vertical direction . different orbital parameters derived with different galactic models can not affect the pn - ism interaction of a21 . the electron densities of a21 in six regions of the bright filaments were derived by @xcite and the mean is @xmath72 @xmath17 . using a relative velocity of @xmath73 km s@xmath0 and an expansion velocity of @xmath73 km s@xmath0 , according to bss and sbs , the density of ism in the vicinity of a21 can be estimated as @xmath74 @xmath17 . according to the model of @xcite , the density of ism at the position is 0.09 cm @xmath68 , which is much smaller than that derived from the models of bss and sbs . bss has pointed out that an asymmetric region surrounding a21 should exist , which would explain the observed shape of the pn and the estimated high ism density . the model of @xcite may underestimate the local density of ism in the vicinity of a21 . because the cs of a21 has been displaced from the center , this pn is in the wzo 3 stage . this pn was found during a search around the hot h - rich white dwarf ton 320 @xcite . with a distance of 532 pc derived by @xcite , this pn has a linear diameter of 4.6 pc , which is bigger than sh 2 - 216 . the [ ] image of ton 320 @xcite shows that this pn is very faint and it has a similar one - side filamentary structure as that of a21 . the bright filaments in the southeast indicate it is interacting with the ism , no faint filaments can be observed in the opposite direction in the same [ ] image . the p.a . of the proper motions of ton 320 is consistent with the direction of the pn - ism interaction region . using the kinematic age 0.11 myr and the @xmath19 velocity of ton 320 , this pn only moved @xmath66 pc away from its birthplace in the vertical direction . because there is no electron density data available for ton 320 , we can not estimate the ism density with the models of bss and bsb . however , using the diffuse ism density @xmath75 @xmath17 at the position of ton 320 derived from the model of @xcite , with a relative velocity of @xmath76 km s@xmath0 and a mean expansion velocity of pns @xmath77 km s@xmath0 @xcite , according to the models of bss and sbs , the mean electron density of ton 320 can be estimated as 3.6 @xmath17 . this electron density is close to those of sh 2 - 216 and puwe 1 . all three of those pns have a large size and extremely low surface brightness , so the estimated very low electron density for ton 320 is reasonable . the very large size and very low density of ton 320 also indicate that it has evolved to its later stage . the cs is displacing from the center of the pn , which indicates that ton 320 is in the wzo 3 stage . the h@xmath2 and [ ] images of a31 @xcite show that its southern rim is brighter and sharp - edged , whereas the northern side is faint and diffuse , which indicates its interaction with the ism . the orbits of a31 indicate it is a thick - disk object and is moving toward the galactic plane . using the kinematic age 0.13 myr and the @xmath19 velocity of a31 , this pn only moved @xmath78 pc away from its birthplace in the vertical direction . the p.a . of the proper motions of this pn is also consistent with the direction of the pn - ism interaction region . using the average filling factor of disk pns @xmath67 @xcite , the mean electron density of a31 can be estimated as 11.4 @xmath17 @xcite . using the relative velocity of @xmath79 km s@xmath0 and expansion velocity of @xmath80 km s@xmath0 , according to the models of bss and sbs , the density of ism in the vicinity of a31 can be estimated as 0.16 @xmath17 which is close to the average density @xmath81 @xmath17 derived from the model of @xcite for the diffuse ism at the position of a31 . the orbit of pg 1034 + 001 was calculated by @xcite with the as91 galactic model . the radial velocity and proper motions of pg 1034 + 001 adopted by @xcite are consistent with those adopted in this paper , but the distance of 155 pc adopted by @xcite is derived based on spectroscopic observation @xcite and is less than our adopted distance of 210 pc . using the method of @xcite and based on its position in the @xmath18 @xmath11 velocity diagram , @xmath26 @xmath28 diagram , and the galactic orbit , @xcite found that pg 1034 + 001 is a thin - disk star . using the improved criteria of @xcite for distinguishing different populations , we find that this star is a thick - disk object . the different population membership determination is mainly due to the adopted different distances for this star . the largest known pn surrounding the do white dwarf pg 1034 + 001 with a diameter of @xmath82 was first found by @xcite based on the investigation of spectra from the sloan digital sky survey ( sdss ; * ? ? ? * ) . a careful inspection of the southern h - alpha sky survey atlas ( shassa ; * ? * ) has shown that this pn is surrounded by an elliptical emission shell with a diameter of @xmath83 , and a fragment of a wider shell with a diameter of @xmath84 is also visible in the northeast @xcite . using the kinematic age 0.54 myr and the @xmath19 velocity of pg 1034 + 001 listed in table [ tb2 ] , this pn moved @xmath85 pc away from its birthplace in the vertical direction ; this distance is only half of its linear size . @xcite noted that the major axis of the halo of the outer shell with a diameter of @xmath83 has a p.a . of @xmath86 which is consistent with the p.a . of the proper motions of @xmath87 . they pointed out that this structure can be understood as an advanced pn - ism interaction , but the p.a . of the proper motions is opposite to that of the strongest emission region in the inner nebula found by @xcite . @xcite also pointed out that the do white dwarf pg 1034 + 001 has probably experienced a final helium - shell flash in its past , which produces the second pn , but according to the simulations of wzo , the outer shell with a diameter of @xmath88 is probably the bow shock structure formed by the interaction between the slow agb wind and the ism ; the inner shell with a diameter of @xmath82 is the pn shell formed by the interaction between the fast post - agb wind and the slow agb wind material . the pn is still within the bubble of undisturbed agb wind material and unaffected by the ism interaction , which indicates it is in the wzo 1 stage . because this pn is still in the wzo 1 stage and no electron density of this pn are available , the models of bss and bsb can not be used to explain the pn - ism interaction and to derive the local density of ism for this pn . ngc 7293 ( helix nebula ) has a bright complex helical structure which is composed of at least two bipolar nebulae , with their poles at different orientations along p.a.s @xmath89 and @xmath90 , respectively @xcite . the very deep h@xmath91 [ ] image of ngc 7293 obtained by @xcite shows that there is a bow - shaped filament at the northeast of the bright helical filaments of this pn , and a faint jet - like feature at the opposite side of the pn is also visible . @xcite calculated the orbit of ngc 7293 with the as91 model and found that this pn is a thin - disk object . in this paper , we also identified ngc 7293 to be a thin - disk pn , which is consistent with the result of @xcite . using the kinematic age 0.03 myr and the @xmath19 velocity of pg ngc 7293 listed in table [ tb2 ] , this pn moved @xmath92 pc away from its birthplace in the vertical direction . @xcite noted that the main shell of ngc 7293 does not show any signs of the pn - ism interaction . they also pointed out that the asymmetric halo of this pn is due to the symmetrically ejected material from the cs which runs into the higher density northeast region of its southwest counterpart . with the average electron density of this pn @xmath93 @xmath17 @xcite , a relative velocity of @xmath73 km s@xmath0 , and an expansion velocity of @xmath94 km s@xmath0 , according to the models of bss and sbs , the density of ism in the vicinity of ngc 7293 can be estimated as @xmath95 @xmath17 which is much larger than @xmath96 derived from the model of @xcite for the diffuse ism . according to the simulations of wzo , the bow - shaped structure in the halo is probably formed by the interaction between the agb wind and the ism . the p.a . of the proper motions of the pn is consistent with the outer bow - shaped region , which indicates a probable wind - ism interaction . the inner bright helical shell is not affected by the ism interaction , and this pn is in the wzo 1 stage . the h@xmath2 and [ ] images of deht 5 @xcite show a large amount of diffuse material surrounding the bright filaments of this pn , which indicates that the pn - ism interaction has destroyed the intrinsic morphology of this pn . the very low expansion velocity of 5 km s@xmath0 also indicates that deht 5 has evolved into its later stage . using the kinematic age of 0.09 myr and the @xmath19 velocity of deht 5 listed in table [ tb2 ] , this pn moved @xmath97 pc away from its birthplace in the vertical direction . using the relative velocity of @xmath4 km s@xmath0 and the expansion velocity of 5 km s@xmath0 , according to the models of bss and sbs , and using the diffuse ism density @xmath75 @xmath17 derived from the model of @xcite , the mean electron density of deht 5 can be estimated as 4.5 @xmath17 which is close to those of sh 2 - 216 , puwe 1 , and ton 320 . this very low density also indicates that deht 5 has evolved into its later stage . the orbits of eight nearby pns are calculated with the as91 and fsc96 galactic models . the small number of pns in the present sample is limited by the available observation data for pns in the galaxy . in our present sample , only pns with direct parallax distances are considered . the errors of the derived orbital parameters are calculated with a monte carlo method . we also compared the orbital parameters derived with different galactic models . we find that the uncertainty of the derived orbital parameters is mainly from the observational errors in the input data . according to the kinematic criteria of @xcite for distinguishing different populations in the galaxy and using the @xmath49 diagram , we find that sh 2 - 216 , deht 5 , ngc 7293 , a21 , and ton 320 belong to the thin - disk population and pg 1034 + 001 and a31 belong to the thick - disk population . puwe 1 probably belongs to the thick - disk population , but with the large errors of its derived orbital parameters , its population classification becomes very uncertain . the present sample of pns is very small , hence we can not come to any conclusion about the population distribution of galactic pns in the disk . population classification for the eight pns in our sample indicates that most of the pns formed are closer to the galactic plane . the dramatic enlargement of the sample of pns with reliable parallaxes will only be expected when the data of the _ gaia _ mission are available . the primary objective of the _ gaia _ mission is to survey more than one billion stars between 6 and 20 mag . the expected accuracies for parallaxes , proper motions and positions are 7 25 @xmath64as down to 15 mag and a few hundred @xmath64as at 20 mag @xcite . the magnitudes of 96% cs in the catalog of @xcite are less than 20.0 ; the precise distances and proper motions for those stars obtained with the _ gaia _ mission can be used to study their kinematics and dynamics in the galaxy . we analyzed the interaction between the pns in our sample and their nearby ism . the pn - sim interaction are observed in the eight pns in our present sample . the p.a.s of the proper motions of the pns are also consistent with the directions of the pn - ism interaction regions . for all eight of the pns in our sample , their kinematic ages are much smaller than the time @xmath30 for them to cross the galactic plane , and the distances that they moved from their birthplaces are all smaller than their linear size . different orbits derived with different galactic models can not affect the pn - ism interaction for each pn . there are two main models to explain the pn - ism interaction . one is presented by bss and sbs . in their models , the pn interacts with the ism directly , and the pn - ism interaction can be observed only when the density of the pn drops below some critical value . the magnetic field is also important in the models of bss and sbs . the other one is the triple - wind " model presented by wzo . in their model , the slow agb wind formed in the early evolution phase of the agb star interacts with the ism , and then the pn interacts with the bow shock formed during the agb wind - ism interaction . so , in the model of wzo , the pn - ism interaction can be observed in the early evolution phase of the pn . the magnetic field is not considered in the model of wzo . for both models , the relative velocity of the pn and the density of the ism in the vicinity of the pn are very important parameters for studying the pn - ism interaction . we use the above - mentioned models to explain the observed pn - ism interaction in the eight pns in our sample . using the models of bss and sbs , the local densities of ism in the vicinity of puwe 1 and a31 are derived and are consistent with those derived from the model of @xcite . for ton 320 and deht 5 , no electron density data are available . using the local densities of ism for these two pns derived from the model of @xcite , the electron densities of these two pns are derived from the models of bss and sbs . compared with the observed electron densities of other pns which have evolved into their later stages , the derived electron densities of these two pns are reasonable . for the above four pns , the models of bss and sbs yield consistent results from the model of @xcite , but for sh 2 - 216 and a21 , the derived local densities of ism in the vicinity of these two pns with the models of bss and sbs are larger than those derived from the model of @xcite . for sh 2 - 216 , considering a magnetic field in the position of this pn in the models of bss and sbs can explain the observed pn - ism interaction . for a21 , an asymmetric region surrounding this pn can explain the predicted high ism density . for these two pns , the model of @xcite will underestimate the local densities of ism . the model of @xcite represents the large - scale and average ism distribution in the galaxy ; it can not predict the local density of ism very precisely . therefore , with the models of bss and sbs , the local ism densities in the vicinity of evolved pns can be derived with the observed pn - ism interaction @xcite . because pg 1034 + 001 and ngc 7293 are still in their early stage of evolution , the models of bss and sbs can not used to derive the ism densities for these two pns . on the other hand , according to the model of wzo , sh 2 - 216 , a21 , and ton 320 are in the wzo 3 stage , pg 1034 + 001 and ngc 7293 are in the wzo 1 stage , and puwe 1 is in the wzo 2 stage . lllllccccc sh 2 - 216 & 04 : 43 : 21.3&@xmath98 : 42 : 05.8&12.630&@xmath99 & @xmath100 & @xmath101&116&@xmath102&@xmath103 + puwe 1 & 06 : 19 : 34.3&@xmath104 : 36 : 42.3&15.534&@xmath105 & @xmath106&@xmath107&141&@xmath108&@xmath109 + a21 & 07 : 29 : 02.7&@xmath110 : 14 : 48.8&15.962&@xmath111 & @xmath112 & @xmath113&166&@xmath114&@xmath115 + ton 320 & 08 : 27 : 05.6&@xmath116 : 30 : 08.5&15.702&@xmath117 & @xmath118 & @xmath119&198&@xmath120&@xmath121 + a31 & 08 : 54 : 13.2&@xmath122 : 53 : 53.0&15.519&@xmath123&@xmath124&@xmath125&228&@xmath126&@xmath127 + pg 1034 + 001&10 : 37 : 03.8&@xmath128 : 08 : 19.2&13.211&@xmath129 & @xmath130&@xmath131&290&@xmath132&@xmath133 + ngc 7293 & 22 : 29 : 38.5&@xmath134 : 50 : 13.7&13.525&@xmath135&@xmath136 & @xmath137&101&@xmath138&@xmath115 + deht 5 & 22 : 19 : 33.7&@xmath139 : 56 : 03.3&15.474&@xmath140&@xmath141&@xmath142&221&@xmath143&@xmath144 + lccccccccc sh 2 - 216 & 100&3.7&128.9&8.0&0.10&0.02&4&0.45&5 + puwe 1 & 20&2.1&365.0&3.1&0.08&0.02&27&0.04&1 + a21 & 9&1.4&540.5&206.0&0.07&0.02&32&0.02&4 + ton 320 & 30&4.6&531.9&&0.05&0.02&&0.11 & + a31 & 16&2.9&621.1&11.4&0.04&0.02&11&0.13&2 + pg 1034 + 001&360&22.0&210.5&&0.07&0.02&&0.54 & + ngc 7293 & 18&1.1&215.5&630.0&0.06&0.02&21&0.03&3 + deht 5 & 9&0.9&344.8&&0.08&0.02&5&0.09&1 + llllccccc sh 2 - 216 & @xmath145&@xmath146&@xmath147&@xmath148&@xmath149&@xmath150&@xmath151 + puwe 1 & @xmath152&@xmath153&@xmath154&@xmath155&@xmath156&@xmath157&@xmath158 + a21 & @xmath159&@xmath160&@xmath161&@xmath162&@xmath163&@xmath164&@xmath165 + ton 320 & @xmath166&@xmath167&@xmath168&@xmath169&@xmath170&@xmath171&@xmath172 + a31 & @xmath173&@xmath174&@xmath175&@xmath176&@xmath177&@xmath178&@xmath179 + pg 1034 + 001&@xmath180&@xmath181&@xmath182&@xmath183&@xmath184&@xmath185&@xmath186 + ngc 7293 & @xmath187&@xmath188&@xmath189&@xmath190&@xmath191&@xmath192&@xmath193 + deht 5 & @xmath194&@xmath195&@xmath196&@xmath197&@xmath198&@xmath199&@xmath200 + llllllll sh 2 - 216 & @xmath201&@xmath202&@xmath202&@xmath203&@xmath204&33.9&44.4 + puwe 1 & @xmath205&@xmath206&@xmath207&@xmath208&@xmath209&35.1&46.5 + a21 & @xmath210&@xmath211&@xmath211&@xmath212&@xmath213&31.7&41.7 + ton 320 & @xmath214&@xmath215&@xmath215&@xmath216&@xmath217&35.1&44.4 + a31 & @xmath218&@xmath219&@xmath220&@xmath221&@xmath222&38.5&50.0 + pg 1034 + 001 & @xmath223&@xmath224&@xmath219&@xmath225&@xmath226&35.7&46.5 + ngc 7293 & @xmath227&@xmath228&@xmath228&@xmath229&@xmath230&34.5&44.4 + deht 5 & @xmath231&@xmath232&@xmath233&@xmath234&@xmath235&33.9&44.4 + the authors thank the referee for his / her comments . this work has been supported in part by the national natural science foundation of china , nos . 10633020 , 10778720 , 10873016 , and 10803007 and by the national basic research program of china ( 973 program ) , no . 2007cb815403 . w. is supported by the young researcher grant of the national astronomical observatories , chinese academy of sciences . this research has made use of the simbad database and vizier catalog access tool , operated at cds , strasbourg , france . allen , c. , & santilln , a. 1991 , , 22 , 255 ( as91 ) balick , b. , & frank , a. 2002 , , 40 , 439 barstow , m. a. , bannister , n. p. , holberg , j. b. , hubeny , i. , bruhweiler , f. c. , & napiwotzki , r. 2001 , , 325 , 1149 benedict , g. f. , et al . 2009 , , 138 , 1969 boffi , f. r. , & stanghellini , l. 1994 , , 284 , 248 borkowski , k. j. , sarazin , c. l. , & soker , n. 1990 , , 360 , 173 ( bss ) de marco , o. 2009 , , 121 , 316 dehnen , w. , & binney , j. 1998 , , 298 , 387 dickey , j. m. , & lockman , f. j. 1990 , , 28 , 215 drew , j. e. , et al . 2005 , , 362 , 753 burand , s. , acker , a. , & zijlstra , a. 1998 , , 132 , 13 esa , 1997 , the hipparcos and tycho catalogues , esa , sp-1200 flynn , c. , sommer - larsen , j. , & christensen , p. r. 1996 , , 281 , 1027 ( fsc96 ) frew , d. j. , madsen , g. j. , otoole , s. j. , & parker , q. a. 2010 , pasa , 27 , 203 gaustad , j. e. , mccullough , p. r. , rosing , w. , & van buren , d. 2001 , , 113 , 1326 gieseking , f. , hippelein , h. , & weinberger , r. 1986 , , 156 , 101 good , s. a. , barstow , m. a. , burleigh , m. r. , dobbie , p. d. , & holberg , j. b. 2005 , , 364 , 1082 gurzadyan , g. a. 1969 , planetary nebulae ( new york : gordon & breach ) harrs , h. c. , dahn , c. c. , monet , d. g. , & pier , j. r. 1997 , in proc . 180 , planetary nebulae , ed . j. habing & h. j. g. l. m lamers(dordrecht : kluwer ) , 40 harrs , h. c. , et al . 2007 , , 133 , 631 hewett , p. c. , irwin , m. j. , skillman , e. d. , foltz , c. b. , willis , j. p. , warren , s. j. , & walton , n. a. 2003 , , 599 , l37 hippelein , h. , & weinberger , r. 1990 , , 232 , 129 hg , e. , et al . 2000 , , 355 , l27 holberg , j. b. , barstow , m. a. & sion , e. m. 1998 , , 119 , 207 hua , c. t. , & kwok , s. 1999 , , 138 , 275 kalberla , p. m. w. , & kerp , j. 1998 , , 339 , 745 kerber , f. , mignani , r. p. , guglielmetti , f. , & wicenec , a. 2003 , , 408 , 1029 kerber , f. , mignani , r. p. , pauli , e. m. , wicenec , a. , & guglielmetti , f. 2004 , , 420 , 207 kerber , f. , mignani , r. p. , smart , r. l. , & wicenec , a. 2008 , , 479 , 155 kwitter , k. b. , jacoby , g. h. , & lawrie , d. g. 1983 , , 95 , 732 kwok , s. , purton , c. r. , & fitzgerald , p. m. 1978 , , 219 , l125 kwok , s. 1982 , , 258 , 280 lasker , b. m. , et al . 2008 , , 136 , 735 lindegren , l. 2010 , in iau symp . 261 , relativity in fundamental astronomy , ed . s. a. klioner , p. k. seidelman & m. h. soffel , 296 meaburn , j. , boumis , p. , lopez , j. a. , harman , d. j. , bryce , m. , redman , m. p. , & mavromatakis , f. 2005 , , 360 , 963 meaburn , j. , lpez , j. a. , & richer , m. g. 2008 , , 384 , 497 monet , d. g. , et al . 2003 , , 125 , 984 pauli , e. m. , napiwotzki , r. , altmann , m. , heber , u. , odenkirchen , m. , & kerber , f. 2003 , , 400 , 877 pauli , e. m. , napiwotzki , r. , heber , h. , altmann , m. , & odenkirchen , m. 2006 , , 447 , 173 phillips , j. p. 1998 , , 340 , 527 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes in fortran : the art of scientific computing , 2rd edn . cambridge univ . press , cambridge ramos - larios , g. , & phillips , j. p. 2009 , , 400 , 575 ransom , r. r. , uyaniker , b. , kothes , r. , & landecker , t. l. 2008 , , 684 , 1009 rauch , t. , kerber , f. , & pauli , e. m. 2004 , , 417 , 647 rauch , t. , ziegler , m. , werner , k. , kruk , j. w. , oliveira , c. m. , vande , p. d. , mignani , r. p. , & kerber , f. 2007 , , 470 , 317 recillas - cruz , e. , & pismis , p. 1981 , , 97 , 398 reynolds , r. j. 1985 , , 288 , 622 sabin , l. , zijlstra , a. a. , wareing , c. , corradi , r. l. m. , mampaso , a. , viironen , k. , wright , n. j. , & parker , q. a. 2010 , pasa , 27 , 166 schnrich , r. , binney , j. , & dehnen , w. 2010 , , 403 , 1829 smith , h. 1976 , , 175 , 419 soker , n. , borkowski , k. j. , & sarazin , c. l. 1991 , , 102 , 1381 ( sbs ) soker , n. , & dgani , r. 1997 , , 484 , 277 tweedy , r. w. , & napiwotzki , r. 1992 , , 259 , 315 tweedy , r. w. , & kwitter , k. b. 1994 , , 433 , l93 tweedy , r. w. , martos , m. a. , & noriega - crespo , a. 1995 , , 447 , 257 tweedy , r. w. , & kwitter , k. b. 1996 , , 107 , 255 villaver , e. , garca - segura , g. , & manchado , a. 2003 , , 585 , l49 wareing , c. j. , zijlstra , a. a. , & obrien , t. j. 2007 , , 382 , 1233 ( wzo ) weinberger , r. 1989 , , 78 , 301 werner , k. , dreizler , s. , & wolff , b. 1995 , , 298 , 567 wu , z. y. , zhou , x. , ma , j. , & du , c. h. 2009 , , 399 , 2146 xilouris , k. m. , papamastorakis , j. , paleologou , e. , & terzian , y. 1996 , , 310 , 603 york , d. g. , et al . 2000 , , 120 , 1579 zacharias , n. , urban , s. e. , zacharias , m. i. , wycoff , g. l. , hall , d. m. , monet , d. g. , & rafferty , t. j. 2004 , , 127 , 3043 zacharias , n. , et al . 2009 , vizier on - line data catalog : i/315

electron - electron interactions are responsible for a correction to the conductance of a diffusive metal , the `` altshuler - aronov correction '' . here we study the counterpart of this correction for a ballistic conductor , in which the electron motion is governed by chaotic classical dynamics . in the ballistic conductance , the ehrenfest time enters as an additional time scale that determines the magnitude of quantum interference effects . the ehrenfest time effectively poses a short - time threshold for the trajectories contributing to the interaction correction . as a consequence , becomes exponentially suppressed if the ehrenfest time is larger than the dwell time or the inverse temperature . we discuss the explicit dependence on ehrenfest time in quasi - one and two - dimensional antidot arrays . for strong interactions , the sign of may change as a function of temperature for temperatures in the vicinity of .

1211.1843

electronic transport in weakly disordered metals is successfully described by the boltzmann theory , in which electrons are treated as effectively classical particles moving freely between scattering events . the wave nature of electrons gives rise to a number of corrections to transport properties , such as the weak localization correction,@xcite the altshuler - aronov interaction correction,@xcite or the universal conductance fluctuations.@xcite weak localization results from the constructive interference of electrons propagating along time - reversed paths.@xcite the physical intuition behind the interaction correction is constructive interference of electron trajectories which are scattered on impurities and friedel oscillations of the electron density.@xcite these quantum corrections become increasingly important as the temperature is lowered , the effective dimensionality of the sample is reduced , or as the disorder level is increased . they have a distinctive and universal dependence on external parameters , such as temperature or magnetic field , which makes them identifiable in experiments . in particular , the two quantum corrections to the conductivity , weak localization and the altshuler - aronov correction , can be distinguished by application of a magnetic field , since weak localization is suppressed by already a very small magnetic field , whereas the altshuler - aronov correction is not . a `` classical analog '' of a disordered metal is realized in high - mobility semiconductor structures with randomly placed large antidots.@xcite the absence of impurities ensures that electrons move ballistically between reflections off the antidots . the reason why these systems are referred to as classical is that the size of the antidots @xmath3 is much larger than the fermi wavelength @xmath4 . as a result , not only the electron s motion through the two - dimensional electron gas , but also the reflection off an antidot is described by classical mechanics . ( in contrast , in a disordered metal , the size of impurities is comparable to @xmath4 , so that the scattering event is strongly diffractive . ) for an irregular arrangement of antidots , the classical dynamics is chaotic . nearby trajectories separate exponentially in time , the exponential separation being characterized by the lyapunov coefficient @xmath5 . the chaotic dynamics is essential for the existence of quantum corrections in this system , as it magnifies the quantum uncertainty of even a minimal wavepacket up to classical dimensions within the short time @xmath6 thus transforming the classical dynamics into quantum - diffractive dynamics on time scales larger than @xmath1.@xcite the time @xmath1 is known as the `` ehrenfest time '' . since wave effects are not operative for times shorter than @xmath1 electrons essentially move along classical trajectories up to the ehrenfest time , the ehrenfest time serves as a short - time threshold for the duration of the trajectories contributing to the quantum corrections in an antidot array . for weak localization , it was found that the correction to the conductivity is exponentially suppressed if @xmath1 is larger than the dwell time @xmath7 , the typical time to be transmitted through the system , or the dephasing time @xmath8.@xcite in contrast , other quantum corrections , such as the universal conductance fluctuations , remain finite if @xmath9.@xcite the goal of this article is to present a theory of the ehrenfest - time - dependence of the altshuler - aronov correction @xmath0 . our analysis significantly extends a previous calculation by kupferschmidt and one of the authors,@xcite which studied the @xmath1 dependence of the interaction correction to the conductance of a ballistic double quantum dot and found that @xmath0 is strongly suppressed if @xmath1 exceeds the dwell time @xmath7 or the inverse temperature @xmath10 . the double quantum dot studied in ref . is the simplest system with nonzero altshuler - aronov correction to the conductance , and is characterized by a long - range interaction , which is spatially homogeneous within each dot . the theory presented here is valid for both short - range and long - range interactions and can be applied to any geometry in which the classical electron dynamics is chaotic although we will focus our discussion on the case of an antidot array . for the general case considered here we confirm the suppression of @xmath0 for @xmath11 and we calculate the precise functional dependence of @xmath0 on @xmath7 and @xmath12 for finite ehrenfest time . the explicit dependence on temperature is characteristic for the interaction correction , which has its origin in virtual processes with an energy transfer larger than temperature . our calculation makes use of a semiclassical formalism that starts from the saddle - point approximation around classical trajectories for the single - particle green function . in this way , the conductance in the absence of electron - electron interactions is written as a double sum over classical trajectories that connect source and drain reservoirs.@xcite weak localization and other quantum corrections to the conductance then follow from special configurations of trajectories , in which the two trajectories in the summation are piecewise paired , and proceed through `` crossings '' at points where the pairing is changed.@xcite in the language of diagrammatic perturbation theory , segments where the trajectories are paired correspond to diffusons or cooperons , whereas the crossings correspond to hikami boxes . the application to interacting electrons requires a modification of the formalism , which will be described in detail below . our analysis applies to a `` ballistic '' conductor , where the label `` ballistic '' is meant to specify that the electrons move along well - defined classical trajectories . in the literature , `` ballistic '' sometimes refers to a different limit , and several calculations of the interaction correction to the conductance have been reported for such `` ballistic limits '' . whereas the original work of altshuler and aronov@xcite addressed a disordered metal with short - range scatterers in the diffusive regime @xmath13 , the theory was generalized to account for the effects of higher temperatures @xmath14 , a regime referred to as `` ballistic''.@xcite the case of a smooth disorder potential , in which scattering is predominantly forward , was considered in ref . . another type of system , where interaction corrections appear , are networks of capacitively coupled ballistic quantum dots,@xcite where , however , ehrenfest - time - related phenomena can be neglected as long as @xmath1 is much smaller than the dwell time in a single quantum dot . interactions also affect the conductance through their effect on the weak localization correction ( dephasing ) . semiclassical studies of the effect of interaction - induced dephasing on weak localization can be found in refs . , for electronic systems and in ref . for bosonic matter waves . in section [ sec : sctheory ] we present our theory of the ehrenfest - time dependence on the interaction correction for a generic ballistic chaotic conductor . in sec . [ sec : antidot ] we then apply our formalism to an antidot array , where the classical electronic motion is diffusive on length scales much larger than the spacing between antidots , and the coulomb interaction is dynamically screened by the diffusively moving electrons . for the antidot array , we find @xmath15 in the limit that the ehrenfest time is larger than dwell - time and inverse temperature . for small ehrenfest times , we recover the results of a disordered metal with quantum impurities , which show a much weaker temperature dependence ( algebraic or logarithmic , depending on dimensionality ) . we conclude in [ sec : conclusion ] . in this section we present the semiclassical description of the interaction corrections for a conductor with a well - defined chaotic classical electron dynamics . we first review the expressions for the interaction corrections to the conductance in terms of the single - particle green function , and then apply the semiclassical approximation methods , taking into account the finite ehrenfest time . for definiteness , we consider a two - dimensional ballistic conductor , such as a ballistic electron gas with an antidot array , in contact with reservoirs at @xmath16 and @xmath17 , see fig . [ fig : vertex ] . without interactions , we can calculate the conductance @xmath18 from the kubo formula,@xcite @xmath19_{\genfrac{}{}{0pt}{}{x'=0}{x = l}},\end{aligned}\ ] ] where @xmath20 denotes the fermi function , @xmath21 is the velocity operator , and @xmath22 and @xmath23 is the retarded and advanced single - particle green function , respectively . retarded and advanced green functions are related as @xmath24 and @xmath17 . in the semiclassical calculation of the conductance , one retarded and one advanced green function are attached to a current vertex located at the interface with the leads . in a semiclassical picture , these green functions are associated with `` retarded '' and `` advanced '' classical trajectories ( solid and dashed in the figure ) , both of which must point into the conductor ( 1 ) . a current vertex combined with two green functions of the same kind is not possible for the calculation of the conductance : after pairing , we have trajectories that go straight into the leads ( 2 ) . on the contrary , for a calculation of the conductivity , the current vertex can be anywhere inside the conductor , and pairing of retarded and advanced trajectories is possible also if two green functions of the same kind are attached to one current vertex ( 3 and 4).,width=278 ] to leading ( first ) order in the interaction strength , the interaction correction @xmath0 is obtained by replacing @xmath22 by @xmath25 and expanding to first order in the interaction @xmath26,@xcite where @xmath27 @xmath28 and with similar expressions for the advanced functions @xmath29 and @xmath30 . in these expressions , @xmath31 and @xmath32 are the retarded and advanced interaction kernels , respectively . the interaction is taken to be zero in the leads , for @xmath33 and @xmath34 . such a structure represents the change of the single - particle green function due to scattering off friedel oscillations of the density matrix @xmath35 ( fock ) or the density @xmath36 ( hartree).@xcite the resulting contributions to @xmath0 are represented diagrammatically as in fig . [ fig : fock ] . we would like to emphasize that we kept here only those diagrams for which one retarded and one advanced green function are connected to each current vertex . these are the relevant diagrams for the calculation of the conductance . this is in contrast to the calculation of the interaction correction to the conductivity @xmath37 , where diagrams with two green functions of the same kind attached to the current vertex play an important role.@xcite ( the conductivity @xmath38 is expressed in a similar way as eq . , but contains integrals over the @xmath39-coordinates as well , rather than fixing them to the contacts at @xmath16 and @xmath17 ) . although the structure of the calculation for conductance or conductivity considerably differs , the final results for these quantities are related by a geometrical factor only . in two dimensions , for a rectangular sample , one has @xmath40 , where @xmath41 is the width of the system . the difference between the conductance calculation and the conductivity calculation is readily seen in the semiclassical language . in that language , green functions are associated with classical trajectories , and only terms in which `` retarded '' and `` advanced '' trajectories are paired contribute . ( for more details , see below ) . since the leads are assumed to be free of disorder and without electron - electron interactions , both the retarded and the advanced trajectories at the positions @xmath42 and @xmath43 in eq . ( [ eq : kubo ] ) must point into the conductor if the conductance is calculated . on the other hand , for a current vertex in the system s interior , pairing of advanced and retarded trajectories is still possible even if two green functions of the same kind are attached to the current vertex , see fig . [ fig : vertex ] . the fact that different diagrams are needed for the calculation of conductance and conductivity is well known , the same is true for the drude conductance and conductance fluctuations of a disordered metal ( see ref . ) . for instance , for the calculation of the drude conductivity @xmath44 in a metal with short - ranged disorder there is no need to dress the diagram with an impurity ladder , while for the classical conductance @xmath45 , the diagram dressed with an impurity ladder , i.e. , a diffuson , is most relevant . for the drude conductivity one might argue that the distance between the current vertices is of the order of the mean free path , since the two green functions decay on this scale . the diffuson in turn is long - ranged , and hence needed to describe propagation from one lead to the other , as required for the conductance . we also note that the calculation of the conductance as it is outlined here is similar to the calculation of the density - density correlation functions . @xcite a subtle point in this regard is the existence of additional corrections in the calculation of the density - density correlation function , namely vertex corrections and the so - called wave - function renormalization . in the description developed below , both of them appear to vanish . for the density - density correlation function , in turn , vertex corrections and the wave function renormalization cancel each other , so that these corrections do not lead to a net change of the result in either case . the conductance @xmath18 depends on the precise locations of antidots , the system boundary , and on the fermi energy . we now employ a semiclassical analysis in order to identify those contributions to the conductance that remain after an average over the fermi energy . starting point is the semiclassical expression of the green function @xmath22 as a sum over classical trajectories @xmath46 from @xmath43 to @xmath42 at energy @xmath47,@xcite @xmath48 here , @xmath49 is the classical action corresponding to the trajectory @xmath46 , which has the properties @xmath50 and @xmath51 where @xmath52 and @xmath53 denotes the momentum at the end and beginning of @xmath46 , respectively , and @xmath54 is the duration of the trajectory . the stability amplitude @xmath55 is given by @xmath56 , with @xmath57 the semiclassical green function further contains an additional phase - shift , the so - called maslov index,@xcite which we omitted because it does not play a role in our considerations . the semiclassical expression for advanced green function follows from eq . ( [ eq : grga ] ) . using the semiclassical green function we express the interaction correction @xmath0 as a fourfold sum over classical trajectories . we refer to these trajectories as `` retarded '' or `` advanced '' , depending on the type of the green function that they originate from . the summation over classical trajectories can be simplified for a system with chaotic classical dynamics : in this case , the classical trajectory and hence the classical action depend very sensitively on the initial conditions . on the other hand , in the semiclassical limit @xmath58 only configurations of trajectories with sum of the actions of the `` retarded '' trajectories _ systematically _ equal to the sum of the actions of the `` advanced '' trajectories up to a difference @xmath59 of the order of @xmath60 contribute substantially to the conductance . this occurs only if the `` retarded '' and `` advanced '' trajectories are piecewise paired , whereby they can exchange `` partners '' only at a `` small - angle encounter'',@xcite at which two pairs meet to within a phase - space distance of order @xmath61 . following the procedure outlined so far , we obtain an expression in terms of trajectory densities which , strictly speaking , is a sum of @xmath76-functions . we then replace the exact trajectory density @xmath77 by a coarse - grained smooth density @xmath78.@xcite the coarse graining takes place with respect to small fluctuations of the initial and final phase space points and/or the positions of the scattering discs or the system s boundaries . in the regime @xmath79 , where the chaotic dynamics has fully developed , the classical dynamics is essentially stochastic , which justifies the coarse graining procedure . the coarse - grained trajectory density @xmath80 can be identified with the probability density @xmath81 that a particle originating at the phase space point @xmath82 is found at the phase space point @xmath67 after a time @xmath62 . ( since we are interested in the regime where temperature is much smaller than fermi energy , we drop the dependence of the classical propagators on @xmath47 . ) for the case of antidot arrays , this probability density is described by a diffusion equation . the drude conductance is obtained by keeping only pairs of classical trajectories that connect source and drain reservoirs . following the steps described above , we find @xmath83_{\genfrac{}{}{0pt}{}{x'=0}{x = l}}\end{aligned}\ ] ] we now turn to the semiclassical calculation for the interaction correction . we start with the fock contribution to @xmath0 , which is given by the two lower diagrams in fig . [ fig : fock ] , together with their counterparts , which are obtained by interchanging the retarded and advanced green functions . we first consider the conductance correction @xmath84 from the lower left diagram and its counterpart , which reads @xmath85_{\genfrac{}{}{0pt}{}{x'=0}{x = l}}\right\rbrace.\end{aligned}\ ] ] after insertion of the semiclassical expression eq . for the green functions , we obtain a sum over one retarded and three advanced trajectories . in the semiclassical limit , a convolution of green functions is customarily calculated using the stationary phase approximation . for this one first needs to determine the configurations of trajectories which make the total action stationary . this results in a factor @xmath86 , where @xmath87 is obtained by inserting the stationary configuration into the total action . integration over quadratic fluctuations around the stationary configurations then renders the prefactor . in the present case the convolution of green functions is accompanied by the interaction propagator . one might expect , that the interaction propagator affects the stationary trajectories , such that they no longer connect to a single classical trajectory , as in the case without interaction . however for the calculation of the conductance , we need to pair the advanced trajectories with the retarded one , see fig . [ fig : diag1 ] . hence , performing the integration over @xmath88 and @xmath89 in eq . within stationary phase approximation , we only take into account stationary configurations that connect to a single classical trajectory . the detailed calculation is carried out in appendix [ sec : convolutionrule ] and has the result @xmath90 where @xmath91 is the coordinate of trajectory @xmath46 after time @xmath62 . the integration over time reflects the freedom to choose @xmath88 and @xmath89 anywhere along the trajectory @xmath46 ; the factor @xmath92 takes into account the action difference at different energies , @xmath93 ( for @xmath94 ) . with @xmath95 we may rewrite @xmath96 where @xmath97 is an integration over phase - space points on the energy shell , and @xmath98 is the initial phase - space point of trajectory @xmath46 . after inserting eqs . and into eq . , and upon applying the semiclassical approximation to the retarded green function as well , the interaction correction @xmath99 is expressed as a double sum over trajectories @xmath46 and @xmath100 running from @xmath43 to @xmath42 . only diagonal terms with @xmath101 are systematically nonzero , so that we only keep these . again making use of the sum rule ( eq . ) we express @xmath84 as an integral over the two intermediate phase space points @xmath102 and @xmath103 . before the exact trajectory densities can be replaced by their coarse - grained versions , we split the classical trajectories into uncorrelated segments using the equality @xmath104 after coarse - graining , the expression for @xmath99 involves the probability densities @xmath105 for the chaotic classical motion . the expression can be further simplified by introducing @xmath106_{x'=0 } , \nonumber \\ p_{\rm out}({{\mathbf}{x } } ) & = & \int dy \int d{{\mathbf}{p}}_{\xi } \int_{0}^{\infty } dt \left [ p({{\mathbf}{x}}\to { { \mathbf}{r}},{{\mathbf}{p}};t ) v_x\right]_{x = l } , \end{aligned}\ ] ] which express the probability that a trajectory at phase space point @xmath107 entered at the left contact or exits at the right contact , respectively . using the equality @xmath108 we finally obtain @xmath109 where we singled out the part containing classical propagators , @xmath110 we now consider the interaction correction @xmath111 that corresponds to the lower right diagram of fig . [ fig : fock ] and its counterpart obtained by switching retarded and advanced labels , @xmath112_{\genfrac{}{}{0pt}{}{x'=0}{x = l}}\right\rbrace.\end{aligned}\ ] ] insertion of the semiclassical expression for the green functions leads to a fourfold sum over two retarded trajectories ( from @xmath43 to @xmath89 and from @xmath88 to @xmath42 ) , and two advanced trajectories ( from @xmath43 to @xmath42 and from @xmath88 to @xmath89 ) . because of the specific requirements for the start and end points of the trajectories , it is not possible , to pair the trajectories one by one for their entire duration . instead , the trajectories need to undergo a `` small - angle encounter '' , in which all four trajectories are close together in phase space for at least part of their length.@xcite the four possible configurations of trajectories are shown in fig . [ fig : diag2 ] , where we take into account the possibilities that none , one , or both points @xmath88 and @xmath89 lie inside the encounter region . their contributions to @xmath0 will be denoted @xmath113@xmath114 , see fig . [ fig : diag2 ] . the summation over classical trajectories with a small - angle encounter follows the procedure outlined in refs . . we refer the reader to appendix [ sec : encrules ] for details , and proceed with the results of that calculation . all four contributions to @xmath0 have the same form as the contribution from the first diagram , see eq . ( [ eq : dgaa - f1 ] ) , but with different expressions for the function @xmath115 . for the contributions 2a2d these expressions read @xmath116 , \\ \label{eq : k2b-1 } k^{2b } & = & -\int d{{\mathbf}{x}}p_{\rm in}({{\mathbf}{x } } ) p({{\mathbf}{x}},{{\mathbf}{x}}_1;\omega ) p_{\rm out}({{\mathbf}{x}}_2)\nonumber\\ & & \mbox { } \times p({{\mathbf}{x}}_2,{{\mathbf}{x}};\tau_{\rm e } ) e^{i\omega\tau_{\rm e}/\hbar } , \\ \label{eq : k2c } k^{2c } & = & -\int d{{\mathbf}{x } } ' p_{\rm in}({{\mathbf}{x}}_1 ) p({{\mathbf}{x}}_2,{{\mathbf}{x}}';\omega ) p_{\rm out}({{\mathbf}{x}}')\nonumber\\ & & \mbox { } \times p({{\mathbf}{x}}',{{\mathbf}{x}}_1;\tau_{\rm e } ) e^{i\omega\tau_{\rm e}/\hbar } , \\ \label{eq : k2d-1 } k^{2d } & = & - \int_0^{\tau_{\rm e } } dt p_{\rm in}({{\mathbf}{x}}_1 ) p({{\mathbf}{x}}_2,{{\mathbf}{x}}_1;t ) \nonumber \\ & & \mbox { } \times e^{i\omega t/\hbar } p_{\rm out}({{\mathbf}{x}}_2).\end{aligned}\ ] ] here @xmath117 is the fourier transform of @xmath81 , @xmath118 taken together , eqs . , , , , , and determine the general result for the fock contribution to @xmath0 for finite ehrenfest time , expressed in terms of classical propagators . let us briefly discuss the effect of the ehrenfest time : interestingly , the contribution @xmath119 does not involve a crossing and therefore shows no dependence on the ehrenfest time . however , it is cancelled by the contribution @xmath120 , if the travel time between @xmath102 and @xmath103 is shorter than the ehrenfest time . thus , adding all contributions together , we indeed find , that effectively only trajectories with a duration longer than the ehrenfest time are responsible for the interaction correction . the hartree contribution to the altshuler - aronov correction is given by the two upper diagrams in fig . [ fig : fock ] . proceeding as in the case of the fock contribution , each green function is written as a sum over classical trajectories , which must then be piecewise paired in order to give a nonvanishing contribution to the interaction correction to the conductance . the resulting configurations of classical trajectories are shown schematically in fig . [ fig : diag3 ] . the trajectory configurations of fig . [ fig : diag3 ] are in one - to - one correspondence to those of figs . [ fig : diag1 ] and [ fig : diag2 ] for the fock contribution to @xmath0 : the diagram of fig . [ fig : diag3]a corresponds to that of fig . [ fig : diag1 ] , whereas the diagrams of fig . [ fig : diag3]b e correspond to those of fig . [ fig : diag2]a d . unlike the fock diagrams , all diagrams for the hartree correction involve a finite - angle crossing of the trajectories , in addition to the small - angle encounter of figs . [ fig : diag3]b e . another important difference is that the action difference @xmath121 for the hartree case depends on the two positions @xmath88 and @xmath89 associated with the interaction vertex . denoting the momenta involved in the finite - angle crossing of the trajectories by @xmath122 and @xmath123 , see fig . [ fig : diag3 ] , the action difference contributes an additional oscillating phase factor @xmath124 . ( no fast oscillating phase factors are associated with the integration over @xmath88 and @xmath89 for the fock diagrams . ) for chaotic classical motion , the directions of the momenta @xmath122 and @xmath123 are random and uncorrelated , while the magnitude @xmath125 is fixed by energy conservation . as a result , only a short - range component of the interaction contributes to the hartree correction , and one finds the same expression for @xmath0 as for the fock contribution , with the replacement@xcite @xmath126 where the brackets @xmath127 indicate an average over the momenta @xmath122 and @xmath123 with @xmath125 . in case of a short - range interaction @xmath128 , one verifies that this replacement rule leads to @xmath129 , in agreement with eqs . and . intuitively , the interaction correction associated with the trajectory configurations of fig . [ fig : diag3 ] can be interpreted as the interference of electrons that follow a classical trajectory connecting source and drain contacts , and electrons that additionally scatter from friedel oscillations.@xcite in the configuration of fig . [ fig : diag3]a the trajectory that contains the scattering from the friedel oscillation is shorter than that of the reference trajectory , whereas it is longer in the configurations of figs . [ fig : diag3]b e . the phase difference between the scattered trajectory and the reference trajectory is precisely compensated by the phase of the friedel oscillation.@xcite a similar interpretation applies to the fock contribution , although here the scattering is from friedel oscillations of the density matrix , not of the electron density itself . for the coulomb interaction @xmath130 , due to the long - range nature , it is never sufficient to deal with the first order in perturbation theory only , and effects of dynamical screening have to be included . hence it is not sufficient to consider only diagrams with a single bare interaction line as in fig . [ fig : fock ] ; instead one has to sum up a ladder of diagrams within the random phase approximation ( rpa ) . this analysis is explained in refs . and it can be carried over to the semiclassical formalism without significant modifications . for the purpose of including higher order interaction contributions , the separation into hartree and fock contributions is no longer meaningful . instead , it is favorable to decompose the interaction into singlet and triplet channels . hereto , we consider the interaction amplitude of a scattering process , where two particles with initial momenta @xmath122 ( @xmath123 ) and spin @xmath46 ( @xmath131 ) interact and depart with final momenta @xmath132 ( @xmath133 ) and spin @xmath100 ( @xmath76 ) . since in the semiclassical theory of transport only paired trajectories are relevant , we may restrict our analysis to the case @xmath134 . the classification into singlet ( @xmath135 ) and triplet part ( @xmath136 ) then amounts to separating the interaction amplitude according to its spin structure as @xmath137 where @xmath138 represents the pauli - matrices ( @xmath139 ) . to lowest order , the interaction amplitude consists of the scattering processes shown in fig . [ fig : interaction0 ] . here , in the left process the interaction transfers a small momentum @xmath140 . such small - angle scattering appears in the fock contribution to the interaction correction . the spin structure of this process belongs to the singlet channel . the right process allows for large - angle scattering , which appears in the hartree contribution to the interaction correction . this process has to be split into singlet and triplet contribution , so that we end up with @xmath141 @xmath142 here we anticipated , that for a diffusive system we may average over the directions of the momenta of the electrons . to include screening , one then sums up the rpa - series in each channel , @xmath143 where the disorder - averaged polarization operator @xmath144 is diagonal in spin space ( note , that the disorder average of the polarization operator does not involve a crossing , and therefore has no @xmath1-dependence ) . the discussion so far is valid for weak interaction ( i.e. interaction parameter @xmath145 ) . for stronger interactions , one should also include fermi liquid effects . then the structure of the screened interaction remains the same , but the bare interaction is now expressed in terms of fermi - liquid parameters @xmath146 , @xmath147 let s turn back to the altshuler - aronov correction . applying the preceding analysis , we find that coulomb interaction is properly included , if we calculate the fock - type diagrams as in fig . [ fig : fock ] , where the interaction is replaced by the effective interaction @xmath148 where the factor @xmath149 comes from the spin summation and accounts for the multiplicity of the triplet channel . the precise relation between @xmath150 and @xmath151 follows from the solution of the diffusion equation and will be clarified in the next section ( see eq . ) . ( @xmath123 ) and spin @xmath46 ( @xmath131 ) , and final momenta @xmath132 ( @xmath133 ) and spin @xmath100 ( @xmath76 ) . since the interaction conserves spin , the left process is proportional to @xmath152 and belongs to the singlet channel , while the right process has the structure @xmath153 and therefore splits into singlet and triplet contribution.,width=278 ] in the preceding section , we developed the semiclassical theory of the interaction correction to the conductance , where we expressed our final results in terms of the classical propagator @xmath154 . the explicit expression of this classical propagator is determined by the geometry of the system under consideration . in semiclassical studies , two prototypical geometries are mainly investigated : ballistic quantum dots , where the classical chaotic motion results from reflections at the boundary of the dot , and antidot arrays , where the placement of artificial macroscopic scatterers leads to a chaotic dynamics . for a single ballistic quantum dot , the interaction correction vanishes . the simplest example of a geometry with a non - zero interaction correction is hence a double quantum dot , which was studied in ref . . we will first show , how the results of the previous section are connected to the results of this reference . the remaining part of this section is then devoted to the interaction correction for antidot arrays , which has not been theoretically studied so far . we consider a double dot system , where two identical dots are connected by a ballistic contact of conductance @xmath155 . the first ( second ) dot is connected to the left ( right ) reservoir by a ballistic contact of conductance @xmath156 . the level density ( i.e. the density of states times the dot s area ) for each dot is @xmath157 per spin . within each dot , the phase space is explored uniformly during the chaotic motion . hence we might replace the integration over phase space by a sum over the dots , @xmath158 where we weigh with the available phase space volume @xmath159 of each dot . the classical propagators are replaced by @xmath160 the probability @xmath161 to be in dot @xmath162 after a time @xmath62 , when the particle initially started in dot @xmath163 , is then calculated from the master equation @xmath164 where the rate matrix @xmath131 has the form @xmath165 here , @xmath166 is the rate for transitions between the dots , and @xmath167 is the rate for escape to the left and right lead . the solution to eq . reads @xmath168 the probability @xmath169 ( @xmath170 ) , that a particle in dot @xmath163 has entered via the left contact ( leaves the system via the right contact ) is given by @xmath171 the bare interaction of the double dot system can be described by a capacitive coupling of the dots @xmath172 where @xmath173 describes the coupling of each dot to an external gate , and @xmath174 is the cross - capacitance between the dots . for the inclusion of screening , we make use of the polarization operator @xmath175 from which we obtain the dynamically screened interaction as @xmath176 for the frequencies of interest one may neglect the first term in this equation , and one obtains @xmath177 using the expressions of this paragraph and the eqs . , , , , , and from the last section , one obtains , after some algebra , the result from ref . , @xmath178 here , @xmath179 and @xmath180 are the characteristic dwell times of the double dot system ( they refer to relaxtion of ( anti)symmetric charge configurations ) . for zero ehrenfest time , one recovers the results known from random matrix theory , while for large ehrenfest times , @xmath181 is suppressed as @xmath182 ( for more details , see ref . , where the stated results are too large by a factor of two ) . we note , that the hartree contribution is zero in this case , due to the long - range nature of interaction . in the following , we now apply the theory developed in section [ sec : sctheory ] to quasi one - dimensional and two - dimensional antidot arrays . the antidot arrays consist of a ballistic electron gas with randomly placed disc - shaped scatterers of size much larger than the fermi wavelength . the classical dynamics in such an antidot array is chaotic , and diffusive on length scales much larger than the disc size @xmath3 or the distance between discs . in particular , since the ehrenfest time @xmath183 because of the large logarithm , and since @xmath184 is comparable to the transport time @xmath185 , the diffusive dynamics applies for timescales down to @xmath1 . a diffusively moving particle quickly loses its memory about the direction of motion , so that the classical propagators @xmath81 depend on the positions @xmath42 and @xmath43 associated with the phase space points @xmath107 and @xmath186 only . this leads to significant simplifications of the general semiclassical expressions for the interaction correction @xmath0 . in order to evaluate @xmath0 in this limit , we start by expressing the integration over momentum at fixed energy as an integration over the corresponding angle @xmath187 , @xmath188 where @xmath189 is the density of states per spin . we then find @xmath190 where @xmath191 depends on the positions only and satisfies a diffusion equation , @xmath192 with diffusion coefficient @xmath193 . for a rectangular sample of dimension @xmath194 , coupled to ideal leads at @xmath16 and @xmath17 and with insulating boundaries at @xmath195 and @xmath196 , the solution of eq . reads @xmath197 with the function @xmath198 the sum over @xmath140 runs over @xmath199 with @xmath200 and @xmath201 with @xmath202 . we also use the fourier - transformed diffusion propagator , @xmath203 finally , the probabilities @xmath204 and @xmath205 that a trajectory originating at position @xmath42 exits the sample on the left or on the right , respectively , are @xmath206 which may be derived from the diffusive flux @xmath207 at position @xmath42 and time @xmath62 , for a particle starting from @xmath43 at time @xmath208 . the drude conductance ( eq . ) is then expressed as @xmath209_{x = l}\ ] ] which gives the familiar result @xmath210 let us now turn to the interaction . for the inclusion of screening effects we need the polarization operator , which , for the low frequencies at which the electron dynamics is effectively diffusive , can be expressed through the diffusion propagator , @xmath211,\ ] ] using eqs . , , , we then find , that the effective interaction can be written as @xmath212 where @xmath213 is given by @xmath214 @xmath215 in the singlet and triplet channel , respectively . due to the divergence of the bare coulomb interaction at small momenta , the interaction in the singlet channel is set by the polarization operator solely . in the triplet channel , the interaction depends on the zero angular harmonic of @xmath216 , which is the only free parameter controlling the interaction strength . for the further calculations , it is convenient to make use of the diffusion equation , to write @xmath217 & = & \left ( d\delta+ \frac{i\omega}{\hbar } \right)p({{\mathbf}{r}},{{\mathbf}{r}}';\tau_{\rm e } ) e^{i\omega\tau_{\rm e}/\hbar}.\end{aligned}\ ] ] with the help of additional spatial integrations over delta functions , which we then replace using eq . , the interaction correction to the conductance then takes the form @xmath218 where the function @xmath219 reads @xmath220 with the short - hand notations @xmath221 and @xmath222 . the technical advantage of the structure of eq . is that each term contains the same diffusive propagators . performing several partial integrations , using @xmath223 , @xmath224 , as well as @xmath225 , we are able to simplify the expression and finally obtain @xmath226 together , eq . and represent the main result for the interaction correction in antidot arrays . for zero ehrenfest time , the time integral of eq . equals @xmath227 and one recovers the results for quantum impurities , obtained by standard diagrammatic perturbation theory ( see ref . , where the symbol @xmath228 of this reference equals @xmath229 and the reference misses a factor two for the triplet contribution ) . if the ehrenfest time is finite , it poses a short - time threshold and only electrons with a travel time larger than @xmath1 contribute to the interaction correction . we now discuss ehrenfest - time dependence of @xmath0 in detail for a quasi-1d and a 2d antidot array . for a quasi-1d antidot array ( width @xmath41 much smaller than length @xmath230 ) , we may simplify the diffusion propagator by taking only the diffusion mode with zero transverse momentum into account . after insertion of the diffusion propagators and the interaction into eqs . and , and using the residue technique for the @xmath231-integration , we find @xmath232 , \end{aligned}\ ] ] where the time @xmath233 is the inverse temperature , @xmath234 is the diffusion time , and the function @xmath235 is expressed as @xmath236 with @xmath237 the summation over @xmath238 in eq . can be written in a closed form ( see eq . in the appendix ) . we will now discuss the dependence of @xmath0 on temperature , system size and ehrenfest time several limiting cases . we first consider the limit @xmath240 corresponding to a large antidot array . in this case , the ehrenfest - time dependence of the interaction correction @xmath0 is determined by the ratio @xmath241 . for @xmath242 one finds the result @xmath243,\ ] ] independent of @xmath1 and known from diagrammatic perturbation theory . here @xmath244 is the riemann zeta function ( see appendix [ sec : appdiscussion ] for details ) . on the other hand , for large ehrenfest times or , equivalently , higher temperatures , @xmath245 , the interaction correction @xmath0 acquires an exponential dependence on temperature @xmath246 , @xmath247 independent of @xmath248 . the crossover between these two limiting cases is shown in fig . [ fig : plotd1 ] for different values of the fermi - liquid interaction constant @xmath248 . we emphasize the influence of the interaction constant in the triplet channel @xmath248 : while for small values of @xmath248 , @xmath0 is always negative and monotonously decaying as temperature is increased , a more interesting behaviour is observed at large interaction strength : at small ehrenfest time and @xmath249 the contribution from the triplet channel dominates and gives rise to a positive sign of the interaction correction . on the contrary , if the ehrenfest time is large , the prefactor of the exponential behaviour shows no dependence on @xmath248 to leading order in @xmath250 and is therefore always negative . hence , at sufficiently large interaction strengths , one observes a sign change of the interaction correction , as the temperature is varied . , @xmath1.,width=278 ] in the limit of zero temperature @xmath0 is a function of the ratio @xmath252 only ( we again refer to appendix [ sec : appdiscussion ] for details ) . at zero ehrenfest time , we have @xmath253 with @xmath235 defined in eq . ( [ eq : gm ] ) . for small @xmath248 , we have @xmath254 . in this parameter range the singlet contribution is dominant , which leads to a negative sign of the interaction correction , and to a monotonous decay as a function of @xmath255 . at larger @xmath248 the triplet contribution competes with the singlet contribution , resulting in a sign change of the interaction correction for sufficiently strong interactions , starting at @xmath256 . for @xmath257 , we find an exponential dependence of @xmath0 on @xmath255 @xmath258 the crossover between the limits @xmath259 and @xmath9 is shown in fig . [ fig : plotl ] for several representative values of @xmath248 . , @xmath1.,width=278 ] finally , in the `` classical '' limit of large ehrenfest times , we have @xmath261 where @xmath262 is defined in eq . ( [ eq : gm ] ) . in this parameter regime the interaction correction has the characteristic exponential suppression @xmath263 . we will now consider a two - dimensional antidot array of dimensions @xmath194 , where we restrict ourselves to the limit of large system size , @xmath264 , @xmath265 . in the large - size limit , the relevant quantity is the conductivity @xmath266 ( although we here write results for the conductivity , we formally calculate the conductance and multiply with the geometrical factor @xmath267 , see also the discussion in sec . [ sec : skeleton ] ) . we may then express eqs . and in momentum space , @xmath268 in the limit of zero ehrenfest time , this expression simplifies to the well - known result of diagrammatic perturbation theory.@xcite the full ehrenfest - time dependence is shown in fig . [ fig : plotd2 ] . for @xmath269 , we have the asymptotic behavior @xmath270\ln\frac{\tau_t}{\tau_{\rm e}},\ ] ] which coincides with the well - known expression of quantum impurities , where the role of the elastic scattering time as a short - time cutoff is taken over by the ehrenfest time . in the opposite limit @xmath271 , we obtain an exponential dependence on temperature , @xmath272 as in the one - dimensional situation , at small @xmath248 the singlet contribution dominates the interaction correction , while at larger @xmath248 the triplet contribution competes , and a sign change of the interaction correction as a function of @xmath273 is observed if @xmath274 . of a two - dimensional antidot array for various values of the interaction strength @xmath248 in the triplet channel.,width=278 ] in this article , we considered the effect of a finite ehrenfest time on the interaction correction of a conductor in which the motion of the electrons is described by chaotic classical dynamics . using semiclassical theory of transport , we derived an expression for the interaction correction containing only the interaction propagator and coarse - grained classical propagators of the electronic motion . we confirm the result of ref . , obtained for a double ballistic quantum dot , that the ehrenfest time enters as a short - time threshold for the interaction correction . in other words , the minimal time it takes to traverse system for trajectories responsible for the interaction corrections is the ehrenfest time . as a specific and experimentally relevant example we applied the formalism to antidot arrays , where the coarse - grained classical dynamics follows a diffusion equation . at zero ehrenfest time , we recovered the well known results of the diagrammatic perturbation theory for a disordered metal.@xcite if the ehrenfest time is large , we found that the interaction correction is exponentially suppressed @xmath275 . while the factor @xmath276 is also present for weak localization , the suppression with temperature is specific to the interaction correction . unlike the dwell time @xmath7 , which governs the ehrenfest - time dependence of weak localization , temperature is a variable that can be easily controlled experimentally without changing the classical dynamics , making the interaction correction a promising experimental signature of the ehrenfest - time dependence of quantum transport . ( we note that weak localization depends on temperature implicitly via its dependence on the dephasing time . however , an independent measurement of the dephasing time that enters into the expression for the ehrenfest - time dependence of the weak localization correction is problematic.@xcite ) a particular signature of the underlying classical motion is a sign change of the interaction corrections for strong enough interactions . associated with this sign change is a non - monotonous temperature dependence of the interaction correction , in the temperature range @xmath277 . as long as only the fock contribution is considered , the sign of the interaction correction is negative . if the hartree contribution more precisely , the triplet channel of interaction is added , there is a competition between hartree and fock - type corrections , and the sign of the total interaction correction at small ehrenfest time may change as a function of the interaction strength . for large ehrenfest time , the fock contribution always dominates , so that the sign of the interaction correction is independent on the interaction strength in that limit . as mentioned above , the sign change of the interaction corrections for systems with small ehrenfest times requires a rather strong interaction in the triplet channel . in particular , in @xmath278 systems the threshold was estimated to be @xmath279 , where @xmath280 is the corresponding fermi liquid constant , whereas in quasi one - dimensional systems the condition reads @xmath281 . let us focus on @xmath278 systems with coulomb interaction , for which the condition is less restrictive . both numerical and experimental results for @xmath280 are available in the literature . in general , @xmath280 is a function of the gas parameter @xmath282 , where @xmath283 is the static dielectric constant and @xmath284 the fermi velocity . to get an estimate for typical values of @xmath285 to be expected in antidot array experiments , we take @xmath286 , which was reported in ref . for antidot arrays fabricated from gaas/ algaas heterostructures . together with the dielectric constant for gaas @xmath287 we obtain @xmath288 . in ref . the constant @xmath280 for a given gas parameter @xmath285 was extracted from experimental data using the results for the interaction corrections to conductivity@xcite . for systems with @xmath289 , typical values of @xmath280 were found to be of the order of @xmath290 . this seems consistent with numerical results obtained in ref . , were systems with moderately large @xmath285 were analyzed . for the maximal value of @xmath291 considered in this paper , the fermi liquid constant decreased further down to @xmath292 . considerably larger negative values up to @xmath293 were inferred for systems with @xmath294 in ref . . we take this as evidence that @xmath278 systems with sufficiently strong triplet channel interactions are realizable , provided the additional antidot structure may be superimposed . relevant values of the gas parameter @xmath285 are likely in the range of @xmath295 . we gratefully acknowledge discussions with maxim breitkreiz , alexander finkelstein , igor gornyi and alexander mirlin . this work is supported by the alexander von humboldt foundation in the framework of the alexander von humboldt professorship , endowed by the federal ministry of education and research and by the german research foundation ( dfg ) in the framework of the priority program 1459 `` graphene '' . here we show how to derive the sum rule eq . . we start by noticing that @xmath296 hence we may write @xmath297 the determinant serves as a jacobian for the transformation @xmath298 , @xmath299\delta[{{\mathbf}{r}}-{{\mathbf}{r}}({{\mathbf}{r}}',{{\mathbf}{p}}';t ) ] \nonumber \\ & & \mbox { } \times f[{{\mathbf}{p}}',{{\mathbf}{p}}({{\mathbf}{r}}',{{\mathbf}{p}}';t),t],\end{aligned}\ ] ] where @xmath300 is the phase space point that a trajectory originating from @xmath301 reaches after time @xmath62 . after insertion of @xmath302 we finally arrive at eq . . in this appendix we derive the convolution rule @xmath303 where @xmath304 is an arbitrary function . ( here we omitted contributions from stationary configurations of trajectories , that can not be connected to a single trajectory . for the calculation of the interaction correction , such contributions drop out upon pairing with the retarded trajectory . ) to this end , we first prove an auxiliary identity for the stability amplitudes . hereto we consider a trajectory @xmath46 that connects @xmath43 with @xmath42 . further , let @xmath88 be a point on trajectory @xmath46 , and @xmath305 ( @xmath306 ) be the part of trajectory @xmath46 connecting @xmath43 with @xmath88 ( @xmath88 with @xmath42 ) . the stability amplitude @xmath55 can be then written as@xcite @xmath307 where @xmath308 is the fermi velocity and @xmath309 ( @xmath310 ) denote displacements perpendicular to the trajectory @xmath46 . the last equation implicitly defines @xmath311 . we then introduce @xmath312 such that @xmath313 , @xmath314 , @xmath315 . then the following identity holds : @xmath316 for the proof of eq . ( [ eq : conva ] ) , we note that @xmath317 measures the change of the final coordinate of @xmath46 induced by a small change of the initial momentum . when we consider @xmath46 to be composed by @xmath305 and @xmath306 , a small change of the initial momentum leads to a change of the intermediate coordinate and momentum , which results in a change of the final coordinate , @xmath318\nonumber\\ = & \left(\frac{\partial r_{\perp \alpha''}}{\partial p_{\perp 1\alpha''}}\right)_{r_{\perp 1 } } \left(\frac{\partial r_{\perp 1\alpha'}}{\partial p'_{\perp \alpha'}}\right)_{r'_{\perp } } \left[\left(\frac{\partial p_{\perp 1\alpha'}}{\partial r_{\perp 1\alpha'}}\right)_{r'_{\perp } } -\left(\frac{\partial p_{\perp 1\alpha''}}{\partial r_{\perp 1\alpha''}}\right)_{r_{\perp}}\right]\end{aligned}\ ] ] the last line yields eq . . in a similar fashion , one verifies that @xmath319 the identity we need for the derivation of the convolution rule ( [ eq : convolution - app ] ) involves the partitioning of a single trajectory @xmath46 into three trajectories @xmath305 ( @xmath320 ) , @xmath306 ( @xmath321 ) and @xmath322 ( @xmath323 ) . in this case , we have @xmath324.\ ] ] to see this , one introduces the trajectory @xmath325 as connection of @xmath46 and @xmath305 and makes use of eqs . and . we now turn to the proof of the convolution rule ( [ eq : convolution - app ] ) . hereto , we define @xmath326 where @xmath327 is arbitrary , but fixed . with the abbreviation @xmath328 we write @xmath329 we then insert the semiclassical expressions for the green functions which expresses the former equation as a sum over trajectories @xmath305 ( from @xmath43 to @xmath330 ) , @xmath306 ( from @xmath330 to @xmath88 ) , and @xmath322 ( from @xmath88 to @xmath42 ) . the integration over @xmath88 is carried out within stationary phase approximation . here we only take into account stationary phase configurations , where @xmath305 , @xmath306 and @xmath322 are connected to a single trajectory . other configurations play no role for the calculation of the conductance , since the three advanced trajectories are paired with a single retarded trajectory . hence the convolution @xmath331 may be written as a sum over trajectories @xmath46 connecting @xmath43 with @xmath42 and a sum over points @xmath332 , for which @xmath46 first passes through @xmath333 and then through @xmath332 . deviations @xmath334 from @xmath332 may be parametrized as @xmath335 where @xmath336 ( @xmath337 ) is the angle of the momentum of trajectory @xmath46 at @xmath332 ( @xmath338 ) , and in turns @xmath339 ( @xmath340 ) represent perpendicular displacements of trajectory @xmath46 at @xmath332 ( @xmath338 ) . we then expand the sum of the actions of trajectories @xmath305 , @xmath306 and @xmath322 up to second order in @xmath341 : @xmath342 where @xmath185 is the duration of @xmath46 between @xmath338 and @xmath332 , and @xmath343 is given by @xmath344 \delta \tilde{r}_{\perp 1}^2+\frac{1}{2}\left[\frac{\partial^2\mathcal{s}_{\alpha''}({{\mathbf}{r}}_1^{(0)},\tilde{{{\mathbf}{r}}}_1^{(0)})}{\partial r_{\perp 1}^2}+\frac{\partial^2\mathcal{s}_{\alpha'''}({{\mathbf}{r}},{{\mathbf}{r}}_1^{(0)})}{\partial r_{\perp 1}^2}\right ] \delta r_{\perp 1}^2\nonumber\\ & + \left[\frac{\partial^2\mathcal{s}_{\alpha''}({{\mathbf}{r}}_1^{(0)},\tilde{{{\mathbf}{r}}}_1^{(0)})}{\partial \tilde{r}_{\perp 1 } \partial r_{\perp 1}}\right]\delta r_{\perp 1 } \delta \tilde{r}_{\perp 1},\end{aligned}\ ] ] where as a consequence of energy conservation only perpendicular displacements need to be considered . the integration over @xmath341 can then be accomplished and using eq . we get @xmath345 where the factor @xmath346 originates from the jacobian of the transformation . ( a possible phase shift from taking the squareroot of eq . is needed to restore the correct maslov index . for our calculation however , the maslov index plays no role and we drop it in our expressions . ) on the other hand , we have @xmath347 with that , we finally obtain @xmath348 multiplying with @xmath349 and integrating over @xmath327 then yields eq . . the summation over classical trajectories with a small - angle encounter as given in fig . [ fig : diag2 ] is performed using the procedure of refs . . we here outline the main points of this calculation . the four trajectories @xmath46 ( from @xmath43 to @xmath89 ) , @xmath100 ( from @xmath88 to @xmath42 ) , @xmath131 ( from @xmath43 to @xmath42 ) , and @xmath76 ( from @xmath88 to @xmath89 ) are piecewise paired as shown in fig . [ fig : encounter ] . we start by noting that the choice of the retarded trajectories @xmath46 and @xmath100 fully specifies the advanced trajectories @xmath131 and @xmath76 , since the linearized chaotic dynamics allows for precisely one unique solution of a trajectory that satisfies the initial and final conditions required for the pairing shown in fig . [ fig : encounter ] . moreover , the products of the stability amplitudes are equal , @xmath350 , so that the product of four green functions required for the calculation of @xmath0 can be written as @xmath351 where @xmath352 is the action difference @xmath353 . the summation over trajectories @xmath46 and @xmath100 is restricted to those trajectories that undergo ( at least ) one small - angle encounter . the action difference @xmath121 has two contributions : one contribution from the length difference of the retarded trajectories @xmath46 and @xmath100 vs. the advanced trajectories @xmath131 and @xmath76 , and one contribution from the different energy @xmath354 associated with the trajectory @xmath76 . the former contribution has been calculated in refs . and equals @xmath355 ; the latter contribution equals @xmath356 , where @xmath357 is the duration of the trajectory @xmath76 . note that the product @xmath355 is independent of the position of the phase space point @xmath358 along the encounter . the trajectories @xmath46 and @xmath100 are enumerated by first picking a phase space point @xmath358 on the trajectory @xmath46 . the poincar surface of section at this point may be parameterized with stable and unstable phase space coordinates , which are denoted @xmath359 and @xmath360 respectively . moving the poincar surface of section along the trajectory , the unstable ( stable ) coordinate grows ( shrinks ) as @xmath361 , where @xmath5 is the lyapunov coefficient . we choose the origin of the coordinate system such that the trajectory @xmath46 pierces the poincar surface of section at coordinates @xmath362 . the point @xmath358 is part of an encounter formed by trajectories @xmath46 and @xmath100 , if @xmath100 passes through the poincar surface of section at phase - space distance @xmath363 and @xmath364 , where @xmath365 is a cutoff scale , below which the chaotic classical motion can be linearized . ( one can always simultaneously rescale the coordinates @xmath366 and @xmath367 , such that the cut - off scale is the same for both coordinates . ) the cut - off scale @xmath365 enters the final results in the combination @xmath368 only , so that the precise value of @xmath365 is unimportant , as long as @xmath365 represents a scale characteristic of the _ classical _ dynamics . one verifies that the choice of the phase space point @xmath358 on the trajectory @xmath46 and of the phase space coordinates for the trajectory @xmath100 also specify the two remaining trajectories @xmath131 and @xmath76 . indeed , since @xmath131 is paired with @xmath46 before the encounter , and with @xmath100 after the encounter , it pierces through the poincar surface of section at the same unstable coordinate as @xmath100 and the same stable coordinate as @xmath46 . similar considerations apply to the trajectory @xmath76 . the summation over trajectories @xmath46 and @xmath100 is then written as @xmath369 where @xmath370 is the phase space point located at phase - space displacement @xmath371 from @xmath107 and the dots indicate an arbitrary function of the end - point coordinates of the trajectories @xmath46 and @xmath100 . the time @xmath372 denotes the duration of the encounter . the factor @xmath372 in the denominator accounts for the fact , that @xmath358 can be chosen anywhere along the encounter . the ends of the encounter are determined by the condition that @xmath373 , or that one of the trajectories involved ends , whichever occurs first . since the phase space coordinates @xmath366 and @xmath367 decreases / increase exponentially upon proceeding along the trajectories @xmath46 and @xmath100 , with a rate given by the lyapunov exponent @xmath5 , one finds that @xmath372 is given by the expression @xmath374 + \min[\lambda^{-1 } \ln(c/|u_{\rm e}|),t_{\alpha}].\ ] ] the four different scenarios , depending on whether @xmath375 and @xmath376 are larger or smaller than @xmath377 and @xmath378 , respectively , correspond to the four contributions to @xmath113@xmath114 to @xmath379 . the same situation also occurs in the calculation of the shot noise , see ref . . , @xmath100 , @xmath131 , and @xmath76 , with timescales and phase - space points as described in the main text.,width=278 ] the next step in the calculation is to replace the exact trajectory densities @xmath77 by the coarse - grained ones . in order to account for correlations inside the encounter , we introduce phase space points @xmath107 and @xmath186 at the beginning and end of the encounter if the phase - space points @xmath102 and @xmath103 associated with the interaction position are not inside the encounter . outside the encounter region we may replace the product of the exact trajectory densities by the product of the classical propagators , whereas inside the encounter only a single classical propagator remains . after coarse graining the classical propagators are insensitive to small phase - space difference between @xmath358 and @xmath370 , so that we may replace @xmath370 by @xmath358 . as a result , the integration over @xmath359 and @xmath360 and the integration over @xmath358 can be performed separately . the integration over @xmath358 can be performed using a convolution rule for the classical propagators , @xmath380 the contribution @xmath113 is then expressed as eq . with @xmath119 replaced by @xmath381 where @xmath117 is the fourier transform of @xmath81 , see eq . ( [ eq : ft ] ) . the encounter time for this contribution is given by @xmath382 . in order to perform the integration over the phase - space coordinate @xmath366 and @xmath367 we make use of the integral identity @xmath383 which holds for an arbitrary function @xmath384 which is a slow function of its argument on the time scale @xmath184 . in this equation , the ehrenfest time is defined as @xmath385 one notes that this definition is consistent with eq . ( [ eq : taue ] ) of the main text , since @xmath386 is a classical action characteristic of the classical motion , which can also be expressed as @xmath387 , where @xmath3 is a length scale characteristic of the classical motion . the equivalence to eq . ( [ eq : taue ] ) then follows since @xmath388 . with the equality ( [ eq : enc1 ] ) the result ( [ eq : k2a-1 ] ) of the main text follows immediately . to prove eq . ( [ eq : enc1 ] ) , one makes use of the identity @xmath389 where the heaviside step function @xmath390 is broadened on the scale @xmath184 . taking derivatives on each side , we obtain @xmath391 these equations can be applied to the left hand side of eq . ( [ eq : enc1 ] ) after writing the integration in terms of positive @xmath366 and @xmath367 and after a variable change that uses @xmath372 and @xmath392 as the integration variables , @xmath393 where we used eq . in the last line . for the contribution @xmath394 , the encounter is bounded to the right by the phase space point @xmath103 of the interaction . in this case , the encounter time is given by @xmath395 , where @xmath376 can take values between zero and @xmath396 . the expression for @xmath397 is again of the form , with @xmath119 replaced by @xmath398 the integration over @xmath366 and @xmath367 in eq . is done with the help of the identity @xmath399 which is proven by first performing the integrations over @xmath366 and @xmath367 , and then using the identity ( [ eq : delta ] ) . the final result is eq . ( [ eq : k2b-1 ] ) of the main text . the derivation of eq . ( [ eq : k2c ] ) of the main text is similar . finally , for the fourth and last contribution @xmath114 , the encounter is bounded by both phase space points @xmath102 and @xmath103 of the interaction vertices . the encounter time is here given by @xmath400 , where @xmath376 and @xmath375 vary between @xmath401 and @xmath396 and between @xmath401 and @xmath402 , respectively . the expression for @xmath114 takes the form , with @xmath119 replaced by @xmath403 the integrations over @xmath366 and @xmath367 can be performed with the help of eq . ( [ eq : delta0 ] ) and yield eq . ( [ eq : k2d-1 ] ) of the main text . in this appendix we add some technical details to the discussion of sec . [ sec : antidot ] . the function @xmath235 of eq . can be cast in the following closed analytic expression @xmath404 where @xmath405 for @xmath406 even ( odd ) . in the regime @xmath407 , we may write this as @xmath408\nonumber\\ & \times\left(1+\mathcal{o}(\sqrt{\tfrac{\tau_t}{\tau_{\rm d}}})\right),\end{aligned}\ ] ] where terms with an additional factor @xmath409 correspond to finite size corrections and will be neglected . after replacing the summation over @xmath406 in eq . by an integration , we obtain @xmath410,\end{aligned}\ ] ] with @xmath411e^{-z}.\ ] ] apart from the prefactor @xmath412 , the interaction correction @xmath0 is a function of the ratio @xmath413 only . the limiting behavior for small and large ratios @xmath413 is given in the main text .

let be the set of real numbers which are normal to base . a well - known result of h. ki and t. linton is that is-complete . we show that the set of reals which preserve under addition is also-complete . we use the characteriztion of given by g. rauzy in terms of an entropy - like quantity called the _ noise_. it follows from our results that no further characteriztion theorems could result in a still better bound on the complexity of . we compute the exact descriptive complexity of other naturally occurring sets associated with noise . one of these is complete at the level . finally , we get upper and lower bounds on the hausdorff dimension of the level sets associated with the noise .

1609.08702

let @xmath5 be a positive integer . every real number @xmath6 has a base @xmath1 expansion @xmath7 , and this expansion is unique if we adopt the convention that a tail of the coefficients @xmath8 can not be equal to @xmath9 . recall @xmath6 is _ @xmath1-normal _ if for every @xmath10 we have that @xmath11 , where @xmath12 . for a real number @xmath13 , define real functions @xmath14 and @xmath15 by @xmath16 and @xmath17 . we let @xmath0 denote the set of reals @xmath6 which are normal to base @xmath1 . we let @xmath18 let @xmath19 be a function from @xmath20 to @xmath20 . we say that @xmath19 _ preserves @xmath1-normality _ if @xmath21 . we can make a similar definition for preserving normality with respect to the regular continued fraction expansion , @xmath22-expansions , cantor series expansions , the lroth series expansion , etc . several authors have studied @xmath1-normality preserving functions . they naturally arise in h. furstenberg s work on disjointness in ergodic theory@xcite . v. n. agafonov @xcite , t. kamae @xcite , t. kamae and b. weiss @xcite , and w. merkle and j. reimann @xcite studied @xmath1-normality preserving selection rules . the situation for continued fractions is very different . let @xmath23 $ ] be normal with respect to the continued fraction expansion . b. heersink and j. vandehey @xcite recently proved that for any integers @xmath24 , the continued fraction @xmath25 $ ] is never normal with respect to the continued fraction expansion . in 1949 d. d. wall proved in his ph.d . thesis @xcite that for non - zero rational @xmath13 the function @xmath14 is @xmath1-normality preserving for all @xmath1 and that the function @xmath15 is @xmath1-normality preserving for all @xmath1 whenever @xmath13 is rational . these results were also independently proven by k. t. chang in 1976 @xcite . d. d. wall s method relies on the well known characterization that a real number @xmath6 is normal in base @xmath1 if and only if the sequence @xmath26 is uniformly distributed mod @xmath27@xcite . d. doty , j. h. lutz , and s. nandakumar took a substantially different approach from d. d. wall and strengthened his result . they proved in @xcite that for every real number @xmath6 and every non - zero rational number @xmath13 the @xmath1-ary expansions of @xmath28 and @xmath29 all have the same finite - state dimension and the same finite - state strong dimension . it follows that @xmath14 and @xmath15 preserve @xmath1-normality . it should be noted that their proof uses different methods from those used by d. d. wall and is unlikely to be proven using similar machinery . c. aistleitner generalized d. d. wall s result on @xmath15 in @xcite . suppose that @xmath30 is a rational number and that the digits of the @xmath1-ary expansion of @xmath31 are non - zero on a set of indices of density zero . he proved that the function @xmath32 is @xmath1-normality preserving . g. rauzy obtained a complete characterization of @xmath3 in @xcite . m. bernay used this characterization to prove that @xmath33 has zero hausdorff dimension @xcite . one of the main results of this paper , stated in corollary [ maincor ] , is to obtain an exact determination of the descriptive set theoretic complexity of @xmath3 . a significance of this is explained at the end of [ intro : results ] below . m. mends france asked in @xcite if the function @xmath14 preserves simple normality with respect to the regular continued fraction for every non - zero rational @xmath13 . this was recently settled by j. vandehey @xcite who showed that @xmath34 is normal with respect to the continued fraction when @xmath6 is normal with respect to the continued fraction expansion and integers @xmath35and @xmath36 satisfy @xmath37 . work was also done on the normality preserving properties of the functions @xmath14 and @xmath15 for the cantor series expansions by the first and third author in @xcite and additonally with j. vandehey in @xcite . however , these functions are not well understood in this context . in any topological space @xmath38 , the collection of borel sets @xmath39 is the smallest @xmath40-algebra containing the open sets . they are stratified into levels , the borel hierarchy , by defining @xmath41 the open sets , @xmath42 the closed sets , and for @xmath43 we let @xmath44 be the collection of countable unions @xmath45 where each @xmath46 for some @xmath47 . we also let @xmath48 . alternatively , @xmath49 if @xmath50 where @xmath51 where each @xmath47 . we also set @xmath52 , in particular @xmath53 is the collection of clopen sets . for any topological space , @xmath54 . all of the collections @xmath55 , @xmath44 , @xmath56 are pointclasses , that is , they are closed under inverse images of continuous functions . a basic fact ( see @xcite ) is that for any uncountable polish space @xmath38 , there is no collapse in the levels of the borel hierarchy , that is , all the various pointclasses @xmath55 , @xmath44 , @xmath56 , for @xmath57 , are all distinct . thus , these levels of the borel hierarch can be used to calibrate the descriptive complexity of a set . we say a set @xmath58 is @xmath44 ( resp . @xmath56 ) _ hard _ if @xmath59 ( resp . @xmath60 ) . this says @xmath61 is `` no simpler '' than a @xmath44 set . we say @xmath61 is @xmath44-_complete _ if @xmath62 , that is , @xmath63 and @xmath61 is @xmath44 hard . this says @xmath61 is exactly at the complexity level @xmath44 . likewise , @xmath61 is @xmath56-complete if @xmath64 . a fundamental result of suslin ( see @xcite ) says that in any polish space @xmath65 , where @xmath66 , and @xmath67 is the pointclass of continuous images of borel sets . equivalently , @xmath68 iff @xmath61 can be written as @xmath69 where @xmath70 is borel ( for some polish space @xmath71 ) . similarly , @xmath72 iff it is of the form @xmath73 for a borel @xmath74 . the @xmath67 sets are also called the _ analytic _ sets , and @xmath75 the _ co - analytic sets_. we also have @xmath76 for any uncountable polish space . h. ki and t. linton @xcite proved that the set @xmath0 is @xmath77-complete . further work was done by v. becher , p. a. heiber , and t. a. slaman @xcite who settled a conjecture of a. s. kechris by showing that the set of absolutely normal numbers is @xmath77-complete . furthermore , v. becher and t. a. slaman @xcite proved that the set of numbers normal in at least one base is @xmath78-complete . k. beros considered sets involving normal numbers in the difference heirarchy in @xcite . let @xmath79 be the set of numbers normal of order @xmath80 in base @xmath1 . he proved that for @xmath5 and @xmath81 , the set @xmath82 is @xmath83-complete ( see @xcite for a definition of the difference hierarchy ) . additionally , the set @xmath84 is shown to be @xmath85-complete . we are interested in determining the exact descriptive set theoretic complexity of @xmath3 and some related sets . the definition of @xmath3 shows that @xmath3 is @xmath75 , since it involves a universal quantification . it is not immediately clear if @xmath3 is a borel set , but this in fact follows from a result of rauzy . specifically , rauzy @xcite characterized @xmath3 in terms of an entropy - like condition called the _ noise_. we recall this condition and associated notation from @xcite . for any positive integer length @xmath86 , let @xmath87 denote the set of functions from @xmath88 to @xmath1 . we call an @xmath89 a _ block function _ of width @xmath86 . as in @xcite we set , for @xmath90 , @xmath91 where @xmath92 is the ( fractional part ) of the base @xmath1 expansion of @xmath6 . we also let for @xmath93 @xmath94 we then define the lower and upper noises @xmath95 , @xmath96 of @xmath6 by : @xmath97 where @xmath98 the upper entropy @xmath96 is defined similarly using @xmath99 where @xmath100 for a fixed @xmath93 we also let @xmath101 and similarly for @xmath102 . rauzy showed that @xmath103 iff it has the maximal possible noise in that @xmath104 . furthermore , @xmath105 iff it has minmal possible noise in that @xmath106 . it is therefore natural to ask for any @xmath107 $ ] , what are the complexities of the lower and upper noise sets associated to @xmath108 . specifically , we introduce the following four sets . let @xmath109 $ ] . let @xmath110 finally , we let @xmath111 thus , @xmath112 , and @xmath113 . the ki - linton result shows that @xmath0 , and thus @xmath114 is @xmath2-complete for any base @xmath1 . recall also the becher - slaman result which shows that the set of reals which are normal to some base @xmath1 forms a @xmath115-complete set . we have the following complexity results . [ mt ] for any @xmath116 , the set @xmath117 is @xmath4-complete and the set @xmath118 is @xmath2-complete . for any @xmath119 $ ] , the set @xmath120 is @xmath2-complete , and the set @xmath121 is @xmath122-complete . for @xmath123 , the set @xmath124 is @xmath4-complete , and the set @xmath125 is @xmath2-complete . as a corollary we obtain the ki - linton result as well as the determination of the exact complexity of @xmath3 . [ maincor ] the sets @xmath0 and @xmath3 are both @xmath2-complete . we have @xmath126 iff @xmath127 , and @xmath105 iff @xmath128 , so the result follows immeduately from theorem [ mt ] . in defining the noise , it is sometimes convenient to use the minor variation @xmath129 that is , the block function predicts the next digit rather than the previous digit . in this case we must start the sum at @xmath130 rather than @xmath27 , but this does not affect any of the limits used in defining @xmath95 or @xmath96 . in proving theorem [ mt ] we will work with @xmath6 in the sequence space @xmath131 where @xmath132 is the set of @xmath133 which do not end in a tail of @xmath9 s . this is a polish space as @xmath134 is a compact polish space ( with the usual product of discrete topologies on @xmath135 ) and @xmath132 is a @xmath136 ( that is , @xmath122 ) subset of @xmath134 . this is permissible as the map @xmath137 given by @xmath138 is continuous . so , for example , given that @xmath139 is @xmath2-complete , where @xmath140 is defined as @xmath118 , except we consider directly @xmath133 , then it follows that @xmath118 is @xmath2-complete . for if @xmath118 were in @xmath141 , then so would be @xmath139 since @xmath142 , that is , @xmath143 . we remark on the significance of complexity classifications such theorem [ mt ] . aside from the intrinsic interest to descriptive set theory , results of this form can be viewed as ruling out the existence of further theorems which would reduce the complexity of the sets . for example , rauzy s theorem reduces the complexity of @xmath3 from @xmath75 to @xmath2 . the fact that @xmath144 is @xmath2-complete tells us that there can not be other theorems which result in a yet simpler characterization of @xmath3 . lastly , we wish to approximate the hausdorff dimension of the sets @xmath145 , and @xmath124 . put @xmath146 . [ hd ] for @xmath147}$ ] we have @xmath148 furthermore @xmath149 before proving the main theorem in [ sec : proofmt ] we show a property of the noise which shows that one must be careful when attempting to construct reals with a desired lower - bound for the noise . if we have a set @xmath150 with density @xmath36 , then for almost all @xmath133 , if @xmath151 is the result of copying @xmath6 to the set @xmath61 and taking value @xmath152 off of @xmath61 , then we easily have that @xmath153 . the next lemma shows that it might be possible to lower the noise by taking non - zero values off of @xmath61 . [ ln ] there is a periodic set @xmath150 , say with period @xmath154 and density @xmath155 , such that if @xmath156 satisfies @xmath157 and @xmath158 , then there is a @xmath159 with @xmath160 and @xmath161 . fix a positive integer @xmath80 , and let @xmath162 . let @xmath61 be the set with period @xmath163 with @xmath164 . supppose @xmath165 with @xmath166 and @xmath167 , where @xmath168 is the density of @xmath61 . we show that there is a @xmath159 with @xmath169 with @xmath170 . consider the @xmath171th block @xmath172 of digits of length @xmath163 , that is @xmath173 . let @xmath174 be the last @xmath80 digits of the block @xmath172 . the idea of the following argument is simply to code @xmath175 via the position of a single digit in @xmath176 . if @xmath177 , let @xmath178 , where @xmath179 , be the binary repsesentation of the integer represented by the base @xmath1 expansion @xmath108 , where we put the least significant digits first . let @xmath180 be the result of translating the digits of @xmath181 according to @xmath182 , @xmath183 . note that @xmath184 will never have more than @xmath185 consecutive @xmath27 s . let @xmath186 . we will describe a particular block function @xmath187 and @xmath159 as above . we describe @xmath188 , which only depends on @xmath175 . we say a sequence @xmath189 is _ canonical _ if it is a subsequence of a sequence of the form @xmath190 where @xmath191 where @xmath192 and @xmath193 is the @xmath80-digit base @xmath1 expansion of @xmath194 ( with least significant digit first ) . we assume here that @xmath195 . thus , a canonical sequence is a way of `` counting '' from @xmath152 to @xmath13 . the block function @xmath196 operates as follows . if @xmath197 is the constant @xmath152 sequence , then @xmath198 . if the first @xmath27 ( reading from the left ) in @xmath108 occurs at position @xmath154 to the right of the midpoint of @xmath108 , then @xmath199 is the next digit of a canonical sequence starting at the position of this @xmath27 , that is , @xmath200 for a canonical sequence @xmath201 of length @xmath202 . if @xmath203 , then @xmath196 checks to see if there is a sequence of @xmath204 consecutive @xmath27 s in @xmath108 . if not , @xmath198 . if so , let @xmath30 be the position which is the start of the first such sequence . note that @xmath205 is large enough so that if @xmath206 , then reading to right from @xmath30 there are enough positions in @xmath108 to see a complete `` cycle '' of a canonical sequence ( that is , some @xmath207 ) . likewise if @xmath208 there is enough room to see a cycle to the left of @xmath30 . then @xmath196 checks to see if the positions to the right ( left if @xmath209 ) of @xmath30 give a cycle of a canonical sequence . if so , we check to see if @xmath108 is the subsequence of the corresponding canonical sequence . if so , @xmath199 is the next digit of this canonical sequence , and if not we set @xmath198 . this completes the definition of @xmath196 . we now define @xmath188 . this consists of a canonical sequence @xmath201 starting at the unique position @xmath210 so that the @xmath80 digits in @xmath211 correspond to the substring @xmath193 of some @xmath207 in @xmath201 . this completes the definition of @xmath212 . we claim that for each @xmath213 , if we let @xmath214 $ ] , then @xmath215 to see this , first note that @xmath196 will predict a @xmath152 at the start of the block @xmath172 ( corresponding to the @xmath152 after a @xmath193 ) , which is the correct value , and then predict a @xmath27 ( corresponding to the start of a @xmath216 sequence ) at the next position , which is incorrect ; note that @xmath172 is large enough so that the first @xmath27 in it is far to the right of the start of the block . given the two initial @xmath152 s in @xmath172 , @xmath196 will continue to predict @xmath152 until @xmath196 reaches the point @xmath213 which is the first @xmath27 ( as @xmath196 can not find the @xmath204 consecutive @xmath27 s it needs to consider outputting a non - zero value ) . at position @xmath30 , @xmath196 will also predict a @xmath152 , which is incorrect . after position @xmath30 , @xmath196 will make correct predictions through the end of @xmath211 , as @xmath217 $ ] is canonical , where @xmath218 is the last position of @xmath211 . thus , @xmath219 , while @xmath220 . so , for @xmath221 , we may choose @xmath86 large enough so that the above construction of @xmath212 works , and we then have @xmath222 . the upper bounds for the complexities of the sets of theorem [ mt ] all follow by straightforward computations from the definitions of these sets . for example , consider @xmath117 . we have @xmath223 since for fixed @xmath224 the set @xmath225 is clopen , this shows that @xmath226 . the following lemma and its proof will be used several times in the proofs for the lower bounds on the complexities in theorem [ mt ] . [ tel ] let @xmath150 be an infinite set with upper density @xmath36 , and let @xmath227 . then for almost all @xmath133 , if @xmath228 is defined by @xmath229 then @xmath230 . fix @xmath231 , and fix a sufficiently fast growing sequence @xmath232 ( we will specify how fast the sequence needs to grow below ) . we also choose the @xmath233 such that @xmath234 , where @xmath235 is the density of @xmath61 among @xmath236 . consider the block of integers @xmath237 . let @xmath238 , so @xmath239 . let @xmath240 be the number of block functions of width @xmath80 . consider one of these block functions @xmath241 . consider the function @xmath242 defined as follows . if @xmath243 , let @xmath244 be the result of copying @xmath108 to @xmath245 and copying @xmath246 to @xmath247 ( that is , for @xmath248 , @xmath249 where @xmath80 is the @xmath86th element of @xmath250 ) . if @xmath251 is the @xmath194th element of @xmath252 , let @xmath253)-s'(p_i ) \mod b$ ] . so , @xmath254 . clearly @xmath255 is a bijection from @xmath256 to @xmath256 . so , the number of @xmath257 for which there are exactly @xmath258 many @xmath259 such that @xmath260)\neq s'(p_i)$ ] is equal to the number of @xmath243 such that @xmath108 has exactly @xmath258 non - zero digits . so , the number @xmath261 of @xmath243 such that @xmath262 ) \neq s'(p_i)\ } \right | } \geq m_k\left(\frac{b-1}{b}-\epsilon_k\right)\ ] ] is at least as big as the number of @xmath243 such that @xmath263 . from the law of large numbers we have that @xmath264 . so , for large enough @xmath265 we have that for all @xmath266 @xmath267 ) \neq t{{}^\smallfrown}s'(p_i ) \big \ } \big | \geq m_k \left(\frac{b-1}{b}-\epsilon_k\right ) \bigg \ } \bigg | \\ & \geq 1-\frac{\epsilon_k}{l_k } \end{split}\ ] ] it follows that for fixed @xmath268 that for all sufficiently large @xmath269 we have that for all @xmath270}$ ] that @xmath271)\neq t{{}^\smallfrown}s'(i ) \big \ } \big | \\ & \geq m_k\left(\frac{b-1}{b}-\epsilon_k\right ) \bigg \ } \bigg | \geq 1- l_k \frac{\epsilon_k}{l_k } = 1-\epsilon_k . \end{split}\ ] ] since @xmath272\left(\frac{b-1}{b}-\epsilon_k\right ) \\ & = dn_k\left(\frac{b-1}{b}\right)\left[1 -\frac{\epsilon_k}{d}-\epsilon_k \frac{b-1}{b}- \frac{n_{k-1}}{dn_k}\right ] + n_k \epsilon_k^2 + n_{k-1}\epsilon_k \\ & \geq \frac{d ( b-1)n_k}{b}(1 - 3b \epsilon_k ) \end{split}\ ] ] ( assuming that @xmath273 ) , it follows that for all large enough @xmath233 and all @xmath270}$ ] we have that @xmath274)\neq t{{}^\smallfrown}s'(i ) \big \ } \big | \\ & \geq \frac{d ( b-1)n_k}{b}(1 - 3b \epsilon_k ) \bigg \ } \bigg | \geq 1-\epsilon_k . \end{split}\ ] ] we assume now that the @xmath233 are sufficiently fast growing to satisfy these inequalities . since @xmath275 , it follows from borel - cantelli that for @xmath276 almost all @xmath133 that for any @xmath93 , there are cofinitely many @xmath80 such that @xmath277)\neq x'(i ) \}| \geq \frac{d ( b-1)n_k}{b}(1 - 3b \epsilon_k)\ ] ] and thus for @xmath276 almost all @xmath133 and all @xmath93 we have that @xmath278)\neq x'(i ) \}| \geq \frac{d(b-1 ) } { b},\ ] ] and so @xmath279 . the next lemma suffices to show that @xmath3 is @xmath2-complete . we give an alternate , somewhat simpler , proof of the lemma after the current proof . however , the first proof more resembles the proofs of the other parts of theorem [ mt ] to follow . [ a3 ] for any @xmath116 , @xmath118 is @xmath2-hard . for @xmath280 , we view @xmath6 as coding the sequence @xmath281 in @xmath282 where @xmath283 . we let @xmath284 . it is well - known that @xmath285 is @xmath2-complete . we define a partition of @xmath286 into disjoint arithmetical sequences as follows . let @xmath287 , be the set of even integers , and in general let @xmath288 . the @xmath289 are pairwise disjoint arithmetic progressions , and @xmath290 . note that @xmath291 . each @xmath292 clearly has density @xmath293 . fix a set @xmath294 such that @xmath295 . let @xmath296 be the density of @xmath297 . let @xmath298 and let @xmath299 be the density of @xmath74 . we will take two fast growing sequences @xmath300 and @xmath301 of natural numbers ( the @xmath302 will grow faster than the @xmath303 ) . we will then set @xmath304 . also , let @xmath305 and @xmath306 . note that @xmath307 . we first define the @xmath303 . assume @xmath308 have been defined . let @xmath309 be large enough so that 1 . [ t1 ] for each @xmath310 with @xmath311 , the density of @xmath312 in @xmath313 is at least @xmath314 . [ t2 ] @xmath315 . [ t3 ] for every @xmath316 and any @xmath317 , if @xmath318 has size at least @xmath319 , and @xmath320 , then we have that @xmath321 , where @xmath322 is the number of @xmath323 such that @xmath108 is @xmath152 off of @xmath61 and we have that @xmath324 properties ( [ t1 ] ) and ( [ t2 ] ) are easily satisfied , and property ( [ t3 ] ) can be satisfied as in the proof of lemma [ tel ] . also , the properties continue to hold if we replace @xmath325 and @xmath326 with @xmath327 and @xmath328 for any postive integer @xmath258 . we next define the sequence @xmath329 , which will be sufficiently fast growing with respect to the @xmath330 . namely , such that for any @xmath194 and any @xmath331 , @xmath332 . we could , for example , take @xmath333 . set @xmath304 and let @xmath305 . we define now the map @xmath334 which will be our reduction from @xmath285 to @xmath118 . we first define a particular real @xmath335 . to do this , consider integers @xmath336 . [ tc ] @xmath337 . for fixed @xmath338 , the number of @xmath339 such that @xmath340 has size at most @xmath341 , since @xmath342 , @xmath343 , and @xmath344 . also , the number of @xmath345 such that @xmath346 for some @xmath347 is at most @xmath341 . this is because @xmath348 , and if @xmath349 then we must have @xmath350 . from the choice of the @xmath302 , this set has size at most @xmath341 . fix for the moment any @xmath351 , and consider the block @xmath352 . let @xmath353 be the set of @xmath354 with @xmath345 , @xmath347 , and with @xmath346 . by claim [ tc ] , consider the set @xmath356 of `` patterns , '' by which we mean elements of @xmath357 . for each pattern @xmath358 , and real @xmath165 , let @xmath359 be the sequence defined by : @xmath360 for each fixed pattern @xmath361 , from properties [ t2 ] , [ t3 ] of the @xmath303 it follows that the @xmath276 measure of the set of @xmath165 such that for all @xmath362 , if @xmath363 then for all @xmath320 we have @xmath364 is at least @xmath365 . since the number of patterns in @xmath356 is at most @xmath366 , we have that the @xmath276 measure of the set of @xmath165 such that for all patterns @xmath361 and all @xmath367 equation ( [ eqmi ] ) holds is at least @xmath368 . by borel - cantelli it follows that for fixed @xmath194 that for @xmath276 almost all @xmath165 that there are cofinitely many @xmath341 such that ( @xmath369 ) for all @xmath320 , and all @xmath370 , and all patterns @xmath371 , if @xmath363 as above , then equation ( [ eqmi ] ) holds . by countable additivity of @xmath276 , for @xmath276 almost all @xmath165 the previous statement holds for all @xmath194 . fix @xmath156 in this measure one set . we now define the continuous map @xmath372 which will be the reduction from @xmath285 to @xmath118 . let @xmath280 code the @xmath373 , where @xmath374 . we define @xmath375 by : @xmath376 the map @xmath377 is clearly continuous . we show that @xmath378 iff @xmath379 . suppose first that @xmath378 , so @xmath380 . then for all @xmath381 we have that for large enough @xmath382 that @xmath383 . the @xmath384 ( of width @xmath27 ) which constantly outputs @xmath152 will then correctly compute @xmath385 for all sufficiently large @xmath382 whenever @xmath381 . for @xmath386 , we will have @xmath387)\neq x(k)\}|= \frac{1}{2^{i+1}}\frac{b-1}{b}$ ] . from this it follows that @xmath388)\neq x(k)\}|= \frac{b-1}{b}\sum_{i \notin j } \frac{1}{2^{i+1}}=s$ ] . so , @xmath389 and so @xmath390 . suppose next that @xmath391 . let @xmath392 be least such that there are infinitely many @xmath341 such that @xmath393 . let @xmath394 be the @xmath392th element of @xmath395 . if @xmath396 is in @xmath395 , then for large enough @xmath397 we have @xmath383 . if @xmath396 is not in @xmath395 , then for large enough @xmath382 we have @xmath398 . also , for infinitely many @xmath341 we have that @xmath398 for all @xmath399 . fix @xmath93 . say @xmath400 . from the definition of @xmath401 , there are cofinitely many @xmath341 such that ( @xmath402 ) holds . intersecting a cofinite and an infinite set gives a @xmath403 such that @xmath393 , ( @xmath402 ) holds , and for all @xmath404 we have that @xmath405 for all @xmath406 . thus , for all @xmath396 which are in @xmath395 we have that @xmath407 for all @xmath406 with @xmath408 . let @xmath409 be the pattern such that @xmath410 iff @xmath411 or @xmath412 $ ] where @xmath413 is the @xmath414th element of @xmath395 . from ( @xmath402 ) applied to @xmath415 and the pattern @xmath416 , and noting that from the definition of @xmath375 that @xmath417 , we have that @xmath418 and thus by property ( [ t2 ] ) of the @xmath303 , the density of @xmath419 such that @xmath420 is at least @xmath421 . since this holds for infinitely many @xmath341 , we have that @xmath422 . thus , @xmath423 . we now present the alternate simpler proof of lemma [ a3 ] . let @xmath424 . for each @xmath425 pick a probability vector @xmath426 such that @xmath427 and construct real numbers @xmath428 such that for any @xmath80 @xmath429 then @xmath430 for every @xmath86 and the constant @xmath152 block function @xmath431 is the minimizer of @xmath432 . let @xmath433 be a sufficiently quickly growing sequence of integers with @xmath434 such that for @xmath435 , @xmath436 , and for all @xmath437 we have @xmath438 put @xmath439 . we view @xmath280 as coding the sequence @xmath440 in @xmath282 where @xmath441 . define @xmath442 and define @xmath375 by @xmath443 the map @xmath377 is continuous . suppose @xmath378 . then @xmath444 . fix @xmath86 and note @xmath445 now for @xmath446 and @xmath447 @xmath448 and @xmath449 for @xmath450 we have @xmath451 on the other hand for @xmath452 @xmath453 thus @xmath454\ } \\ & \qquad + \frac{j-1}{(j-1)^2 } + \frac{2}{j-1 } + \frac{1}{j}\\ & = \lim_\ell \left ( s + \limsup_j \max{\left \ { \frac{1-s-1/b}{m(j-1 ) } , \frac{1-s-1/b}{m(j ) } \right \}}\right ) = s.\end{aligned}\ ] ] now suppose @xmath391 . then @xmath455 . fix @xmath86 and note @xmath456 which implies @xmath457 we next show the lower bound for @xmath120 . the proof is similar to that for @xmath118 . [ lema2 ] for any @xmath458 $ ] the set @xmath120 is @xmath2-complete . we reduce the set @xmath459 to @xmath120 . let @xmath294 be such that @xmath460 , where we recall @xmath296 is the density of @xmath297 . let the sequences @xmath461 , @xmath462 be as in lemma [ a3 ] , and as in that lemma let @xmath463 , @xmath305 . we use also the notion of pattern defined as follows . for @xmath464 let @xmath465 and let @xmath356 be the set of functions @xmath154 from @xmath466 to @xmath467 such that if @xmath468 then @xmath469 for all @xmath470 . although @xmath466 now has size greater than @xmath471 , this restriction on @xmath154 implies that @xmath472 . arguing as in lemma [ a3 ] , for each @xmath473 there is a @xmath276 measure one set @xmath474 of @xmath165 such that for cofinitely many @xmath341 , all @xmath320 , all @xmath475 , and all patterns @xmath361 , if we define @xmath359 by @xmath476 then if @xmath477 we have @xmath478 we fix @xmath165 in all of the measure one sets @xmath474 . we define @xmath334 by : @xmath479 we show @xmath377 is a reduction from @xmath480 to @xmath120 . if @xmath481 , then there is an @xmath392 such that for infinitely many @xmath341 we have @xmath482 . let @xmath483 be the @xmath392th element of @xmath395 . we may assume that @xmath401 is such that for every @xmath194 that @xmath484 . it follows that for @xmath485 and any @xmath486 that for all sufficiently large @xmath341 that @xmath487 it follows that for any @xmath488 and @xmath486 that for all sufficiently large @xmath341 with @xmath482 that @xmath489 it follows by considering the trivial @xmath93 which always outputs @xmath152 that @xmath490 thus , @xmath491 . suppose next that @xmath492 . let @xmath381 . for all sufficiently large @xmath341 we have that @xmath398 for all @xmath493 where @xmath494 and @xmath495 . for such @xmath341 , consider the pattern @xmath361 such that @xmath496 if @xmath468 , and for @xmath345 with @xmath495 and @xmath347 we have @xmath497 , where @xmath413 is the @xmath414th element of @xmath395 . from the properties of @xmath401 we have that for all @xmath498 that @xmath499 this gives that @xmath500 , and so @xmath501 . we next show that the set @xmath121 is @xmath122-complete for any @xmath458 $ ] . [ lema4 ] for any @xmath458 $ ] , the set @xmath121 is @xmath122-complete . let @xmath502 be the set @xmath503 . let @xmath150 be a set of density @xmath504 $ ] . we define the integers @xmath505 inductively , and set @xmath506 $ ] . given @xmath507 , we let @xmath309 be large enough so that for each @xmath508 , we have that for all @xmath320 that @xmath509 this is possible by law of large numbers and lemma [ tel ] . by borel - cantelli , for @xmath276 almost all @xmath280 we have that @xmath510 where @xmath511 is the result of replacing @xmath512 with @xmath201 . fix @xmath513 in this measure one set . we define @xmath514 by : @xmath515 if @xmath492 , then there are infinitely many @xmath341 such that @xmath516 and thus @xmath517 . from the definition of @xmath401 , there is a tail of these @xmath341 for which @xmath518 thus , for any @xmath384 we have that @xmath519 we may assume that @xmath520 with @xmath341 , and it follows that @xmath521 , so @xmath522 . if @xmath481 , then @xmath385 is @xmath152 for all large enough @xmath80 . this gives that @xmath523 , and so @xmath524 . we next show the lower - bound for @xmath117 . [ inflower ] for @xmath116 , the set @xmath117 is @xmath4-hard . fix @xmath116 . let @xmath525 . @xmath526 is a @xmath4-complete set , and so it suffices to reduce @xmath526 to @xmath117 . let @xmath294 be such that @xmath527 . let the @xmath297 partition @xmath286 as before , so @xmath528 . we will define a fast growing sequence @xmath505 , and we will also let @xmath529 . each @xmath171 codes a triple @xmath530 , where @xmath531 . we will also define a certain sufficiently fast growing function @xmath532 ( this will be the map @xmath533 of claim [ u2 ] ) . also as in the proof of lemma [ a3 ] we will fix a particular @xmath335 from a certain @xmath276 measure one set which will guide the construction of the reduction map @xmath377 . the construction will be similar to that of lemma [ a3 ] , the main difference being that at some points of the construction instead of copying @xmath152s to parts of the block @xmath172 we will copy a portion of @xmath401 repeated with period @xmath534 . [ u1 ] let @xmath535 , a disjoint union , and assume @xmath61 , @xmath74 , @xmath536 have period @xmath154 ( that is , @xmath537 , @xmath538 , @xmath539 have period @xmath154 ) . let @xmath61 , @xmath74 , @xmath536 have densities @xmath540 , @xmath541 , @xmath542 respectively . then for @xmath276 almost all @xmath335 ( in fact , if @xmath401 is normal in base @xmath1 ) we have the following . there is an @xmath93 such that for any @xmath543 with @xmath544 and @xmath545 of period @xmath154 , we have that @xmath546 . let @xmath401 be normal in base @xmath1 . let @xmath547 . consider sequences @xmath548 for some integer @xmath171 . for large enough @xmath171 , the probability that a sequence @xmath549 will have the property that @xmath550 has size at least @xmath551 is at least @xmath552 . this follows by the argument of lemma [ tel ] . we call such an @xmath244 good . since @xmath401 is normal in base @xmath553 , it follows that for large enough @xmath554 that the number of @xmath555 such that @xmath556 is good is at least @xmath557 . for such @xmath554 we have that @xmath558 let @xmath559 be the block function such that @xmath560 , so @xmath196 is simply guessing that @xmath561 will be @xmath562 . it follows that if @xmath563 is such that @xmath564 is normal , then for large enough @xmath554 we have that @xmath565 thus , @xmath566 . since @xmath567 was arbitrary , @xmath546 . for @xmath165 and @xmath568 , we define @xmath569 by @xmath570 for @xmath571 , and for @xmath572 we set @xmath573 . thus , after the first @xmath80 digits of @xmath401 , we repeat the digits of @xmath401 periodically with period @xmath154 . [ u2 ] let @xmath574 , @xmath575 . then there is a @xmath576 which is a power of @xmath577 and an @xmath578 such that if @xmath579 are disjoint and of period @xmath154 with densities @xmath540 , @xmath541 , @xmath542 respectively and with @xmath535 , then for any @xmath580 , we have that @xmath581 } } \geq 1-\epsilon'$ ] , where @xmath582 is the set of @xmath583 such that for any @xmath584}$ ] if @xmath585 is defined by @xmath586 then for any @xmath400 we have @xmath587\ \ , \big [ \frac{1}{k } \geq ( d_a+d_b)\frac{b-1}{b}(1 -\epsilon)-j_0 d_c \big ] . \end{split}\ ] ] furthermore , this holds for all @xmath588 . fix @xmath589 , @xmath400 , @xmath547 . we show that for large enough @xmath590 , and @xmath591 , a disjoint union with @xmath592 having period @xmath154 , and @xmath593 , that the probability a @xmath594}$ ] satisfies @xmath595 is as close to @xmath27 as desired . given @xmath594}$ ] , let @xmath596}$ ] be defined as @xmath597 above except we put @xmath598 for @xmath599 $ ] . the argument of lemma [ tel ] shows that as @xmath154 grows , with probability approaching @xmath27 in choosing @xmath600)$ ] we have that @xmath601 from the finite version of fubini it follows that for any @xmath602 that for large enough @xmath154 that with probability at least @xmath603 in chosing @xmath604)$ ] that for probability at least @xmath603 in choosing @xmath605)$ ] we have that if @xmath606 is defined from @xmath607 and @xmath608 , then @xmath609 holds . choose @xmath608 in this set of measure at least @xmath603 . let @xmath132 be the set of @xmath605)$ ] such that if @xmath610 ( with @xmath611 ) then @xmath401 satisfies @xmath609 . so , @xmath132 has measure at least @xmath603 . from the law of large numbers we have that the probability that a @xmath165 has the property that for all @xmath612 @xmath613 is at least @xmath614 , where @xmath615 as @xmath616 . we choose @xmath617 small enough so that @xmath618 . it follows that for all @xmath619 that with probability at least @xmath603 , @xmath620 has the property that with probability at least @xmath614 , @xmath621 has the property that if @xmath622 is formed from @xmath623 , @xmath608 , @xmath611 as above , then for all @xmath624 @xmath625 since @xmath626 it follows that for any @xmath619 that with probability at least @xmath627 , @xmath628 has the property that if @xmath606 is obtained from @xmath401 and @xmath629 as above , then for all @xmath624 we have that equation ( [ equ ] ) holds . if we set @xmath630 , then we get the inequality of equation [ equ ] replacing @xmath631 with @xmath632 for any @xmath633 $ ] . consider then @xmath634 , which is defined as @xmath606 except we use the given @xmath629 instead of the @xmath152 function . we have that @xmath635 and @xmath636 can only disagree if @xmath637 . thus , there can be at most @xmath638 many @xmath639 where such a disagreement occurs . the inequality of the claim then follows . [ u2b ] for every @xmath640 there is an @xmath641 such that for any @xmath642 , and any @xmath643 , for all large enough @xmath171 we have that if @xmath644 , @xmath645 , and @xmath646 , then @xmath647 , where @xmath582 is the set of @xmath583 such that if @xmath648 then for any @xmath400 we have @xmath649\ \ , \big [ \frac{1}{k } \geq d_a\left(\frac{b-1}{b}\right)(1 -\epsilon ) \big ] . \end{split}\ ] ] it is enough to fix @xmath392 , @xmath589 , @xmath650 , @xmath567 , @xmath651 , and @xmath154 and we show that for large for large enough @xmath171 the stated property holds . note that there are @xmath652 many choice for @xmath653 . it is enough to fix a choice for @xmath654 are show that for large enough @xmath171 the property holds . this , however , follows immediately from the argument of lemma [ tel ] . from claim [ u2 ] we have the following . let @xmath655 be a sufficiently fast growing sequence of powers of @xmath577 ( exactly how fast the sequence needs to grow will be specified below ) . [ u3 ] for almost all @xmath165 we have the following . for any @xmath589 and @xmath400 , for all large enough @xmath171 we have that if @xmath656 , a disjoint union of sets of period @xmath657 , where @xmath658 and @xmath659 for some @xmath660 , and if @xmath661 and @xmath662 , @xmath663 as in claim [ u2 ] , @xmath664 as in claim [ u2b ] , then if @xmath597 is defined as in claim [ u2b ] , then we have : \(1 ) ( large @xmath341 case ) if @xmath665 then @xmath666\ \ , \big [ \frac{1}{k - b_n } \ , | \ { i \in [ b_n , k ) & \colon e(u'(i - j),\dots , u'(i-1 ) ) \neq u'(i)\ } | \\ & \geq ( d_a+d_b)\left(\frac{b-1}{b}\right)\left(1-\frac{1}{j}\right ) - \frac{j}{2^n } \big ] . \end{split}\ ] ] \(2 ) ( general case ) @xmath666 \ \ , \big [ \frac{1}{k - b_n } \\ & \geq d_a\left(\frac{b-1}{b}\right)\left(1-\frac{1}{n}\right ) \big ] . \end{split}\ ] ] where @xmath540 , @xmath541 are the densities of @xmath61 , @xmath74 in @xmath667 . it is enough to fix @xmath589 , @xmath668 , and show that almost all @xmath401 have the desired property for these values . by borel - cantelli it is enough to show that the probability that @xmath669 satisfies the conclusion of the claim it at least @xmath670 . fix @xmath671 , and we show , assuming @xmath672 is sufficiently large compared to @xmath671 , that the interval @xmath673 has this property . there are at most @xmath171 many partitions @xmath656 of the type stated in the claim ( since the choice of @xmath674 is determined by @xmath660 ) , so it enough to fix @xmath674 and show that with probability at least @xmath675 in choosing @xmath676 the statement of the claim holds . note that the @xmath341 in the claim satisfy @xmath677 , so there is a bound @xmath678 depending only on @xmath171 , such that if @xmath661 , @xmath662 , then @xmath679 . we will assume that @xmath680 for all @xmath171 , and in particular we will choose @xmath672 so that @xmath681 . this is possible as @xmath682 depends only on @xmath683 and not the value of @xmath672 . similarly we may fix @xmath661 and show that with probability at least @xmath684 in choosing @xmath685 we satisfy the claim . in case ( 1 ) , that is , @xmath686 , the conclusion follows immediately from claim [ u2 ] , assuming that @xmath687 . in applying claim [ u2 ] we use @xmath688 . in case ( 2 ) , the conclusion follows immediately from claim [ u2b ] , assuming again that @xmath672 is sufficiently large compared to @xmath671 ( the interval @xmath689 of claim [ u2b ] becomes @xmath667 here ) . we now fix @xmath335 in the measure one set described in claim [ u3 ] . we also fix the fast growing sequence @xmath690 . as we said in claim [ u3 ] , we take @xmath691 , where @xmath678 is the maximum of @xmath692 , where @xmath693 , from claim [ u2 ] and @xmath694 from claim [ u2b ] . claim [ u3 ] then gives the following property of @xmath401 and the @xmath671 . @xmath695 : for any @xmath589 , @xmath400 , for all large enough @xmath171 if @xmath656 where @xmath696 , @xmath697 where @xmath698 , then if @xmath662 ( as in claim [ u2 ] ) and @xmath597 is defined from @xmath401 and @xmath154 as in claim [ u2b ] , then for any @xmath661 if we set @xmath699 we have : 1 . if @xmath665 then @xmath700 . 2 . @xmath701 . we now return to the proof of lemma [ inflower ] . given @xmath702 , we define a function @xmath703 as follows . we let @xmath704 be @xmath705 if @xmath706 for all @xmath707 . otherwise , set @xmath708 . in general , we define @xmath709 the function @xmath710 does the following . the input @xmath194 sets the `` width '' of the search , that is , it will search over the @xmath711 for @xmath707 . the input @xmath341 sets the initial start of the search in that the search will begin at the @xmath712 . the search checks to see if all of these are equal to @xmath152 . if so , it will output @xmath713 , denoting a successful search , and then replace @xmath341 with @xmath714 and begin a new search at @xmath715 . the output @xmath716 records the new value @xmath714 , and the output @xmath717 records the new `` height '' of the search , will in this case is set back to @xmath152 . if not all the values @xmath712 are @xmath152 , then the @xmath341 value remains the same and we increment the height of the search . the means we will search the values @xmath718 where @xmath494 , @xmath341 is the current value of @xmath719 , and @xmath720 , which is the current height . for a given @xmath341 , we keep incrementing the height @xmath80 and see if @xmath721 . if so , the search is successful , and we then increment @xmath341 to @xmath714 and reset the height @xmath80 to @xmath152 . otherwise , we continue to increment the height @xmath80 and continue the search . the search is attempting to verify , step by step , that @xmath722 if this condition holds for some @xmath723 , then @xmath719 will tend to @xmath724 with @xmath201 . if this condition fails for @xmath723 , then for @xmath339 the least integer @xmath725 such that @xmath726 we have that @xmath719 will be equal to @xmath339 for all large enough @xmath201 ( the search will `` get stuck '' at @xmath339 ) . recall that @xmath727 is a sufficiently fast growing sequence , so that @xmath401 and the @xmath727 satisfy claim [ u3 ] . as before , we let @xmath529 . we view @xmath171 as coding a triple of integers which we write as @xmath728 . we define the map @xmath729 as follows . let @xmath702 . recall the @xmath297 are the pairwise disjoint arithmetical sequences of lemma [ a3 ] , so @xmath297 has density @xmath730 . also , @xmath731 is such that @xmath732 . for @xmath386 , we will just copy @xmath401 to @xmath297 . it remains to specify @xmath733 for @xmath734 where @xmath311 . for @xmath735 , where @xmath381 , we define @xmath733 through the following cases ( the definition of @xmath569 is given right before claim [ u2 ] ) . 1 . if @xmath386 , we set @xmath736 . 2 . if @xmath381 and @xmath737 , we set @xmath736 if @xmath738 . 3 . if @xmath311 , @xmath737 , and @xmath739 , then if @xmath740 ( unsuccessful search at step @xmath741 for @xmath742 ) we set @xmath736 . if @xmath311 , @xmath737 , @xmath739 , and @xmath743 ( successful search at step @xmath741 ) , we set @xmath744 , where @xmath745 is defined in claim [ u2 ] . we show that @xmath377 is a reduction from the @xmath4 set @xmath526 to the set @xmath746 first assume that @xmath747 . fix @xmath547 . let @xmath392 be large enough so that @xmath748 where @xmath296 is the density of @xmath297 . let @xmath589 be large enough so that @xmath749 . we can do this as @xmath747 . let @xmath150 be given by @xmath750 that is , the @xmath742 search at step @xmath741 is successful . from the definition of @xmath589 and the properties of @xmath751 we have that @xmath61 is infinite ( that is , there are infinitely many @xmath201 such that @xmath752 ) . for any @xmath753 , @xmath754 in @xmath395 , and @xmath735 , we have that @xmath755 . it follows from claim [ u1 ] , and the fact that the @xmath671 grow sufficiently fast , that there is an @xmath93 ( with say @xmath756 ) such that for all large enough @xmath753 we have that @xmath757 since @xmath486 was arbitrary , we have that @xmath758 , that is , @xmath759 . assume next that @xmath760 . let @xmath392 be least so that @xmath761 . so , for any @xmath341 , @xmath762 has a limiting value @xmath763 as @xmath201 goes to infinity ( i.e. , the width @xmath392 search starting at @xmath341 will always gets stuck ) . note that if @xmath764 , then @xmath765 will reach its limiting value at or before when @xmath762 does , that is , the @xmath766 search will get stuck at or before when @xmath767 gets stuck . so , for all @xmath341 we have that for all sufficiently large @xmath171 with @xmath768 , and @xmath769 , that @xmath770 ( unsuccessful search at step @xmath741 ) . consider @xmath771 where @xmath772 . so , for each @xmath341 we have that for all large enough @xmath171 with @xmath769 that either @xmath773 , in which case @xmath774 , or else @xmath739 , in which case ( since @xmath775 ) we also have that @xmath776 . thus , for any @xmath341 we have that for large enough @xmath171 that @xmath777 is either of the form @xmath778 or else of the form @xmath779 where @xmath780 . that is , the periods @xmath154 used in truncating @xmath401 in the block @xmath172 go to infinity with @xmath171 . it follows from @xmath695 ( where the @xmath392 there is the current @xmath781 ) that for any @xmath547 and any @xmath400 that for all large enough @xmath171 that in either case , @xmath782 or @xmath783 ( in which case @xmath784 for large enough @xmath171 ) , we have that for all @xmath80 with @xmath785 that if @xmath786 for any @xmath341 and all large enough @xmath171 . for any @xmath486 we have for all large enough @xmath171 that @xmath787 we may assume the @xmath671 grow sufficiently fast so that @xmath788 and @xmath789 as @xmath790 ( note here that @xmath678 is defined independently of @xmath671 ) . then for large enough @xmath171 we have the above inequality for @xmath36 holds using @xmath791 for all large enough @xmath80 . thus , @xmath792 , and so @xmath793 . for each @xmath147}$ ] , we know that @xmath794 and @xmath795 which implies @xmath796 . since @xmath117 and @xmath118 do not contain @xmath797 we must introduce the following machinery to compute the hausdorff dimension of these sets . for @xmath133 let @xmath798 be the set of weak-@xmath799 limit points of the sequence of measures @xmath800 where @xmath801 is the shift on @xmath134 . this is a closed convex subset of the shift - invariant probability measures on @xmath134 which we denote by @xmath802 . we say a point @xmath133 is generic for a measure @xmath276 if @xmath803 . for a closed convex subset @xmath804 define @xmath805 to be the set of @xmath133 such that @xmath806 . note @xmath807 is the set of generic points for @xmath276 . recall the measure theoretic entropy of a shift - invariant measure @xmath276 on @xmath134 is @xmath808 \log \mu[b].\end{aligned}\ ] ] colebrook proved the following result in @xcite . in an analogous way to real numbers we can define the noise of a measure @xmath276 as @xmath810 \max_{0 \leq d \leq b-1 } \frac{\mu[d { { } ^\smallfrown}b]}{\mu[b]}.\end{aligned}\ ] ] it is clear that the upper and lower noise of each element in @xmath805 is @xmath811 and @xmath812 respectively . in general @xmath813 thus @xmath814 furthermore @xmath815 now consider @xmath816 the uniform measure on @xmath134 and @xmath817 the dirac point mass at @xmath818 . let @xmath819 be the convex hull of @xmath820 . then @xmath819 is a closed convex subset of @xmath802 and we have @xmath821 . thus @xmath822 and @xmath823 which implies @xmath824 . we also have the following lower bound on @xmath825 @xmath826 note the same lower bound holds for @xmath827 since @xmath828 this second inequality follows since if @xmath829 we can find @xmath830 $ ] such that @xmath831 where @xmath816 is the uniform measure on @xmath134 and @xmath832 since @xmath816 is the measure of maximal entropy . we have that @xmath833 is equivalent to @xmath834 and @xmath835 is equivalent to @xmath836 but we do not have a general expression for @xmath837 one approach to finding this supremum is to restrict our attention to measures which are @xmath80-step markov , that is measures of the form @xmath838 & = \rho(b_1 , \cdots , b_k ) \prod_{i=1}^{\ell - k } p_{b_i \cdots b_{i+k-1 } , b_{i+1 } \cdots b_{i+k}}\end{aligned}\ ] ] where @xmath839 is a probability distribution on @xmath840 ( which we view as a @xmath841 matrix ) and @xmath285 is a @xmath842 matrix of non - negative real numbers such that @xmath843 a @xmath80-step markov chain @xmath276 with stationary distribution @xmath839 and transition matrix @xmath285 has entropy @xmath844 and noise @xmath845 \max_{0 \leq d \leq b-1 } \frac{\mu[d { { } ^\smallfrown}b]}{\mu[b ] } \\ & = 1 - \sum_{b \in b^k } \mu[b ] p_{d b_1 \cdots b_{k-1 } , b_1 b_2 \cdots b_k}.\end{aligned}\ ] ] thus @xmath846 . if we can compute this supremum over all stochastic matrices @xmath285 with steady state @xmath839 we have improved lower bounds on @xmath847 . problems of this type are unfortunately quite difficult in general . this problem is tractable for small @xmath80 however , and we now consider an easy special case . for @xmath850 and @xmath285 with identical rows ( so we now write @xmath851 instead of @xmath356 ) we have the associated markov measure @xmath276 is bernoulli which implies @xmath852 now if @xmath853 then @xmath854 and we can without of loss of generality take this maximum to be at @xmath855 . therefore @xmath856 clearly this supremum is attained when @xmath857 since @xmath858 . thus , we must maximize @xmath859 subject to the constraints @xmath857 , @xmath860 and @xmath861 , which is equivalent to maximizing @xmath862 subject to the constraints @xmath860 and @xmath863 . the maximizer of @xmath864 subject to the constraint @xmath863 occurs when @xmath865 . since @xmath866 we have @xmath867 , so for each @xmath868 we have @xmath860 . thus we have @xmath869 in order to obtain an upper bound on @xmath870 we use the argument of m. bernay @xcite which showed @xmath871 . to do this define the sets @xmath872 and note @xmath873 m. bernay proved the following lemma .

we present a database of time - series photometry of the orion nebula cluster obtained with two ground - based telescopes at different longitudes to provide simultaneous coverage with the 13-d _ chandra _ observation of the cluster . the resulting database of simultaneous optical and x - ray light curves for some 800 pre main - sequence ( pms ) stars represents , by a factor of hundreds , the largest synoptic , multi wavelength - regime , time - series study of young stars to date . this database will permit detailed analyses of the relationship between optical and x - ray variability among a statistically significant ensemble of pms stars , with the goal of elucidating the origins of pms x - ray production . in this first paper , we present the optical observations , describe the combined x - ray / optical database , and perform an analysis of time - correlated variability in the optical and x - ray light curves . we identify 40 stars ( representing 5% of our study sample ) with possible time - correlated optical and x - ray variability . examples of both positive and negative time - correlations are found , possibly representing x - ray flares and persistent coronal features associated with both cool and hot surface spots ( i.e. magnetically active regions and accretion shocks ) . we also find two possible examples of white - light " flares coincident with x - ray flares ; these may correspond to the impulsive heating phase in solar - analog flares . however , though interesting , these represent unusual cases . more generally , we find very little evidence to suggest a direct causal link between the sources of optical and x - ray variability in pms stars . the conclusion that accretion is a primary driver of x - ray production in pms stars is not supported by our findings .

astro-ph0606079

the x - ray luminosities of solar - type stars , when they are very young , are prodigious . indeed , it is now well established that low - mass pre main - sequence ( pms ) stars can emit x - rays at levels up to @xmath1 times that of the present - day sun ( e.g. * ? ? ? but after two decades of research since t tauri stars ( ttss ) were first identified as strong x - ray sources ( e.g. * ? ? ? * ) , we still do not know the origins of x - ray production in young stars . while flares observed in the x - ray light curves suggest that the x - rays are produced mainly in solar - analog magnetic reconnection flares , the absence of an obvious rotation - activity relation in these stars prevents a clear association of the x - rays to dynamo - generated fields @xcite . the flares can be so powerful that non - solar geometries of the magnetic fields must be considered , including field lines linking the star with the protoplanetary disk @xcite . in addition , the role of accretion is unclear , and may affect the production of x - rays either directly or indirectly . for example , energetic shocks associated with accretion flows at or near the stellar surface may enhance x - ray production , as has been suggested in the detailed spectroscopic analyses of tw hya @xcite and bp tau @xcite . at the same time , the mass - loading of magnetic field lines may actually inhibit x - ray production in accreting systems , as suggested by @xcite . these issues are reviewed more fully in @xcite . we have undertaken a synoptic study of the orion nebula cluster ( onc ) combining a nearly continuous , 13-d _ chandra _ observation with simultaneous , multi - wavelength , time - series photometry in the optical . a specific motivation is to establish the incidence of time - correlated x - ray and optical variability in order to begin disentangling the roles of magnetic and accretion activity in the production of x - rays . for example , accretion shocks on the surfaces of classical ttss ( cttss ; ttss that still possess accretion disks ) produce strong variability at optical wavelengths , caused by time - variable accretion , rotational modulation of accretion hot spots , " or both @xcite . if x - rays are produced near the sites of accretion shocks , then one might expect that changes in the strength of these shocks will induce changes in the strength of the x - ray emission , and that x - ray and optical variability might therefore be correlated in time . alternatively , x - rays may originate in coronal structures associated with magnetically active regions on these stars , in which case we might expect that x - ray emission will be more pronounced when the magnetically active regions ( which appear as dark photospheric spots in the optical ) face toward us , and least pronounced when these regions face away from us ; that is , optical and x - ray variability will be anti - correlated in time . more complex behavior may result from the superposition of these effects , emphasizing the need for time - resolved measurements . more generally , by establishing the frequency with which x - ray and optical variations are co - temporal , we seek to constrain the extent to which the sources of the x - ray and optical variability may be co - spatial or otherwise causally linked . detailed analyses of simultaneous , time - resolved x - ray and optical variability have previously been reported for three pms stars : the cttss v773 tau @xcite and bp tau @xcite , and the weak - lined tts ( wtts ) v410 tau @xcite . in none of these were the optical and x - ray variability found to be time - correlated . unfortunately , in all three cases the statistical significance of the results were limited by the short duration of the x - ray observations : v773 tau was observed by rosat for 6.7 hr , bp tau was observed by rosat for 8 hr , and v410 tau was observed by _ chandra _ for 7.2 hr . the combined x - ray / optical dataset reported here includes some 800 members of the onc and thus represents , by a factor of hundreds , the largest attempt to date to study the relationship between x - ray and optical variability in pms stars . we will report results from this database in stages , focusing first on statistical properties of the study sample as a whole . in this first paper , we establish the incidence of _ time - correlated _ optical and x - ray variability . a companion paper ( * ? ? ? * hereafter paper ii ) focuses on _ time - averaged _ measures of optical and x - ray variability , and on the relationship between these variability indicators and basic stellar properties . this database will then form the basis for follow - up studies focusing on fine - grained analyses of individual objects . in [ data ] , we present the x - ray and optical time - series database that forms the basis for these studies and describe the methods that we employ in this paper to search for time - correlated x - ray / optical variability . in [ corr - time ] , we identify 40 candidate stars with possible time - correlated optical and x - ray variability , and classify them according to common light - curve characteristics to guide follow - up study of these interesting objects . importantly , these candidates represent at most @xmath2% of our study sample ; @xmath3% of the stars in our sample exhibit no evidence for time - correlated optical and x - ray variability . as we discuss in [ discussion ] , these results imply that the sites of optical and x - ray variability in pms stars are not in the vast majority of cases instantaneously one and the same . a direct causal link between the sources of optical and x - ray variability in pms stars is not supported by our study . we summarize our findings in [ conclusions ] . as the foundation for the synoptic x - ray / optical database used in this and follow - up studies , we use the deep x - ray observation of the onc obtained by collaboration . details of the coup dataset are described elsewhere @xcite . briefly , the onc was observed by the _ chandra _ acis detector ( energy passband of 0.510 kev ; * ? ? ? * ) from 8 jan 2003 to 21 jan 2003 for a total of 849 ks ( 9.8 days ) , representing the deepest x - ray observation of a star - forming region ever obtained . there were five small gaps of @xmath4 ks each ( when the spacecraft passed through the van allen belts ) , but the data stream is otherwise continuous over a total timespan of 13.2 days . within the acis 17 @xmath5 17 field of view centered on the trapezium , 1616 point sources were identified , of which @xmath6 are associated onc stars @xcite . for each of these , @xcite performed a spectral extraction and analysis from which they derived the source s x - ray luminosity @xmath7 and other parameters . most importantly for our present purposes , they also performed an analysis of photon arrival times , providing measures of each source s x - ray variability and yielding for each source a nearly continuous x - ray light curve spanning 13.2-d . the coup database also includes a compilation of a number of other physical characteristics for each source from photometric and spectroscopic catalogs already in the literature . for our purposes in this and follow - up studies , the most relevant of these is the equivalent width ( ew ) of the infrared triplet from @xcite , which serves as a measure of the mass accretion rate in each object . to complement this unique and rich database of x - ray light curves and ancillary data , we organized a simultaneous , multi - band , ground - based monitoring campaign , with the aim of obtaining well - sampled optical light curves that overlap well with the x - ray light curves . using small telescopes at two observing sites to minimize diurnal observing gaps , we repeatedly imaged the same field as that observed by _ acis , cycling through the @xmath0 filters , and alternating between short and long exposure times in each filter to maximize the dynamic range of the observations . constraints in scheduling the ground - based observations together with uncertainties in the scheduling of the _ chandra _ observations prevented us from obtaining optical data at the start of the _ chandra _ exposure . the optical observations begin 6 days after the start of the 13.2-d x - ray observation , and continue for 2 days beyond , for a total overlap of 7.2 d. we observed with the wiyn 0.9-m telescope at the kitt peak national observatory ( kpno ) in arizona , usa , over the period ut 2003 jan 1525 . due to poor weather , no data were obtained on the night of ut 2003 jan 18 . at the same time , we observed with the 1.5-m cassini telescope in loiano , italy , over the period ut 2003 jan 1522 , with no data obtained on the nights of ut 2003 jan 16 , 17 , 20 , 21 . exposure times were varied slightly to compensate for variable observing conditions , but were otherwise similar at both sites . typical exposure times were 5 s in @xmath0 for short exposures , 420 s for long exposures in @xmath8 and @xmath9 , and 720 s for long exposures in @xmath10 and @xmath11 . we typically obtained a total of @xmath12 measurements per filter per night , in both short and long exposures , with a cadence of @xmath13 hr@xmath14 . a log of the optical observations is presented in table [ obs - table ] . the images were reduced , and instrumental magnitudes for all point sources extracted , using standard iraf procedures . differential light curves were determined from psf photometry using an algorithm for inhomogeneous ensemble photometry @xcite as implemented in @xcite for observations of high - nebulosity regions such as the onc . systematic offsets between the kpno and loiano observations were adjusted by comparing the instrumental magnitudes of non - variable stars . all differential light curves are available in electronic format from the authors . calibrated mean optical magnitudes for each star are reported in the coup database . in order to cross - identify the optical sources with the x - ray sources in the coup database , we first determined the optical identifications of the stars from the catalogs of @xcite and @xcite using a 15 matching radius . we then used these optical identifications to associate each object with its x - ray counterpart in the coup database using the table of cross - identifications provided by @xcite . in this way , we were able to unambiguously associate 814 optical sources with x - ray counterparts in the coup database . these 814 objects span a very large range of apparent magnitudes and colors , with @xmath15 and @xmath16 , and a correspondingly large range of spectral types and extinctions , b3 @xmath17 spty @xmath17 m7 and @xmath18 @xcite . this sample is thus representative of the underlying onc population ; only the very massive , heavily obscured , and sub - stellar populations are excluded . an unavoidable consequence of this breadth of stellar properties is that objects do not appear with uniform frequency in our optical light curves . for example , the brightest objects appear only in the short - exposure light curves , being saturated in the long - exposure frames . similarly , the coolest and the most heavily reddened objects appear only in the long - wavelength passbands . in addition , some objects near the bright and faint extremes of our survey have more sparsely sampled light curves , as variations in observing conditions ( e.g. seeing , sky transparency , background ) cause these objects to be lost from some exposures . thus , any individual object in our final database possesses between 1 and 8 different optical light curves ( @xmath0 filters , long and short exposures ) with time sampling patterns that are very similar but , for the reasons already noted , not identical . of course , objects whose characteristics are intermediate to the extremes discussed above possess the most complete set of optical light curves and the most complete time sampling . the stars in our database exhibit a large variety of behaviors in their optical and x - ray variability . [ coup1499][coup132 ] show the optical and x - ray light curves for a selection of objects as a representative visual summary of the range and types of variability observed . here we describe the methods we use to measure and to correlate variability in the x - ray and optical light curves in our database . to quantify the time - averaged x - ray variability of the stars in our database , we use the bayesian block ( bb ) analysis provided by the coup @xcite . briefly , the bb analysis @xcite segments each light curve into the maximum number of time blocks such that the differences in the mean flux levels of the blocks are statistically significant . the bb analysis thus yields a robust measure of the maximum and minimum flux levels in the source s segmented x - ray light curve ( bb@xmath19 and bb@xmath20 , respectively ) ; we use the statistic @xmath21bb@xmath19/bb@xmath20@xmath22 to quantify the magnitude of variability in the x - ray light curve . as discussed in paper ii , 556 ( 68% ) of the 814 stars in our database are x - ray variable at greater than 99.9% confidence according to the bb statistic , with a range of amplitudes @xmath23bb@xmath19/bb@xmath20@xmath24 . the six objects displayed in figs . [ coup1499][coup132 ] are representative of this range , with @xmath25bb@xmath19/bb@xmath20@xmath26 . to characterize the time - averaged optical variability of these stars , we use the @xmath27 statistic described by @xcite . briefly , the @xmath27 statistic is similar to the standard @xmath28 statistic , however the @xmath27 statistic simultaneously considers the measurements from all of the light curves of a given object . as such , the @xmath27 statistic is more robust against outliers , especially when multiple light curves are available in a given passband ( i.e. long and short exposures ) . as discussed in paper ii , 358/814 ( 44% ) of the stars in our database are optically variable at greater than 99.9% confidence according to the @xmath27 statistic , with a range @xmath29 , corresponding to peak - to - peak variability in the @xmath8 band ranging from a maximum of @xmath30 mag to a minimum of @xmath31 mag ( upper limit ) . the six objects displayed in figs . [ coup1499][coup132 ] are representative of this range , with @xmath32 . with this range of variability amplitudes , and with variability timescales ranging from hours to days , the optical variability of our sample is typical of what has been observed before in optical variability studies of ttss ( e.g. * ? ? ? we defer to paper ii a full analysis of the relationships between these time - averaged variability indicators , but note here that the light curves shown in figs . [ coup1499][coup132 ] are representative of the behavior observed in our database as a whole . in general , we find examples of optical and x - ray variability in every possible permutation : we find stars that are highly variable in both optical and x - ray light ( e.g. fig . [ coup1499 ] ) , stars that are strongly variable in the optical but non - variable in x - rays ( e.g. fig . [ coup75 ] ) and vice - versa ( e.g. fig . [ coup132 ] ) , and stars that are non - variable altogether . for each of the stars in our database , we calculate kendall s @xmath33 statistic to quantify the degree to which variability in the x - ray and optical light curves are correlated in time . kendall s @xmath33 is a standard non - parametric statistical test for the degree of association between two variables , and has a known probability distribution . @xmath33 test rank - orders the time - matched optical magnitudes and x - ray fluxes , computes the degree to which the optical and x - ray ranks " are correlated , and calculates the probability that the measured correlation is consistent with the null hypothesis of no correlation . measurement errors are ignored in computing @xmath33 . to allow for the possibility that time - correlated variability may be a function of wavelength , we calculate kendall s @xmath33 and its associated null - hypothesis probability by correlating each star s x - ray light curve separately against each of its associated optical light curves . where multiple light curves are available in a given passband ( i.e. long and short exposures ) , we select the light curve with the smallest ratio of mean photometric error to total number of measurements , @xmath34 , i.e. the light curve with the best combination of signal - to - noise and time sampling . for a given optical light curve , we match each optical data point with the corresponding time bin in the star s x - ray light curve . in total , our search for time - correlated variability in our sample of 814 stars involves 1944 different @xmath0 light curves . in the analysis that follows , we take kendall s @xmath33 probabilities of less than 0.01 as suggestive of time - correlated optical and x - ray variability . the sign of the @xmath33 statistic indicates whether the optical and x - ray light curves are positively or negatively time - correlated . none of the examples shown in figs . [ coup1499][coup132 ] exhibit statistically significant time - correlated optical and x - ray variability . for instance , at the same time that coup 1499 ( fig . [ coup1499 ] ) shows variability of @xmath30 mag in the optical , its x - ray light curve is almost completely non - variable . to be sure , this star is one of the strongest x - ray variables in our database , with @xmath35(bb@xmath19/bb@xmath20 ) @xmath36 . the sources of x - ray and optical variability in this star are evidently not simultaneously one and the same . [ coup901][coup1363 ] show similar examples , while figs . [ coup1466][coup132 ] exhibit the reverse behavior , with strong x - ray flaring events that are not simultaneously reflected in the optical . such a lack of time - correlated optical and x - ray variability is a general feature of the @xmath37 stars in our database , as we now discuss . in this section we present the results of our analysis of time - correlated optical and x - ray variability in our database of 814 coup sources with simultaneous optical and x - ray light curves . we identify 40 candidate stars whose optical and x - ray light curves appear to be correlated or anti - correlated in time . these represent the first such cases yet reported for pms stars , and we classify these objects according to possible physical origins for the observed variability . more generally , however , time - correlated variability evidently occurs very rarely in our study sample . following the procedure described in [ corr ] , we have searched for time - correlated variability in the optical and x - ray light curves of the 814 stars in our database . forty stars show evidence for the x - ray light curve being time - correlated with at least one optical light curve on the basis of a non - parametric kendall s @xmath33 test ( [ corr ] ) . these 40 stars are listed in table [ correlated - table ] , and their x - ray and optical light curves shown in figs . [ coup28][coup1608 ] . relative to our entire study sample of 814 stars , the fractional incidence of time - correlated variability represented by these 40 stars is @xmath38 . in our study sample , 178 stars are neither optically nor x - ray variable according to the bb and @xmath27 statistical measures discussed in [ methods ] ( see also paper ii for a thorough analysis and discussion of these time - averaged variability indicators ) . if we exclude these 178 stars from the comparison sample ( leaving 636 stars ) , the 40 stars exhibiting time - correlated variability then represent a fractional incidence of @xmath39% . in any case , time - correlated optical and x - ray variability is evidently an uncommon phenomenon in our sample . from statistical arguments , it is probable that a number of these candidates will be spurious . estimating the number of expected false positives accurately is complicated by the fact that a given star may possess multiple optical light curves that are not statistically independent . we may , however , estimate an upper limit to the expected number of false positives , given that we adopted a kendall s @xmath33 null - hypothesis probability of 0.01 as our criterion for identifying these candidates , and that our search for time - correlated variability in our sample of 814 stars involves 1944 different @xmath0 optical light curves ( [ corr ] ) . this implies 1920 false positives . as a reality check on the kendall s @xmath33 statistic , we have additionally performed a monte carlo simulation in which we applied the kendall s @xmath33 test to one million random pairings of our optical and x - ray light curves , and we find that kendall s @xmath33 probabilities of less than 0.01 occur with a frequency of 1.4% , very close to the expected value of 1.0% . the candidates listed in table [ correlated - table ] include a total of 63 @xmath0 light curves that are found to be time - correlated with their associated x - ray light curves according to kendall s @xmath33 , and thus represent a @xmath40 detection of time - correlated variability over the expected number of false positives . it is not possible on the basis of these statistical arguments alone to determine which of these candidates are the false positives and which are the true time - correlated variables . in many cases , the appearance of the x - ray and optical variations are quite dissimilar , so the statistical correlation may often not arise from a physical correlation . some candidates might nonetheless be regarded as more credible than others based on the number of their optical light curves that pass the kendall s @xmath33 test . for example , coup 122 ( fig . [ coup122 ] ) has four optical light curves ( @xmath0 ) and all four of them are identified as time - correlated with the x - ray light curve ( see table [ correlated - table ] ) . in contrast , coup 1309 ( fig . [ coup1309 ] ) also has four optical light curves , none of which appear to be convincingly correlated with the x - ray light curve , and indeed only one of them is identified as time - correlated by the kendall s @xmath33 statistic . in other cases , only one optical light curve is identified as being time - correlated with the x - ray light curve even though the other optical light curves share a similar morphology . for example , in coup 152 ( fig.[coup152 ] ) the @xmath8- and @xmath10-band light curves are not identified as time - correlated , despite a very similar morphology to the @xmath9-band light curve . common sense suggests that this is not a physically meaningful result ; instead , it is likely that small , random noise differences among the optical light curves allow just one of them to pass the 0.01 kendall s @xmath33 probability threshold . such a precarious correlation is less credible than , e.g. , the case of coup 122 ( fig . [ coup122 ] ) noted above . the reader may wish to apply other subjective criteria to the light curves shown in figs . [ coup28][coup1608 ] . we thus do not attempt to identify specific candidates in table [ correlated - table ] as definitive . rather , we emphasize that these 40 candidates have been identified on an objective , statistical basis and that an estimated upper limit on the number of false positives suggests that @xmath41 of these candidates may represent _ bona fide _ cases of time - correlated optical and x - ray variability . this implies an overall incidence of time - correlated variability in our sample of @xmath42% . time - correlated optical and x - ray variability is evidently an uncommon phenomenon in our sample . if x - ray production in pms stars is connected to accretion related processes , we might expect time - correlated optical / x - ray variability to be more common among actively accreting stars , as these stars may be more likely to have their optical variability dominated by accretion . following previous studies ( e.g. * ? ? ? * ; * ? ? ? * ) , we take the accretors " to be those stars with strongly in emission , i.e. , ew [ ] @xmath43 , and non - accretors " to be those with clearly in absorption , i.e. , ew [ ] @xmath44 . ew [ ] measurements have been reported for 493 stars in our study sample . of these , 151 stars are accretors and 145 stars are non - accretors , as defined here ( the remainder show indeterminate values close to 0 ) is an imperfect indicator of accretion . as shown by @xcite , some pms stars with clear accretion signatures in h@xmath45 do not manifest this accretion activity in . it is therefore possible or even likely that some low - level accretors will go undetected in . the criterion that we adopt for identifying the presence of accretion is thus in practice a conservative one ; stars with ew [ ] @xmath43 and ew [ ] @xmath44 are securely identified as accretors and non - accretors , while some stars with indeterminate values close to 0 may be weakly accreting objects that will go unidentified as such . ] . eighteen of the 40 stars in table [ correlated - table ] have ew [ ] measurements allowing them to be clearly classified as accretors or non - accretors ( table [ correlated - table ] ) . these 18 stars are roughly evenly divided in terms of their accretion properties ( 10 accretors , 8 non - accretors ) . thus there is no indication that accretors are more likely than non - accretors to exhibit time - correlated optical and x - ray variability . in this section , we discuss the 40 stars that we have identified as candidates for time - correlated x - ray / optical variability ( table [ correlated - table ] ) . classifying them according to common light curve characteristics , we suggest physical interpretations for the origins of the observed variability . note that , due to the often complex nature of the observed variability , some stars are not classified while others may be included in more than one category ( see table [ correlated - table ] for an overview ) . this discussion is presented to highlight what may represent the first examples in the literature of time - correlated optical and x - ray variability among pms stars , and to guide follow - up investigation of these specific objects . we remind the reader that in fact the overwhelming majority of the stars in our study sample do not exhibit evidence for time - correlated optical and x - ray variability . several stars show periodic or quasi - periodic modulations in their optical light curve(s ) on time scales that appear to be consistent with optical variability periods ( @xmath46 ) reported for them in the literature . the persistence of the periodicity suggests that it is related to the stars rotation periods and that we are seeing either dark spots in regions of magnetic activity , or accretion hot spots , rotating in and out of view . at the same time , these stars x - ray light curves exhibit typical coronal flare - like morphologies that are coincident in time with extrema in the optical light curves . this type of behavior is found both in cases for which the x - ray / optical correlations are positive and negative , and among both accretors and non - accretors . sources for which an x - ray flare coincides with optical maximum ( positive correlations ) include , , , , and . sources for which an x - ray flare coincides with optical minimum ( anti - correlations ) include , , and . additional stars that can be included in this category are those with no previously reported @xmath46 but which still show gradual ( and possibly periodic ) modulations in at least one optical light curve . sources for which an x - ray flare coincides with optical maximum ( positive correlations ) include , , and . sources for which an x - ray flare coincides with optical minimum ( anti - correlations ) include , , and . finally , there are some stars whose time - correlations are dominated by other behavior ( see groups b / c below ) but which in addition exhibit x - ray flares near optical maximum or minimum . sources for which an x - ray flare coincides with optical maximum include and . sources for which an x - ray flare coincides with optical minimum include and . it is possible that these stars represent examples of physical associations between regions of x - ray flaring and stellar spots either hot accretion spots ( positive correlations ) or dark magnetic spots ( anti - correlations ) . however , it is also possible that these stars represent a more general class of objects that show periodic rotational modulation and x - ray flares with no preferred correlation with rotational phase . for example , if we assume a typical @xmath46 of 6 d , and that x - ray flares occur randomly in time with a typical duration of 0.3 d , then the probability of finding either a correlation or anti - correlation is @xmath47% . moreover , among the ten cases of positive correlations included above , only three show clear evidence of active accretion . in any case , detailed follow - up of these stars in comparison to the full sample of coup sources showing x - ray flares should help resolve these interpretations . we have found several stars that show gradual and periodic or quasi - periodic variations in both their optical and x - ray light curves ; in particular , the x - ray light curves do not exhibit the flare - like features that characterize the objects in group a. assuming that modulation in the x - ray light curve indicates the presence of a stable structure ( within the time span of the _ chandra _ observation ) that moves in and out of view as the star rotates , its origin could be a bright accretion spot or a magnetically active region on the stellar surface . in the first case , a positive correlation with the optical light curves might be expected . in the second , an x - ray bright coronal feature accompanied by a photospheric dark spot would give rise to anti - correlated behavior . there are nine cases for which the x - ray and optical light curves produce a net anti - correlation : , , , , , , , and . with the exception of the last object , which showed weak signatures of active accretion in the observations of @xcite , the available ew [ ] measurements are consistent with these stars all being non - accretors , and therefore implying that the photometric and x - ray variations are not related to accretion . all but one star ( coup 1521 ) have previously published optical periods , and the variability observed in our optical light curves is fully consistent with those periods . in addition , the x - ray light curves of three of these stars ( coup 161 , 1355 , and 1521 ) have been independently found by @xcite to be periodic , with x - ray periods @xmath48 that are equal to the optical periods ( i.e. @xmath49 ; see table [ correlated - table ] ) , corroborating the interpretation that the x - ray variability arises from rotational modulation . as discussed by @xcite , this rotational modulation implies that the x - ray emission is produced in close proximity to the stellar surface , such that the x - ray emitting region is occulted as it rotates to the back side of the star . these objects thus represent compelling cases of stars having stable coronal structures that are in close spatial proximity to and likely to be physically linked with dark , magnetically active regions on their surfaces . interestingly , for and , @xcite find @xmath50 ( see table [ correlated - table ] ) . that is , the light curves exhibit two x - ray modulations for each optical modulation . this is particularly evident in ( fig . [ coup139 ] ) . of course , this behavior may simply be the result of randomly repeating flares and/or randomly elevated x - ray emission levels , as discussed for the group a stars . another interpretation is that the stellar coronae in these cases have two x - ray bright regions , separated by @xmath51 in longitude , but that we only see a dark , magnetically active region on the stellar surface associated with one of them . [ coup1590 ] ) , which exhibits an x - ray light curve that appears to oscillate between two different amplitudes , may be a case in point . in addition to these stars with anti - correlated light curves , there are 14 stars that exhibit positive correlations in their x - ray / optical light curves . they are : , , , , , , , , and . @xcite have searched for periodicity in the x - ray light curves of eight of these 14 stars , and have identified two of them ( and ) as having definitive periodic x - ray light curves , with periods equal to ( or half of ) the published @xmath46 . they also identified another three of these stars ( , , and ) as having likely " periodic x - ray light curves . thus , the incidence of periodic x - ray modulation among objects with positively correlated optical / x - ray light curves appears to be high . the positive correlation between the x - ray and optical light curves suggests that these cases represent stable x - ray bright structures physically associated with photospheric hot spots , rotating in and out of view . these hot spots may be the footpoints of mass accretion streams on the stellar surface . alternatively , the observed positively correlated x - ray / optical light curves may represent examples of x - ray faint coronal structures i.e . coronal holes that are co - spatial with dark , magnetically active regions on the stellar surface . however , an interpretation of accretion hot spots is more strongly supported by the ew [ ] properties of these 14 stars , which include some of the strongest accretion signatures found in the entire study sample ( table [ correlated - table ] ) . among these , ( fig . [ coup250 ] ) is particularly interesting , having been identified by @xcite as having a periodic x - ray light curve . this star may thus be a good example of a star having stable structures that are physically associated with accretion hot spots on their surfaces . follow - up modeling and analysis of the color information contained in our multi - band optical light curves should help to clarify whether we are in fact observing hot spots in this object . many of the x - ray flares seen in the coup dataset resemble solar long - duration flares seen with the _ goes _ and _ yohkoh _ satellites which cover the _ chandra _ spectral band . the peak luminosities and durations , and hence total energies , of the coup flare plasma emission are orders of magnitude higher than seen in solar flares . but it is reasonable to suggest that , as in solar flares , the long - duration soft ( _ chandra _ band ) x - ray flares are preceeded by brief impulsive phases when particle acceleration , radio gyrosynchrotron , hard ( @xmath52 kev ) x - ray emission , and optical white - light " continuum increases occur ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the impulsive phase has , on rare occasions , been seen in magnetically active stars , notably the rs cvn systems @xmath53 gem @xcite and hr 1099 @xcite and dme stars prox cen @xcite and ev lac @xcite . two stars in our study sample and exhibit rather irregular optical light curves that show punctuated increases in brightness that are positively correlated with short - duration flares in the x - ray light curves . while these correlations may be coincidences , we think it is worthwhile to highlight these objects since they may represent the best candidates for the occurrence of white - light magnetic flares . we note that has also been identified as a candidate in groups a and b above due to the particularly complex nature of the observed variability . in particular , we caution that the optical brightening event " near day 10.5 may simply be related to the longer - term , quasi - periodic optical variability of this object . on the other hand , while we have very little data for , what we do have is tantalizing . a single optical measurement , brighter than the mean brightness level by @xmath54 , occurs precisely in the time bin of a very brief x - ray flare that at its peak is @xmath55 times brighter than the mean x - ray flux level for this object . we have carefully scrutinized the optical data for this star and believe the observed optical brightening event to be credible . in contrast to the strong rotation - activity relationship " that is clearly present on the main sequence @xcite , a series of studies based on _ chandra _ observations of a large sample of pms stars in the orion nebula cluster ( onc ) failed to find any direct correlation between the stars rotation rates and their x - ray luminosities @xcite . a possible explanation for the lack of a pms rotation - activity relationship is that the x - rays originate not ( or not only ) from dynamo - generated magnetic processes , but also ( or instead ) from energetic processes related to accretion . if x - rays are produced near the sites of accretion shocks , then we may expect that x - ray and optical variability would be time - correlated . however , at most @xmath2% of the stars in our sample show credible , statistically significant , time - correlated optical and x - ray light curves ( [ corr - incidence ] ) , and this time - correlated variability is just as likely to occur among non - accreting stars as it is among accretors ( [ corr - accretion ] ) . our findings are consistent with previous studies of simultaneous x - ray and optical variability in pms stars @xcite , with a sample that is hundreds of times larger . our results do not negate the possibility that accretion is indirectly related to x - ray production at some level . for example , the presence of an accretion disk may alter the stellar magnetospheric environment ( e.g. via large star - disk magnetic loops ; * ? ? ? * ) such that large reconnection flares or other energetic processes take place in a way that is not possible in the absence of a disk ( e.g. * ? ? ? * ) . such events , displaced and disconnected from magnetically active regions on the stellar surface , might not be expected to produce time - correlated behavior in x - rays and the optical . in addition , we have in fact found a small number of stars that exhibit possibly accretion - related x - ray emission , in the form of x - ray flares and x - ray bright features that may be co - spatial with accretion hot spots on the stellar surface ( [ credible ] ) . the strong optical / x - ray brightening event observed in the fu ori like outburst of v1647 ori ( mcneil s nebula ) serves as a particularly dramatic example of time - correlated optical / x - ray variability that is likely to be accretion driven @xcite . nonetheless , though intriguing , such cases are evidently very rare . similarly , any physical connection of x - ray emission to magnetically active surface regions ( i.e. cool spots ) does not for the vast majority of stars in our sample manifest itself in a time - correlated way ( [ credible ] ) . this may indicate that the x - ray emitting coronae in pms stars have spatial structures that little reflect the magnetic structure of the underlying photosphere . a possible explanation for this is that x - ray emission takes place on very large spatial scales relative to the photosphere , and may not be confined to the small surface regions ( spots ) from which the optical variability presumably originates . while this picture is at odds with the small coronae implicit in the finding of rotationally modulated x - ray emission in @xmath56% of coup sources by @xcite , such an interpretation could be consistent with the very large flaring loop sizes inferred on a small number of coup sources by @xcite . more generally , our findings are consistent with recent analyses of the x - ray spectra of cttss , and with recent measurements of the magnetic field structures of wttss . in an analysis of high - resolution x - ray spectra of the cttss bp tau , cr cha , su aur , and tw hya , @xcite conclude that both accretion shocks and coronal activity contribute to the observed x - ray emission , but that coronal activity is by far the dominant contributor in all cases except tw hya . these results , together with our finding that time - correlated optical and x - ray variability occurs only very rarely , support the general conclusion that neither surface accretion shocks nor magnetically active surface ( cool ) spots are the exclusive sites of x - ray production in the vast majority of pms stars . we have presented a combined analysis of simultaneous optical and x - ray light curves for a sample of more than 800 pre main - sequence ( pms ) stars in the orion nebula cluster ( onc ) . this dataset represents an increase of several hundred - fold , both in the number of objects studied and in the total duration of the x - ray and optical time - series data , over all previous investigations of simultaneous optical and x - ray variability in pms stars . we identify 40 stars that exhibit the first examples to our knowledge of time - correlated optical and x - ray variability in pms stars . among these 40 stars , we find examples of both positive and anti - correlations . positive correlations may represent x - ray dark coronal structures associated with cool , magnetically active surface spots , or may represent x - ray bright structures associated with hot accretion spots . anti - correlations may represent , e.g. , magnetic flaring events associated with cool surface spots . in any event , the implication is that the x - ray emitting regions in these stars are in close proximity to and thus likely to be physically linked with cool and/or hot spots on their surfaces . follow - up modeling and analysis of the color information contained in our multi - band optical light curves should help to clarify whether we are observing cool or hot spots in each object . in addition , we identify two possible examples of white - light " flares , perhaps corresponding to the impulsive heating phase in solar - analog flares . however , these cases represent at most @xmath2% of our study sample . the remaining @xmath3% of our study sample exhibits no such time - correlated variability . evidently , time - correlated optical and x - ray variability is rare in pms stars . we conclude that the sites of optical and x - ray variability in pms stars are not for the vast majority of stars in our study sample instantaneously one and the same . surface accretion shocks and cool magnetic spots are evidently not the exclusive or dominant sites of x - ray production in most pms stars . it is a pleasure to acknowledge the _ orion ultradeep project ( coup ) team , supported by _ guest observer grant sao go34009a ( e. feigelson , pi ) . this work is also supported by nsf grant ast0349075 to k.g.s . we wish to thank c.deliyannis for generously sharing his wiyn 0.9-m time to make the optical observations presented here possible . we also thank s.wolk for assisting in the planning for the optical monitoring runs . facility : cxo(acis ) rccl 2003 jan 15 & @xmath10 long & 4 & 2452654.69829 + & & & 2452654.74710 + & & & 2452654.78729 + & & & 2452654.82935 + & @xmath10 short & 4 & 2452654.69343 + & & & 2452654.74240 + & & & 2452654.78264 + & & & 2452654.82473 + & @xmath8 long & 4 & 2452654.68066 + & & & 2452654.72999 + & & & 2452654.77007 + & & & 2452654.81229 + & @xmath8 short & 4 & 2452654.67623 + & & & 2452654.72604 + & & & 2452654.76613 + & & & 2452654.80839 + & @xmath9 long & 4 & 2452654.68900 + & & & 2452654.73799 + & & & 2452654.77822 + & & & 2452654.82024 + & @xmath9 short & 4 & 2452654.68497 + & & & 2452654.73398 + & & & 2452654.77403 + & & & 2452654.81634 + & @xmath11 long & 4 & 2452654.70725 + & & & 2452654.75627 + & & & 2452654.79855 + & & & 2452654.83816 + & @xmath11 short & 5 & 2452654.70319 + & & & 2452654.75232 + & & & 2452654.79217 + & & & 2452654.79445 + & & & 2452654.83439 + 2003 jan 16 & @xmath10 long & 5 & 2452655.70634 + & & & 2452655.75586 + & & & 2452655.79487 + & & & 2452655.83286 + & & & 2452655.87096 + & @xmath10 short & 5 & 2452655.70174 + & & & 2452655.75128 + & & & 2452655.79037 + & & & 2452655.82835 + & & & 2452655.86639 + & @xmath8 long & 5 & 2452655.71509 + & & & 2452655.76489 + & & & 2452655.80322 + & & & 2452655.84131 + & & & 2452655.87962 + & @xmath8 short & 5 & 2452655.71109 + & & & 2452655.76104 + & & & 2452655.79944 + & & & 2452655.83746 + & & & 2452655.87558 + & @xmath9 long & 5 & 2452655.72068 + & & & 2452655.77256 + & & & 2452655.81079 + & & & 2452655.84938 + & & & 2452655.88714 + & @xmath9 short & 5 & 2452655.72455 + & & & 2452655.76876 + & & & 2452655.80707 + & & & 2452655.84568 + & & & 2452655.88341 + & @xmath11 long & 5 & 2452655.73077 + & & & 2452655.78041 + & & & 2452655.81872 + & & & 2452655.85701 + & & & 2452655.89490 + & @xmath11 short & 5 & 2452655.72684 + & & & 2452655.77667 + & & & 2452655.81491 + & & & 2452655.85327 + & & & 2452655.89116 + 2003 jan 17 & @xmath10 long & 7 & 2452656.64191 + & & & 2452656.68023 + & & & 2452656.72500 + & & & 2452656.76363 + & & & 2452656.80270 + & & & 2452656.84125 + & & & 2452656.88129 + & @xmath10 short & 7 & 2452656.63677 + & & & 2452656.67572 + & & & 2452656.72057 + & & & 2452656.75909 + & & & 2452656.79805 + & & & 2452656.83668 + & & & 2452656.87652 + & @xmath8 long & 7 & 2452656.65095 + & & & 2452656.69519 + & & & 2452656.73391 + & & & 2452656.77238 + & & & 2452656.81125 + & & & 2452656.85048 + & & & 2452656.89037 + & @xmath8 short & 7 & 2452656.64706 + & & & 2452656.68480 + & & & 2452656.72993 + & & & 2452656.76837 + & & & 2452656.80753 + & & & 2452656.84657 + & & & 2452656.88620 + & @xmath9 long & 7 & 2452656.65643 + & & & 2452656.70081 + & & & 2452656.74172 + & & & 2452656.78003 + & & & 2452656.81670 + & & & 2452656.85609 + & & & 2452656.89592 + & @xmath9 short & 7 & 2452656.66025 + & & & 2452656.70467 + & & & 2452656.73783 + & & & 2452656.77625 + & & & 2452656.82052 + & & & 2452656.86029 + & & & 2452656.90020 + & @xmath11 long & 7 & 2452656.66661 + & & & 2452656.71092 + & & & 2452656.74953 + & & & 2452656.78802 + & & & 2452656.82649 + & & & 2452656.86648 + & & & 2452656.90659 + & @xmath11 short & 7 & 2452656.66254 + & & & 2452656.70706 + & & & 2452656.74577 + & & & 2452656.78419 + & & & 2452656.82269 + & & & 2452656.86271 + & & & 2452656.90252 + 2003 jan 19 & @xmath10 long & 1 & 2452658.80562 + & @xmath10 short & 3 & 2452658.80091 + & & & 2452658.84655 + & & & 2452658.88929 + & @xmath8 short & 2 & 2452658.81034 + & & & 2452658.85586 + & @xmath9 short & 1 & 2452658.81810 + & @xmath11 short & 2 & 2452658.82642 + & & & 2452658.87448 + 2003 jan 20 & @xmath10 long & 3 & 2452659.69923 + & & & 2452659.73927 + & & & 2452659.78394 + & @xmath10 short & 3 & 2452659.69449 + & & & 2452659.73461 + & & & 2452659.77938 + & @xmath8 long & 3 & 2452659.70758 + & & & 2452659.74787 + & & & 2452659.79236 + & @xmath8 short & 3 & 2452659.70375 + & & & 2452659.74389 + & & & 2452659.78848 + & @xmath9 long & 3 & 2452659.71525 + & & & 2452659.75586 + & & & 2452659.80028 + & @xmath9 short & 3 & 2452659.71140 + & & & 2452659.75189 + & & & 2452659.79629 + & @xmath11 long & 3 & 2452659.72508 + & & & 2452659.76987 + & & & 2452659.80822 + & @xmath11 short & 3 & 2452659.72127 + & & & 2452659.76597 + & & & 2452659.80436 + 2003 jan 21 & @xmath10 short & 1 & 2452660.87434 + & @xmath8 short & 1 & 2452660.88349 + & @xmath11 short & 1 & 2452660.89865 + 2003 jan 22 & @xmath10 long & 1 & 2452661.70622 + & @xmath10 short & 2 & 2452661.70159 + & & & 2452661.86023 + & @xmath8 long & 1 & 2452661.83375 + & @xmath8 short & 1 & 2452661.71098 + & @xmath9 long & 1 & 2452661.84161 + & @xmath9 short & 1 & 2452661.83778 + & @xmath11 short & 1 & 2452661.84603 + 2003 jan 23 & @xmath10 long & 4 & 2452662.63928 + & & & 2452662.67797 + & & & 2452662.71604 + & & & 2452662.75484 + & @xmath10 short & 5 & 2452662.63462 + & & & 2452662.67339 + & & & 2452662.71156 + & & & 2452662.75033 + & & & 2452662.79970 + & @xmath8 long & 6 & 2452662.64782 + & & & 2452662.68635 + & & & 2452662.72483 + & & & 2452662.76400 + & & & 2452662.81349 + & & & 2452662.85621 + & @xmath8 short & 6 & 2452662.64387 + & & & 2452662.68253 + & & & 2452662.72065 + & & & 2452662.75986 + & & & 2452662.80971 + & & & 2452662.85213 + & @xmath9 long & 5 & 2452662.65582 + & & & 2452662.69412 + & & & 2452662.73251 + & & & 2452662.77315 + & & & 2452662.82160 + & @xmath9 short & 5 & 2452662.65175 + & & & 2452662.69017 + & & & 2452662.72875 + & & & 2452662.76936 + & & & 2452662.86015 + & @xmath11 long & 4 & 2452662.66371 + & & & 2452662.74026 + & & & 2452662.78107 + & & & 2452662.82975 + & @xmath11 short & 4 & 2452662.65992 + & & & 2452662.69818 + & & & 2452662.73643 + & & & 2452662.77733 + 2003 jan 25 & @xmath10 long & 1 & 2452664.81536 + & @xmath10 short & 2 & 2452664.81050 + & & & 2452664.85198 + & @xmath8 long & 1 & 2452664.82516 + & @xmath8 short & 1 & 2452664.82035 + & @xmath9 long & 1 & 2452664.83348 + & @xmath9 short & 1 & 2452664.82927 + & @xmath11 long & 1 & 2452664.84101 + & @xmath11 short & 1 & 2452664.83732 + 2003 jan 15 & @xmath10 long & 3 & 2452655.31530 + & & & 2452655.37790 + & & & 2452655.43330 + & @xmath10 short & 4 & 2452655.31070 + & & & 2452655.31140 + & & & 2452655.33080 + & & & 2452655.43200 + & @xmath8 long & 3 & 2452655.33870 + & & & 2452655.40080 + & & & 2452655.44700 + & @xmath8 short & 2 & 2452655.39840 + & & & 2452655.44550 + & @xmath9 long & 3 & 2452655.34910 + & & & 2452655.41040 + & & & 2452655.45690 + & @xmath9 short & 2 & 2452655.40820 + & & & 2452655.45560 + & @xmath11 long & 2 & 2452655.36310 + & & & 2452655.42120 + & @xmath11 short & 2 & 2452655.37050 + & & & 2452655.41910 + 2003 jan 16 & @xmath11 long & 1 & 2452656.46640 + & @xmath11 short & 1 & 2452656.46530 + 2003 jan 18 & @xmath10 long & 3 & 2452658.31120 + & & & 2452658.36380 + & & & 2452658.41440 + & @xmath10 short & 3 & 2452658.31000 + & & & 2452658.36230 + & & & 2452658.41320 + & @xmath8 long & 3 & 2452658.32980 + & & & 2452658.37810 + & & & 2452658.43280 + & @xmath8 short & 3 & 2452658.32860 + & & & 2452658.37690 + & & & 2452658.42740 + & @xmath9 long & 3 & 2452658.34020 + & & & 2452658.38810 + & & & 2452658.44180 + & @xmath9 short & 3 & 2452658.33900 + & & & 2452658.38670 + & & & 2452658.44060 + & @xmath11 long & 3 & 2452658.34880 + & & & 2452658.39840 + & & & 2452658.45470 + & @xmath11 short & 2 & 2452658.39700 + & & & 2452658.45020 + 2003 jan 19 & @xmath10 long & 4 & 2452659.28790 + & & & 2452659.33810 + & & & 2452659.38800 + & & & 2452659.46500 + & @xmath10 short & 1 & 2452659.46330 + & @xmath8 long & 4 & 2452659.30670 + & & & 2452659.35630 + & & & 2452659.40160 + & & & 2452659.48030 + & @xmath8 short & 4 & 2452659.30560 + & & & 2452659.35520 + & & & 2452659.39990 + & & & 2452659.47910 + & @xmath9 long & 4 & 2452659.31510 + & & & 2452659.36440 + & & & 2452659.41030 + & & & 2452659.48820 + & @xmath9 short & 4 & 2452659.31400 + & & & 2452659.36320 + & & & 2452659.40920 + & & & 2452659.48710 + & @xmath11 long & 4 & 2452659.32480 + & & & 2452659.37350 + & & & 2452659.41900 + & & & 2452659.49700 + & @xmath11 short & 4 & 2452659.32300 + & & & 2452659.37210 + & & & 2452659.41790 + & & & 2452659.49570 + 2003 jan 22 & @xmath10 long & 2 & 2452662.29290 + & & & 2452662.36200 + & @xmath10 short & 2 & 2452662.29150 + & & & 2452662.35710 + & @xmath8 long & 1 & 2452662.30880 + & @xmath8 short & 1 & 2452662.30770 + & @xmath9 long & 1 & 2452662.31990 + & @xmath9 short & 1 & 2452662.31870 + & @xmath11 long & 1 & 2452662.33070 + & @xmath11 short & 1 & 2452662.32950 + rllrcrcclc 28 & @xmath0 & @xmath10 & 1.6 & @xmath57 & 4.41 & & a + 54 & @xmath0 & @xmath10 & @xmath581.0 & @xmath58 & & & a + 97 & @xmath11 & @xmath11 & @xmath583.4 & @xmath58 & & & + 112 & @xmath0 & @xmath59 & @xmath580.7 & @xmath57 & 6.52 & poss . & ab + 122 & @xmath0 & @xmath0 & 0.0 & @xmath57 & 9.19 & poss . & ab + 139 & @xmath0 & @xmath11 & 0.9 & @xmath58 & 9.04 & @xmath60 & b + 147 & @xmath61 & @xmath61 & @xmath581.0 & @xmath57 & & & b + 152 & @xmath0 & @xmath9 & & @xmath58 & & & a + 161 & @xmath0 & @xmath9 & 1.4 & @xmath58 & 5.46 & 1 & b + 197 & @xmath0 & @xmath62 & & @xmath57 & & & + 226 & @xmath0 & @xmath63 & 1.2 & @xmath57 & 10.98 & 1 & b + 241 & @xmath0 & @xmath8 & 2.8 & @xmath57 & 9.81 & & + 250 & @xmath0 & @xmath11 & @xmath589.6 & @xmath57 & 6.76 & 1 & abc + 298 & @xmath59 & @xmath11 & 0.0 & @xmath58 & 6.14 & & a + 309 & @xmath61 & @xmath61 & & @xmath57 & & & + 325 & @xmath0 & @xmath11 & 0.0 & @xmath57 & & & a + 346 & @xmath59 & @xmath11 & 0.0 & @xmath57 & & & + 501 & @xmath0 & @xmath8 & @xmath5825.2 & @xmath57 & 16.33 & poss . & ab + 566 & @xmath59 & @xmath11 & & @xmath57 & & & ab + 597 & @xmath0 & @xmath0 & 4.5 & @xmath57 & & & a + 718 & @xmath0 & @xmath9 & & @xmath58 & 5.74 & & ab + 816 & @xmath11 & @xmath11 & & @xmath57 & & & c + 1071 & @xmath0 & @xmath11 & 1.6 & @xmath57 & 0.96 & & b + 1076 & @xmath0 & @xmath10 & & @xmath57 & 11.68 & & + 1143 & @xmath0 & @xmath10 & & @xmath58 & 13.00 & & b + 1212 & @xmath61 & @xmath9 & & @xmath57 & & & + 1252 & @xmath0 & @xmath11 & 1.6 & @xmath58 & 11.68 & & + 1264 & @xmath0 & @xmath11 & @xmath581.8 & @xmath58 & 6.63 & & + 1292 & @xmath0 & @xmath64 & @xmath583.0 & @xmath57 & & & ab + 1309 & @xmath0 & @xmath8 & & @xmath58 & & & + 1316 & @xmath0 & @xmath62 & @xmath581.8 & @xmath58 & & & a + 1328 & @xmath59 & @xmath8 & & @xmath58 & & & + 1355 & @xmath59 & @xmath11 & 0.0 & @xmath58 & 10.36 & 1 & b + 1384 & @xmath0 & @xmath64 & 1.9 & @xmath58 & 2.85 & 1 & a + 1387 & @xmath0 & @xmath0 & 0.0 & @xmath58 & 6.42 & & a + 1463 & @xmath0 & @xmath62 & & @xmath58 & 10.66 & 1 & b + 1521 & @xmath0 & @xmath0 & & @xmath58 & & & b + 1572 & @xmath59 & @xmath11 & 0.0 & @xmath58 & & & + 1590 & @xmath0 & @xmath11 & @xmath581.2 & @xmath58 & 7.10 & @xmath60 & b + 1608 & @xmath0 & @xmath9 & @xmath581.3 & @xmath57 & 4.21 & & a +

we describe a new type of dynamical model for hot gas in galaxy groups and clusters in which gas moves simultaneously in both radial directions . the observational motivations for this type of flow are compelling . x - ray spectra indicate that little or no gas is cooling to low temperatures . bubbles of hot gas typically appear in chandra x - ray images and xmm x - ray spectra within kpc of the central elliptical galaxy . these bubbles must be buoyant . furthermore , the elemental composition and total mass of gas - phase iron observed within kpc of the center can be understood as the accumulated outflow of most or all of the iron produced by type ia supernovae in the central galaxy over time . this gaseous iron has been circulating for many gyrs , unable to cool . as dense inflowing gas cools , it produces a positive central temperature gradient , a characteristic feature of normal cooling flows . this gas dominates the local x - ray spectrum but shares the total available volume with centrally heated , outflowing gas . circulating flows eventually cool catastrophically if the outward flowing gas transports mass but no heat ; to maintain the circulation both mass and energy must be supplied to the inflowing gas over a large volume , extending to the cooling radius . the rapid radial recirculation of gas within kpc results in a flat core in the gas iron abundance , similar to many group and cluster observations . we believe the circulation flows described here are the first gasdynamic , long - term evolutionary models that are in good agreement with all essential features observed in the hot gas : little or no gas cools as required by xmm spectra , the gas temperature increases outward near the center , and the gaseous iron abundance is about solar near the center and decreases outward . .2 in

astro-ph0407367

now that we have become accustomed to referring to the hot gas in groups and clusters of galaxies as `` cooling flows '' , no cooling gas has been observed . although energy is lost by x - ray emission , the absence of spectral evidence for gas cooling to low temperatures has been well - documented in many observational papers ( e.g. peterson et al . 2001 ; xu et al . the hot gas in groups and clusters of galaxies may cool , but at a much lower rate than previously thought . this has led to numerous theoretical models in which the gas is assumed to be reheated in some fashion by powerful radio jets , by rising bubbles , by shocks , by magic etc . most or all of these models have been unsatisfactory at some level . the most common difficulty of heated flows is not reproducing the observed radial temperature and density profiles ( e.g. brighenti & mathews 2002 , 2003 ) , which are smooth and approximately similar for galaxy clusters of all scales . less theoretical effort has been expended in understanding the radial abundance gradients in the hot gas , but it is fair to say that little progress has been made so far on this important problem . in this paper we show that a combination of heating and radial mass redistribution can satisfy these stringent observational constraints and simultaneously explain the observed radial metallicity profiles . while the flows described here retain some useful features of traditional cooling flows , we refer to them as `` circulation flows '' since gas flows in both directions simultaneously . cooling to low temperatures is greatly or entirely eliminated . the primary source of heat that drives the circulation is assumed to be moderate agn activity in the cores of luminous elliptical galaxies that reside at or near the center of the diffuse x - ray emission in groups and poor clusters . our theoretical and observational studies of the viralized hot gas in galaxy clusters have been motivated in several ways . first , we have maintained that hot gas in galaxy groups is more likely to reveal the physical nature and dynamical evolution of the gas than hotter gas in rich clusters . we assume that some of the groups are old and , if cosmically isolated , are less likely to have been upset and mixed by recent mergers or powerful radio sources . second , we have argued that the radial variation of the gaseous iron abundance in these groups retains an imprint of the origin and dynamical evolution of the gas since it received its iron . finally , at the relatively low temperatures of gas in galaxy groups , @xmath3 kev , the influence of thermal conductivity @xmath4 on the scale of the flow can be ignored . in many galaxy groups and clusters the gas - phase iron abundance has a noticeable peak centered on the central elliptical galaxy that extends out to @xmath1 kpc . the iron abundance increases to @xmath5solar at the center . xmm spectra indicate that 70 - 80 percent of this iron was formed in type ia supernovae . of particular interest for our discussion here is the realization that the total amount of gas - phase iron in this central region is comparable to the total amount of iron that could be created by all type ia supernovae in the central galaxy since its stars were formed . this is only possible if little or none of the iron - enriched gas has cooled . it is also significant that the region of enhanced iron emission around the central galaxy is much larger in extent than the half - light radius of the optical galaxy . to explain these important features , we describe a simple time - dependent model for the long term evolution of hot gas in galaxy groups in which gas enriched by type ia supernovae inside the central galaxy is heated and buoyantly transported far into the surrounding gas where it is stored over time . meanwhile , most of the hot x - ray emitting gas loses energy and flows inward , as in a standard cooling flow . the cooling inflow receives both iron - enriched gas and energy from clouds or bubbles of heated gas that are moving outward . to avoid flows with catastrophic cooling or discordant temperature profiles , it is essential that the inflowing gas receive both energy and mass from the central regions . the physical nature of this heating @xmath6 compression , sound wave dissipation , weak shocks , etc , is left unspecified for the time being . for successful flows the mass and energy deposition must be spatially broad , not concentrated near a single radius in the flow . this can be accomplished if inflowing gas arriving near the central galaxy is heated to a variety of entropy levels and ultimately floats upward approximately to that radius where its entropy matches that of the surrounding , inflowing gas . the discussion below is a time - dependent generalization of the steady state circulation flows that we discussed in a previous paper ( mathews et al . 2003 ) . flows in which gas moves in both radial directions at each radius are notoriously difficult to reproduce faithfully even with the most sophisticated multidimensional numerical codes because numerical diffusion across the eulerian grid blurs the distinction between the counter - streaming flows . for this reason , in this initial study of the time - dependent evolution of radially circulating gas , we study the evolution of the inward flowing gas with a standard eulerian code , but simplify our treatment of the outflowing gas that is difficult to resolve on computational grids . the physical processes by which gas near the central elliptical galaxy is heated are left unspecified . some of the less important aspects of bubble dynamics that we discussed in our study of steady state circulation flows such as momentum exchange by the drag interaction of rising bubbles , expansion of individual bubbles , etc are either not treated in detail here or are represented by source terms in the equations for the inflowing gas . since the dynamical time for buoyant gas is much less than the radiative cooling time , we assume that heated outflowing clouds move rapidly to their final destination . nevertheless , we do not expect that the detailed inclusion of these complications will alter the basic character or the success of the circulation flows described here . we show that time - dependent circulation flows can quite naturally produce all of the major observed radial profiles in the hot gas : its temperature , density and metallicity . in addition to these important attributes , our circulation flows are compatible with the x - ray observation that very little if any gas cools to low temperatures . a common feature of well - studied x - ray luminous galaxy groups is an extended region of enhanced iron emission surrounding the central elliptical ( e.g. buote et al . 2003a , b ; sun et al . 2003 ; xue , bhringer & matsushita 2004 ; kim & fabbiano 2004 ) . typically , the iron abundance is near - solar within 30 - 40 kpc of the center and decreases outward within 100 - 150 kpc . the ratio of @xmath7/fe elements suggest that a large fraction of this iron has been produced by type ia supernovae ( sniae ) . today the stars in this region from which the sniae presumably occurred are located almost entirely within the central e galaxy with half - light radius @xmath8 kpc , much smaller than the scale of iron enrichment in the surrounding gas . for example , the x - ray luminous group ngc 5044 consists of a large elliptical galaxy surrounded by a cloud of much smaller galaxies extending to at least 400 - 500 kpc . at a distance of 33 mpc , the central galaxy has luminosity @xmath9 @xmath10 , effective radius @xmath11 kpc , and total stellar mass @xmath12 @xmath13 where @xmath14 is the stellar mass to light ratio . the galaxy is surrounded by a huge cloud of hot gas extending to at least 300 kpc . if the hot gas is assumed to have a single phase , its temperature rises from @xmath15 kev at the center to a maximum of @xmath16 kev at 40 - 80 kpc , then decreases slowly beyond ( buote et al . similar temperature profiles are observed in many other groups and poor clusters ( sun et al . 2003 ; loken et al . 2002 ) . however , within about 30 kpc the x - ray spectrum is fit better with two or more gas temperatures spanning the range @xmath17 kev ( buote et al . the low temperature component at @xmath18 kev dominates the emission in this region but only occupies @xmath19 of the volume out to 20 or 30 kpc . this suggests that the remaining volume is occupied by the hotter gas which must be buoyant if it is in pressure equilibrium . this conclusion is directly supported by the pronounced spatial irregularities visible in the chandra image of ngc 5044 ( buote et al . 2003a ) . the iron abundance in ngc 5044 decreases from @xmath20 solar near the center to @xmath21 solar at r = 150 kpc and remains approximately uniform beyond ( buote et al . 2003b ; buote , brighenti & mathews 2004 ) . the total mass of iron observed within 100 kpc is @xmath22 @xmath13 and @xmath7/fe abundance ratios indicate that 70 to 80 percent of this iron has been produced in type ia supernovae . it is interesting to estimate the total mass of iron produced by sniae in the central e galaxy over cosmic time , @xmath23 in @xmath13 ( mathews 1989 ) where @xmath24 is in gyrs and @xmath25 gyrs is the current age of the universe . the iron yield of each type ia supernova is @xmath26 @xmath13 . we assume a type ia supernova rate of @xmath27 in snu units ( rate per 100 years per @xmath28 ) . the current type ia rate in early type galaxies is @xmath29 ( cappellaro , evans & turatto 1999 ) . the @xmath30 variation is consistent with the @xmath31 variation observed by pain et al . ( 2002 ) out to redshift 0.5 , although the exponent 1.2 is uncertain . suppose for example that most of the stars in ngc 5044 formed at cosmic time @xmath32 gyrs ( redshift @xmath33 ) and a 1 gyr delay preceded the first type ia supernova ( gal - yam & maoz 2004 ) . in this case the integration above begins at time @xmath34 gyrs corresponding to redshift @xmath35 . the integration predicts that @xmath36 @xmath13 of iron has been created by all sniae in the central elliptical ngc 5044 . it is remarkable that this estimate is comparable to the total mass of iron observed in the central iron peak out to 100 kpc , @xmath37 @xmath13 , suggesting that rather little snia iron is currently in the stars . at @xmath38 kpc the iron abundance in ngc 5044 plateaus at @xmath39 solar ( buote , brighenti & mathews 2004 ) so it is natural to assume that the background gas was pre - enriched everywhere by type ii supernovae in the early universe to @xmath40 solar . the total iron mass from this background within 100 kpc is only @xmath41 @xmath13 , suggesting that 80 percent of the iron in the central peak is from type ia supernovae , which is consistent with x - ray spectra . while the uncertainty of past supernova rates certainly tempers our estimate of the total iron mass from sniae , its similarity to the observed mass has important implications for the global gas dynamics . very recently , de grandi et al . ( 2004 ) have shown that the mass of the excess iron surrounding cluster - centered elliptical galaxies correlates with their optical luminosity , just as we would expect from the integration above . furthermore , most of the gas - phase iron in the central iron peaks of clusters and groups is observed to have an snia abundance signature within about 100 kpc of the central elliptical galaxy ( de grandi & molendi 2001 ) . one possible explanation for the extended regions of iron enrichment is that the iron was created in sniae within the central e galaxy , @xmath42 kpc , but was subsequently distributed outward by merging events into the much larger region where it is observed today . such mixing due to gas - phase mergers would need to be accompanied by an increased entropy if the iron - rich gas is to remain at a larger radius . if a merger radially mixes the gas without strong heating , after a dynamical time the gas will return to its original radius in the entropy structure and the original metallicity gradient will be restored . however , we expect that most of the shock energy dissipated ( and entropy created ) in mergers occurs closer to the virial radius which is very much larger than the @xmath43 kpc region enriched in iron . central increases in gas - phase metallicity are only observed in the most relaxed groups and clusters they are clearly an indication of an old system . if mixing due to mergers occurred long ago , not enough snia iron would have been available to account for current observations . in our previous gasdynamical studies including iron enrichment ( reviewed in mathews & brighenti 2003 ) , we found that the radial distribution of gas - phase iron from both type ii and early - time type ia supernovae can extend well beyond the optical radius @xmath44 of the central elliptical galaxy . this spatial spreading of the iron can result from early galactic winds if the central elliptical formed monolithically , or from sniae in stars in the central elliptical located well beyond @xmath44 where both the stellar and gas densities are low . nevertheless , these models can not spread the iron to radii as large as @xmath1 kpc where it is typically observed . in a recent paper ( buote , brighenti & mathews 2004 ) we have shown that very little of the gas - phase iron in the ngc 5044 group has been supplied by outflows from the large number of low luminosity group member galaxies . the radial distribution of stellar mass in these non - central group galaxies is approximately constant within the @xmath1 kpc region where the gas iron enrichment has a strong gradient . it is clear from our arguments in that paper that the mass of iron in the central peak within @xmath45 kpc can not be ascribed to starburst outflows or winds associated with other members of the ngc 5044 group . it is difficult to escape the conclusion that a very large fraction of the iron in @xmath43 kpc was produced by type ia supernovae associated with stars that are currently within the large central elliptical , but where were these stars when the iron was produced ? perhaps the extended iron enrichment occurred as the stars within the central e galaxy were assembled . this seems unlikely , however , since the dynamical time for the assembly to occur , @xmath46 gyr , is very much less than the @xmath47 gyr time required to create @xmath48 @xmath13 by type ia supernovae . stars in galaxies undergoing dynamical friction spend most of their total lifetimes in the assembled elliptical galaxy , not in the assembly itself . in this paper we consider an enrichment process in which iron from sniae in the central elliptical galaxy is heated by agn activity and conveyed by buoyant transport to @xmath43 kpc in the surrounding hot gas where it is currently observed . we show below that the agn energetics required to centrally heat the gas are reasonable and consistent with other properties of galaxy groups containing hot gas . furthermore , to store most or all of the iron from type ia supernovae over @xmath47 gyrs , it is essential that very little of the iron - rich gas cools . this is possible since the failure of cooling flows to cool is now widely accepted . while we have used the ngc 5044 group in our discussion , the same argument could be applied to many other similar galaxy groups and poor clusters with @xmath3 kev : rgh 80 ( xue , bhringer & matsushita 2004 ) , ngc 507 ( kim & fabbiano 2004 ) , mkw 4 ( osullivan et al . 2003 ) and ngc 1399 ( buote 2000a , b ) . in every case the hot gas surrounding the central elliptical has an extended region that can be fit better with more than one local temperature and is enriched to @xmath49 solar out to @xmath50 kpc by the products of type ia supernovae . the poor clusters m87 ( gastaldello & molendi 2002 ) and centaurus ( sanders & fabian 2002 ) also have extended multi - temperature , iron - rich regions surrounding the central elliptical galaxy . we consider a spherically symmetric cooling flow that flows in to some small radius @xmath51 where it is heated . plumes and bubbles of heated gas rise buoyantly upstream in the cooling flow until their entropy matches that of the inflowing gas , provided there is no energy exchange between the bubbles and the ambient flow . the rising gas bubbles displace the inflowing gas and reduce the volume through which it flows . larger bubbles rise faster than smaller ones since they experience less drag per unit mass ( cf . equation 13 of mathews et al . the volume filling factor @xmath52 available to the inflowing gas is difficult to estimate because it depends on the unknown bubble size distribution . therefore we use observations of ngc 5044 and rgh 80 as guides to parameterize @xmath53 with the following expression @xmath54 where @xmath55 and @xmath56 kpc is the maximum radius for which the x - ray spectrum shows two temperature components ( buote et al . this form for @xmath53 is also similar to the self - consistent filling factors we calculated for steady state circulation flows ( mathews et al . 2003 ) . in reality @xmath52 may be a function of both radius and time , but we demonstrate below that the inflowing solutions are surprisingly insensitive to our choice for @xmath53 , and presumably also its time variation . the volume filled with inflowing gas in a shell of thickness @xmath57 is @xmath58 where @xmath59 is the area available to the cooling inflow at radius @xmath60 . in this first time - dependent calculation of mass recirculation in cooling flows , we treat the central galaxy as a source of heating and iron enrichment at radius @xmath61 kpc . this is a reasonable approximation since the scale of the central galaxy is much less than the observed extent @xmath62 kpc of the region of snia - enriched gas surrounding it . the equation of continuity for the cooling inflow is @xmath63 + |{\dot m}_h| \right\}.\ ] ] the source term on the right represents the rate that mass is recirculated per unit volume by rapidly buoyant bubbles arriving from the galactic core . here @xmath64 is the rate that mass arrives at the heating radius @xmath51 and @xmath65 is the ( generally smaller ) rate that new mass is created in the central galaxy by stellar mass loss and type ia supernovae ( snia ) respectively . the specific stellar mass loss rate is @xmath66 where @xmath25 gyrs is the current age of the universe . the mass loss rate from snia is @xmath67 where @xmath68 is the stellar mass to b - band light ratio and @xmath69 @xmath13 is the mass ejected in each snia event . the coefficient @xmath70 in the source term for the equation of continuity is the normalized probability that mass heated at radius @xmath51 is transported to a remote volume @xmath71 in the flow . in our study of steady state circulation flows we showed that the time for gas bubbles heated at @xmath51 to move out to their final radius was @xmath72 of the flow time back to @xmath51 . the time required for heated bubbles to reach their final radius is comparable to the dynamical time in the cluster potential which is short compared to the cooling ( or inflow ) time ; @xmath73 is a defining characteristic of so - called cooling flows . the normalization for the circulation probability is @xmath74 where @xmath51 and @xmath75 are the minimum and maximum radii that define the region of circulation . this suggests a simple probability @xmath76 that depends only on the radius , @xmath77 . for the mass redistribution we consider parameterized functions of the form @xmath78 where @xmath79 @xmath80 is an adjustable scale factor and @xmath81 is either 1 or 2 but @xmath82 may have any positive value . the normalization can be expressed in terms of incomplete gamma functions , @xmath83 \right\}^{-1}\ ] ] where @xmath84 and @xmath85 . the probability distribution has a maximum at @xmath86 . the density of iron requires a continuity equation of its own . if @xmath87 is the proton number density in the hot gas , the density of hydrogen is @xmath88 where @xmath89 is the proton mass . the total density is @xmath90 where @xmath91 is the mean molecular weight for full ionization and @xmath92 allows for the mass of helium . since @xmath93 the density of element @xmath94 ( i.e. , iron ) is @xmath95 . the gas dynamical evolution of the iron is most easily described by regarding the product @xmath96 as a single variable , @xmath97 @xmath98 + |{\dot m}_h| z_h \right\}.\ ] ] here @xmath99 is the adopted mean stellar iron abundance in the central galaxy ( rickes et al . 2004 ) and we take @xmath100 as the solar abundance . each snia remnant is assumed to contain @xmath26 solar masses of iron . the production of iron by type ia supernovae is about ten times greater than that due to stellar mass loss . although the temperature and density in rising bubbles should both match those in the surrounding gas when their entropies become equal , the mass added by recirculated gas in some remote volume @xmath71 must cause the local gas density to increase with time . if the density increases in some region of the flow , the increased radiative cooling there increases the density further . eventually , the gas may cool catastrophically by runaway radiative losses . the dense , cooled gas then falls in the gravitational potential , and after time becomes supersonic , increasing @xmath108 enormously . to avoid these undesirable consequences , it is necessary to heat the inflowing gas as it moves past the rising bubbles . this heating can be due to @xmath109 work done by the expanding bubbles , the power expended by the drag force on the rising bubbles , and thermal mixing of some bubble gas with the local inflow . the thermal evolution of the bubbles may not be purely adiabatic . in this discussion we wish to avoid a detailed discussion of these complicated bubble - flow interactions and simply assume that a power @xmath110 is transferred to the cooling flow where @xmath111 is a characteristic luminosity expended in heating gas at @xmath51 and @xmath112 is a dimensionless factor of order unity . the magnitude of this heating luminosity will be justified below . we also assume for simplicity that this distributed heating is transferred to the cooling flow with the same probability distribution @xmath70 as the mass recirculation . obviously , some very complicated bubble physics is subsumed with this assumption , some of which were discussed in our previous paper . however , by combining these detailed interactions into a single energy source term , we can explore new gas dynamical possibilities that must be verified by more detailed calculations in the future . the evolution of the energy density @xmath113 , including this heating and radiation losses , is described by @xmath114 @xmath115 + |{\dot m}_h| \right\ } + \epsilon_h l_h { dp \over dv}\ ] ] where @xmath112 is of order unity and @xmath116 is the usual coefficient for optically thin radiative cooling ( sutherland & dopita 1993 ) . the terms on the right hand side represent respectively compressive heating , cooling by radiative losses , addition of thermal energy density associated with the ( non - heating ) mass deposition and genuine heating provided by the mass deposition . when @xmath117 the recirculated mass is assumed not to heat or cool the gas , so the specific thermal energy @xmath118 contributed by the recirculated gas is identical to that of the local gas . consequently , the energy density equation must have a source term analogous to that in equation ( 5 ) . this third term on the right side of equation ( 18 ) has little effect on the solutions . while the spatial distribution @xmath70 of the bubble heating term @xmath119 need not be identical to that of the recirculated mass , it is the simplest assumption that involves the fewest number of additional parameters and seems appropriate for this initial study of time - dependent circulation . the power expended in heating the gas at @xmath51 is @xmath120 { 3 k t_h \over 2 \mu m_p } \langle h \rangle\ ] ] where @xmath121 is the temperature of the inflowing gas at @xmath51 and @xmath122 is a dimensionless mean heating factor required to recirculate the gas . the mean heating factor @xmath122 can be estimated by examining how the incoming gas is heated at @xmath51 and assuming adiabatic bubble evolution . bubbles are formed at @xmath51 with temperature @xmath123 where @xmath121 is the cooling flow temperature at the heating radius . [ virialized gas lost from stars in the central galaxy is also assumed to have a pre - heating temperature @xmath124 ; in general @xmath108 is considerably larger than @xmath125 . ] the bubbles are always in pressure balance with the surrounding gas , @xmath126 and so @xmath127 as bubbles move out from @xmath51 , their entropy is assumed to be constant , @xmath128 some deviation from constant entropy is expected if the bubbles radiate or if they lose mechanical energy in heating the surrounding gas , so our adiabatic assumption is a limiting case . eliminating @xmath129 from the equations above gives an equation for the bubble density at @xmath130 @xmath131 and , if the entropy is conserved , @xmath132 adiabatically rising bubbles expand and cool . by examination of the last two equations above , as outflowing bubbles of constant specific entropy approach the radius in the flow where the flow entropy matches that of the bubbles , @xmath133 , the densities and temperatures also become equal , i.e. @xmath134 and @xmath135 . the specific thermal energy of gas adiabatically transported in this manner to distant regions of the flow is identical to that in the surrounding gas and no heating occurs in this idealized approximation . we shall see below , however , that the increase in gas mass in the flow due to the deposition of outflowing gas actually results in a higher gas density , which is further increased by ( non - adiabatic ) radiative cooling . alternatively , if bubbles of hot gas thermally merge with their environment at some smaller radius where @xmath136 , the surrounding flow will be heated . from the last two equations above the entropy @xmath137 of heated bubbles at @xmath51 is @xmath138 where @xmath139 . this last equation associates the heating parameter @xmath140 with a unique bubble entropy and therefore with a unique radius @xmath60 in the cooling flow , assuming that the entropy of the cooling inflow increases monotonically with radius and that the bubbles remain adiabatic . in ngc 5044 and other similar cooling flows , as well as for the circulation flows considered here , @xmath141 is a good approximation , so @xmath142 where we assume @xmath143 . the mean heating factor is approximately @xmath144 @xmath145\ ] ] where @xmath146 is the quantity in square brackets in equation ( 12 ) . figure 1 shows a sample of normalized circulation probabilities @xmath147 that we consider below with each curve labeled with the parameters @xmath148 and the mean heating factor @xmath122 . in our previous paper on steady state circulation flows we considered in detail the dynamics of buoyant bubbles , including the exchange of momentum between the cooling inflow and the counter - streaming heated bubbles ( mathews et al . we showed that this momentum exchange is not particularly important for a wide range of bubble sizes . in addition we discussed in detail the contribution of the heated gas to the overall x - ray emission . the x - ray emission from rising bubbles is small but observable , but emission from the hottest gas that ends up at the largest radii , is necessarily the lowest because of its low density . in the standard single - temperature interpretation of the x - ray emission from ngc 5044 and other similar flows , the observed temperature is dominated by that of the dense inflowing gas , as we have shown in our previous paper on circulation flows ( mathews et al . finally , in this earlier paper we discussed in detail the various heating processes by which the rising bubbles can share energy with the ambient gas . we will see below that energy exchange between gas flowing in opposite directions is critical to the success of time - dependent circulation flows . gas dynamical solutions are found by solving equations ( 5 ) , ( 15 ) , ( 16 ) and ( 18 ) in spherical symmetry using a 1d zeus - like eulerian code ( stone & norman 1992 ) . we employ a computational grid of 100 logarithmically increasing zones between @xmath61 kpc and @xmath149 kpc in which the first zone is 215 pc in width . outflow boundary conditions are imposed at @xmath51 and the boundary at @xmath75 is fixed and reflecting since it is far beyond the cooling radius ( @xmath150 kpc ) where the radial flow is negligible . our first objective is to establish a dynamically relaxed cooling flow without mass recirculation , i.e. @xmath151 . this flow can serve as a smooth starting point for flows with circulation . solutions with @xmath151 also allow us to explore the effect that various filling factors @xmath53 have on the flow . the initial conditions for these pure cooling ( and non - circulating ) flows are the temperature and density profiles observed in ngc 5044 ( buote et al . the gas is initially at rest . these initial profiles are shown with dotted lines in figure 2 . three cooling flows are calculated for @xmath152 years between @xmath51 and @xmath75 for three fixed filling factors having @xmath153 , 0.5 and 1 , but all with @xmath56 kpc ( see equation 3 ) . after @xmath154 gyrs the computed gas density and temperature profiles , shown with solid lines in figure 2 , remain in reasonably good agreement with the profiles observed in ngc 5044 today , shown with dotted lines . while this is reassuring , it has no particular significance since there is no reason to expect that the profiles observed today should persist over such a long time , particularly since our flows include no heating . some modest heating does seem to be required in @xmath43 kpc where the computed temperature profiles are @xmath155 kev below those observed . gas beyond about 100 kpc does not change appreciably during the entire calculation although information moves at the sound speed from @xmath51 to @xmath75 in only 0.95 gyr . both observed and computed temperature profiles @xmath156 exhibit maxima well beyond the optical half - light radius , @xmath157 kpc , of the central elliptical galaxy . this thermal feature , which is observed in the hot gas on all scales from galaxies to rich clusters , is a characteristic signature of cooling by radiative losses . the overall approximate agreement of our simple cooling flows with ngc 5044 indicates that , to a first approximation , the x - ray emission from this group is dominated by inflowing gas that resembles a conventional cooling flow . the observed entropy , pressure and density profiles plotted in figure 2 with dotted lines have been interpreted assuming that the filling factor is unity . however , the computed solid line curves in figure 2 are the actual variables in the gas flowing for each assumed @xmath53 . therefore , to compare the computed flows ( solid lines ) in figure 2 with observations , the computed density profiles would need to be adjusted downward by @xmath158 and the entropy upward by @xmath159 . both these adjustments tend to bring the solid curves into better agreement with the observations at small radii . the most useful result in figure 2 is the insensitivity of the inflowing gas density and temperature profiles to large variations in the filling factor @xmath53 . to the accuracy that we require for our goals here , the computed gas profiles are identical for all three @xmath53 . the similarity of flows with different filling factors is particularly advantageous because of the difficulty in determining the filling factor either from first principles or from very detailed numerical calculations ; fortunately @xmath53 can be found from observations . in our study of steady state circulation flows we were able to determine the filling factor for the return flow in a self - consistent manner only if all bubbles were assumed to have the same mass ( mathews et al . however , in view of the insensitivity to @xmath53 , none of the conclusions that we reach below depends strongly on the particular filling factor @xmath53 chosen or its possible variation with time , provided it does not deviate radically from the three @xmath53 illustrated in figure 2 that bracket the known observations of galaxy groups . therefore , in the remaining flows discussed below we adopt a fixed filling factor with parameters @xmath160 and @xmath56 kpc . as discussed earlier , we consider the possibility that most or all of the gas that flows into the central galaxy , as well as gas expelled from evolving stars , is heated and buoyantly transported , with its iron , into a large volume of the surrounding hot gas . heating near @xmath51 is assumed to be continuous and isotropic , creating bubbles that flow out in every direction . in this initial study of time - dependent circulation flows we exploit the defining property of cooling flows that the dynamical time is much less than the cooling time . indeed , as we showed in mathews et al . ( 2003 ) , the dynamical time for heated gas to rise buoyantly is much less than the radiative cooling time that governs the incoming flow . in this section we describe circulation flows in which bubbles of gas heated near radius @xmath61 kpc float outward without heating the incoming flow . in view of the variety of complicated ways that rising bubbles can heat the inflowing gas , it is useful to consider flows with no heating as a useful intermediate step toward more realistic flows . these idealized bubbles rise adiabatically to their final radius and deposit gas into the cooling flow at a radius where the entropy and specific thermal energy of the bubble are identical to those in the ambient flow . non - heated circulation flows can be studied by allowing recirculation of mass , @xmath161 , but not heat energy , i.e. @xmath112 is set to zero in equation ( 18 ) so there is no distributed energy source . each flow ( and those discussed subsequently ) begins at cosmic time @xmath162 gyrs , corresponding to a redshift of 2.5 , with initial conditions taken from the fully - relaxed @xmath163 solution in figure 2 . the spatial evolution of the iron abundance @xmath164 , is also followed . we find that such non - heated flows cool catastrophically and deviate strongly from the observations . such cooling is not desired in our solutions since we wish to retain all of the snia iron in the hot gas as the observations suggest . if the mass circulation pattern @xmath70 is too concentrated near some radius @xmath165 , the increased gas density there leads to rapid radiative losses and eventual cooling . such a case is illustrated in figure 3 in which the recirculation and heating parameters for @xmath70 are ( @xmath166 ) = ( 2,18,10,0 ) . for this @xmath70 the recirculated mass is deposited in an annulus @xmath47 kpc wide located at @xmath167 kpc . as it grows in mass , this shell cools and settles inward in the gravitational potential . at time 3.6 gyrs , as shown in figure 3 , some gas has cooled to @xmath168 k and the flow does not subsequently recover from this state . therefore , the commonly held notion that bubbles of heated gas rise adiabatically to some radius where their entropy @xmath169 matches that of the local cooling flow @xmath170 can not be sustained because the local entropy of the flow is reduced as recirculated mass is introduced . even if the mass in an annular shell increases at constant temperature , the gas density increases and local radiative losses are stimulated ; both processes decrease the local gas entropy . this is an important consideration that leads to the failure of all models described in this section . our discussion in 3 relating the heating factor @xmath140 to a unique radius @xmath60 must necessarily be viewed as a semi - quantitative idealization . the iron abundance in figure 3 is distributed within a region between the current location of the dense shell and the somewhat larger radius @xmath165 at which @xmath70 has a maximum . a similar localized cooling can develop if the mass recirculation is more evenly distributed just between @xmath51 and some radius @xmath171 less than the cooling radius . if the mass deposition decreases sharply at @xmath171 , the cooling inflow from @xmath172 can be locally decelerated and compressed near @xmath171 . this sort of flow , not illustrated here , can result in radiative cooling to low temperatures in a gaseous shell . such compressive cooling was anticipated in our earlier work on steady circulation flows and resembles the cooling galactic drips discussed by mathews ( 1997 ) . circulation flows without distributed heating ( @xmath117 ) also fail even if the mass redistribution @xmath70 is very broad . such a flow with parameters ( @xmath166 ) = ( 1,1.5,30,0 ) is illustrated in figure 4 at time 8.0 gyrs when the flow has cooled catastrophically near @xmath61 kpc . in this flow the entropy @xmath173 out to @xmath174 kpc has been dramatically reduced because of radiative losses associated with the increased density . as the density increases , the gas pressure gradient must become more negative to balance the gravitational potential , causing the gas pressure at small radii to increase . the gas pressure actually develops a maximum just beyond @xmath51 within which gas is forced inward . the entire gas flow within @xmath174 kpc is approaching that of freefall and the mass flux at @xmath51 , @xmath108 , has become very large . the high inflow velocity and the rapid recirculation of gas from @xmath51 to larger radii cause the iron abundance to be low and flat - topped as seen in figure 4 . fortunately , catastrophic cooling is easily eliminated by introducing a distributed heating term in the energy equation , @xmath175 . for simplicity , we consider heating that is distributed over the flow with the same normalized probability distribution @xmath70 that describes the mass circulation . even with this constraint many circulation flow models are in satisfactory agreement with observation . the time - dependent flows we describe here are often in a quasi - steady state , stable over long periods of time , so it is interesting to consider how the interbubble gas is heated by counter - streaming bubbles in the steady state limit . as a matter of principle , an ensemble of outflowing and expanding bubbles does not heat the surrounding gas by performing @xmath6 work provided conditions are approximately steady state ( mathews et al . 2003 ) . certainly a single bubble that suddenly appears and rises into an atmosphere of hot gas displaces its volume in the ambient gas at every radius and moves some ambient gas higher in the gravitational potential . in this case @xmath6 work is done as the cloud expands and moves outward . however , a single buoyant cloud is not a steady state phenomenon , it is inherently time - dependent . an approximate steady state is possible only if many clouds are flowing through each shell of volume @xmath176 at any time . in this case the volume filled with rising bubbles @xmath177 is nearly constant ; as bubbles exit from the surface at @xmath178 , new ones appear at @xmath60 , approximately conserving the number and total volume of bubbles . individual bubbles expand adiabatically as they move from @xmath60 to @xmath179 and their internal energy decreases accordingly . however , this expansion does not continuously heat the interbubble gas since , in the strict steady state sense , this work was already done by similar bubbles at an infinite time in the past . since the volume of the interbubble gas does not change in steady flow , @xmath180 and the surrounding gas experiences no continuous @xmath181 heating . instead , as bubbles move through @xmath71 the ambient cooling flow gas is subsonically rearranged transversely along equipotential surfaces , conserving its total volume . however , some work is done against viscous forces during this rearrangement . as a result , each bubble experiences a drag force as it moves upstream through the interbubble gas . work done against viscous drag is a continuous source of spatially distributed heating in the surrounding gas . figure 5 illustrates two heated circulation flows that evolved without catastrophic radiative cooling from 2.65 gyrs to 13.7 gyrs and which are in reasonable agreement with the observed properties of ngc 5044 and other similar x - ray luminous groups . the final temperature and iron abundance profiles of a flow described with circulation parameters ( @xmath166 ) = ( 1,1.5,45,1.6 ) are shown as light solid and dashed lines in figure 5 . the parameters for this flow were chosen to fit the observed temperature profile of ngc 5044 , but the iron abundance does not exceed about 0.7 solar , somewhat lower than that observed in this group ( buote et al . the iron abundance observed in ngc 5044 , which rises to @xmath182 solar at @xmath183 kc , is better fit with the circulation flow ( @xmath166 ) = ( 1,1.1,41,1.9 ) which is shown with heavy solid and dashed lines in figure 5 . because of the slightly larger heating @xmath184 in this flow , the inflow velocities are lower as the gas approaches @xmath51 and the iron abundance there is correspondingly larger . for both flows in figure 5 the total mass of iron within 100 kpc is about @xmath185 @xmath13 since we use the same snia rate as in equations ( 1 ) and ( 2 ) . evidently , if we relaxed our constraint that the recirculation distributions @xmath70 of mass and heat energy are identical and allowed them to vary with time , it would be possible to match both the abundance and gas temperature profiles for any observed group . the density , pressure , flow velocity and entropy profiles for the two flows shown in figure 5 are acceptable fits to the ngc 5044 observations . indeed , we find similar flows with a range of parameters ( @xmath166 ) , indicating that our results are robust . it is particularly significant that the computed temperature profiles @xmath156 in figure 5 have maxima and positive gradients @xmath186 at small @xmath60 . in our previous paper on circulation flows ( mathews et al . 2003 ) we showed that the x - ray emission from hotter buoyant clouds of much lower density does not significantly alter the single phase temperature profile established by the denser inflowing gas . however , in the diffusely heated 1d and 2d cooling flows that we previously solved with numerical hydrocodes , we invariably found monotonically decreasing temperatures @xmath187 in the centrally heated region ( e.g. brighenti & mathews 2002 ) . in those calculations we heated all the gas throughout a volume near the center of the flow , including gas at all densities , and our specific objective was not to create or follow bubbles in the heated region . some larger buoyant bubbles flowed out from the heated region but they were not resolved in part because of spurious numerical diffusion . we avoid numerical diffusion here by assigning a fraction @xmath188 of the volume to the buoyantly outflowing heated gas . but the outwardly increasing temperature near the center in figure 5 is made possible only by the combined influence of the redistribution of heated mass and the cooling inflow . the profiles shown in figure 5 are not momentary transients that happen to be similar to the observations at time @xmath25 gyrs . while these flows are fully time - dependent , the gas density and temperature profiles remain highly stable during the last @xmath189 gyrs of the calculations . these flows are therefore quasi - steady and the main secular change is the slowly increasing iron abundance @xmath190 . finally , we emphasize that the circulation of mass that we describe here can not be viewed as normal convection , in which the entropy profile is flat , the temperature fluctuations are small , and the temperature gradient is negative . figure 6 shows the time variation of several global parameters for the two circulation flows illustrated in figure 5 . the mass flow rate @xmath191 at @xmath51 , the stellar mass loss rate @xmath192 and the approximate power expended in heating the gas at @xmath51 , @xmath111 , in units of @xmath193 erg s@xmath194 all slowly decrease with time . the current heating luminosity , @xmath195 erg s@xmath194 , is comparable to the x - ray luminosity of ngc 5044 , @xmath196 erg s@xmath194 , and this is expected in order to maintain the flow without radiative cooling to low temperatures . this level of heating @xmath110 corresponds to a rather low level agn , which is consistent with other observations of ngc 5044 . in mathews et al . ( 2003 ) we showed that the heating of the cooling inflow by bubbles due to work done against drag forces and @xmath109 expansion ( if applicable ) are comparable . we now estimate the collective heating by bubbles in ngc 5044 due to drag interactions . assuming that the bubble velocity @xmath197 is much greater than that of the cooling inflow , the terminal velocity is @xmath198 where @xmath199 is the dimensionless drag coefficient and @xmath200 . the rate that energy is lost from a single bubble of radius @xmath201 is then @xmath202 . the space density of bubbles is @xmath203 . the heating rate due to all bubbles is @xmath204 erg s@xmath194 , assuming @xmath205 , coherent bubbles of radius 0.5 kpc formed at @xmath61 kpc and taking @xmath206 as the density profile observed in ngc 5044 . the bubbles are assumed to be in pressure equilibrium and the integration extends from @xmath51 to the radius of equal entropy where @xmath207 . in spite of the obviously approximate nature of this estimate , it is within a factor of 2 or 3 of the heating luminosities @xmath110 required for the flows illustrated in figure 5 . therefore the amount of distributed heating we require for successful flows is comparable to that expected from bubble - flow interactions . in our models the heating is continuous , but it may be possible to create similar circulation flows with intermittent heating near the center . it is likely , however , that the intermittency if it exists has a rather short period or duty cycle . the dynamical outflow time for bubbles of size @xmath201 at radius @xmath60 is @xmath208 . for bubbles of radius @xmath209 , at @xmath210 kpc , @xmath211 km s@xmath194 and the bubble rise time @xmath212 years is rather short . if gas is heated periodically , the period would have to be less than this and may in fact be very much less . hot bubbles appear in the chandra image of ngc 5044 well within 10 kpc from the center , and the x - ray image of ngc 5044 is similar to those of other elliptical galaxies with comparable @xmath213 . if mass circulation is a common feature of all flows of this type , the implication is that there are no truly quiescent supermassive black holes . finally , we note that the heating due to type ia supernovae is small compared to @xmath111 . for example , at time @xmath214 the total snia heating luminosity is @xmath215 ergs s@xmath194 where @xmath216 ergs is released with each supernova . evaluating at @xmath217 , we find @xmath218 erg s@xmath194 which is very much less than @xmath110 . for simplicity we have ignored supernova heating in equation ( 18 ) . one of the strangest features occasionally observed in cooling flows is a central minimum in the iron abundance ( johnstone , et al 2002 ; sanders & fabian 2002 ; schmidt , fabian & sanders 2002 ; blanton , sarazin & mcnamara 2003 ; dupke & white 2003 ) . when present , these minima typically lie within the central @xmath219 kpc of the flows , i.e. where conditions are dominated by the central elliptical galaxy . such minima or a central flattening in @xmath190 which is even more common would not be expected in conventional cooling flows since the local frequency of iron - producing type ia supernovae should increase with the stellar density toward the very center of the flow . ( the stellar density is more centrally peaked than the gas density . ) efforts to understand the central dips in the iron abundance in terms of differential cooling of abundance inhomogeneities have been discussed by morris & fabian ( 2003 ) . in the circulation flows described here it is possible to produce small central iron abundance minima if the redistribution probability @xmath70 varies with time . for example , a small central dip in the iron abundance results if the iron ( and mass ) deposition @xmath70 is concentrated near @xmath220 kpc , just preceding the flow time for gas in this region to return to @xmath51 by the current time @xmath221 gyrs . central regions with very low iron abundance are difficult to form in circulation flows because ( 1 ) the cooling inflow necessary to explain the observed gas temperature profile continuously recirculates iron - enriched gas within @xmath2 kpc and , since most of the snia iron is produced at early times , ( 2 ) relatively little iron is available during the last cooling inflow time to create a strong reversal in the iron abundance profile . for these same reasons , central iron deficiencies are not easy to produce by allowing the iron mass and energy redistribution probabilities @xmath70 to differ . nevertheless , because of the continuous radial recirculation it is quite natural for the central gas iron abundance in circulation flows to become rather constant within @xmath222 kpc where the snia - producing stars are very strongly peaked . similar flat or slowly varying iron abundance cores are commonly observed in many groups and clusters . although the heated flows we describe here are generally successful , much additional work will be necessary before they can be fully accepted . we have not identified the physical mechanism(s ) that heat the gas near the center , but they are likely to be associated with the central black hole . these heating processes may be related to the common difficulty in observing hot gas in close proximity to the central black hole , constraining the accretion rate to less than 10 percent of the bondi rate ( loewenstein et al . if gas is heated near the black hole , less accretion is expected . it is likely that cosmic rays and magnetic fields are relevant since the cores of almost all massive elliptical galaxies are sites of low luminosity nonthermal radio emission ( e. g. sadler et al . powerful radio jets are typically found in rich clusters , not in ( more numerous ) small groups that contain central ellipticals of comparable optical luminosity . jets that deposit energy asymmetrically and to very large radii are an unlikely heating source for groups like ngc 5044 and rgh 80 . by contrast , the heating required in our circulation flows is isotropic and requires a modest , approximately continuous agn power . in our discussion of steady state circulation flows ( mathews et al . 2003 ) , we describe limits on the size and mass of bubbles if heating occurs at some radius @xmath51 in the flow . provided they remain coherent , larger bubbles can move to larger radii , but the size of new bubbles can not exceed the radius @xmath51 where they are formed and this in turn limits the global rate that mass can be buoyantly transported outward . we discussed these difficulties in the context of steady state circulation and they did not critically impair the circulation model . in this first paper on time - dependent circulation we have not discussed again all these important considerations . the insensitivity of circulation flows to the filling factor also lessens the urgency for a fully self - consistent model for the rising bubbles , even if that were possible . nevertheless , as we computed these circulation flows , our concept of the heating radius @xmath51 became more flexible . we consider @xmath51 to be only a representative radius where the gas is heated . larger ( smaller ) clouds may be heated and formed at larger ( smaller ) radii . in reality the heating probably occurs throughout a volume that may vary with time . a critical measure of the success of circulation flows is the apparent rise in the gas temperature toward a maximum well outside the central elliptical galaxy . this @xmath156 is possible only if the emission from the high density cooling inflow dominates the observed temperature profile in the inner flow , even when the mass flux carried by the rising bubbles transports nearly the same mass outward . we showed that this is indeed the case for steady state circulation flows ( mathews et al . 2003 ) . to see why this is also true for time - dependent circulations , we estimate the ratio of the bubble to cooling flow bolometric x - ray emissivities at @xmath223 at time @xmath214 , @xmath224 the bubble density at @xmath51 can be estimated from the steady state equation for mass conservation , @xmath225 } { 1 \over u_{bh}}\ ] ] where @xmath226 is the typical velocity of rising bubbles near @xmath51 . using the flow parameters for the ( @xmath166 ) = ( 1,1.5,30,2.0 ) flow at time @xmath217 , we find @xmath227 gm @xmath228 where @xmath229 is the bubble velocity in units of 100 km s@xmath194 . the bubble temperature @xmath230 k follows from pressure equilibrium . with these values we find @xmath231 if @xmath232 , a typical initial bubble velocity in the steady flows of mathews et al . therefore , emission from the cooling inflow dominates the observed @xmath156 and the bolometric x - ray emission near the center of this flow . x - ray spectral features from heated gas may be emitted mostly by somewhat denser gas at intermediate temperatures , @xmath233 . there are two stages of heating in our circulation flows . near the center @xmath234 we assume a strong and concentrated heating by an agn that rather sharply raises the entropy , creating buoyant bubbles of hot gas . this is regarded as the primary heating mechanism . a second heating process occurs when the rising bubbles share some of this thermal energy with the inflowing gas over a large region of the flow . without this distributed heating the entropy profile in the gas fails to match the observations ( as in figure 4 ) . distributed heating occurs because of bubble - flow drag or because some of the outflowing gas thermally merges with the incoming flow before it reaches the radius where the initial entropy of the bubble equals that of the ambient flow . this distributed heating helps determine the location of the temperature maximum observed in the flow . finally , it is not surprising that the central heating exceeds @xmath111 by some factor @xmath235 because ( 1 ) the mean heating factor @xmath122 is underestimated if some of the bubble energy is prematurely transferred to the inflowing gas , ( 2 ) some of the bubble energy is lost by radiation and this is necessary to explain the superiority of two - temperature fits to the x - ray spectra , and ( 3 ) the kinetic energy acquired by the bubbles is extracted from the galaxy - group potential and is independent of @xmath111 . x - ray observations of the perseus cluster show buoyant cavities that have moved through distances that are 10 - 15 times larger than that of the cavity diameters ( e.g. fabian et al . the coherence of these large bubbles is remarkable , and they appear to be transporting mass and heated gas to large radii in the flow . since small buoyant regions experience more drag as they move upstream through the cooling flow , their rate of progress may not be sufficient to convey the mass and energy that we require in the models described here . clearly , observational and computational studies of the physical coherence of heated regions will be necessary to fully validate circulation flows . the dynamical coherence of heated bubbles may result from internal magnetic fields that are force free . it is possible that some gas cools . for ngc 5044 , @xmath61 kpc , taken literally , is about @xmath236 so we expect @xmath21 @xmath13 yr@xmath194 from stellar mass loss within @xmath51 . even if this gas cools radiatively in the usual manner , such a small cooling rate would not have been detected in our xmm observations ( buote et al . however , cooling in this core region is likely to be accelerated by electron - dust grain cooling and emit x - rays at a much reduced level ( mathews & brighenti 2003 ) . central ( dust - enhanced ) cooling may consume @xmath237 percent of the iron produced by sniae . in this paper we describe a new model for the dynamical evolution of hot gas in groups and clusters of galaxies in which gas flows in both radial directions simultaneously . the incoming gas resembles a traditional cooling flow and dominates the x - ray emission . the outgoing gas consists of an ensemble of heated buoyant bubbles that rise to distant regions where they merge with the inflowing gas . we have not calculated the dynamics of the rising bubbles of heated gas in detail , and this must be an objective for the future . instead , we constructed a schematic time - dependent model guided in part by observations and by the detailed bubble - flow interactions described in our previous study of steady state circulation flows ( mathews et al . it is noteworthy and fortunate that the density and temperature profiles of the cooling , inflowing gas are not sensitive to the filling factor @xmath53 , which is easier to determine reliably from the observations than from dynamical models for the bubbles . this insensitivity may explain the approximate similarity of @xmath156 and @xmath238 profiles in clusters of all sizes . we have shown here that the steady state circulation flows we proposed earlier ( mathews et al . 2003 ) can be generalized to include time variation . an essential feature of successful circulation flows is that both mass and energy must be distributed from the center throughout a large volume of the flow within the cooling radius . the outward increase in gas temperature near the center of the flow , a characteristic feature of most galaxy clusters , results largely from a conventional cooling inflow , modified somewhat by additional energy received from locally outflowing bubbles . after evolving for many gyrs , circulation flows are in accord with the important observation that most or all of the iron produced by type ia supernovae in the central elliptical galaxy has been stored in a region typically extending to @xmath1 kpc . from the perspective of the circulation flow hypothesis , a record of the enrichment - heating history of groups is retained in currently observed iron and temperature profiles of the hot gas and it may be possible to retrieve this information by comparison with more accurate circulation flows . the central peak in the radial iron distribution within @xmath0 kpc is very sensitive to the ( negative ) cooling inflow velocity at small radii . if this velocity is high , the iron abundance remains low and flat - topped . iron abundance profiles that peak sharply toward the center , such as that observed in ngc 5044 , can occur only if the flow velocity is reduced to a very low level by ( recent ) large heating near the center . we have shown that the central iron peak is more pronounced when the mass redistribution and heating are more strongly peaked at smaller radii . concentrated peaks in the iron abundance within @xmath1 kpc are also typically observed in many rich clusters : sersic 159 - 03 ( kaastra et al 2001 ) , cygnus a ( smith et al . 2002 ) , abell 1795 ( ettori et al . 2002 ) and abell 2029 ( lewis , stocke & buote 2002 ) . we speculate that these regions are also mostly enriched by the circulation of iron from type ia supernovae in the central galaxies and there is some observational support for this ( e.g. ettori et al . 2002 ) . if so , this may indicate a characteristic heating luminosity @xmath239 ergs s@xmath194 for all clusters . however , if there is an upper limit on the heating luminosity @xmath111 , perhaps set by the physical conditions near the central black hole , it is possible that the mass inflow rate @xmath240 in rich clusters is too large for the inflowing gas to be heated in the way we describe here . this may give rise to an additional more violent energy releases in the form of powerful radio jets which are rare in galaxy groups that have more modest mass cooling rates @xmath240 . we believe the circulation flows described here are the first gasdynamic , long - term evolutionary models that are in good agreement with all essential features of the hot gas : the gas temperature maximum at several @xmath44 , the approximately linear rise of entropy with radius , the dominance of snia products in the x - ray spectra , the total iron mass within 100 kpc and its radial distribution . the relative constancy of these observed profiles during many gyrs of our calculated flows is consistent with the similarity of the observed x - ray properties among galaxy groups and clusters . finally , little or no gas cools in circulation models , and this is consistent with the much decreased or absence of cooling gas in x - ray spectra of both galaxy groups and clusters . .4 in studies of the evolution of hot gas in elliptical galaxies at uc santa cruz are supported by nasa grants nag 5 - 8409 & atp02 - 0122 - 0079 and nsf grants ast-9802994 & ast-0098351 for which we are very grateful . blanton , e. l. , sarazin , c. l. & mcnamara , b. r. 2003 , apj , 585 , 227 brighenti , f. & mathews , w. g. 2003 , apj 587 , 580 brighenti , f. & mathews , w. g. 2002 , apj 573 , 542 buote , d. a. 2000a , apj 539 , 172 buote , d. a. 2000b , mnras , 311 , 176 buote , d. a. et al . 2003a , apj , 594 , 741 buote , d. a. et al . 2003b , apj , 595 , 151 buote , d. a. , brighenti , f. & mathews , w. g. 2004 , apj ( in press ) ( astro - ph/0404430 ) cappellaro , e. , evans , r. & turatto , m. 1999 , a&a , 351 , 459 de grandi , s. , ettori , s. , monghetti , m. & molendi , s. 2004 , a&a , 419 , 7 de grandi , s. & molendi , s. 2001 , apj , 551 , 153 dupke , r. a. & white , r. e. iii 2003 , apj , 583 , l13 ettori , s. , fabian , a. c. , allen , s. w. & johnstone , r. m. 2002 , mnras , 331 , 635 conference , august 2000 ( astro - ph/0103392 ) fabian , a. c. et al . 2000 , mnras , 318 , l65 gal - yam , a. & maoz , d. 2004 , mnras , 347 , 942 gastaldello , f. & molendi , s. 2002 , apj , 572 , 160 kim , d .- w . & fabbiano , g. 2004 , apj , submitted ( astro - ph/0403105 ) johnstone , r. m. , allen , s. w. , fabian , a. c. & sanders , j. s. 2002 , mnras , 336 , 299 kaastra , j. s. et al . 2001 , a&a , 365 , l99 lewis , a. d. , stocke , j. t. & buote , d. a. 2002 , apj , 573 , l13 loewenstein , m. , mushotzky , r. f. , angelini , l. , arnaud , k. a. & quataert , e. 2001 , apj 555 , l21 loken , c. , norman , m. l. , nelson , e. , bryan , g. l. & motl , p. 2002 , apj , 579 , 571 mathews , w. g. 1989 , aj , 97 , 42 mathews , w. g. & brighenti , f. 2003 , araa , 41 , 191 mathews , w. g. 1997 , aj , 113 , 755 mathews , w. g. , brighenti , f. , buote , d. a. & lewis , a. d. 2003 , apj , 596 , 159 mathews , w. g. & brighenti , f. 2003 , apj 590 , 5 morris , r. g. & fabian , a. c. 2003 , mnras , 338 , 824 navarro , j. , frenk , c. & white , s. 1997 , apj 490 , 493 osullivan , e. , vrtilek , j. m. , read , a. m. , david , l. p. & ponman , t. j. 2003 , mnras , 346 , 525 pain , r. et al . 2002 , apj , 577 , 120 peterson , j. r. , paerels , f. b. s. , kaastra , j. s. , arnaud , m. , reiprich , t. h. , fabian , a. c. , mushotzky , r. f. , jernigan , j. g. & sakelliou , i. 2001 , a&a 365 , l104 rickes , m. g. , pastoriza , m. g. & bonatto , ch . 2004 , a&a , 419 , 449 sadler , e. m. , slee , o. b. , reynolds , j. e. & ekers , r. d. 1994 , _ the first stromlo symposium : the physics of active galaxies _ , asp conf . 54 , 335 sanders , j. s. & fabian , a. c. 2002 , mnras , 331 , 273 schmidt , r. w. , fabian a. c. , & sanders , j. s. 2002 , mnras , 337 , 71 smith , d. a. , wilson , a. s. , arnaud , k. a. , terashima , y. & young , a. j. 2002 , apj 565 , 195 stone , j. m. & norman , m. l. 1992 , apjs , 88 , 253 sun , m. et al . 2003 , apj , 598 , 250 sutherland , r. s. , & dopita , m. a. 1993 , apjs , 88 , 253 xu , h. et al . 2002 , apj 579 , 600 xue , y .- j . , bhringer , h. & matsushita , k. 2004 , a&a , ( in press ) ( astro - ph/0403026 ) . the dotted lines show the density and temperature profiles observed in ngc 5044 that serve as initial conditions for the computed profiles ; the dotted line pressure and entropy profiles are derived from the observations . the filling factors correspond to @xmath153 , 0.5 and 1.0 , all with @xmath56 kpc . in descending order from the top the panels show profiles for the gas entropy @xmath242 , the gas density @xmath243 ( solid lines ) and pressure @xmath244 ( dashed lines ) in units of @xmath245 dyne , the gas temperature @xmath246 , the gas velocity @xmath247 . and the flow filling factor @xmath52 . the low amplitude velocity profiles with @xmath153 and @xmath160 are fortuitously identical . , title="fig : " ] .7 in is concentrated near @xmath248 kpc , ( @xmath166 ) = ( 2,18,10,0 ) . the flow began at cosmic time @xmath162 gyrs and is shown at @xmath249 gyrs when severe cooling occurs . the dashed lines show the gas pressure and the iron abundance in solar units . the dotted lines show the observed @xmath238 and @xmath156 profiles for ngc 5044 . , title="fig : " ] .7 in is a broad function extending out to several hundred kpc , ( @xmath166 ) = ( 1,1.5,30,0 ) . the flow is shown at time @xmath250 gyrs when catastrophic cooling occurred at the inner radius . the dashed lines show the gas pressure and the iron abundance in solar units . the dotted lines show the observed @xmath238 and @xmath156 profiles for ngc 5044 . , title="fig : " ] .7 in that give good agreement at time @xmath221 gyrs with the observed iron abundance and temperature profiles observed in ngc 5044 . the dashed lines show the gas pressure and the iron abundance in solar units . heavy and light lines correspond to probabilities @xmath70 defined by ( @xmath166 ) = ( 1,1.1,41,1.9 ) and ( @xmath166 ) = ( 1,1.5,45,1.6 ) respectively . the dotted lines show the observed @xmath238 and @xmath156 profiles for ngc 5044 . , title="fig : " ] .7 in ) = ( 1,1.1,41,1.9 ) ( light lines ) and ( 1,1.5,45,1.6 ) ( heavy lines ) . shown are the mass inflow rate at @xmath51 @xmath191 in @xmath13 yr@xmath194 ( long dashed lines ) , the total stellar mass loss rate @xmath192 in @xmath13 yr@xmath194 in the central e galaxy ( short dashed lines ) , and the heating luminosity @xmath251 in units of @xmath193 ergs s@xmath194 ( solid lines ) . , title="fig : " ] .7 in

we present splitting functions in the triple collinear limit at next - to - leading order . the computation was performed in the context of massless qcd+qed , considering only processes which include at least one photon . through the comparison of the ir divergent structure of splitting amplitudes with the expected known behavior , we were able to check our results . besides that we implemented some consistency checks based on symmetry arguments and cross - checked the results among them . studying photon - started processes , we obtained very compact results . lpn14 - 099 + ific/14 - 54

1408.4821

studying the collinear limit of scattering amplitudes is a key point for improving theoretical computations . this is a requirement to compare theoretical predictions with the highly - accurate results provided by the lhc and other experiments at colliders . in this context , understanding the singular behavior of scattering amplitudes in the multiple - collinear limit is necessary to obtain higher order qcd corrections to several processes . it is a well known fact that , in certain general kinematical configurations , strict collinear factorization is fulfilled @xcite . these factorization properties establish that it is possible to isolate the singular behavior of scattering amplitudes into universal factors called splitting amplitudes @xcite . at the level of squared amplitudes , the altarelli - parisi ( ap ) kernels or splitting functions control the infrared ( ir ) singular behavior . moreover , splitting functions govern the evolution of parton distributions functions ( pdf ) and fragmentation functions ( ff ) through the dglap equation @xcite . also , they are the main ingredient in parton - shower generators and they are crucial for subtraction methods @xcite to compute physical cross sections at higher orders in perturbation theory . double - collinear splitting amplitudes and squared amplitudes have been computed at one - loop @xcite and two - loop level @xcite . splitting kernels beyond the double collinear limit , however , are known at the tree - level only @xcite , although some specific results for the triple collinear limit of one - loop amplitudes are already available @xcite . in this article , we present explicit results for unpolarized squared splitting amplitudes at next - to - leading order ( nlo ) in qcd for triple collinear processes involving at least one photon . in the context of dimensional regularization ( dreg ) @xcite , we work in conventional dimensional regularization ( cdr ) to make explicit the ir divergent structure , with @xmath0 the number of space - time dimensions . the corresponding squared splitting amplitudes in other schemes can be obtained by computing scalar - gluon contributions or by using transition rules if we are only interested in the @xmath1 terms . transition rules for scattering amplitudes were previously discussed in the literature @xcite , while we treated this topic for the double - collinear limit in ref . . we will not enter into this discussion in this paper . the outline of the paper is the following . in section [ sec : multiple ] we describe the multiple - collinear kinematics and introduce some relevant properties and notation . after some brief comments about qcd with photons , we review the double - collinear splitting functions for processes with photons ( both at the amplitude and squared - amplitude level ) . then , we present explicit results for nlo unpolarized splitting kernels with one and two photons in section [ sec : partons ] . to be more specific , we compute @xmath2 , @xmath3 and @xmath4 at nlo in qcd . in section [ sec : photons ] we show results for triple splitting kernels involving one photon which splits into three partons : @xmath5 and @xmath6 . in this section , we also present a discussion about the functional structure of these objects , using the photon - started splitting processes to handle more compact expressions . finally we present the conclusions and perspectives in section [ sec : conclusions ] . to explore the multiple collinear limit , we consider an @xmath7-particle process in which a subset of @xmath8 particles become simultaneously collinear . we label the collinear particles with @xmath9 . if @xmath10 is the four - momentum associated with particle @xmath11 , and assuming @xmath12 ( massless on - shell partons ) , then the subenergies @xmath13 and @xmath14 , with @xmath11 and @xmath15 , are the only dimensional relevant quantities involved in the collinear limit . we define a pair of light - like vectors ( @xmath16 , @xmath17 ) , such that ^ = p_1,m^ - n^ [ ptilde ] approaches the collinear direction in the multiparton collinear limit , and @xmath18 parametrizes how the collinear limit is approached , with @xmath19 . the longitudinal - momentum fractions @xmath20 of the collinear partons are z_i = , i c , and fulfil the constraint @xmath21 . we work in the physical - gauge where the factorization properties are explicit , with d _ ( k , n ) = - _ + [ bosonpol ] the physical polarization tensor of a gauge vector boson ( gluon or photon ) with momentum @xmath22 . the auxiliary gauge vector @xmath7 in is taken identical to the light - like vector @xmath7 introduced in . on the other hand , let s recall some universal properties of the collinear limit of scattering amplitudes . strict collinear factorization is fulfilled to all orders in the time - like region ( i.e. where @xmath23 for all @xmath24 ) . for simplicity , we limit ourselves to this region for the rest of the paper and leave the computation of factorization breaking effects in the space - like region @xcite for a future publication . when particles @xmath25 to @xmath8 become collinear , the one - loop matrix element @xmath26 factorizes according to & & ^(1)_a_1 a_m(p_1 , , p_m ; ) + & + & ^(0)_a_1 a_m(p_1 , , p_m ; ) , which is valid for the most singular terms in the collinear limit . the factors @xmath27 and @xmath28 , also called splitting matrices , encode the singular behavior in the multiple collinear limit at tree - level and one - loop , respectively . as discussed in ref . @xcite , they turn out to be process - independent in the time - like region as they depend on the momenta and quantum numbers ( flavour , spin , and color ) of the collinear partons only . centering in @xmath28 , it can be expressed as ^(1)_a_1 a_m & = & ^(1)div._a_1 a_m + ^(1)fin._a_1 a_m , [ descomposicionsp ] where all the infrared divergences are contained in @xmath29 , which can be expanded as ^(1)div._a_1 a_m ( p_1, ,p_m ; ) = * _ i_*^(1)_a_1 a_m(p_1, ,p_m ; ) ^(0)_a_1 a_m(p_1 , , p_m ; ) , [ descomposicionsp2 ] with * _ i_*^(1)_a_1 a_m(p_1, ,p_m ; ) & = & c _ g_^2 ( ) ^- \ { _ i , j=1 ( i j)^|m _ i _ j ( ) ^- + & + & _ i , j=1^|m _ i _ j ( 2 - ( z_i ) ^- - ( z_j ) ^- ) + & - & ( _ i=1^|m ( _ i - _ i^ ) - ( _ a - _ a^ ) - ( _ 0 - _ 0^ ) ) } , [ sp1div ] where the color matrix of the collinear particle with momentum @xmath30 is denoted by @xmath31 , @xmath32 counts the number of collinear final state qcd partons and @xmath33 refers to the total number of qcd partons in the splitting process . we also define c_= , as the usual one - loop @xmath34-dimensional volume factor . we can appreciate that @xmath35 in collinear splittings which are started by non - qcd partons . we order the final state particles such that @xmath36 are the colored ones while the remaining ones are singlets under @xmath37 transformations . the flavour coefficients @xmath38 and @xmath39 are @xmath40 and @xmath41 . up to @xmath1 , the regularization scheme ( rs ) dependence is controlled by the coefficients @xmath42 and @xmath43 . explicitly , we have _ i^=_0^=0 , in conventional dimensional regularization ( cdr ) , while _ q^&=&_|q^=c_f/2 , + _ g^&=&_0^/2=c_a/6 , in dimensional reduction ( dr ) . moreover , we consider unrenormalized quantities , so @xmath44 is the bare qcd coupling . in the curly bracket of , the contribution proportional to @xmath45 is of ultraviolet origin , and it will be removed by working at the level of renormalized matrix elements and splitting matrices . the factor @xmath46 in is a matrix in the color space . however , since color conservation implies that the color charge of the parent parton is given by _ i t_i ^(0)_a_1 a_m = ^(0)_a_1 a_m _ a , and since we consider processes with at most @xmath47 qcd partons , the color algebra can be carried out in closed form and @xmath46 becomes proportional to the unit matrix , namely , it becomes a @xmath48-number that we denote @xmath49 . the square of the splitting matrix @xmath50 , summed over final - state colors and spins and averaged over colors and spins of the parent parton , defines the @xmath8-parton unpolarized splitting function @xmath51 , which is a generalization of the customary altarelli - parisi double collinear splitting function . fixing the normalization of the tree - level splitting function by [ p0def ] _ a_1 a_m^(0 ) = ( ) ^m-1 , then it is possible to use and to present the nlo correction as _ a_m^(1 ) & = & ( ) ^m-1 ( + h.c . ) + & = & 2 re ( i^(1)_a_1 a_m(p_1, ,p_m ; ) ) _ a_1 a_m^(0 ) + ( ^(1)fin._a_1 a_m + c.c . ) , [ equaciondescomposicion1 ] with the aim of exposing explicitly the divergent structure of the splitting function . then , working with the finite remainder , we classify each term according to its transcendental weight . so , the second step of the decomposition that we carried out is to express the finite contribution to the nlo unpolarized splitting kernel as ^(1)fin._a_1 a_m & = & c^a_1 a_m , [ equaciondescomposicion2 ] where @xmath52 denotes a basis of a function s space with transcendental weight @xmath11 and @xmath53 are the corresponding coefficients . here @xmath54 is a normalization factor which depends on the process and includes all the couplings . in this section we present explicit results for the nlo corrections to triple collinear splitting functions involving photons . we describe the corresponding model through the lagrangian density given by _ qcd+qed&= & l_qcd - ^ g_e _ q e_q _ ij |^i_q^^j_q a _ - f^f _ , where @xmath55 are color indices , @xmath56 is the electromagnetic coupling ( i.e. the absolute value of electron charge ) , @xmath57 is the charge of quark s flavour @xmath58 ( @xmath59 and @xmath60 ) and @xmath61 is the gauge - field strength tensor for the abelian group @xmath62 . the photon - quark interaction is proportional to the identity matrix @xmath63 in the color space . besides that , the computation is performed in the time - like region . with this kinematic choice , we can ensure strict factorization properties ( as claimed in ref . @xcite ) . centering in the triple collinear limit and using the notation previously introduced , we have @xmath64 but @xmath32 is not fixed . since we are interested in qcd corrections , we consider first splitting processes where the parent parton is a qcd parton . for @xmath65 , we have the splitting @xmath66 with @xmath67 a quark or gluon . since @xmath68 , the one - loop correction @xmath69 is ultraviolet finite and the only allowed unpolarized splitting function at nlo with @xmath65 is @xmath70 . when @xmath71 , with a photon in the final state , there are more possibilities . the relevant splittings functions are @xmath72 , @xmath73 , and @xmath74 where @xmath75 . finally , splitting processes initiated by photons will be discussed in section [ sec : photons ] . as mentioned in the introduction , the computation is performed using cdr scheme in order to simplify the treatment of intermediate expressions . it is possible to obtain dr scheme results up to @xmath1 by replacing the flavour coefficients and @xmath43 in . also , we could obtain the exact expressions in other schemes by computing the corresponding scalar - gluon contributions @xcite , although in this work we are only interested in @xmath1 corrections . we start showing results for the @xmath2 splitting function at nlo . then , we express @xmath3 in terms of the previous one , in order to simplify the results . finally , we compute @xmath4 using the techniques mentioned above . before starting with the triple - collinear splitting functions , let s show some results for the double - collinear limit of scattering amplitudes with photons . this topic has been analysed in ref . @xcite where we also discussed some technical details related with dreg schemes . in the context of qcd+qed , we have only two double - collinear splitting processes : @xmath76 and @xmath77 . and @xmath78 . , scaledwidth=75.0% ] the corresponding feynman diagrams required to perform the computation are shown in fig . [ fig : diagramas_doblephotones ] . using them , we obtained both splitting amplitudes and unpolarized splitting functions up to nlo accuracy . the results for the lo contributions are ^(0)_q_1 _ 2 & = & * i d * ^(0)_q_1 |q_2 & = & * i d * |u(p_1)()v(p_2 ) , and ^(1)_q_1 _ 2 & = & c_g_^2 c_f ( ) ^- , + ^(1)_q_1 |q_2 & = & c_g_^2 c_f ( ) ^- ^(0)_q_1 |q_2 , for the qcd nlo corrections in the time - like region and using cdr scheme with @xmath0 . it is interesting to notice that @xmath77 only involves rational functions ( aside from the global prefactor ) and is extremely simple . in the last part of this article , after presenting the full nlo corrections to triple - splitting functions started by photons , we will discuss the origin of these compact expressions , but we can anticipate that it is related with the kind of integrals involved in the computation . in the double collinear limit , it is difficult to appreciate this fact because there is only one kinematical scale involved ( @xmath79 ) . in any case , since it is a three - particle process , computations can require triangle integrals at most . but only triangles with a lcg propagator can introduce hypergeometric functions which depend on @xmath80 ; otherwise the integral is completely independent of the momentum fraction . moreover , the presence of momentum fractions inside non - rational functions is a consequence of having lcg propagators inside the loop integrals . since this kind of functions is absent in @xmath81 , then all lcg propagators inside the loop should cancel . as a partial summary of the situation , there is a connection between gauge constraints and having on - shell qcd particles ( charged under a _ non - abelian _ group ) and just an off - shell parent photon ( which is an _ abelian _ field ) . for the sake of completeness , we show the corresponding splitting functions for the double - collinear limit with photons . at lo we have ^(0)_q_1 _ 2 & = & g_e^2 e_q^2 = g_e^2 e_q^2 _ q_1 _ 2^(0 ) , + ^(0)_q_1 |q_2 & = & g_e^2 e_q^2 = _ q_1 |q_2^(0 ) , where @xmath82 , with the parameter @xmath83 used to switch between fdh / hv ( @xmath84 ) and cdr ( @xmath85 ) schemes . choosing cdr scheme , the nlo corrections can be written as ^(1)_q_1 _ 2 & = & c_g_^2 c_f ( ) ^- + c.c . , + ^(1)_q_1 |q_2 & = & c_g_^2 c_f ( ) ^- ^(0)_q_1 |q_2 + c.c . , where we adapted the original notation of @xcite in order to be compatible with the one used in this article . let s consider the process @xmath2 . in terms of color structure , this is the simplest case because it is proportional to the identity matrix @xmath63 . the corresponding splitting amplitude at tree - level is given by sp^(0)(a_1;a)_q_1 _ 2 _ 3 & = & |u(p_1 ) ( ( p_3 ) _ 3 + 2 p_1 ( p_3 ) ) ( p_2 ) u ( ) + ( 23 ) . [ splittingloq - qaa ] splitting amplitude . this splitting amplitude is symmetric under the exchange of the two final state photons . ] in order to reduce the output size and to show explicitly that the splitting function is dimensionless , we use the notation x_i & = & , with @xmath86 a reordering of the indices set @xmath87 and the special case @xmath88 . moreover , we define _ i,j^i , j & & x_i z_j + x_i ( z_j-1 ) , + outgoing particles , and @xmath89 . using this notation , it is possible to write the unpolarized splitting function at lo as ^(0)_q_1 _ 2 _ 3 & = & ( ( 1 + ) + ( -1 ) ( ( x_1-+ ) .. + & + & .. x_3+z_1-z_2 + 2 ) ) + ( 2 3 ) . going to the nlo correction , we use the decomposition suggested in and . the factor predicted by catani s formula is i^(1)_q_1 _ 2 _ 3 ( p_1 , p_2 , p_3 ; ) & = & 2 c _ g_^2 c_f ( ) ^- , and we found agreement with the divergent contribution in our results . the rational coefficient is given by c^(0 ) & = & + + + & + & + + + & + & ( + ) - + + ( 2 3 ) , with the global prefactor c^q & = & e_q^4 g_e^4 g_^2 c_f . at transcendental weight 1 , the basis is composed by f^(1)_1 & = & ( x_1 ) , + f^(1)_2 & = & ( x_3 ) . it is important to notice that these logarithms depend on ratios of kinematical variables , which ensures that this contribution can be integrated if we need to compute @xmath90lo corrections . due to symmetry considerations , we can represent the results as ^(1)fin._q_1 _ 2 _ 3 . |_w=1 & = & c^q , where c^(1)_1 & = & + + & + & , + c^(1)_2 & = & ( + + - . + & - & . ( + ) -4 ) + & + & ( 2 ^2,0_0,2-+(- ) . --2 ) + & + & ( - ) . the splitting is fully symmetric under the exchange of particles 2 and 3 , then let s exploit this symmetry to minimize the basis . at weight 2 : ^(1)fin._q_1 _ 2 _ 3 . |_w=2 & = & c^q , where the corresponding basis is f^(2)_1 & = & ^2 ( z_1 ) , + f^(2)_2 & = & - , + f^(2)_3 & = & ( x_1)(z_1 ) , + f^(2)_4 & = & + ( x_1)(x_2 ) , + f^(2)_5 & = & r(x_1,x_2 ) + 2 + 2(x_1)(x_2 ) , + f^(2)_6 & = & --- + & - & ( x_2 ) ( ) , + f^(2)_7 & = & + ( ) ( ) + ( x_1 ) ( z_1 ) , with @xmath91 . the first three components of the basis are symmetric under the exchange of particles 2 and 3 . on the other hand , in the previous list we introduced the function ( x_1,x_2 ) & = & -(x_1 ) ( x_2 ) - - , which is going to appear in all the remaining splitting functions , and whose origin will be explained in the last part of this article . also , we can appreciate that @xmath92 when @xmath93 ( @xmath94 ) . following with the presentation of the results , the coefficients associated with the previous basis are given by c^(2)_1 & = & , + c^(2)_2 & = & -- + & + & + + , + c^(2)_3 & = & - , for the explicitly symmetric contribution , and c^(2)_4 & = & - , + c^(2)_5 & = & -+ + & - & + + & - & + + -4 , + c^(2)_6 & = & - , + c^(2)_7 & = & ( + ) + & -&+ + & + & , for the others . the natural following step is to replace one photon by a gluon . this will produce more complicated expressions , but it is possible to relate them with the @xmath2 splitting . in order to do that , let s take a look at fig . [ fig : diagramas_q - qga ] . at lo , we have exactly the same kinematic structure , which implies that sp^(0)(a_1,_2;a)_q_1 g_2 _ 3 & = & |u(p_1 ) u ( ) + & = & _ * t*_^_2_a_1 a sp^(0)(a_1;a)_q_1 _ 2 _ 3 , [ splittingloq - qga ] and the corresponding unpolarized lo splitting function is simply _ q_1 g_2 _ 3^(0 ) & = & c_f _ q_1 _ 2 _ 3^(0 ) , in terms of the one associated to @xmath2 . splitting amplitude . the first 5 diagrams are associated with the _ abelian _ contribution already presented in @xmath95 . the last 2 diagrams lead to _ non - abelian _ ( na ) contributions.,scaledwidth=80.0% ] to obtain the nlo corrections , we use the decomposition suggested in and . the factor predicted by catani s formula is i^(1)_q_1 g_2 _ 3 ( p_1 , p_2 , p_3 ; ) & = & ( ) ^- + & = & i^(1)_q_1 _ 2 _ 3 ( p_1,p_2,p_3 ; ) + c_a i^(1,c_a)_q_1 g_2 _ 3 ( p_1 , p_2 , p_3 ; ) , where i^(1,c_a)_q_1 g_2 _ 3 ( p_1 , p_2 , p_3 ; ) = ( ) ^- ( 1-x_3 ^ - -z_2 ^ - ) , with @xmath96 and we found agreement with the divergent contribution in our results . before writing explicit results , let s take a look to the corresponding nlo feynman diagrams contributing to the splitting amplitude . if we just consider the topology , we can appreciate that many diagrams are already present in fig . [ fig : diagramas_q - q2a ] . since a photon has been replaced by a gluon , there is an extra color matrix . but in these diagrams , the kinematic structure is unchanged and only some color factors need to be modified . at amplitude level , in @xmath2 there was a global color factor @xmath97 related with a gluon of color @xmath98 inside the loop . however , in @xmath3 we must insert the color matrix @xmath99 associated with the external gluon , which leads to two possibilities : * @xmath100 if the external gluon is attached to a fermion line inside the loop , or ; * @xmath101 when the external gluon interacts with an external fermion ( i.e. a fermionic line which is not inside a loop ) . it is important to notice that both configurations include a contribution proportional to @xmath102 . on the other hand , there are some diagrams which include a triple - gluon vertex in fig . [ fig : diagramas_q - qga ] whose topology is different from the ones associated with the @xmath2 process . in summary , we conclude that it is possible to express the @xmath3 splitting function in terms of @xmath2 . to make this relation explicit at nlo , we write _ 3^(1 ) & = & d_a ^(1,d_a)_q_1 g_2 _ 3 + c_a ^(1,c_a)_q_1 g_2 _ 3 + c.c . , with ^(1,d_a)_q_1 g_2 _ 3 & = & , + ^(1,c_a)_q_1 g_2 _ 3 & = & re ( i^(1,c_a)_q_1 g_2 _ 3 ( p_1 , p_2 , p_3 ; ) ) ^(0)_q_1 g_2 _ 3 + ^(1,c_a)fin._q_1 g_2 _ 3 , where the global prefactor is given by c^q g & = & e_q^2 g_e^2 g_^4 c_f . centering in the finite nlo contribution , we can write the corrections proportional to @xmath103 using as ^(1,c_a)fin._q_1 g_2 _ 3 & = & c^q g . after these considerations , let s show the explicit results . in first place , the rational coefficient is given by c^(0,c_a ) & = & + + & - & - ( 2 3 ) , which is totally antisymmetric when interchanging particles 2 and 3 . since the rational contribution for the @xmath2 kernel is totally symmetric , it is clear that @xmath104 is related with the na diagrams associated with the non - abelian triple - vertex . at weight 1 , the basis is composed by f^(1,c_a ) & = & ( x_3 ) , which only depend on ratios of kinematical variables . due to symmetry considerations , we can rewrite this contribution as ^(1,c_a)fin._q_1 g_2 _ 3 . |_w=1 & = & c^q g , where c^(1,c_a)_sym & = & + + & + & + + , + c^(1,c_a)_asym & = & ( -x_2 - 1 ) + + & - & + + & + & + & + & + & - & + & + & . finally , if we go to transcendental weight 2 , it is possible to perform the expansion ^(1,c_a)fin._q_1 |_w=2 & = & c^q g ^7_i = 1 c^(2,c_a)_i f^(2,c_a)_i , using the set of functions f^(2,c_a)_1 & = & + ( 1-z_3 ) ( ) + ( x_3 ) ( ) , + f^(2,c_a)_2 & = & - , + f^(2,c_a)_3 & = & -+ ( ) ( ) , + f^(2,c_a)_4 & = & --+(x_2 ) ( z_2 ) , + f^(2,c_a)_5 & = & 2- + & + & ( x_2 ) ( 1-z_3)+ ( ) ( ) , + f^(2,c_a)_6 & = & -- , + f^(2,c_a)_7 & = & r(x_2,x_3 ) , where the associated coefficients are given by c^(2,c_a)_1 & = & ( --+ ) + & + & + , + c^(2,c_a)_2 & = & ( + + + ) + & + & --8 , + c^(2,c_a)_3 & = & ( + + -+ ) + & - & -- , + c^(2,c_a)_4 & = & ( ( 1-z_1 ) ( z_3-z_2)-)+ + & - & + , + c^(2,c_a)_5 & = & + ( 2 3 ) , c^(2,c_a)_6 & = & -(+)+ + & - & + , + c^(2,c_a)_7 & = & ( + ) + & - & ( - ) + + & -&+ + & + & - . note that these coefficients mix both symmetric and antisymmetric contributions , as happened for the component of transcendental weight 1 . finally , we arrive to the last available configuration in the triple collinear limit with photons . starting with the leading order , the splitting amplitude associated with the process @xmath4 reads sp^(0)(a_1,a_2;)_q_1 |q_2 _ 3 & = & |u(p_1 ) ( - ) v(p_2 ) , [ splittinglog - qqba ] while the unpolarized splitting function is given by _ 3^(0 ) & = & + & + & ( 12 ) , with @xmath82 , as defined for @xmath2 splitting . we can appreciate that this expression is totally symmetric when interchanging particles 1 and 2 . at amplitude level , there is an additional minus sign coming from the change @xmath105 . on the other hand , it is useful to define ^(0)_q_1 |q_2 _ 3 .|_^0 & = & + ( 12 ) = . |_=0 , because this expression will allow to simplify nlo results . splitting amplitude.,scaledwidth=80.0% ] before presenting the results , let s compare this process with @xmath3 . the particles involved are the same , but we exchange the parent quark with a final state gluon . the corresponding feynman diagrams are shown in fig . [ fig : diagramas_g - qqba ] . excluding self - energy corrections to the parent parton , the other diagrams are in one - to - one correspondence in both processes . moreover , both sets of diagrams are related by the exchange @xmath106 . but the parent parton is an off - shell particle , and there are differences if process starts with a gluon or a quark . so , we do not expect to have any crossing transformation relating @xmath107 and @xmath108 beyond tree - level . we will show a conclusive counterexample in the next section . the ir divergent structure of this splitting is given by i^(1)_q_1 |q_2 _ 3 ( p_1 , p_2 , p_3 ; ) & = & ( ) ^- , according to catani s formula and it agrees with our results . to express the results in a compact way , we decompose it according to the color structure . so , the finite contribution to the unpolarized splitting function is expanded as _ q_1 |q_2 _ 3^(1)fin . & = & c^q |q , with the global prefactor c^q |q & = & , and ^(1,n_f)fin._q_1 |q_2 _ 3 & = & c^(0,n_f ) , + ^(1,d_a)fin._q_1 |q_2 _ 3 & = & c^(0,d_a ) + ^2_i=1 c_i^(1,d_a ) f_i^(1 ) + c_1^(2,d_a ) f_1^(2,d_a ) , + ^(1,c_a)fin._q_1 |q_2 _ 3 & = & c^(0,c_a ) + c_1^(1,c_a ) f_1^(1 ) + _ j=1 ^ 4c_j^(2,c_a ) f_j^(2,c_a ) , where we classify each contribution according to its transcendental weight . we can appreciate that the part proportional to @xmath109 is purely rational . since it is originated from diagrams which contain fermionic loops , only _ standard _ bubbles ( i.e. without the extra lcg propagator ) can contribute to @xmath109 terms . but these bubbles have single @xmath110 poles which are completely absorbed into @xmath111 , including also the corresponding logarithms . so , only the rational terms proportional to @xmath109 survive after the subtraction procedure . now let s show the explicit results . starting with the rational terms , we have c^(0,n_f ) & = & - ^(0)_q_1 |q_2 _ 3 .|_^0 , + c^(0,c_a ) & = & ^(0)_q_1 |q_2 _ 3 .|_^0 + ( + ) + & + & , + c^(0,d_a ) & = & ( x_1 x_2-z_1 z_2-^1,0_0,1 ^ 2,0_0,2)(- ) + & - & ( + + + 8 ) + & - & - , where we can appreciate that the expressions are symmetric under the transformation @xmath112 . the basis for the contribution of transcendental weight 1 is given by f^(1)_1 & = & ( x_1 ) , + f^(1)_2 & = & ( x_3 ) , and the associated coefficients are c^(1,d_a)_1 & = & ( --)- + & - & + , c^(1,d_a)_2 & = & + & - & + & - & , for the abelian part , and c^(1,c_a)_1 & = & + & + & , for the @xmath103 terms . finally , if we go to transcendental weight 2 , we can write the term proportional to @xmath113 making use of f^(2,d_a)_1 & = & r(x_1,x_3 ) , + c^(2,d_a)_1 & = & + & - & , while for the component proportional to @xmath103 we require f^(2,c_a)_1 & = & -2 -2 + 2 + & + & 2 ( x_2 ) ( 1-z_1 ) + ( 12 ) , + f^(2,c_a)_2 & = & ( x_1 ) ( x_2 ) , + f^(2,c_a)_3 & = & --(x_1 ) ( z_1 ) + ( 12 ) , + f^(2,c_a)_4 & = & ( ) ( ) -(x_2 ) ( 1-z_1 ) , with the associated coefficients c^(2,c_a)_1 & = & , + c^(2,c_a)_2 & = & + , c^(2,c_a)_3 & = & 2- , + c^(2,c_a)_4 & = & + + + & + & + + & + & . it is interesting to notice that these expressions are very compact , especially if we compare them with the analogous expansion for @xmath2 and @xmath3 . this fact is related with the presence of a initial - state vector particle , as we discuss in the next section . finally , we also considered nlo qcd corrections to splittings functions started by photons . these objects might be relevant to describe the decay of a virtual photon into three on - shell massless particles up to nlo accuracy in qcd . following the discussion presented in the previous sections , we consider the two photon - started triple splitting processes : @xmath5 and @xmath6 . moreover , let s recall that @xmath114 vanish at tree - level for all @xmath115 due to decoupling identities , which implies that @xmath114 must be finite beyond lo . besides presenting the results , we take advantage of their simplicity in order to extract some interesting conclusions about their functional structure . beginning with the lo splitting amplitude , we have sp^(0)(a_1,a_2)_q_1 possible to appreciate that this amplitude is just the color stripped contribution to @xmath116 . explicitly , sp^(0)(a_1,a_2;)_q_1 |q_2 _ 3 & = & _ * t*_^_a_1 a_2 sp^(0)(a_1,a_2)_q_1 |q_2 _ 3 , which trivially implies that the unpolarized splitting function is given by _ q_1 |q_2 _ 3^(0 ) & = & _ q_1 symmetry under the exchange @xmath112 is obviously inherited . splitting amplitude . we neglect self - energy corrections to the parent photon because they contribute at a higher qed order.,scaledwidth=80.0% ] when we considered @xmath4 in section [ sec : partons ] , we arranged the different contributions to the nlo splitting function in order to identify abelian and non - abelian terms . we saw that purely abelian terms were given by compact expressions which can be written in terms of logarithms and just one weight 2 function . since @xmath5 is proportional to the abelian part , its splitting function is extremely simple . in fig . [ fig : diagramas_a - qqba ] we show all the feynman diagrams required in the computation . notice that self - energy corrections to the incoming photon are not taken into account because they contribute to higher - orders in qed . following the same recipe for the other processes , we start studying the ir divergent structure . according to catani s formula , we have i^(1)_q_1 complete agreement with our results . on the other hand , after performing the full computation we notice that the nlo splitting function can be expressed as _ q_1 |q_2 _ 3^(1)fin . & = & 2 c_f c^q |q , as it was expected based in a naive feynman diagram comparison between these processes . here we introduced the global prefactor c^q |q & = & , in order to simplify the notation . to conclude this section , let s make a comment about crossing identities for splitting functions ( and amplitudes ) . in section [ sec : partons ] , we presented @xmath2 and the simplified weight 2 contribution to the nlo correction could be expressed in terms of 11 functions . however @xmath117 is proportional to just one of these functions . since crossing relations are ( rational ) variable transformations , they can not modify the dimension of the space of functions which span the transcendental weight 2 contribution . we can explain this situation reasoning as follows . to obtain @xmath5 we need to perform the exchange @xmath118 . this implies changing the parent parton , which is a _ distinguished _ particle from a kinematical point of view ( i.e. it is slightly off - shell and this fact involves that self - energy corrections can not be neglected ) . to be more explicit , in @xmath2 the parent quark is off - shell and outgoing photons are on - shell , but in @xmath5 we have one off - shell photon and the quark that has been moved to the final state loses its virtuality . so , strictly speaking , kinematics forces us to consider that the particle content of these splittings is not the same . finally , we arrive to the last non - trivial triple splitting process with photons : @xmath6 . the corresponding feynman diagrams at nlo are shown in fig . [ fig : diagramas_a - qqbg ] . splitting amplitude.,scaledwidth=66.0% ] the lo splitting amplitude can be written as sp^(0)(a_1,a_2,_3)_q_1 |q_2 g_3 & = & |u(p_1 ) ( - ) v(p_2 ) + & = & _ * t*_^_3_a_1 a_2 sp^(0)(a_1,a_2)_q_1 |q_2 _ 3 , [ splittingloa - qqbg ] where we used previous results to obtain the last line . in fact , using that relation with the @xmath5 splitting , the corresponding splitting function at lo turns out to be _ q_1 |q_2 g_3^(0 ) & = & _ q_1 |q_2 _ 3^(0 ) = 2 c_a c_f _ 3^(0 ) , which can easily be related with different splitting functions interchanging photons and gluons . note that we are not changing kinematics but only the color structure . in order to express results for the nlo correction , we follow the same strategy used with @xmath4 . since there are contributions proportional to both @xmath103 and @xmath119 , we treat them separately . the contribution proportional to @xmath113 is just the one coming from @xmath120 ( with a factor 2 of difference due to the definition of @xmath113 ) . to check the global ir divergent structure , we extract the @xmath110 poles from the splitting function and compare them with the expression provided by catani s formula , i.e. i^(1)_q_1 |q_2 g_3 ( p_1 , p_2 , p_3 ; ) & = & i^(1)_q_1 |q_2 _ 3 ( p_1 , p_2 , p_3 ; ) + & + & ( ) ^- ( x_3 ^ --x_1 ^ --x_2 ^ - ) , finding a complete agreement . note that we used the operator @xmath121 from to simplify the results . moreover , following this idea , we write the nlo correction to the splitting function as _ q_1 |q_2 g_3^(1 ) & = & d_a ^(1,d_a)_q_1 |q_2 _ 3 + c_a ^(1,c_a)_q_1 |q_2 g_3 + , with ^(1,d_a)_q_1 |q_2 g_3 & = & + & = & , [ ecuacionp1daa - qqbarg1 ] where we introduced the global prefactor c^q |q g & = & . to treat the finite contribution , we express it as _ q_1 given by _ q_1 |q_2 g_3^(1,d_a)fin . & = & ^(1)fin._q_1 |q_2 _ 3 = c^q |q g _ q_1 |q_2 _ 3^(1,d_a)fin . . the novel contributions originated in the non - abelian part are contained inside @xmath122 . classifying these terms according to its transcendental weight , we get ^(1,c_a)fin._q_1 |q_2 g_3 & = & c^(0,c_a ) + c^(1,c_a ) f^(1 ) + c^(2,c_a ) f^(2,c_a ) + ( 1 2 ) , [ peso2_aqqbg ] where the rational coefficient is given by c^(0,c_a ) & = & --8 + & + & , and f^(1 ) & = & , + f^(2,c_a ) & = & r(x_1,x_2 ) , are the functions which expands the spaces of transcendentality 1 and 2 , respectively . it is crucial to appreciate that @xmath123 is the same function involved in the @xmath5 ( aside from a permutation of the kinematical variables ) . we will return to this point in the next subsection . finally , in order to get the full nlo correction to this splitting function , the coefficients listed in are given by c^(1,c_a ) & = & + & + & - , + c^(2,c_a ) & = & ^(0)_q_1 |q_2 _ 3 .|_^0 . again , it is possible to appreciate that this splitting can not easily be related to @xmath4 . in particular , in section [ sec : partons ] we showed that 7 functions were required to expand the component of transcendental weight 2 for @xmath124 . however , @xmath125 involves a @xmath126-dimensional transcendental weight 2 space . since the results shown in this section are very simple ( compared with those presented in section [ sec : partons ] ) , we can go further and discuss about the structure of triple splittings started by photons . let s start with @xmath5 . if we perform a direct computation without making an @xmath110-expansion , the nlo contribution to the splitting function can be expressed as _ 3^(1 ) & = & a_1^(4)(z_j , x_k ; ) i_1^(4 ) + _ i=1 ^ 3 a^(2)_i(z_j , x_k ; ) i_i^(2 ) + ( 1 2 ) , where coupling constants and color factors are absorbed inside the coefficients @xmath127 . here @xmath128 and @xmath129 are rational functions of the kinematical variables @xmath130 and @xmath110 . branch - cuts are defined by the feynman integrals @xmath131 ( box ) and @xmath132 ( bubbles ) . moreover , using @xmath0 , the explicit expression for the box integral is i^(4)_1 & = & _ q = ( ) ^- + & & , while bubbles are given by i^(2)_1 & = & _ q = ( ) ^- , + i^(2)_2 & = & _ q = ( ) ^- , + i^(2)_3 & = & _ q = ( ) ^- , where we defined _ q = - , in order to simplify the notation . we can appreciate that these integrals are known to all orders in @xmath110 which implies that the same can be said about @xmath120 . if we study the integrals involved in nlo corrections to @xmath6 , we realize that an analogous situation takes place . in other words , it turns out to be possible to write _ g_3^(1 ) & = & a_1^(4)(z_j , x_k ; ) i_1^(4 ) + a_2^(4)(z_j , x_k ; ) i_2^(4 ) + & + & _ i=1 ^ 3 a^(2)_i(z_j , x_k ; ) i_i^(2 ) + ( 1 2 ) , where i^(4)_2 & = & _ q = ( ) ^- + & & = . i^(4)_1 |_2 3 , is just a standard one - mass box integral . the corresponding coefficients can be computed to all orders in @xmath110 , but their explicit expressions are a bit cumbersome . after motivating the results for @xmath120 and @xmath133 , it is interesting to appreciate that i^(4)_1 = ( ) ^- , which explains the origin of the function @xmath134 . in other words , it originates from the hypergeometric functions involved in the _ standard _ scalar boxes ( extracting squared logarithms introduced by _ standard _ bubbles ) . but the important fact is that this one appears in all triple - collinear splitting functions with photons and also in the antisymmetric part of @xmath135 ( computed in ref . so , the presence of this function is due to purely kinematical reasons . another interesting fact is that to span triple collinear splitting functions started by photons , we only used boxes ( maximal topological complexity ) and bubbles ( minimal non - trivial topology ) : triangles do not contribute to this computation . there is a last point which deserves to be discussed . we have seen that splittings started by gluons are simpler than quark - started ones . in same sense this sounds reasonable because spinors only satisfy dirac s equation , while physical gluons should verify transversality and gauge invariance . explicitly , when we compute splitting amplitudes we start with an off - shell amputated amplitude @xmath136 and then we project it over an on - shell spinor or polarization vector , @xmath137 or @xmath138 respectively . but spinors satisfy u ( ) = 0 , while on - shell massless polarization vectors verify ( ) = 0 , + n ( ) = 0 , where @xmath7 is a null - vector which defines the light - cone gauge . on the other hand , we can write _ q a_1 a_m & = & a^(1,i)_(amp , q ) . |_lcg u_i ( ) , [ splittinggeneralquarkstarted ] + _ g a_1 a_m & = & a^(1,)_(amp , g ) . ( ) , [ splittinggeneralgluonstarted ] for quark and gluon - started splitting amplitude , respectively . so , roughly speaking , we are imposing 2 restrictions on @xmath139 and just one on @xmath140 . we would like to understand also which is the difference between gluon and photon - started splitting functions . the last ones are extremely compact and , moreover , they do not involve integrals with lcg propagators . if we compare the feynman diagrams required for @xmath5 and @xmath4 , we can see that non - abelian interactions vanish completely in the first case . moreover , changing the coupling constant and removing the global color factor , nlo qcd corrections to @xmath5 are exactly the same that the corresponding qed nlo corrections . since qed is an abelian theory , we can use a covariant gauge to compute virtual corrections and the result remains unchanged . however the same explanation can not be directly applied to @xmath6 because of na diagrams , although the central idea is related with gauge invariance . note that splitting functions are computed using a fixed physical - gauge ( lcg in particular ) in order to simplify factorization properties in the collinear limit . if we restrict the analysis to the computational procedure , we find out that we are computing nlo qcd corrections to an amputated scattering amplitude with the incoming particle slightly _ off - shell _ , and then making a projection over an _ on - shell _ spinor / polarization vector . when the process is started by a qcd parton , there is a color flux across the off - shell parent leg . if the parent parton is a photon then _ i=1^|m _ * t*_^i = 0 , because photons are color singlets . this implies that we can attach the amputated amplitude to a pair of colorless fermions ( for instance , an electron - positron pair ) and reconstruct the full nlo qcd correction to the _ on - shell _ physical scattering amplitude . since physical scattering amplitudes are gauge invariant , we can use covariant gauge and only _ standard _ scalar integrals are required in the computation . in other words , we can write ^(1)_e^- e^+ a_1 a_2 a_3(k_1,k_2;p_1,p_2,p_3 ) . |_lcg & = & |v(k_2)(-g_e ^)u(k_1 ) a^(1,)_(amp , ) . |_lcg + & = & -g_e c_pol , [ ecuacionprueba1 ] where we are using physical momenta for the process @xmath141 ( @xmath142 could be a qcd parton ) , which implies that @xmath143 ( massless fermions on - shell ) and @xmath144 ( vector equation ) . here @xmath145 is an arbitrary normalization factor which only depends on the kinematics . due to gauge invariance , ^(1)_e^- e^+ a_1 a_2 a_3(k_1,k_2;p_1,p_2,p_3 ) = a^(1)_e^- e^+ a_1 a_2 a_3(k_1,k_2;p_1,p_2,p_3 ) . |_lcg , which implies a^(1,)_(amp , ) . |_lcg & = & a^(1,)_(amp , ) . this relation is true for every @xmath146 which fulfils physical conditions . in particular , we might use k^_1=^ , k^_2= n^ , or in the inverted order . however , if we restrict external polarizations to be physical and use one of the previously mentioned choices , then = ^_(,n ) , where we are applying the well - known mapping of polarization vectors into spinor chains defined in the spinor - helicity formalism ( for instance , see ref . so , relying in and these results , we conclude that it is possible to use covariant gauge in the computation of loop diagrams inside a photon - started splitting amplitude . and , for the triple collinear limit , this fact directly implies that we can make the replacement @xmath147 inside the loop integrals associated with the feynman diagrams presented in figs . [ fig : diagramas_a - qqba ] and [ fig : diagramas_a - qqbg ] . this proof can be generalized for splittings with more colored particles in the final state . in this article we present the triple collinear splitting functions at nlo accuracy for processes which involve photons . computations are performed using dreg to make ir / uv divergences explicit and we choose to work with cdr scheme in order to simplify the analytic treatment of the expressions . the results are organized in such a way that the divergent structure is exposed explicitly . moreover , this structure completely agrees with the predicted behavior according to catani s formula . besides that , the classification of terms according to their transcendental weight and exchange symmetries allows us to obtain very compact results . these splitting functions are computed in the time - like region ( tl ) where strict collinear factorization is fulfilled @xcite . we have considered processes started by qcd partons and also by photons . in particular , we used these results to explore the possibility of crossing - like relations among splitting functions beyond lo . since the parent parton is off - shell , crossing - symmetry is broken and it is not possible to relate splittings with the same particle content . the situation is different for the double collinear limit at lo because it involves a single energy scale . anyway , even for double - collinear configurations , the study of splittings started by photons reveals that it is not possible to establish such a connection in the context of higher - order corrections in qcd@xmath148qed . besides providing explicit results for all triple - splitting functions with at least one photon , we explore the simplifications that occur in photon - started processes . we showed that results can be expressed in terms of standard scalar boxes and bubbles . having all qcd partons on - shell imposes additional constraints which force the cancellation of lcg denominators inside loops . we give a proof of this fact based in the spinor - helicity formalism and gauge invariance . finally , we would like to discuss about the importance of higher orders in the @xmath149-expansion of the triple collinear splitting functions . one of the main motivations to compute these objects is related with @xmath150lo corrections to physical cross - sections . the counter - terms involved in subtraction - like methods require the convolution of splitting functions with some factors which contain @xmath149-poles . for example , at nlo , the typical form of the initial - state collinear counter - term is @xcite d^cnt_a_1 a_2 x & = & _ b dz _ a_1 b p(a_1,b)^(0)(z,)d_b a_2 x^(0 ) , which contains @xmath151 contributions arising from @xmath152 terms in the splitting functions . the codes that we employ to obtain the results shown in this article can be extended to compute @xmath152 terms ( or even higher orders ) , although explicit analytic expressions might be extremely lengthy . anyway , numerical results for the counter - terms could be calculated provided that we know the corresponding @xmath149-expansion of the involved master integrals . we would like to thank stefano catani for making useful comments in an early stage of this project . also , we would like to thanks m. v. garzelli for the fruitful discussions about the implementation of numerical checks . this work is partially supported by ubacyt , conicet , anpcyt , the research executive agency ( rea ) of the european union under the grant agreement number pitn - ga-2010 - 264564 ( lhcphenonet ) , by the spanish government and eu erdf funds ( grants fpa2011 - 23778 and csd2007 - 00042 consolider ingenio cpan ) and by gv ( prometeuii/2013/007 ) . s. catani , d. de florian and g. rodrigo , jhep * 1207 * ( 2012 ) 026 [ arxiv:1112.4405 [ hep - ph ] ] . f. a. berends and w. t. giele , nucl . b * 306 * ( 1988 ) 759 . m. l. mangano and s. j. parke , phys . * 200 * ( 1991 ) 301 . s. catani and m. h. seymour , nucl . b * 485 * ( 1997 ) 291 [ erratum - ibid . b * 510 * ( 1998 ) 503 ] [ hep - ph/9605323 ] . z. bern , v. del duca and c. r. schmidt , phys . b * 445 * ( 1998 ) 168 ; z. bern , v. del duca , w. b. kilgore and c. r. schmidt , phys . d * 60 * ( 1999 ) 116001 . z. bern , g. chalmers , l. j. dixon and d. a. kosower , phys . * 72 * ( 1994 ) 2134 ; z. bern , l. j. dixon , d. c. dunbar and d. a. kosower , nucl . b * 425 * ( 1994 ) 217 . z. bern and g. chalmers , nucl . b * 447 * ( 1995 ) 465 . d. a. kosower and p. uwer , nucl . b * 563 * ( 1999 ) 477 . z. bern , l. j. dixon and d. a. kosower , jhep * 0408 * ( 2004 ) 012 . s. d. badger and e. w. n. glover , jhep * 0407 * ( 2004 ) 040 . a. vogt , s. moch and j. vermaseren , acta phys . b * 37 * ( 2006 ) 683 [ hep - ph/0511112 ] . a. vogt , s. moch and j. a. m. vermaseren , nucl . b * 691 * ( 2004 ) 129 [ hep - ph/0404111 ] . s. moch , j. a. m. vermaseren and a. vogt , nucl . b * 688 * ( 2004 ) 101 [ hep - ph/0403192 ] . j. m. campbell and e. w. n. glover , nucl . b * 527 * ( 1998 ) 264 . s. catani and m. grazzini , phys . b * 446 * ( 1999 ) 143 . v. del duca , a. frizzo and f. maltoni , nucl . b * 568 * ( 2000 ) 211 . t. g. birthwright , e. w. n. glover , v. v. khoze and p. marquard , jhep * 0505 * ( 2005 ) 013 . t. g. birthwright , e. w. n. glover , v. v. khoze and p. marquard , jhep * 0507 * ( 2005 ) 068 . s. catani and m. grazzini , nucl . b * 570 * , 287 ( 2000 ) [ hep - ph/9908523 ] . s. catani , d. de florian and g. rodrigo , phys . b * 586 * ( 2004 ) 323 . c. g. bollini and j. j. giambiagi , nuovo cim . b * 12 * , 20 ( 1972 ) . g. t hooft and m. j. g. veltman , nucl . phys . b * 44 * ( 1972 ) 189 . z. kunszt , a. signer and z. trocsanyi , nucl . b * 411 * ( 1994 ) 397 [ hep - ph/9305239 ] . p. draggiotis , m. v. garzelli , c. g. papadopoulos and r. pittau , jhep * 0904 * ( 2009 ) 072 [ arxiv:0903.0356 [ hep - ph ] ] . m. v. garzelli , i. malamos and r. pittau , jhep * 1001 * ( 2010 ) 040 [ erratum - ibid . * 1010 * ( 2010 ) 097 ] [ arxiv:0910.3130 [ hep - ph ] ] . g. f. r. sborlini , d. de florian and g. rodrigo , jhep * 1401 * ( 2014 ) 018 [ arxiv:1310.6841 [ hep - ph ] ] . j. r. forshaw , m. h. seymour and a. siodmok , jhep * 1211 * ( 2012 ) 066 [ arxiv:1206.6363 [ hep - ph ] ] .

anisotropies in the electromagnetic emission produced by dark matter annihilation or decay in the extragalactic sky are a recent tool in the quest for a particle dark matter evidence . we review the formalism to compute the two - point angular power spectrum in the halo - model approach and discuss the features and the relative size of the various auto- and cross - correlation signals that can be envisaged for anisotropy studies . from the side of particle dark matter signals , we consider the full multi - wavelength spectrum , from the radio emission to x - ray and gamma - ray productions . we discuss the angular power spectra of the auto - correlation of each of these signals and of the cross - correlation between any pair of them . we then extend the search to comprise specific gravitational tracers of dark matter distribution in the universe : weak - lensing cosmic shear , large - scale - structure matter distribution and cmb - lensing . we have shown that cross - correlating a multi - wavelength dark matter signal ( which is a direct manifestation of its particle physics nature ) with a gravitational tracer ( which is a manifestation of the presence of large amounts of unseen matter in the universe ) may offer a promising tool to demonstrate that what we call dm is indeed formed by elementary particles .

1312.4835

the presence of a large amount of dark matter ( dm ) in the universe is now supported by means of a numerous and converging astrophysical and cosmological probes . these probes allow to determine that the average , horizon - scale , dm contribution to the energy density of the universe is six times larger than the baryonic density . in a spatially flat @xmath0cdm scenario the dm energy density is about 25% of the critical density of the universe @xcite . on smaller scales , the dm is observed to be distributed quite anisotropically : it forms a hierarchical network of cosmic structures , from the large - scale galaxy clusters down to the small - scale galaxies and their inner parts . this observational picture is consistent with the theoretical understanding of cosmic structure formation through gravitational instability , based on a dm component which is relatively cold . the cold dark matter ( cdm ) paradigm , with cdm evolving in an expanding universe is able to reproduce the statistical properties of the large scale structure ( lss ) of the universe , and predicts a galaxy formation pattern which is consistent , at least to a large extent , to observations . although the common picture of cdm structure formation may be posed under discussion at very small scales , and even though the true role of baryons if the galaxy formation process has only recently started to be investigated , nevertheless the general picture of the lss formation is robust ( for recent reviews on the subject , see , e.g. , @xcite ) . the most common interpretation of the dm in the universe relies on the simple and quite natural paradigm of particle excitation in the hot early plasma , with subsequent decoupling and formation of a cosmological abundance of dm particles . cdm is then naturally explained in terms of relic particles which need to be relatively massive ( in order to be cold at decoupling ) and weakly interacting ( such that they freeze - out with the correct relic abundance ) . although alternative hypothesis are possible ( axions or axion - like particles , non - thermal or non - symmetric dm production ) the wimp mechanism offers a viable and natural solution to the presence of dm in the universe . dm evidence is currently of pure gravitational origin : galaxy - cluster dynamics , rotational curves of spiral galaxies , gravitational lensing observations , hydrodynamical equilibrium of hot gas in galaxy clusters , the same energy budget of the universe and the theory of structure formation , are all probes based on the gravitational effects induced by the presence of large amounts of dm . however , if dm is composed by elementary particles , vestiges of the early thermal phase , it is expected to produce also non - gravitational signals , specifically related to its particle - physics nature . a large number of astrophysical dm signals are currently explored : except for direct detection , which relies on the direct scattering of the dm particles with a low - background detector in an underground laboratory , all other foreseen signals are of indirect origin and rely on the products of annihilation ( or decay ) of the dm particle in the astrophysical environment , either galactic or extra - galactic . these indirect signals comprise charged galactic exotic cosmic - rays species ( electrons / positrons , antiprotons , antideuterons ) , the whole spectrum of electromagnetic emission ( from radio to gamma - rays ) and neutrinos . the indirect detection signals are intrinsically anisotropic , since they are produced by dm structures in the universe , and structures are present potentially at any scale : galaxy clusters on the largest scales , individual galaxies inside clusters , subhalos inside galaxies . especially electromagnetic signals , which more directly trace the distribution of the dm that produces them , are expected to exhibit some level of anisotropy . bright enough dm objects ( at any given wavelength ) would appear as resolved sources : an example could be a radio or gamma - ray halo surrounding and comprising a galaxy cluster or a single galaxy ( dwarf galaxies , being dm dominated objects , are interesting targets for this search ) or even galactic subhalos which could appear as isolated objects in the radio or gamma - rays sky ) . unresolved dm sources , i.e. with brightnesses below the detector sensitivity , produce instead a diffuse flux of photons which contains a subdominant level of anisotropy , due to the sum of a large number of faint sources related to the anisotropic dm distribution . in this case , the effect of anisotropies could be identified on a statistical basis , by means of studying observables related to angular correlations of the spatial map of the signal , the most typical and relevant one being the two - point angular power spectrum . we comment that also the diffuse neutrino signal would be endowed of the anisotropic feature , since neutrinos are necessarily produced in the same processes that lead also to prompt gamma - rays : however , the mechanisms of neutrino detection introduce a larger spatial uncertainty , as compared to photon detectors , and moreover the detection rates for neutrinos are significantly lower than for photons , due to the weak nature of neutrino interactions . indirect detection searches of the anisotropic dm sky can follow three directions : the study of anisotropies of a single electromagnetic signal , which represents an auto - correlation observable ; the cross - correlation of anisotropies in two different electromagnetic signals ; the cross - correlation of an electromagnetic signal with gravitational tracers of the dm distribution in the universe . the first option has already been attempted both for gamma - rays @xcite ( also on the data recently reported by the fermi collaboration @xcite ) and for the radio signal @xcite ( for some discussions in the x - ray band , see @xcite ) . different wavelengths are emitted by dm through different physical mechanisms : gamma - rays may be promptly produced in the dm annihilation / decay event through production of mesons ( mostly neutral pions ) or baryons , which subsequently decay , or as final - state radiation of charged particles ; x - rays ( and gamma - rays ) are produced by inverse - compton scattering of the electrons / positrons produced by dm annihilation / decay on the interstellar radiation fields and on the cmb ; radio waves are due to synchrotron emission of the same electrons / positrons in the magnetic fields present in the cosmic structure where they are produced . the different electromagnetic signal therefore probe differently the astrophysical properties of the emitting dm structure . prompt gamma - ray emissions most directly retain the spatial features of the dm density field , while the radiative signals feel also the electrons / positrons spatial diffusion and energy loss , as well as the distribution of magnetic fields . on the other hand , the angular resolution of gamma - rays detectors is lesser than the one accessible to radio ( and x - ray ) telescopes . each signal has therefore different features , and the possibility to exploit all types of signal can potentially help in extracting a true dm signal . let us notice here that the relative size of the electromagnetic emission at different wavelengths from dm particles , can significantly change depending on the particle dm properties . for example , a dm particle that can annihilate into hadrons , typically produces softer spectra both of gamma - rays and of electrons / positrons , as compared to leptonic annihilations : this reflects on the spectral features of both the prompt and radiative signals , and to some extent also on the spatial features of the radiative signal , since diffusion and energy losses of electrons / positrons depend on energy . similar arguments apply to the case of decaying dm . the spectral features then depend also on the dm mass , since non - relativistic annihilations ( or decays ) set an upper bound on the maximal accessible energy at the value of the dm mass ( or half the value of the dm mass for dm decay ) . all these features can then affect differently the relative size also of the auto - correlation signals . the second possibility , i.e. cross - correlation between different electromagnetic signal , has not yet been studied : by employing the signal coming at two different wavelengths , we can probe in an independent and richer way the different responses of a dm signal to the dm particle properties ( as discussed above ) , and to some extent also to the properties of the dm structures ( again , for the reasons discussed above on the way an electromagnetic signal is produced ) . in the following we will therefore discuss the potentialities for a dm signal in the different cross - correlation channels and try to quantify the mutual impact in the signal extraction . we will explicitly discuss radio - gamma , radio - x and x - gamma cross - correlations . the third option is potentially the most relevant one , since it attempts to directly correlate two distinctive features of particle dm : an electromagnetic signal , which is a typical and unambiguous manifestation of the dm being an elementary particle , from one side ; the direct probe of the existence of dm in the universe , which is seen through the gravitational effects induced by the presence of dm in the universe , on the other side . a positive signal in this cross - correlation channel would provide direct evidence that what is measured by means of gravitational probes is indeed due to dm in terms of an elementary particle and is not , e.g. , a manifestation of alternative theories of gravity . this third approach has been pioneered in ref . @xcite , where the cross - correlation of the gamma - rays emission with the weak - lensing cosmic shear has been studied and shown to represent a viable and promising novel channel of investigation . in this paper we extend the cross - correlation studies to two additional gravitational probes : tracers of the large scale structure and the very recent observation of the cmb lensing signal . we also extend the analyses to comprise not only gamma - rays , but also the radio and x - rays emissions from dm , in order to establish the potentialities of the full range of possibilities . in this paper we concentrate on the theory of extragalactic dm anisotropies . we discuss the features , sizes and comparisons of the various auto- and cross - correlation signals in the anisotropic sky , for dm annihilations and decays and looking into the whole electromagnetic spectrum , with the aim of exploring the full range of possible signals and correlations , and of setting the ground for future analyses . first , we will carefully outline the formalism needed to perform computations of extragalactic dm anisotropies . we then review some channels already discussed in the literature ( gamma - rays autocorrelations , radio autocorrelations and gamma rays / cosmic shear cross - correlation ) , and extend the range of investigation to propose new channels : for the signals related to the particle physics nature of dm , we explicitly discuss the auto- and mutual cross - correlations in the whole electromagnetic spectrum , by adding to the study the gamma / radio case and the auto- and cross - correlation signals involving x - rays ; for the signals able to correlate the gravitational probes of dm in the universe to the particle - physics ability to produce photon fluxes , we extend the idea to investigate the gamma - rays / cosmic - shear correlation , originally proposed in ref . @xcite , to further adopt radio and x - rays emissions from one side , and to embrace large - scale - structure tracers @xcite and cmb lensing from the other side . a second step of this complex type of analysis will be a detailed study of the different , various and intricate astrophysical emitters at the different wavelengths ( like active galactic nuclei , star forming galaxies , etc . ) which also contribute to the anisotropic sky and represent the background over which the signal we are discussing have to be confronted , as well as an accurate assessment of the observational capabilities . we leave these parts of the analysis to future works . the paper starts with a review of the general formalism apt to derive the two - point angular power spectrum of correlation signals , and details the steps relevant for studying both dm annihilation ( which depends on the square of the dm density field ) and dm decay ( which instead depends linearly on the dm density ) cases . then discusses the features of the relevant ingredients which enter the correlation signal : the three - dimensional power spectrum , both for the auto- and cross - correlation , and the window functions for each electromagnetic and gravitational signal . angular power spectrum predictions are then shown and discussed for all types of correlations signals : auto correlations in the radio , x and gamma - rays bands ; cross - correlations among them ; cross - correlations between the radio or gamma - rays emission and the gravitational probes ( cosmic shear , cmb lensing , lss observables ) . results will be presented for benchmark cases and for both annihilating and decaying dm . in this section we review the general formalism to derive the angular power spectrum of the auto- and cross - correlation of any given pair of intensity fields , which in our analysis are given by the radio , x - ray , gamma - ray dm fluxes , or the cosmic shear , lss , cmb - lensing maps . the source intensity @xmath1 along a given direction @xmath2 can be written as : i_g ( n ) = d g(,n)w ( ) [ eq : int ] where @xmath3 denotes the radial comoving distance ( function of the redshift @xmath4 ) , @xmath5 is the density field of the source , and @xmath6 is the window function ( which is function of the distance but does not depend on the specific direction @xmath2 ) . we can also define a normalized version of the window function as @xmath7 , such that the average intensity is obtained by integration of the normalized window function along the distance : @xmath8 . the window function brings the information on how the actual observable related of the signal under study is distributed in redshift . it will act as a weight for the different redshift contributions of the 3d power spectrum when computing the angular power spectrum . its specific form varies depending on whether we are studying the cosmological emission of an electromagnetic signal from dm annihilation ( or decay ) , or instead the images of background galaxies ( which are lensed by the structures ) in the case of cosmic shear , or the cmb flux at the large scattering surface in the case of cmb lensing , or the distribution of astrophysical sources and of their light emission ( typically in the visible , infrared or radio wavelength ) in the case of lss tracers . the specific form of @xmath9 for all the cases under study will be reported in a dedicated section below . by defining the intensity fluctuation as @xmath10 and expanding the fluctuation field in terms of spherical harmonics , @xmath11 , the ( dimensionless ) @xmath12 coefficients can be expressed as : a_m = dn i_g(n)y_m^(n ) = dn d f_g(,r ) w()y_m^(n ) [ eq : alm ] where the orthonormal relations for @xmath13 have been used and where @xmath14 . a spatial fourier transformation of @xmath15 and the use of the rayleigh expansion of a plane wave into spherical harmonics and spherical bessel functions @xmath16 , allows to derive : a_m & = & dnd f_g(,k ) e^ikrw()y_m^(n ) + & = & dnd f_g(,k ) w()y_m^(n ) + & = & d w ( ) f_g(,k ) j_(k ) y_m^ ( ) [ eq : alm2 ] where we used @xmath17 . the angular power spectrum ( ps ) is defined as @xmath18 , where @xmath19 label the two signals for which the correlation is studied ( e.g. , @xmath20 gamma - rays and @xmath21 cosmic - shear ) , and the brackets @xmath22 denote ensemble average . clearly , the cross - correlation angular ps @xmath23 , as can be easily derived from the relation @xmath24 , while auto - correlation occurs for @xmath25 . by using eq . ( [ eq : alm2 ] ) , the correlation angular ps takes the form : c_^(ij ) & = & d w_i ( ) d^ w_j(^ ) f_g_i(,k)f_g_j^(^,k^ ) j_(kr ) j_^(k^r^ ) y_m ( ) y_^m^^ ( ) + & = & d w_i ( ) d^ w_j(^ ) dk p_ij(k,,^ ) j_(kr ) j_^(k r^ ) y_m ( ) y_^m^^ ( ) + & = & d w_i ( ) d^ dk k^2 w_j(^ ) p_ij(k,,^ ) j_(kr ) j_(k r^ ) + & = & w_i ( ) w_j ( ) p_ij(k=/ , ) [ eq : clfin ] where in the second step we introduced the definition of the three - dimensional ( 3d ) power - spectrum @xmath26 through @xmath27 , and in the last step we assumed the limber approximation @xcite to hold for such ps . the next step is to compute the explicit form for the specific 3d power spectra @xmath26 , which are the fourier transform of the two - point correlation functions ( 2pcf ) in real space : @xmath28 . following ref . @xcite , we assume that the density field can be expressed as the sum of independent seeds ( i.e. , of discrete masses , in the case of the gravitational tracers , or the ensuing electromagnetic emission ) : f(x)=_af(m_a , x - x_a)=dm d^3x _ a^3(x-x_a)(m - m_a ) f(m , x - x ) [ eq : denseed ] where @xmath29 labels the seeds , the @xmath30 s are dirac - delta functions , and we take the mass @xmath31 to be the parameter which characterizes the seeds . the seed density can be expressed as : @xmath32 , where @xmath22 denotes the ensemble average over of all possible seed distributions . with these definitions , the 2pcf reads : ^(2)(x , y)=dm_1dm_2 d^3x_1d^3x_2 _ a ^3(x_1-x_a)(m_1-m_a)_b ^3(x_2-x_b)(m_2-m_b ) f_1(m_1,x - x_1)f_2(m_2,y - x_2 ) the correlation of a seed @xmath29 with mass @xmath33 at position @xmath34 with a different seed @xmath35 with mass @xmath36 at position @xmath37 is provided by the seed-2pcf @xmath38 , and it is not difficult to see that : _ a ^3(x_1-x_a)(m_1-m_a)_b ^3(x_2-x_b)(m_2-m_b ) = [ 1+_s^(2)(m_1,m_2,x_1 , x_2)]+^3(x_1-x_2)(m_1-m_2 ) [ eq:2corr ] which leads to : ^(2)(x , y ) & = & dm d^3x_1 f_1(x - x_1,m ) f_2(y - x_1,m ) + & + & dm_1dm_2 d^3x_1d^3x_2 f_1(x - x_1,m_1 ) f_2 ( y - x_2,m_2 ) _ s^(2)(m_1,m_2,x_1,x_2 ) [ csi2x ] coming back to the fact that the 3d power spectrum @xmath26 is the fourier transform of @xmath39 , and writing @xmath40 in terms of their fourier transforms @xmath41 , one obtains : p_ij(k ) = dm f_i^(k|m)f_j(k|m ) + dm_1dm_2 f_i^(k|m_1 ) f_j(k|m_2)p^(s)(k , m_1,m_2)[eq : pss ] where the power spectrum of the seed distribution @xmath42 is the fourier transform of @xmath38 . in our analysis , we often refer to mass density fluctuations . in this case , the seed-2pcf is the ( homogeneous and isotropic ) linear correlation function of matter ( notice that it has to be the linear one , since in eq . ( [ eq : denseed ] ) we wrote the density field as a linear superposition of seeds ) , @xmath43 . the corresponding power spectrum is the customary @xmath44 . for other objects considered in the following , i.e. dm halos and astrophysical sources , we will assume ( as usually done ) that their 2pcf @xmath45 can be related to the linear correlation function of matter by means of @xmath46 where @xmath47 is the linear bias between the object @xmath48 and matter . thus using @xmath49 , we finally arrive at the decomposition of the 3d power - spectrum into the one - halo and two - halo terms : p_ij(k ) = p_ij^1h(k ) + p_ij^2h(k ) where : p_ij^1h(k ) & = & dm f_i^(k|m)f_j(k|m ) [ eq : ps1halo ] + p_ij^2h(k ) & = & p^lin(k)[eq : ps2halo ] notice that the average @xmath50 of the density field of the source is given by : |g(z)=g(z , n ) = dmd^3x g(x|m , z ) , [ eq : gave ] which implies that at small @xmath51 ( where @xmath52 ) the terms in the square - brackets in eq . ( [ eq : ps2halo ] ) are of order 1 ( except in the case of a significant bias ) . the 2-halo term is thus normalized to the standard linear matter ps at small @xmath51 , which motivates the normalization of the window function introduced above . in principle , instead of adopting the above halo - model formalism , one could directly take the results from n - body simulations , generate the corresponding maps of the emissions and then extract the angular power spectrum . on the other hand , the small scales are currently not covered by simulations and in order to make a realistic prediction of the signal ( in particular in the annihilating dm scenario ) one has to introduce some prescription to include them . results based on simulations only ( without the introduction of small scales ) can be significantly biased . the halo - model approach is instead based on tuning three - functions ( describing the number of halos @xmath53 , the mass concentration of halos @xmath54 , and the halo profile @xmath55 ) to the simulation results , where they exist , and defining some appropriate extrapolations in the unsampled regime . the computation is therefore quite neat once those functions have been drawn , and allows for an easier physical insight . since it is probably unlikely that n - body simulations will cover scales corresponding to the wimp free - streaming mass in the forthcoming years , such a phenomenological approach will remain a prime way to compute the wimp angular ps . the next step is to explicitly derive the three - dimensional power spectra , defined in eq . ( [ eq : pss ] ) , for the different cases of auto - correlation and cross - correlation involving particle dm signals . this requires to adopt , for each specific case , the appropriate intensity - field function @xmath56 . we will only consider a dm source with a spatial profile directly tracing the dm density profile ( or its square in the annihilating scenario ) . this is typically the case for prompt emission of gamma - rays , while for the radio emission , and for the x - ray or gamma - ray signal produced by inverse compton processes , the radiative emissions associated to electrons and positrons injected by dm can have a different shape either because of the spatial profile of the fields of interaction ( the magnetic field for the synchrotron emission which produces the radio signal ; the radiation fields , like cmb or starlight , which are responsible for the x - ray , or also gamma - rays , production through inverse compton scattering ; both fields contribute to energy losses ) , or because of the diffusion of @xmath57 before emission . in this case the density field of the emission is no longer proportional to the dm density ( or its square ) and has to be accordingly modified . this modification typically affects only anisotropies on very small scales ( see the full formalism in ref . @xcite ) and , for simplicity , will be neglected here except for some approximate estimate in the case of the radio emission . in the cases of lensing and decaying dm , the density field of the source is directly given by the density distribution @xmath58 : @xmath59 , and therefore @xmath60 , where the @xmath61 is the density contrast . this is obvious for decaying dm , while in the lensing case it stems from the fact that gravitational lensing is due to the potential wells of the large scale structure , which are related to the matter distribution @xmath55 by the poisson equation . therefore , from eqs . ( [ eq : ps1halo ] ) and ( [ eq : ps2halo ] ) , the non - linear matter ps @xmath62 is obtained by summing up the following one - halo and two - halo terms : p_^1h(k ) & = & _ m_min^m_max dm v(k|m)^2 [ eq : psdelta1 ] + p_^2h(k ) & = & ^2p^lin(k ) [ eq : psdelta2 ] where @xmath63 is the fourier transform of @xmath64 . in the rest of the paper , we will adopt the halo mass function @xmath53 of ref . @xcite , the halo concentration from ref . @xcite down to @xmath65 and extrapolating @xmath54 at smaller masses following ref . @xcite , and a nfw halo density profile @xcite . the halo bias @xmath66 is taken from ref . we compare our halo - model - based predictions for the matter ps with latest results from high - resolution @xmath67-body simulations @xcite in fig . [ fig : ps1 ] . a very good agreement is achieved at low redshift , which is the most relevant epoch for dm purposes , as we will see in the following . for astrophysical sources ( as , e.g. , the lss tracer considered below ) , which are better characterized by their luminosity @xmath68 rather than the mass , the formalism described above can still be adopted by simply replacing the mass function @xmath53 with the luminosity function @xmath69 . approximating astrophysical sources as point sources , we have @xmath70 , and eqs . ( [ eq : ps1halo ] ) and ( [ eq : ps2halo ] ) give : p_ss^1h(k , z ) & = & _ _ min(z)^_max(z ) d ( , z)()^2 + p_ss^2h(k , z ) & = & ^2 p^lin(k , z).[eq : psb ] since the term in the square - brackets does not depend on @xmath51 ( because of the point - source approximation ) and eq . ( [ eq : gave ] ) is now @xmath71 , the two - halo term is just a rescaled version of @xmath72 with the rescaling factor due to the bias ( so possibly varying with @xmath4 ) . the one - halo term is constant in @xmath51 , namely it is a poisson - noise " term . the power spectrum of lss tracers can be thus computed starting from the luminosity function of the population under investigation as described above ( for gamma - ray emitters see , e.g. , ref . however for a given survey might be not so easy to separate different populations ( or impossible when dealing with unresolved contributions , unless using theoretical arguments ) . to keep the discussion general , here we adopt a common description for all types of galaxies , using a halo occupation distribution of galaxies @xmath73 and weighting the contributions of different redshifts through the redshift distribution function @xmath74 . this means that the density field of galaxies can be expressed as @xmath75 , where @xmath76 and @xmath55 is the host - halo density profile . the expression of @xmath73 is taken from ref . @xcite . in this approximated scenario , we have : p_gal , gal^1h(k ) & = & _ m_min^m_max dm v(k|m)^2 [ eq : pslss1 ] + p_gal , gal^2h(k ) & = & ^2p^lin(k ) . [ eq : pslss2 ] in the case of annihilating dm , the signal scales with the square of the density field : @xmath77 . in order to derive the associated angular ps , we can use the formalism outlined in the previous section where now we square the expression given in eq . ( [ eq : denseed ] ) ( also here @xmath78 , as in the decaying dm case ) . the correlation function would then lead to a term proportional to : _ i ^3(x_1-x_i)(m_1-m_i)_j ^3(x_2-x_j)(m_2-m_j)_k ^3(x_3-x_k)(m_3-m_k)_l ^3(x_4-x_l)(m_4-m_l ) which in turn can be expanded into four- , three- , and two - point seed correlation functions . however , since dm halos ( described by @xmath79 ) are mutually exclusive , an annihilation can only occur if the two particles are within the same halo . this implies that all the terms other than the one- and two - halo contributions ( which are the same appearing in eq . ( [ eq:2corr ] ) ) vanish . therefore , eq . ( [ csi2x ] ) does not formally change , with @xmath80 now being @xmath81 . the auto - correlation ps @xmath82 is thus given by the sum of the two following terms : p_^2 ^ 2 ^ 1h(k ) & = & _ m_min^m_max dm ( ) ^2 [ eq : psh1 ] + p_^2 ^ 2 ^ 2h(k ) & = & ^2p^lin(k ) [ eq : psh2 ] where @xmath83 is the fourier transform of @xmath84 . our results are only very mildly dependent on the upper mass cutoff ( which we set to be @xmath85 ) and the clustering on large scales has been well sampled by means of simulations and observations . a crucial and uncertain ingredients is instead the dm clustering at very small masses ( minimum halo mass @xmath86 , concentration parameter , and substructure scheme ) . concerning substructures , we include them by adopting the scheme described in ref . @xcite , namely by replacing @xmath87 with @xmath88 , where @xmath89 is a boost function associated to subhalos which multiplies the smooth halo density profile . in the left panel of fig . [ fig : ps2 ] , we show three specific examples of 3d power spectra , which refer to dm models endowed with different features : two models with @xmath90 ( which is the typical wimp free - streaming mass ) with substructures described by adapting the boost - function @xmath91 to reproduce the via lactea ( vl ) simulation @xcite or the virgo collaboration ( vc ) results @xcite ) ; one model with @xmath92 ( which is the minimum halo mass currently inferred from dynamical measurements ) without substructures . having normalized the power spectra to be ( roughly ) equal on large scales , the effect of substructures mainly reflects into boosting the power on mpc - scales . in the vc scenario the signal from largest halos is indeed strongly boosted by the contribution of subhalos , while in the vl scenario the effect is much milder . -body simulations presented in @xcite , at different redshifts.,scaledwidth=49.0% ] . _ right : _ cross - correlation 3d ps of annihilating / decaying dm with lensing and galaxy tracers ; the total , 1-halo and 2-halo terms are explicitly shown.,title="fig:",scaledwidth=45.0% ] . _ right : _ cross - correlation 3d ps of annihilating / decaying dm with lensing and galaxy tracers ; the total , 1-halo and 2-halo terms are explicitly shown.,title="fig:",scaledwidth=45.0% ] in the case of cross - correlations of electromagnetic dm signals among themselves , the situation is similar to the one discussed in the previous section for auto - correlations : for decaying dm , since all signals depend linearly on the dm density , the relevant matter ps is @xmath62 , whose one- and two - halo terms are given by eqs . ( [ eq : psdelta1],[eq : psdelta2 ] ) , while for annihilating dm the matter ps is @xmath82 , with its component given by eqs . ( [ eq : psh1],[eq : psh2 ] ) . in the case of cross - correlation between lensing ( which depends linearly on the dm density ) and an electromagnetic signal produced by a decaying dm particle , we are again concerned with the @xmath62 power spectrum , expressed through eqs . ( [ eq : psdelta1],[eq : psdelta2 ] ) . the case of cross - correlation of astrophysical sources with decaying dm is analogous to the case involving astrophysical sources and lensing signal , both given by : p_s^1h(k , z ) & = & _ _ min(z)^_max(z ) d(,z ) v(k|m ( ) ) + p_s^2h(k , z ) & = & p^lin(k , z).[eq : psbd2 ] notice that this requires a relation between the source luminosity @xmath68 and the host - halo mass @xmath31 . in the approximate description introduced above , we have instead : p_gal,^1h(k ) & = & _ m_min^m_max dm v(k|m)^2 [ eq : pslssdelta1 ] + p_gal,^2h(k ) & = & p^lin(k ) . [ eq : pslssdeta2 ] in the case of cross - correlation between a gravitational tracer and a signal from dm annihilation , the matter power spectrum @xmath93 involves terms of the type : _ i ^3(x_1-x_i)(m_1-m_i)_j ^3(x_2-x_j)(m_2-m_j)_k ^3(x_3-x_k)(m_3-m_k ) , for the same reasons discussed above in connection with the auto - correlation of annihilating dm , only the terms in the r.h.s . of eq . ( [ eq:2corr ] ) survive , leading again to eq . ( [ csi2x ] ) for the two - point correlation function ( see also appendix b in ref . this allows to write the one- and two - halo terms of the ps of cross - correlation with lensing or decaying dm as : p_^2 ^ 1h(k ) & = & _ m_min^m_max dm v(k|m ) [ eq : pssh1 ] + p_^2 ^ 2h(k ) & = & p^lin(k ) , [ eq : pssh2 ] the ps in the case of cross - correlation of annihilating dm with astrophysical sources is : p_s^2 ^ 1h(k , z ) & = & _ _ min(z)^_max(z ) d(,z ) + p_s^2 ^ 2h(k , z ) & = & p^lin(k , z ) , [ eq : psbh ] and in the approximate scenario is : p_gal,^2 ^ 1h(k ) & = & _ m_min^m_max dm v(k|m ) [ eq : psannlss1 ] + p_gal,^2 ^ 2h(k ) & = & p^lin(k ) , [ eq : psannlss2 ] a few examples of power spectra for the different cross - correlation cases are shown in fig . [ fig : ps2 ] , for both annihilating and decaying dm with lensing or galaxy tracers . ghz for radio , @xmath94 kev for x - rays , @xmath95 gev for gamma - rays . they have been normalized to the total average intensity @xmath96 such that the integral of the curves is equal to 1 for all cases . _ right : _ window functions vs. redshift for cosmic shear , cmb lensing , and two populations of galaxies ( detected in the 2mass and nvss surveys ) tracing lss . , title="fig:",scaledwidth=45.0% ] ghz for radio , @xmath94 kev for x - rays , @xmath95 gev for gamma - rays they have been normalized to the total average intensity @xmath96 such that the integral of the curves is equal to 1 for all cases . _ right : _ window functions vs. redshift for cosmic shear , cmb lensing , and two populations of galaxies ( detected in the 2mass and nvss surveys ) tracing lss . , title="fig:",scaledwidth=45.0% ] the final step is to determine the window functions of the different observables . let us start with the gravitational tracers . the window function of the lensing signal takes the form ( see , e.g. , ref . @xcite ) : w()=h_0 ^ 2_m[1+z()]_^d( ) [ eq : w(z ) ] where @xmath97 is the hubble constant , @xmath98 the matter - density parameter and @xmath99 denotes the redshift distribution of the background sources ( which are lensed by the structures ) , normalized to unity area . in the case of cmb lensing , the source is provided by the last scattering surface . approximating it as a surface with infinitesimal width located at @xmath100 , we can use the same eq . ( [ eq : w(z ) ] ) with @xmath101 . in this case we obtain : w()=h_0 ^ 2_m[1+z ( ) ] [ eq : wcmb ] in the case of astrophysical sources , as e.g. the lss tracers , we have ( see , e.g. , ref . @xcite ) : w(e , z ) = e^- . [ eq : wastro ] however , in this paper , following the discussion outlined in the previous section , we employ a common description for all types of galaxies with a window function given by @xmath102 . for the case of decaying dm we again have @xmath60 and the window function is given by ( see , e.g. , ref . @xcite ) : w(e , z ) = e^- [ eq : wdec ] where @xmath103 is the dm density parameter , @xmath104 is the critical density of the universe , @xmath105 and @xmath106 denote the mass and decay lifetime of the dm particle , respectively , and @xmath107/de$ ] is the number of photons ( at radio , x - ray , or @xmath108-ray frequency , depending in the specific signal under study ) emitted per decay - event in the energy range @xmath109 . finally , @xmath110 is the optical depth for absorption . it can be relevant in the gamma - ray case and is mainly due to pair production on the extragalactic background light emitted by galaxies in the ultraviolet , optical , and infrared bands . for radio and x - rays , the absorption can have an impact only in the very low frequency part of the bands . notice that the factor @xmath111 comes from the normalization of @xmath112 , such that @xmath113 . notice also that we are considering the differential energy flux as the intensity @xmath114 under study : thus we quote differential ( in energy ) window functions for the dm electromagnetic signals . in the case of annihilating dm , the signal scales with @xmath115 , and to define the corresponding window function we can make use of the so - called clumping factor @xmath116 , defined as : ^2(z)==_m_min^m_max dm d^3x , [ eq : clumpfact ] the window function for annihilating dm then takes the form ( see , e.g. , ref . @xcite ) : w(e , z ) = ( 1+z)^3 ^ 2(z ) e^- , [ eq : wann ] where @xmath117 is the velocity - averaged annihilation cross sections times relative velocity ( which we assume to be the same in all dm structures ) and @xmath118/de$ ] is the number of photons emitted per annihilation event in the energy range @xmath109 . and the window functions shown in fig . [ fig : winfun ] . for the radio - radio case , we also show ( with the thinner dotted line ) the effect of a cored dm profile replacing the nfw distribution ( as a possible results of particle diffusion ) . , title="fig:",scaledwidth=45.0% ] and the window functions shown in fig . [ fig : winfun ] . for the radio - radio case , we also show ( with the thinner dotted line ) the effect of a cored dm profile replacing the nfw distribution ( as a possible results of particle diffusion ) . , title="fig:",scaledwidth=45.0% ] let us turn now to the discussion of some examples of auto- and cross - correlation angular power spectra . to be definite , and for illustrative purposes , we focus on a benchmark dm candidate with mass @xmath119 gev , and final state of annihilation / decay into quarks @xmath120 . since we will report dimensionless window functions and angular power spectra ( i.e. quantities relative to the total average intensity ) , the actual values of the annihilation cross section or the decay lifetime are not relevant in the discussion . for definiteness , the substructure description is chosen to follow the vc scenario mentioned above . concerning lss tracers , we will consider two specific examples , given by the population of galaxies observed by the 2mass @xcite and nvss @xcite surveys . the parameters entering in the window function @xmath74 of the two cases are taken from ref . @xcite and ref . @xcite . in the case of weak - lensing shear , we consider the distributions of background sources that will be relevant for the dark energy survey ( des ) telescope and for the euclid satellite . the source distribution for des @xcite can be described by @xmath121 , with @xmath29 , @xmath35 , and @xmath122 provided in table 1 of ref . @xcite , and @xmath123 fixed by the normalization condition @xmath124 . for euclid @xcite , we can use @xmath125 $ ] , where @xmath126 with @xmath127 being the median redshift of the survey and @xmath128 again fixed by the normalization . in the right panel of fig . [ fig : ps2 ] we show the cross - correlation 3d ps of annihilating / decaying dm with lensing and lss tracers , computed as described in the previous sections . notice that the ps for the cross - correlation between lensing with decaying dm is the customary matter power spectrum @xmath62 ( described in eq . ( [ eq : psdelta1],[eq : psdelta2 ] ) ) . the other cases ( lss with decaying dm and the cases with annihilating dm ) lie above it , having more power on the structure scales @xmath129 . this is particularly true in the annihilating case and can be also understood by comparing fig . [ fig : ps1 ] with the left panel of fig . [ fig : ps2 ] in the vc scheme . the window functions are shown in fig . [ fig : winfun ] . the left panel refers to the particle dm cases and shows the redshift dependence of the window functions for the radio , x - ray , and gamma - ray emissions , in both annihilating and decaying scenarios . the radio signal is due to synchrotron - emission produced by electrons and positrons ( induced by the dm annihilations / decays ) interacting with a magnetic field chosen to be constant and with strength @xmath130 . the x - ray flux instead arises from inverse compton ( ic ) of the same electrons and positrons on the cmb , with ensuing up - scatter of the cmb photons . the window functions depend on the photon energy , as is clear from eqs . ( [ eq : wdec],[eq : wann ] ) : for definiteness , we choose some reference frequency / energy , such that the different emissions are all involving gev particles ( photons in the gamma - ray case , while @xmath57 in the radio and x - ray case ) originated from the annihilations / decays . the reference frequencies are : @xmath131 ghz for radio , @xmath94 kev for x - rays , @xmath95 gev for gamma - rays . the plot shows that a multi - wavelength approach can be quite powerful . indeed , the different emissions exhibit rather different window functions : each one therefore can provide different and complementary information . in particular , the radio case is more peaked at low redshift , while the x - ray window function has a flatter shape , with the gamma - ray case being somewhat in between . this is due to the fact that the cmb becomes much more dense in the past ( the photon energy - density goes with the well known @xmath132 scaling ) , making the energy losses for ic very effective at higher redshifts : this depletes the radio signal at high @xmath4 , while , in the x - ray case , the corresponding increase of the ic emissivity compensates the effect . we also notice that the decaying and annihilating dm cases produce very similar window functions for radio emission , while for x - rays and gamma - rays the window function is flatter in the annihilating case as compared to the decaying case . this occurs because of the effect of clustering , which enhances the signal at larger redshifts . the window functions for weak - lensing shear ( des and euclid populations ) , cmb lensing , and lss ( 2mass and nvss galaxy catalogs ) are shown in the right panel of fig . [ fig : winfun ] . they have quite different shapes , as a consequence of the different redshift evolution of the relevant tracers which produce the observed signal . since the cross - correlation power spectrum depends on the overlap of the window functions of the two correlated signals , as is shown in eq . ( [ eq : clfin ] ) , the different gravitational tracers will be more / less efficient in the cross - correlation with particle dm signals . comparing the two panels of fig . [ fig : winfun ] allows to understand which cross - correlation combinations could be optimal ( although experimental limitations may occur , like limited statistics in one observational channel or angular resolutions ) . notice that 2mass and nvss are sort of two opposite limits for galaxy tracers ( and this is the why we choose these examples ) : 2mass coverage peaks at very low - redshifts , below 0.2 , while nvss has a broad redshift distribution . the weak - lensing shear is peaked at intermediate - low redshift , which makes this observables a good candidate for cross - correlation with particle dm signals , as proposed in ref . the cmb window function instead extends over a very wide range of redshift ( in principle , up to the last scattering ) : this is also why it appears significantly lower , as compared to the other cases in fig . [ fig : winfun ] , since we have normalized all window functions to the total average intensity ( i.e. the curves shown in fig . [ fig : winfun ] are normalized to 1 ) . in the following , we also normalize all the angular ps to the total averaged intensity , as reported in eq . ( [ eq : clfin ] ) , in order to make the various cases more easily comparable . the results will therefore show the fractional size of the anisotropy signal ( multiplied by @xmath133 ) with respect to the average isotropic intensity field . the angular power spectra of annihilating dm are presented in the left panels of fig . [ fig : corr1],[fig : corr2],[fig : corr3 ] , while the decaying cases are reported in the right panels of the same figures . fig . [ fig : corr1 ] shows the auto- and cross - correlations of the electromagnetic signals . the most anisotropic cases involve the radio emission . this is because the radio window function is peaked at low - redshift : the signal therefore comes from a relatively low number of halos and thus is quite anisotropic . the flatter window function of x - rays implies that in this case the number of objects contributing to the intensity field is larger , thus making the emission smoother and the ensuing ps smaller . the gamma - rays case is intermediate to radio and x - rays : as shown by its window function in fig . [ fig : winfun ] , the gamma - rays emission is peaked at low - to - intermediate redshifts and this implies a relatively larger anisotropy in the intensity , as compared to x - rays . we also notice that the radio ps flattens more rapidly at @xmath134 with respect to gamma and x - ray ps . the 1-halo term takes over at large multipoles and leads to constant @xmath135 ( poisson - noise ) when sources are point - like , while instead it leads to a decreasing @xmath135 when structures start to be resolved . the flattening is therefore again related to the redshift distribution . indeed the inner structure of closer objects , which are more important in the radio case , is resolved at larger angular scale with respect to more distant objects which contribute to the bulk of the emission in the gamma and x - ray cases . we finally notice that at small angular scales , annihilating dm provides much more anisotropy as compared to decaying dm , due to its dependence on @xmath115 and to the subhalo scheme adopted . as mentioned above , we are taking an approximate description for radiative emissions ( both ic and synchrotron ) where all the power is radiated at the same place of @xmath57 injection , and therefore the density field of the emission is proportional to the dm density ( for decaying dm ) or dm density squared ( for annihilating dm ) . one of the neglected effects is the diffusion of @xmath57 , which might lead to a smoothing of the cuspy behaviour of the density profile into a more cored profile ( for a more extended discussion on the impact of this approximation on radio anisotropies , see ref . if we still assume that all the power is radiated within the object , we can model the effect of diffusion on the final @xmath57 distribution by replacing the nfw profile @xmath136 ( where @xmath137 and @xmath138 is the scale radius ) with a cored density , like e.g. , @xmath139 . the thick and thin dotted lines in the left panel of fig . [ fig : corr1 ] show the case with an nfw and cored distribution , respectively . the difference is not dramatic ( notice that this can also be seen as an estimate for the case where the dm distribution itself is cored , although the power spectrum we have been using is based on results of n - body simulations , thus fully consistent only with a cuspy profile ) . summarizing the behavior of the auto- and cross - correlation power spectra of the electromagnetic signals among themselves , we can state that the radio emission exhibits the strongest anisotropy , both among the auto - correlation signals and in combination with the other emissions . there are almost two orders of magnitude between the radio / radio auto - correlation signal as compared to the x - rays / x - rays auto - correlation ( which exhibit also the smallest among all the correlations of electromagnetic signals ) and one order of magnitude larger than the gamma / gamma autocorrelation . interesting prospects are present for the cross - correlation radio / gamma , which is a factor of 3 - 5 larger than the gamma / gamma case . these general features occur both for the annihilating and decaying dm signals , with more power at large multipoles for annihilating dm as compared to the decaying case , and with slightly more separation among the different cases again for annihilating dm as compared to decaying dm . since we are dealing here with angular power spectra normalized to the average intensity , the actual feasibility of detection will depend also on the absolute normalization level accessible by the different detectors , which is detector specific . being concerned in this paper with the theoretical properties of the auto- and cross - correlation signals and with the assessment of their mutual impact , we are not adopting here any specific experimental figure of merit : results shown in fig . [ fig : corr1],[fig : corr2],[fig : corr3 ] can then be folded with the individual detector capabilities . in fact , the experimental ability to disentangle an anisotropy signal will also depend on the specific features of the detectors and on the astrophysical backgrounds , which also produce an anisotropic electromagnetic emission . photon detectors at different wavelengths have intrinsically different angular resolution : e.g. radio telescopes can resolve very fine details , and have an angular resolution which is much better than gamma - rays detectors . they will therefore be a suitable instrument for studying large multipoles @xmath140 . on the other hand , at lower multipoles gamma - rays detector may be more suitable than interferometric radio telescopes . the combination of the information coming from the auto - correlation signal at different wavelengths , as well as cross - correlations of different signals , may therefore be a relevant tool to identify and characterize a dm signal . concerning astrophysical backgrounds , a large number of electromagnetic emitters are present , like e.g. active galactic nuclei , star forming galaxies , etc . : their emission is anisotropic as well . the physical location of these astrophysical sources is to some extent correlated to the dm structures : nevertheless , they possess different properties both in redshift distribution ( relevant for the ensuing window functions ) and in spectral features . these differences can be potentially used to attempt a separation between these backgrounds and the dm signals discussed in this paper . examples of the impact of astrophysical background have discussed in the case of the gamma - rays auto - correlation @xcite , of the radio auto - correlation @xcite and of the cross - correlations gamma / shear @xcite . in this paper we are concerned on assessing the size and the relative impact of dm signals . a detailed analysis of the background sources is beyond the scope of the present paper , and for the cases for which studies are present we refer to the quoted references . gev ) and gravitational tracers ( cosmic shear , cmb lensing and lss tracers ) , for annihilating dm ( left panel ) and decaying dm ( right panel ) . the computation is performed using the 3d ps models reported in fig . [ fig : ps2 ] and the window functions shown in fig . [ fig : winfun ] . , title="fig:",scaledwidth=45.0% ] gev ) and gravitational tracers ( cosmic shear , cmb lensing and lss tracers ) , for annihilating dm ( left panel ) and decaying dm ( right panel ) . the computation is performed using the 3d ps models reported in fig . [ fig : ps2 ] and the window functions shown in fig . [ fig : winfun ] . , title="fig:",scaledwidth=45.0% ] the cross - correlation of the gamma and radio emissions with gravitational tracers is shown in figs . [ fig : corr2 ] and [ fig : corr3 ] , respectively . for definiteness , also here we have considered the gamma - rays emission at @xmath95 gev and the radio emission art @xmath131 ghz . we notice that the strongest correlation occurs with the 2mass population , while the lowest is with the cmb lensing . this can be again understood by looking at the redshift dependence of the window functions . to have a good overlapping with the electromagnetic dm source , the gravitational tracer has to be peaked at relatively low redshift . moreover , as we already discussed , the closer is the emission the more anisotropic it appears ( and the ensuing angular ps flattens sooner , as a function of the multiple @xmath141 ) . from the window function behavior of fig . [ fig : winfun ] , we see that the radio signal and the 2mass tracer are both strongly peaked at very low redshifts : this fact enhances the cross - correlation both because of the large overlap and because closer sources are fewer and therefore more anisotropic . this is clearly seen in the cross - correlations results of fig . [ fig : corr3 ] , where the angular ps for radio/2mass is significantly large . in the left panel of fig . [ fig : corr3 ] we show again the effect of a cored distribution , which is even milder than for the auto - correlation ps ( this occurs because this effect does not affect the lss tracers and so now modifies only one of the two fields entering in the ps computation ) . also the gamma/2mass cross - correlation is relatively large , although smaller than the for the radio case . this is again due to the good overlap of the gamma - rays and 2mass window functions , nevertheless to a lower extent than in the radio case . the cross - correlations with cosmic shear ( originally proposed for gamma / shear in ref . @xcite ) and with nvss have very similar angular ps , which could potentially offer a tool to disentangle a dm signal from astrophysical backgrounds . in the case of cross - correlation with the cosmic - shear , a tomographic approach is feasible , and represents an unique opportunity to test the different redshift scaling we have been discussing @xcite . the cross - correlation with the cbm lensing observable is the smallest , due to the fact that cmb - lensing sources are distributed in a much deeper interval of redshift , as compared to the electromagnetic emitters . we finally notice that the decaying dm cross - correlation angular ps ( but this applies also for the auto - correlation case ) , even though they have a behavior similar to the annihilating case , nevertheless exhibit a slower growth of @xmath142 , which means less power at small scales : this can be understood by comparing the annihilating and decaying 3d power spectra in fig . [ fig : ps2 ] at @xmath129 . summarizing the behavior of the cross - correlation power spectra of the electromagnetic signals with the gravitational tracers , we can state that again the radio emission exhibits the strongest anisotropy signal . the largest effect occurs ( both for radio and gamma - rays ) in the cross - correlation to the low - redshift 2mass population : the angular ps spectrum in this case is about one order of magnitude stronger than the cross - correlations with cosmic shear and nvss , and about three orders of magnitude larger than the cross - correlation with the cmb - lensing observable . we caution again that however such differences do not straightforwardly translate into the actual experimental capabilities . ghz ) and gravitational tracers ( cosmic shear , cmb lensing and lss tracers ) , for annihilating dm ( left panel ) and decaying dm ( right panel ) . the computation is performed using the 3d ps models reported in fig . [ fig : ps2 ] and the window functions shown in fig . [ fig : winfun ] . in the left panel , we also show ( with thinner lines ) the case of a cored dm distribution ( as a possible results of particle diffusion ) . , title="fig:",scaledwidth=45.0% ] ghz ) and gravitational tracers ( cosmic shear , cmb lensing and lss tracers ) , for annihilating dm ( left panel ) and decaying dm ( right panel ) . the computation is performed using the 3d ps models reported in fig . [ fig : ps2 ] and the window functions shown in fig . [ fig : winfun ] . in the left panel , we also show ( with thinner lines ) the case of a cored dm distribution ( as a possible results of particle diffusion ) . , title="fig:",scaledwidth=45.0% ] in this paper we have discussed extragalactic anisotropies in the electromagnetic emission produced by dm annihilation or decay as a promising tool to search for a dm signal . we have first reviewed the formalism needed to compute a generic 2-point angular power spectrum by following the halo - model description of clustering of structures in the universe . this formalism was then applied to realize estimates involving relevant emissions induced by particle dm annihilations or decays . we have discussed the features and the relative size of the various auto- and cross - correlation angular power spectra that can be envisaged for anisotropy studies . from the side of dm signals we have considered the full multi - wavelength spectrum , including the synchrotron emission at radio frequencies , the ic radiation in the x - ray and gamma - ray bands , as well as the prompt emission of gamma - rays . the angular power spectra of auto - correlation of each of these signals and of the cross - correlation between any pair of them is presented . as a way to enhance the capability of detection of such non - gravitational signals of dm ( and to improve their disentanglement from other astrophysical backgrounds ) we introduce their cross - correlation with maps tracing the gravitational potential . we have analyzed this possibility studying specific gravitational tracers of dm distribution in the universe : weak - lensing cosmic shear , large scale structure matter distribution and cmb - lensing . we have shown that cross - correlating a multi - wavelength dm signal ( which is a direct manifestation of its particle physics nature ) with a gravitational tracer ( which is a manifestation of the presence of large amounts of unseen matter in the universe ) may offer a prime tool to demonstrate that what we call dm is indeed formed by an elementary particle . work supported by the infn research grant _ astroparticle physics project fa51 _ and by the _ strategic research grant : origin and detection of galactic and extragalactic cosmic rays _ funded by torino university and compagnia di san paolo . n.f . acknowledges support of the spanish micinn consolider ingenio 2010 programme multidark csd200900064 . s. camera , m. fornasa , n. fornengo and m. regis , astrophys . j. * 771 * ( 2013 ) l5 [ arxiv:1212.5018 [ astro-ph.co ] ] . just when this paper was submitted , an independent analysis discussing cross - correlation of gamma - rays produced by dark matter annihilation with galaxy catalogs appeared : s. i . ando , a. el . benoit - levy and e. komatsu , arxiv:1312.4403 [ astro-ph.co ] . d. n. limber , ap.j . ( 1953 ) 117 , 134l ; n. kaiser , ap.j . 388 ( 1992 ) 272 ; n. kaiser , ap.j . 498 ( 1998 ) 26 . r. k. sheth and g. tormen , mon . not . * 308 * ( 1999 ) 119 [ arxiv : astro - ph/9901122 ] . j. c. munoz - cuartas , a. v. maccio , s. gottlober and a. a. dutton , mon . not . * 411 * ( 2011 ) 584 [ arxiv:1007.0438 [ astro-ph.co ] ] . j. s. bullock , t. s. kolatt , y. sigad , r. s. somerville , a. v. kravtsov , a. a. klypin , j. r. primack , a. dekel , mon . not . soc . * 321 * ( 2001 ) 559 - 575 . [ astro - ph/9908159 ] . j. f. navarro , c. s. frenk and s. d. m. white , astrophys . j. * 490 * ( 1997 ) 493 [ arxiv : astro - ph/9611107 ] . a. cooray and r. k. sheth , phys . * 372 * ( 2002 ) 1 [ astro - ph/0206508 ] . r. takahashi _ et al . _ astrophys . j. * 761 * ( 2012 ) 152 . [ arxiv:1208.2701 [ astro-ph.co ] ] . m. kamionkowski , s. m. koushiappas and m. kuhlen , phys . rev . d * 81 * ( 2010 ) 043532 [ arxiv:1001.3144 [ astro-ph.ga ] ] . z. zheng , a. a. berlind , d. h. weinberg , a. j. benson , c. m. baugh , s. cole , r. dave and c. s. frenk _ et al . _ , astrophys . j. * 633 * ( 2005 ) 791 [ astro - ph/0408564 ] . t. abbott _ et al . _ [ des collab . ] , astro - ph/0510346 . _ , astron . astrophys . * 479 * ( 2008 ) 9 . r. laureijs _ et al . _ , arxiv:1110.3193 ; l. amendola _ et al . _ [ euclid theory working group collab . ] , arxiv:1206.1225 ; http://www.euclid-ec.org .

using a sample of 43 bright ( , distance kpc ) rr lyrae stars ( rrls ) from the quest survey with spectroscopic radial velocities and metallicities , we find that several separate halo substructures contribute to the virgo overdensity ( vod ) . while there is little evidence for halo substructure in the spatial distribution of these stars , their distribution in radial velocity reveals two moving groups . these results are reinforced when the sample is combined with a sample of blue horizontal branch stars that were identified in the sdss , and the combined sample provides evidence for one additional moving group . these groups correspond to peaks in the radial velocity distribution of a sample of f type main - sequence stars that was recently observed in the same directon by segue , although in one case the rrls and f star groups may not lie at the same distance . one of the new substructures has a very narrow range in metallicity , which is more consistent with it being the debris from a destroyed globular cluster than from a dwarf galaxy . a small concentration of stars have radial velocities that are similar to the virgo stellar stream ( vss ) that was identified previously in a fainter sample of rrls . our results suggest that this feature extends to distances as short as kpc from its previous detection at kpc . none of the new groups and only one star in the sample have velocities that are consistent with membership in the leading tidal stream from the sagittarius dwarf spheroidal galaxy , which some authors have suggested is the origin of the vod .

0807.1735

over the past several decades , a large number of investigations of the stars and the globular clusters in the galactic halo have concluded that the halo formed chaotically and over a long period of time through the accretion and disruption of dwarf galaxies ( e.g. , searle & zinn 1978 , norris 1986 ; carney et al . 1990 ; majewski 1992 ; mackey & gilmore 2004 ; see also freeman & bland - hawthorn 2002 for a review ) . there is evidence , however , that the inner and outer parts of the halo have sufficiently different properties that at least two mechanisms are required to explain the halo ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? accretion appears to have dominated the formation of the outer halo , but the flattening and prograde rotation of the inner halo and the old ages of the globular clusters in the inner halo suggest that its formation involved the dissipative merging of gas clouds at an early epoch ( see * ? ? ? * and references therein ) . while this dual halo picture is supported by substantial evidence , it continues to be important to see what fraction of the halo can be explained by , for example , an accretion - only scenario . the most direct evidence for accretion comes from the detection of halo substructures consisting of co - moving streams of stars . the best documented of these are the streams originating in the sagittarius ( sgr ) dwarf spheroidal ( dsph ) galaxy , which cross much of the sky ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * among others ) . several other substructures have been recently detected in the halo of the milky way ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * among others ) and in the halo of m31 @xcite . these streams provide strong evidence in favor of the hierarchical model of the formation of galaxies , in which large galaxies such as the milky way formed through multiple mergers with smaller building blocks . simulations of the hierarchical picture predict that the milky way has experienced a large number of minor mergers ( see for example * ? ? ? * ) and show that the ones that occurred during the past few gyrs . may be recognizable today as stellar streams . while these simulations appear to be consistent with the current observations of the outer halo @xcite , the search for more streams and better documentation of the known ones remains important . we report here an investigation of the halo substructure in the direction of virgo . the first sign of an overdensity of halo stars in this direction came from the quest rr lyrae star ( rrls ) survey @xcite , which revealed a small overdensity near 13 hr right ascension ( @xmath4 ) , which was the west boundary of the survey as it existed then . as the survey was extended , it became clear that this feature is considerably larger and is centered near @xmath5 hr @xcite . in the jargon of the quest survey it became known as the _ `` 12.4 hr clump '' _ ( declination ( @xmath6 ) was not specified because of the narrow range of the survey around @xmath7 ) . the rrls in the clump cover a range of distances from the sun ( @xmath8 ) with the greatest concentration at @xmath9 kpc @xcite . using photometry from the sdss , newberg et al . ( 2002 ) independently discovered an overdensity in f type main - sequence stars in approximately the same direction and at a similar distance , which they called s297 + 63 - 20.5 . they suggested that it may be related to the small group of rrls identified earlier by @xcite , and several authors have subsequently suggested that the 12.4 hr clump of rrls and s297 + 63 - 20.5 are the same substructure or at least related . also using the sdss photometric catalogue , juric et al ( 2005 , 2008 ) later identified a very large overdensity of main sequence stars toward virgo ( the _ virgo overdensity _ ( vod ) ) . on the sky , the vod encompasses both the 12.4 hr clump and s297 + 63 - 20.5 . but , according to @xcite , the vod is spread along the line of sight from @xmath10 to @xmath11 kpc , which is a larger range than either the 12.4h clump or s297 + 63 - 20.5 . the spectroscopy of rrls in the 12.4 hr clump by @xcite provided the first evidence of a moving group in the virgo direction , which they called the _ virgo stellar stream _ ( vss ) . the recent spectroscopy of f type main - sequence stars in s297 + 63 - 20.5 by @xcite revealed a rich structure in velocity space , but with a major peak that nearly coincides with the velocity of the vss . they tentatively conclude that s297 + 63 - 20.5 , the vss , and the vod are likely associated , if not the same . on the other hand , several authors @xcite have argued that the virgo substructures are parts of sgr streams , while others @xcite have rejected this hypothesis and suggested instead that they are due to a separate merger event involving a dwarf galaxy . we are carrying out a program to observe spectroscopically a large number of quest rrls in order to measure their radial velocities and metallicities . rrls are arguably the halo tracer that can be most precisely located in space , and with the addition of radial velocities of even modest precision ( @xmath12 @xmath13 ) , they become powerful probes of phase space . this wonderful property of rrls comes with a price , for they are expensive observationally . complete light curves are necessary to obtain both mean magnitudes ( hence , precise distances ) and ephemerides . the later are necessary to isolate the motion of the star in the sky from the pulsational velocity of the star by means of the fitting of a radial velocity curve . in this paper , we present spectroscopic observations of bright rrls ( @xmath14 , equivalent to d @xmath15 kpc ) in the part of the quest survey that overlaps with the vod . after examining these data for halo substructure , we discuss the relationships between our results and the previous detections of substructures in virgo and also consider the possibility that they are related to streams from the sgr dsph galaxy . the rrls surveys by @xcite and by @xcite provide important data on parts of the vod that are not covered by the quest survey . according to @xcite , the vod occupies @xmath16 sq . degrees of the sky , and its rough limits are shown in figure [ fig - zone ] . notice that the sdss does not extend south of @xmath17 at @xmath18 ( see figure 6 in * ? ? ? thus , it is possible that the vod extends farther in that direction . the region surveyed by quest is a @xmath19-wide band centered at @xmath20 . it passes over part of the vod , as shown in the figure [ fig - zone ] . the dot - dashed line in figure [ fig - zone ] shows the southern limit of a second quest catalog ( vivas et al . in preparation ) which is centered at @xmath21 . eight stars from this second region are included in our sample . the region in common between the vod and the quest survey is roughly limited by galactic longitude in the range @xmath22 . at the declination of the quest survey , those limits are equivalent to @xmath23 . we have therefore selected for spectroscopy all rrls in the quest catalog @xcite within these limits in @xmath4 , and with magnitude @xmath24 . the constraint in magnitude was imposed by the limitations of our major instrument , the smarts 1.5 m telescope ( see [ sec - observations ] ) . the magnitudes of this sample correspond to a range of distance from the sun of @xmath25 to @xmath26 kpc ( using @xmath27 , following vivas & zinn 2006 ) , which partly overlaps the range estimated for the vod ( @xmath28 kpc , * ? ? ? figure [ fig - quest ] shows the extinction corrected magnitude @xmath29 ( equivalent to distance ) of part of the quest rrls in the northern galactic hemisphere , as a function of @xmath4 . the box encloses the sample of 44 stars studied in this work . within the box , all the stars in the first quest catalog were observed spectroscopically , but only 8 rrls ( out of 38 ) from the second quest catalog were observed . for reference , we have indicated with solid symbols the location of the rrls that probably belong to the vss @xcite . the stars in our sample are located in the same area of the sky as the vss , but are closer to the sun . most of them ( the ones with @xmath30 kpc ) are located within the same region of the sky and distance range as the vod . notice also in figure [ fig - quest ] that part of the tidal stream of the sgr dsph galaxy lies in the background , at @xmath31 kpc . we obtained a total of 88 spectra of the 44 rrls with 3 different telescopes between 2001 and 2007 . the instrumental setups of the different observing runs are summarized in table [ tab - telescopes ] , and the individual observations are detailed in table [ tab - observations ] . most of the spectra were obtained with the smarts 1.5 m telescope at cerro tololo interamerican observatory , chile ( denoted as `` smarts '' in table [ tab - observations ] ) . we also used observations obtained with the 1.5 m telescope in la silla , chile ( denoted as `` eso '' ) and with the hydra multifiber spectrograph at wiyn in kitt peak national observatory , usa . most of the spectra were taken at blue wavelengths , covering the balmer lines ( beginning in h@xmath32 ) and the ca ii h and k lines . however , we took advantage of available time from other projects at the wiyn telescope and obtained spectra using a setup which covered the red part of the spectra , including the ca triplet lines . the wiyn observations are denoted in table [ tab - observations ] as `` wiyn - b '' and `` wiyn - r '' for blue and red spectra respectively . for each target star , we tried to obtain at least two spectra taken at different phases during the pulsation cycle . thirty five out of the 44 rrls have two or more spectra available . table [ tab - observations ] contains i d ( same quest i d as in vivas et al . 2004 ) , julian date , exposure time , telescope , phase of observation ( see [ sec - ephemerides ] ) , heliocentric radial velocity with its error ( see [ sec - rv ] ) , and metallicity ( see [ sec - feh ] ) . the 8 stars from the unpublished second part of the quest catalog have i d numbers higher than 700 . the spectrum of star # 254 indicates that is too cool to be a rr_c _ star . this is not surprising because a small fraction of the quest rr_c _ stars are expected to be misidentifications of other types of variable stars , mainly variable blue stragglers and w uma eclipsing binaries ( see * ? ? ? star # 254 was excluded from any further analysis . to avoid excessive broadening in the spectral features due to the changing velocity of the star during the pulsational cycle , exposure times were kept short , usually under 30 minutes . a few stars have exposure times as long as 60 minutes , which is equivalent to @xmath33 of the pulsation cycle of a typical rrls . in all observing runs , we observed several stars from the list of @xcite which are both radial velocity and pseudo - equivalent width standards . since good ephemerides are essential for fitting the radial velocity curves , it is important to improve them as much as possible . the quest survey has a large number of epochs ( an average 32 for our targets ) , which should yield precise ephemerides . in @xcite , the time of maximum light ( hjd@xmath34 ) for the rr_c _ stars was set by the observation with the brightest magnitude , which is not the best procedure because it can be skewed by photometric error and by poor sampling near maximum light . in addition , for some rr_c _ stars , it was necessary to list two possible periods because both produced reasonable light curve shapes to the eye . to obtain better light curve parameters , we have fitted a light curve template to each rr_c _ star in our sample . a sine curve produced a poor fit because a typical rr_c _ stars has a phase difference between maximum and minimum light of @xmath35 . consequently , we constructed a light curve template based on 9 of the best observed quest rr_c _ stars , which did not suffer from period aliases or from a poorly measured hjd@xmath34 . this template was then fitted using @xmath36 minimization , while varying the amplitude , phase at maximum light , magnitude of maximum light , and period around the values listed in the quest catalogue . we also explored the parameters around the @xmath37 alias periods . the ephemerides for the rr_ab _ stars in the quest catalogue were more secure because they were already determined from light curve fitting and because period aliases were less of a problem for them . however , the period was not allowed to vary in the fitting . this time it was allowed to vary in the fitting , but the resulting light curves are not significantly different from the old ones . only in 5 cases were the changes in period equal to or slightly larger than @xmath38 days . the ephemerides that yielded the best fits to the light curves for the rr_c _ and rr_ab _ stars have been adopted here and are listed in table [ tab - velocities ] . the metallicities of the rrls have been measured from the spectrograms following the methodology and calibration of @xcite , which is based on freeman & rodgers ( 1975 ) modification of the @xcite @xmath39s method . the method , which involves plotting the pseudo - equivalent width of the ca ii k line ( w(k ) ) , corrected for interstellar absorption , against the mean pseudo - equivalent widths of the @xmath32 , @xmath40 , and @xmath6 balmer lines ( w(h ) ) , is described in detail in @xcite . in general , the type c rrls are less numerous than the type ab in old stellar populations , e.g. globular clusters , and because they are harder to detect than type ab , their number in rrls surveys of the field suffer more from incompleteness . @xcite did not produce a [ fe / h ] calibration for type c rrls , which were ignored in his study of the metallicities and kinematics of field rrls . there is a sufficient number of type c stars in our sample that it is important to measure [ fe / h ] from our spectrograms . the average type c variable is @xmath41k hotter and has a @xmath42 dex larger surface gravity than the average type ab variable , although the types overlap in both quantities , particularly when the type ab variables are hottest during their pulsations . the layden calibration may therefore require no modification when measuring type c rrls . this is supported by kemper s ( 1982 ) demonstration that the @xmath39s method , which was devised for type ab variables , can be applied to type c , and more directly by @xcite , who found that both types of variables in same globular cluster define a tight sequence in a plot of k line strength against hydrogen line strength with some intermingling of the types along the sequence . however , @xcite defined their k line and h line strengths differently than did @xcite . to see if layden s calibration can be applied to the type c , we observed both type c and type ab variables in the globular cluster m3 with the hydra multiobject spectrograph on the wiyn telescope . while the fiber setup targeted a large number of variables of both types in this variable - rich cluster , the combination of partly cloudy weather and variable fiber throughputs limited the number of usable spectra to 8 type c stars and 10 type ab . analyses of high dispersion spectrograms of red giants @xcite and rrls @xcite and the tightness of the red giant branch in the color - magnitude diagram of m3 indicate that there is very little if any variation in [ fe / h ] among the stars in this cluster . for our purposes , we assume that there is no real star to star scatter and investigate whether layden s calibration yields the same [ fe / h ] when applied separately to the two types of rrls . our measurements for the m3 rrls in the layden pseudo - equivalent width system are plotted in figure [ fig - feh ] , where different symbols are used to depict the type ab and type c stars . the mean [ fe / h ] obtained from all 10 of the type ab variables is @xmath43 with a standard deviation of 0.27 . the large scatter is undoubtedly due to the low s / n of some of the spectrograms . the 5 type ab rrls ( filled squares in figure [ fig - feh ] ) for which we have the highest s / n spectrograms ( @xmath44 at the caii k line ) yield @xmath45[fe / h]@xmath46 with a more reasonable std . dev . of 0.14 . the 8 type c variables yield @xmath45[fe / h]@xmath47 ( std . @xmath48 ) , and the 5 that have spectrograms with @xmath49 yield @xmath45[fe / h]@xmath50 ( std . dev . @xmath51 ) . the high s / n samples differ in @xmath45[fe / h]@xmath52 by @xmath53 , which suggests that the offset , if any , between the type ab and c variables is small . since these mean values agree well with zinn & west s ( 1984 ) result for m3 ( [ fe / h ] = -1.66 with an internal precision of @xmath54 ) , layden s calibration appears to be equally applicable to type c and type ab rrls . the metallicities that we have derived from the spectrograms of our program stars are listed in the last column of table [ tab - observations ] . the majority of the spectrograms have @xmath55 , and from the experiences with the m3 variables , we estimate that the precision obtained from one spectrogram is @xmath56 for the type ab variables and @xmath57 for the type c. the lower precision for the type c rrls is consequence of the convergence of the constant [ fe / h ] lines at large w(h ) ( see figure [ fig - feh ] ) . a few spectrograms have @xmath58 , and the [ fe / h ] values derived from them are marked with colons in table [ tab - observations ] . layden s calibration breaks down for type ab variables that are observed on the rising branch , where the strength of the k line is reduced by the relatively large effective gravity and shock waves may produce emission in the cores of the h lines that reduce their equivalent widths . we have discarded measurements [ fe / h ] that were made at phases where these effects are likely to be important . the surface gravity variations of the type c variables are milder , and the equivalent widths of the h lines are not affected by emission . none of their measurements have been discarded . radial velocities for all blue spectra were obtained by fourier cross - correlation ( iraf s _ fxcor _ task ) with radial velocity standard stars of a and f spectral type , in the spectral range @xmath59 . the radial velocity standards were separated in two groups according to their effective temperature . each target star was cross - correlated with 6 to 14 different standard spectra of the group with more similar effective temperature . for each spectrum , we obtained a mean radial velocity , @xmath60 , by weighting the result from each standard by the cross - correlation error returned by _ fxcor_. this error depends on both the s / n of the spectra and the similarity between the target star and the standard . in order to estimate the radial velocity errors produced by the different telescope - spectrograph combinations , we cross - correlated each spectrum of the radial velocity standards with the spectra of the other standards , taking care to avoid cross - correlating two spectra of the same star . reassuringly , the differences between the mean velocities from these correlations and the literature values were close to zero for all of the standards with each instrumental setup . we adopted the standard deviations of these differences as the minimum error in velocity that we can achieve with a particular setup for our target stars ( 16 , 8 , and 12 km / s for the smarts , wiyn blue , and eso 1.5 m spectra , respectively ) . because these errors include the uncertainties in the zero - points of the wavelength calibrations , we believe that these values better reflect the true errors than the means of the cross - correlation errors from _ fxcor _ , which are smaller . cross - correlation was not used to calculate the radial velocities of the red hydra spectra because the subtraction of the strongest night - sky lines left residuals in some wavelength regions . instead , gaussian line profiles were fitted to the unblended paschen lines of hydrogen and/or the two strongest lines of ca ii , which however at the resolution of the spectrograms are each blended with a paschen line . the ca ii lines were only measured in spectra that were taken at the coolest phases of the type ab variables when the paschen lines are much weaker than the ca ii lines . before computing the radial velocities from the ca ii lines , we made slight adjustments to their wavelengths to take into account the blending . these were estimated from high s / n spectrograms of bright stars of known velocity whose spectra resemble the type ab variables at minimum light . to test this method , we obtained with the smarts 1.5 m telescope red spectra of 6 bright type ab rrls during cloudy weather . the velocities obtained from the measurements of the ca ii lines did not systematically deviated from the ones measured from the paschen lines alone , and the systemic velocities ( see below ) obtained for these stars agreed with values in the literature . for the red hydra spectra , the standard deviation of the mean of the velocities given by the individual lines ranged from 4 to 10 @xmath13 , which is consistent with the results obtained from similar spectra of bright stars of known velocity . the best technique for obtaining the radial velocity of the center of mass ( the systemic velocity , @xmath61 ) of a type ab rrls is to integrate the radial velocity curve given by weak metal lines over the whole pulsational cycle , and the results obtained for rrls in globular clusters ( e.g. * ? ? ? * ) are consistent with the radial velocities of the clusters to within the internal velocity dispersions of a few @xmath13 . in our low resolution spectrograms , the weak lines produce poorly determined peaks in the cross - correlations , whereas well - defined ones are produced by the balmer lines and other strong features . furthermore , we could not obtain sufficiently large numbers of spectra to determine the velocity curve for each star . for the rr_ab _ , we adopted instead the method described by @xcite , which involves measuring the velocities at a few phases ( @xmath62 ) from low resolution spectrograms and then fitting them with the radial velocity curve of the ab variable x arietis that @xcite obtained from measurements of @xmath63 . observations made near maximum light ( @xmath64 and @xmath65 ) were not used in these fits because the radial velocity curve undergoes a large discontinuity in this region ( see for example * ? ? ? * ) . the velocity curve that is determined from the h lines has a larger amplitude and differs in other ways from the one given by the metals lines ( see * ? ? ? * ; * ? ? ? nonetheless , layden s ( 1994 ) comparisons of his measurements of @xmath61 with literature values for the same stars indicates that this method produces results that are consistent with the errors of the measurements and the fits of the velocity curve . in the case of the template star x ari , the systemic velocity is equal to the h line velocity at @xmath66 , but there is some small star to star variation in this value . in our calculation of the errors in @xmath61 , we have included terms to account for this variation and for the variation in the amplitudes of the radial velocity curve ( see * ? ? ? * ) . for rr_c _ stars , which have much smaller velocity amplitudes than the rr_ab _ , there are at most small offsets between the velocity curves given by the metal lines and the h lines . to measure @xmath61 , we fitted a template constructed by combining the radial velocity curves of t sex and dh peg ( see * ? ? ? * ) . figure [ fig - lc ] shows the best fit for each star having two or more observations . these plots include all of the radial velocity measurements for a star , even if they were not used in the fits because they are close to @xmath67 . in each case , these points lie within the expected range , which provides a check that our measurements do not suffer from large systematic errors . notice also that six of the stars were observed with two different telescopes , and the consistency of the measurements , as seen in figure [ fig - lc ] , is also an indication that large systematic errors are not present . in the fitting of the radial velocity curves , the data were weighted by their errors . the mean difference between the data points and the radial velocity curve , @xmath68 , in most cases was of the same size , or smaller , than the typical errors in the individual measurements ( see table [ tab - velocities ] ) . for the rr_c _ , we calculated the error in the systemic velocity by combining the individual @xmath69 in the usual manner for calculating the @xmath70 of a weighted mean value . for both types of stars , if @xmath68 was larger than @xmath71 we adopted the former as our final error in the systemic velocity . the mean error in the systemic velocities of all , but one , rrls range from 7 to 28 @xmath13 , with a mean value of 17 @xmath13 . one star , the type c # 174 , had an unusually bad fit of the radial velocity curve and its final error is 42 @xmath13 . finally , we calculated the radial velocity measured by an observer at the sun who is at rest with the galactic center , @xmath72 by @xmath73 our results are presented in table [ tab - velocities ] , which lists : i d , coordinates @xmath4 and @xmath6 , the extinction corrected mean v magnitude , the type of rr lyrae star , the period and julian date at maximum light ( see [ sec - ephemerides ] ) , distance from the sun ( d ) , number of spectra available and number of spectra actually used in the fitting of the radial velocity curve , the error in the fit of the radial velocity curve ( @xmath68 ) , the systemic velocity ( @xmath61 ) and its error ( @xmath71 ) , and the velocity in the galactic rest frame , ( @xmath72 ) . as noted above , the vod encompasses the entire range in @xmath4 and perhaps a larger range of distance than our sample of rrls . the spatial distribution of our sample of rrls ( see the box in figure [ fig - quest ] ) does not reveal any regions of exceptionally high density that are strong candidates for halo substructure . there is a roughly uniform band of stars that crosses the region at distances between 7 and 10 kpc and also a region of relatively high density near @xmath74 kpc within @xmath75 . this second feature is the near side of the large over - density in rrls that we have referred to as the `` 12.4 hr clump '' @xcite , which at 19 kpc contains the vss @xcite . figure 12 in @xcite shows the location of the 12.4 hr clump in the first band of the quest survey , but only two of the stars identified there as probable clump members are in our sample . the addition of rrls from the 2nd quest band makes the clump even more prominent , and there are 9 stars in our sample that now appear to be outliers of the clump . if the band of stars at @xmath76 kpc and the 12.4 hr clump are real halo substructures , many of the stars should have similar values of @xmath72 . figure [ fig - vgsr_histo ] is a histogram of @xmath77 of the whole sample of rrls . the mean velocity is @xmath78 @xmath13 , and the standard deviation is @xmath79 @xmath13 . these values disagree with recent measurements of samples of bhb stars in the halo ( see below ) . for a sample of 1170 bhb stars , @xcite found a line - of - sight velocity dispersion , @xmath80 @xmath81 , and @xmath82 @xmath13 . @xcite found a similar @xmath83 in their sample of bhb selected from the 2mass survey , although they noted that the velocity dispersion increased to @xmath84 @xmath13 if they selected only the most metal poor stars in their sample to avoid thick disk contamination . the shapiro - wilk test for normality @xcite indicates that the @xmath72 distribution of our sample does not deviate significantly from a normal distribution . it seems safe , therefore , to use the student t - test and the f - test , which reject with @xmath85 confidence that our sample has the same mean velocity and velocity dispersion as the bhb sample of @xcite . however , the velocity dispersion of our sample is not significantly different from the value @xcite obtained for their most metal - poor bhb stars . while these differences may be another sign that rrls and bhb stars in the halo have different kinematics ( see below and * * ) , the halo substructure in our sample is definitely a major contributor to its relatively large velocity dispersion and its positive mean velocity . while not statistically different from a normal curve , the histogram in figure [ fig - vgsr_histo ] is clearly unexpected for a random selection of halo objects , for it has a very prominent peak at @xmath86 @xmath13 . in total , there are eight rrls , or 19% of our sample that have @xmath87 @xmath81 , which is very unlikely to be random occurrence ( 1 in 10,000 , if @xmath80 @xmath13 and@xmath82 ) . this excess of rrls with large positive @xmath72 is most likely due to a halo substructure . other likely substructures , which have less extreme values of vgsr , are revealed when distance and velocity are used in concert . in figure [ fig - bin ] , we present histograms of @xmath72 for these 3 distance intervals , where we have selected dividing lines that avoid putting stars on the boundaries . the histograms in figure [ fig - bin ] do not resemble the normal curves expected of random selections of halo stars , and in each case the shapiro - wilk test rejects normality with greater than 98% confidence . when considering the @xmath88 kpc interval , it is important to recall that quest survey is probably incomplete at the bright magnitudes of these relatively nearby stars because of saturation of one or more detectors in the quest mosaic camera ( see * ? ? ? detector saturation was not a problem for the stars in the more distant intervals , and the quest survey is expected to be @xmath89 and @xmath90 complete for the type ab and type c variables , respectively . for these intervals at least , the departures of the @xmath91 distributions from the expected halo curves are therefore probably due the presence of halo substructure . to identify possible groups of stars , we implemented with only a few modifications the algorithm described by @xcite ( the stellar pair search ) . in this technique , the distances of each star from all others in the sample are calculated from the cartesian galactic coordinates @xmath92 and @xmath93 . pairs of stars are ones that are separated by less than a critical distance , @xmath94 and have a difference in radial velociticy less than a critical velocity , @xmath95 . pairs having one star in common form a group . a group may be formed by many connected pairs , and the minimum number of stars in a group , @xmath96 , is 3 ( two pairs connected by one star ) . to study the statistical significance of the groups we performed monte carlo simulations of random samples of stars having the radial distribution of the rrls in the halo . specifically , we used the density radial profile described in @xcite for the direction @xmath97 , @xmath98 , which has a slope of @xmath99 and takes into account the flattening of the inner halo . we assigned to each simulated star a random @xmath4 and @xmath6 within the same limits of our sample , and a random @xmath77 drawn from a gaussian distribution with @xmath80 @xmath13 and @xmath82 . each simulated sample was introduced in the group detection algorithm described above . the experiment was repeated many times , and we counted how many times the simulated samples produced groups with characteristics similar to the groups in the real sample of rrls . the choices for the parameters @xmath94 and @xmath95 require some discussion . as we described above , the vod covers the whole range of right ascension of the sample studied here , which is @xmath100 . any substructure in our data may be as large as this . at 7 kpc from the sun , @xmath101 is equivalent to 5 kpc , and at 12 kpc from the sun , it is 8 kpc . on the other hand , the uncertainty in distance of the rrls is @xmath102 @xcite , which suggests a lower limit of @xmath103 kpc . the streams from either destroyed globular clusters or dwarf galaxies are expected to be cold substructures , with velocity dispersion @xmath104 @xmath13 . given the errors in @xmath61 , @xmath95 should be larger than 17 @xmath13 . since halo substructures are unlikely to have uniform sizes or velocity dispersions , we ran the group detection algorithm and the monte carlo simulations using combinations of @xmath94 from 1 to 4 kpc in steps of 0.5 kpc and @xmath95 from 20 to 30 @xmath13 in steps of 5 @xmath13 . a group in the rrls data was only considered real when less than 5% of the 5000 monte carlo simulations ( a @xmath105 detection ) produced a group with @xmath106 members within @xmath12 @xmath13 of @xmath107 , the absolute value of the mean velocity of the group of real rrls . the simulations therefore counted random groups near the @xmath72 of the suspected group in the real data and ones near @xmath108 . if a group was detected with different sets of parameters , as happened in most cases , we report here the parameters that gave the lowest probability of random occurrence and/or the maximim number of group members . we restricted our sample to the 33 stars that have @xmath109 kpc since the incompleteness that is caused by the different saturation levels of the ccds of the quest camera @xcite may effect the interpretation of results at closer distances . two significant groups were found among our rrls ( figure [ fig - groups ] ) . as we suspected from the histograms in figures [ fig - vgsr_histo ] and [ fig - bin ] , there is a group with large positive velocity in our data . the maximum number of members ( 8 rrls ) was obtained with @xmath110 kpc and @xmath111 @xmath13 . not one of the 5000 monte carlo simulations produced a similar group . the stars making up this group are spread in @xmath8 from @xmath112 to @xmath113 kpc ( see figure [ fig - groups ] ) . four of the more distant stars in this group ( @xmath114 kpc ) are part of the near side of the `` @xmath115 hr clump '' . the unusually high velocities of these stars and that the fact that some of them are probably part of the `` @xmath115 hr clump '' , unquestionably an over - density in rrls , suggest that they constitute a halo substructure that may be part of the vod ( see [ sec - dis ] ) . the combination of large @xmath72 and a spread in @xmath8 is consistent with a stellar stream that obliquely cuts the line - of - sight ( see fig . 15 in * ? ? the other significant group contains 9 rrls . while it is detected with several choices for @xmath94 and @xmath95 , the values of 1.5 kpc and 25 @xmath13 , respectively , yield the lowest probability of a random occurrence ( only 0.7% of the monte carlo simulations produced similar groups , a @xmath116 detection ) . the relatively small value of @xmath94 of this group indicates that it is more concentrated in space than the high velocity group ( see figure [ fig - groups ] ) . it lies @xmath117 kpc from the sun and has a mean @xmath72 of @xmath118 @xmath13 . this group may be due to a stellar stream that cuts the line - of - sight at nearly a right angle , which would explain the narrow distribution in @xmath8 and the relatively low @xmath72 of the stars . if any of the features described above are real substructures , then they should also appear in the distributions of the other types of stars that make up very old stellar populations . while bhb stars and rrls are , of course , samples of horizontal branch stars that differ primarily in temperature , there are good reasons to suspect that they may not always trace the same stellar populations . the majority of the globular clusters in the outer halo have red hb s , while the metal - poor globular clusters in the inner halo have primarily blue hb s ( the dependence of the second parameter of hb morphology on galactocentric distance ) . consequently , there are large differences in the frequencies of rrls and bhb stars between the clusters in the inner and outer halos @xcite . the majority of the dwarf spheroidal satellite galaxies of the milky way and m31 have red hb s and high frequencies of rrls and low frequencies of bhb stars @xcite . @xcite have suggested that this variation in the rrls / bhb star ratio and the kinematic disconnect between the inner and outer halos ( see [ sec - intro ] ) may be responsible for the kinematic differences that they found between samples of rrls and bhb in the same region of the galactic halo . these observations and the trend in rrls / bhb suggest that the rrls may be a denser tracer of substructure in the outer halo than bhb stars . since a small fraction of outer halo globular clusters and dsph galaxies have blue hb s , each with minority populations of rrls , and since a small population of bhb stars exists in the red hb clusters , some correlation in phase space is expected between these types of stars in the outer halo . on the other hand , an exact correspondence would be surprising . the sample of bhb stars considered here was drawn primarily from the survey by @xcite that is based on photometry and spectroscopy from the sdss . this database of bhb stars is less complete than our rrls sample because the sdss spectroscopy targeted primarily quasars and galaxies . the recent , more complete sample of bhb stars studied by @xcite unfortunately does not overlap with the virgo region . from the @xcite catalog , we selected all bhb stars ( 35 in total ) within the same range of @xmath4 as the rrls sample . in @xmath6 we expanded the limits to @xmath119 . this is still a small range compared with both the width of the region in @xmath4 , and the size of the vod . we also constrained the distance range to @xmath120 kpc . @xcite estimate that their measurements have an average error in velocity of about @xmath121 @xmath13 . it has been established both from observations of globular clusters and theoretical modelling that there is a relation between the luminosity of bhb stars and their temperature in the sense that bluer bhb stars are fainter than red bhb stars ( see for example * ? ? ? when estimating the distances to their bhb sample , @xcite adopted instead the same @xmath122 for all bhb stars on the basis of some observational evidence that they realized was in conflict with theoretical computations . @xcite noted that the adoption of the variation in @xmath122 with effective temperature that is suggested by the computations would lead to smaller distance moduli by an average of @xmath123 mag . @xcite have used the color - magnitude diagrams of globular clusters to derive a relationship between the @xmath124 colors of bhb stars and their @xmath125 values , which is anchored on the same @xmath125 for rrls that we are using . a comparison , after making the necessary transformations between photometric systems , of the @xcite relationship to the one employed by @xcite reveals that the difference in distance modulus is near zero at the blue edge of the instability strip and again small at the highest temperatures , but is @xmath126 mag ( a factor of @xmath127 in distance ) at @xmath128 or @xmath129 . because only a fraction of the sample of bhb stars observed by @xcite _ may _ be effected by distance errors , we have not attempted to recalibrate their distance scale . none of the following comparisons depend critically on an exact match of the bhb and rrls distance scales . in a few of the fields that were observed with the hydra multi - object spectrograph on the wiyn telescope , we identified potential bhb stars from the quest photometry . the three stars that listed in table [ tab - bhb ] appear to be true bhb stars . their @xmath130 and @xmath131 colors , as measured by the sdss , are consistent with the low gravities of bhb stars on the basis of figure 10 in @xcite . > from our spectra , we measured the full widths of the @xmath32 , @xmath40 , and @xmath6 balmer lines at 0.2 below the continuum level , which are well - known diagnostics of surface gravity at the temperatures of bhb stars ( eg . * ? ? ? in addition to the candidate bhb stars , we measured the wiyn spectra of two bhb stars in the globular cluster m3 that are similar in @xmath132 color to the candidates and of the bright a and f type stars that we employed as radial velocity standards with the wiyn observations . the candidate bhb stars have smaller h - line widths than the main - sequence stars and ones that are similar to the bhb stars in m3 . to permit direct comparison with the @xcite sample of bhb stars , we adopted their method to derive the distances to these bhb stars ( see table [ tab - bhb ] ) . in figure [ fig - bin_bhb ] , we have plotted histograms of @xmath72 for the bhb stars from @xcite and table [ tab - bhb ] that lie in the same @xmath8 intervals as the ones in figure [ fig - bin ] . in this diagram , there may be more smearing between the intervals than for the rrls because the absolute magnitudes of the bhb stars are more uncertain ( see above ) . neither of the two features identified in the rrls sample is recognizable in the bhb star data alone , which could be due to the incompleteness of the sample or to low frequencies of bhb stars in these groups . however , several bhb stars have @xmath72 values that coincide with the peaks in the rrls histogram . indeed , when we introduced the combined sample of rrls and bhb stars into the group detection algorithm described above , the same two groups identified in the rrls sample alone appeared . the group with negative velocities now contains 14 members , 5 of which are bhb stars . the new mean velocity is @xmath133 @xmath13 , which we adopt as the velocity of the group . only 5% of the monte carlo simulations produced similar groups . the group with high positive velocities has 10 members ( 2 bhb stars ) . similar groups were obtained in only 2 out of 5000 simulations ( 0.04% ) . the combined sample of rrls and bhb stars also contains a third , small but nonetheless significant , group having a mean radial velocity of @xmath134 @xmath13 . the group , which was detected with @xmath135 kpc and @xmath111 @xmath13 , contains only 3 members , all of which are bhb stars located in the very narrow intervals of @xmath136 to @xmath137 kpc and @xmath138 to @xmath139 . the monte carlo simulations found simialr groups in @xmath140 of the cases ( a @xmath141 detection ) . this group may be related to main sequence stars with similar velocities ( see below ) and with 2 rrls at essentially the same distance and velocity but farther west ( @xmath142 ) . since this group lies near the faint limit of our sample , it may be the near side of a larger substructure . the recent investigation of the vod region by @xcite has provided radial velocity measurements for many stars with the colors of f type stars in two segue fields , at ( @xmath143 ) = ( @xmath144 ) and ( @xmath145 ) , one of which overlaps with the region we are considering here ( see figure [ fig - zone ] ) . @xcite make the very reasonable assumption that the vast majority of these stars are probably near the main - sequence turnoff in the color - magnitude diagram . because such stars have a significant spread in absolute magnitude , their distances are much less precisely known than either the rrls or the bhb stars . the big advantage of using these stars is that they are much more numerous than either rrls or bhb stars . below we examine their data for evidence for the major groups described above . [ [ group - at - v_gsr215-rm - kms-1 . ] ] group at @xmath146 @xmath13 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this group is probably the most remarkable feature seen in our data since very few stars with such high velocities are expected in a sample of halo stars ( 1 or 2 in our sample of rrls ) . the 10 rrls and bhb stars in this group have a mean velocity of @xmath147 @xmath13 , and a standard deviation of @xmath148 @xmath13 , which is not much larger than the typical error in our radial velocities ( @xmath149 @xmath13 ) . as noted above , this group appears to be spread along the line of sight , and it may be significant that in the sample of rrls studied by @xcite , there are 3 stars with @xmath72 values consistent with membership in this group that have @xmath8 between 16 and 17 kpc . in their discussion of the f - type stars , @xcite briefly mention an excess of high velocity stars , which we suspect is related to this group of rrls and bhb stars . in their figure 9 , @xcite present for each of their fields histograms of @xmath72 for the f stars in two ranges in @xmath150 magnitudes . in the histogram for the brighter group , which according to the distance scale , then the distance uncertainty may be as large as 40% ] of @xcite corresponds to 11 to 14 kpc , there are 8 f type stars with @xmath72 consistent with assignment to our high velocity group . it is probably significant that all 8 of these stars are found in the more northern of the two fields ( see figure [ fig - zone ] ) , the one that overlaps with the region studied here ( 2 stars of even higher @xmath72 are present in this field as well ) . since neither one of the two f - star fields has any stars with @xmath151 @xmath13 , there is an asymmetry toward large positive velocities in the same region of the sky where we find one in the rrls sample . among the fainter f type stars , which lie at distances between 14 and 18 kpc , there are only 3 stars in the 2 fields combined with @xmath72 within the range of this group . we have been able to determine metallicities for 6 of the 8 rrls in our sample that appear to be members of this group . they have a mean @xmath152 } \rangle = -1.55 $ ] , with a standard deviation of 0.15 dex , which is of the same size as the uncertainty in our [ fe / h ] measurements . this narrow distribution is much more consistent with the debris from a destroyed globular cluster than from a dwarf galaxy . two of the 3 high - velocity rrls measured by @xcite have metallicities that are similar to the metallicities of the brighter rrls . the other one has a much lower ( [ fe / h]@xmath153 ) , which may be a sign that it is unrelated . we are observing more rrls in the range @xmath154 kpc , with the expectation that more velocity and metallicity measurements will reveal the origin of this substructure . [ [ group - at - v_gsr-49-rm - kms-1 . ] ] group at @xmath155 @xmath13 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this is the most populated group that we find in our data , containing 9 rrls and 5 bhb stars with velocities in the range @xmath156 @xmath13 . the mean velocity of this group is @xmath157 @xmath13 , with a standard deviation of @xmath158 @xmath13 . the contamination from unrelated halo stars is expected to be relatively high at the low velocity of this group , and assuming the normal distribution in figure [ fig - vgsr_histo ] for such field stars , we estimate that only about 50% of the stars in this velocity interval are probably members of a stream . thus , our estimates for both the mean velocity and the velocity dispersion are likely to be skewed by this contamination . the 9 rrls in this group have @xmath159 } \rangle = -1.72 $ ] , with a standard deviation of 0.28 dex . there is not a strong peak in the metallicity distribution that would suggest that the large dispersion is caused by contamination by non - members of an otherwise narrow distribution . @xcite found a significant peak at @xmath160 @xmath13 in the velocities of the f - type stars , which considering the uncertainties is close to the peak that we find . while this suggests membership in the same halo substructure , it is not altogether clear that they overlap spatially . this feature is most clearly evident in the southern of the two fields investigated by @xcite , the one that is outside the region investigated here , and among the bluer and brighter sample of f - type stars ( @xmath161 kpc if they have @xmath162 ) . on the other hand , the rrls in this group have @xmath163 kpc , which place them in the foreground of the f - type stars , unless the f - type stars are less luminous than suspected . there is better spatial coincidence between the f - type stars and a group of fainter rrls that were measured by @xcite , which have similar @xmath72 . the relationships between the f - type stars , the rrls in @xcite , a sample of m giants , and the tidal stream from the sgr dwarf galaxy are discussed at length by @xcite , who suspect that this feature may be a new halo substructure that is unrelated to sgr . the feature that we have identified appears to be unrelated to sgr as well ( see below ) . more observations are required to clarify its relationship to the feature described by @xcite . [ [ group - at - v_gsr-171-rm - kms-1 . ] ] group at @xmath164 @xmath13 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the 3 bhb stars in the group have basically the same values of @xmath8 , 11.2 kpc , but given the distance uncertainty a spread of @xmath165 kpc can not be excluded . @xcite identify a peak in their f star data at @xmath166 @xmath13 , and in this case , there are good reasons to believe that the features are the same . a peak is seen in both of their fields , although it is more prominent in the southern field , the one that is outside our region , than in the northern . it is only seen in the brighter f stars that have @xmath167 . according to the distance scale of @xcite these stars lie at @xmath8 between 11 and 14 kpc , which agrees well with the distances of the bhb stars . the f star data show a very prominent peak at @xmath168 @xmath13 in both fields , and @xcite consider this to be the velocity of s297 + 63 - 20.5 feature . on the basis of the spatial coincidence between s297 + 63 - 20.5 and the vss and the near coincidence in @xmath72 , they conclude that these features are probably the same or physically associated . however , @xcite note that the @xmath169 @xmath13 offset in @xmath72 between s297 + 63 - 20.5 and the vss requires explanation . the peak at @xmath168 @xmath13 is seen in the faintest and the reddest group of f stars , which have @xmath150 between 20 and 20.5 . @xcite note , however , that the most of these stars actually have @xmath170 , which corresponds to @xmath171 kpc on their distance scale . since the core of the vss is located at about 19 kpc @xcite , it is possible that @xmath169 @xmath13 offset is due to a gradient in @xmath72 with @xmath8 . some evidence for this is provided by the data for the brighter and bluer f star data ( see figure 10 in @xcite ) , which has a prominent peak at @xmath172 @xmath13 . the histograms in figures [ fig - bin ] and [ fig - bin_bhb ] indicate there are 5 stars ( 2 rrls and 3 bhb stars ) having a mean velocity of @xmath173 @xmath13 which , within uncertainties , is identical to the peak seen in the bright f star data . all 5 rrls and bhb stars are located near the far distance limit of our sample at @xmath174 kpc and within the near boundary of the `` 12.4 hr clump '' of rrls . the 5 stars are indeed identified as a group by our algorithm ( with parameters @xmath175 kpc , @xmath176 @xmath13 ) , but by itself this group has only low significance because @xmath177 of the monte carlo simulations produce similar groups . while a chance coincidence can not be ruled out , the similarity in @xmath178 between this group and the peak in the bright f star data suggests that they are related . the group of rrls and bhb stars provides a hint that the vss may extend to distances as short as @xmath179 kpc . a gradient in @xmath72 , with @xmath72 becoming less positive with increasing @xmath8 , is expected of a stellar stream that obliquely cuts the line of sight ( see * ? ? ? it is also possible , however , that observational error and sample selection may also explain the @xmath72 offset . we are exploring the possible extension of the vss and the possibility of a gradient by measuring more quest rrls at distances between 13 to 18 kpc ( duffau et al . 2008 , in preparation ) . soon after the first detections of over - densities of stars in the direction of virgo , several authors ( eg . * ? ? ? * ; * ? ? ? * ) speculated that they may be debris from the disruption of the sgr dsph galaxy . even then there was some evidence to contrary , for @xcite had noted that the color of the turnoff in s297 + 63 - 20.5 was redder than the color of the turnoff in the sgr streams . @xcite later argued that the sgr streams were not responsible for the vss or the majority of the other stars in their sample because models of the streams produced very few stars at the observed positions and the predicted values of @xmath72 did not match the ones observed . the realization by @xcite that the vod has huge dimensions on the sky motivated @xcite to reexamine the sgr connection . through detailed modeling of the streams , they showed that ones that assume an oblate shape for the gravitational potential of the galaxy can explain the vod as the leading stream from sgr as it plunges toward the sun and crosses the galactic plane in the solar neighborhood . this model does not , however , provide a good match to our observations , those of @xcite , and more generally the 12.4 hr clump of rrls . in the left - hand panels of figure [ fig - sgr ] , we have plotted the rrls in the first band of the quest survey that lie within the @xmath4 range of our sample . closed symbols depict the rrls that have measured velocities from @xcite , @xcite , and the present study . these three datasets cover different range of distances , @xmath180 , @xmath181 and @xmath182 kpc respectively . the small crosses are the particles in the stream models of @xcite under the assumptions that the potential is oblate , spherical , or prolate . the panels on the right compare the model predictions for @xmath72 as a function of @xmath8 with the observed velocities . the quest survey detects the sgr leading arm at @xmath183 kpc in the range @xmath184 ( see figure [ fig - quest ] and vivas et al . 2005 ) and the 12.4 hr clump in the ranges @xmath185 kpc and @xmath186 ( see fig . 12 in * ? ? ? note that there is not a blending together of these features ( see figure [ fig - quest ] ) , as one would expect if they are parts of the same stream , but instead they are separated by a region of low density . this is not a selection effect introduced by the observing technique ( see * ? ? ? if the leading arm crosses the galactic plane near the sun , then the stream stars at small @xmath8 will have large , negative values of @xmath72 because they have fallen from @xmath31 kpc on orbits that are nearly aligned with the line - of - sight from the sun . this prediction of large negative velocity at small @xmath8 and the continuity of the stream are two characteristics of this picture that should not be model dependent . @xcite note that their models reproduce the models of @xcite , which because they are readily available , we will use here to illustrate the salient points of this picture . each of the @xcite models matches reasonably well the positions and the velocities of the rrls at @xmath31 kpc . however , both the oblate and the spherical models predict smoothly varying densities of rrls over wide ranges in @xmath8 , which are in conflict with the observed region of relatively low density of rrls between 20 and 40 kpc ( see also * ? ? ? these same models predict that @xmath72 becomes more negative as @xmath8 decreases and reaches quite large @xmath187 at @xmath188 kpc . only one rrls in our sample and one in the sample observed by @xcite have values of @xmath72 that are consistent with these models . also , none of the bhb stars in the region have highly negative velocities ( see figure [ fig - bin_bhb ] ) . in the prolate model , the leading stream crosses the galactic plane far from the sun , which explains why it predicts relatively few stars in the diagrams at @xmath188 kpc . we conclude on the basis of these comparisons that none of the halo substructures that we have detected are probably due to the leading arm . moreover , based on the current sgr models , our observations imply that the shape of the galactic potential can be neither oblate nor spherical . @xcite note that in their oblate model the trailing stream of sgr crosses through the region of the vss and coincides with the vss in @xmath72 ( @xmath189 km / s ) . this is a plausible explanation for the vss , if it is still viable after the oblate model is reconciled with the observations presented here and with other recent observations which indicate that the leading arm does not cross the galactic plane in the solar neighborhood @xcite . kinematical information is crucial to detect substructures in regions where no obvious spatial overdensities are present . taken together , our rrls sample , the bhb star sample of @xcite , and the f star sample of @xcite paint a consistent picture that several separate halo substructures exist at @xmath190 kpc in the part of the vod that is covered by the quest survey . the rrls sample of @xcite and the same bhb and f star samples show that some of these substructures extend to larger @xmath8 . none of these substructures seem to be related with the sgr leading tail . the analysis of the sdss photometric data by @xcite make a compelling case that the region of highest stellar density in s297 + 63 - 20.5 lies at @xmath191 kpc . this is identical to within the errors of the region of highest density in the 12.4 hr clump of quest rrls , which is where @xcite discovered the vss in velocity space . the near coincidence in @xmath178 between the vss and the largest peak in @xmath72 that @xcite observed in s297 + 63 - 20.5 provides strong evidence that they are the same . as noted by @xcite , the small offset ( @xmath192 @xmath13 ) could be due to measurement error and/or by the small sizes of the samples of spectroscopically observed stars . it is also possible that it is due to a gradient in @xmath72 with @xmath8 ( see above ) . @xcite found the same velocity peak in their two spectroscopic fields , which indicates that the s297 + 63 - 20.5/ vss has considerable angular extent on the sky . other evidence for this is provided by the recent rrls surveys by @xcite and @xcite . @xcite found 2 clumps of rrls south of the vod region ( see figure [ fig - zone ] ) at coordinates @xmath193 and @xmath194 and distances 16 and 19 kpc respectively . velocity information is not yet available for these clumps . @xcite identified an excess of candidate rrls having @xmath195 @xmath13 in the north galactic cap ( @xmath196 ) . the results presented here indicate that the s297 + 63 - 20.5/vss feature may also extend along the line of sight to distances as close as @xmath197 kpc . we conclude that the vod contains at least one relatively large structure , s297 + 63 - 20.5 /vss , which is intersperced with smaller halo substructures that are unrelated to s297 + 63 - 20.5/vss or to each other . while these smaller structures might be peculiar to the vod region , it is more likely that they are common in the outer halo . they are better documented in the vod direction simply because the presence of s297 + 63 - 20.5/ vss has drawn attention to this region . in the case of our sample of rrls , the spatial distribution of the stars provided only very marginal evidence of any substructure @xcite . the picture changed dramatically when we used spatial and velocity data together . there is already strong evidence from the spatial distribution of stars in the sdss that much , if not all , of the outer halo was accreted from the destroyed of satellite galaxies @xcite . our investigation of the vod region suggests that the addition of velocity data will make the case even stronger . this research was supported by the national science foundation under grants ast 00 - 98428 and ast 05 - 07364 . the availability of the smarts 1.5 m telescope and its service observing mode made this project feasible . we thank charles bailyn for his efforts to setup the smarts consortium and fred walter and suzanne tourtellotte for their assistance with smarts operations and data handling . we are indebted to the service mode observers sergio gonzalez , alberto pasten , claudio aguilera and alberto miranda for their fine operation of the telescope and the spectrograph . we also thank the anonymous referee for helpful suggestions to the manuscript . , r. , vivas , a. k. , gallart , c. , & winnick , r. 2004 , in astronomical society of the pacific conference series , vol . 327 , satellites and tidal streams , ed . f. prada , d. martinez delgado , & t. j. mahoney , 92 to @xmath201 ( from * ? ? ? the box encloses the bright subsample ( @xmath202 ) toward virgo for which we obtained spectroscopy . the @xmath200 symbols indicate the 8 rrls from the second quest catalog with available spectroscopy that are included in our analysis . the large spatial substructures found among the fainter rrls ( see * ? ? ? * ) are indicated : the sgr stream , the globular cluster pal 5 and its tails , and the `` @xmath203 clump '' , which contains the vss . we have indicated with solid circles the rrls having velocities consistent with the vss @xcite . ] . the two significant groups found in our data are indicated by solid symbols , one of them have stars with very high positive velocities and the other one has stars with velocities around @xmath204 @xmath13 . ] lccccccl eso / la silla & 1.5 m & b&c & 600 & 3300- 5500 & 3.1 & 1/1 & 2001 jun 17 + kpno & wiyn & hydra & 600 & 7100 - 10000 & 2.9 & 5/3 & 2003 mar 11 + ctio & smarts 1.5 m & r - c & 600 & 3532 - 5300 & 4.3 & 42/28 & 2003 may 1 - 9 + ctio & smarts 1.5 m & r - c & 600 & 3532 - 5300 & 4.3 & 24/14 & 2004 feb - jun + ctio & smarts 1.5 m & r - c & 600 & 3532 - 5300 & 4.3 & 5/4 & 2006 jun + ctio & smarts 1.5 m & r - c & 600 & 3532 - 5300 & 4.3 & 4/2 & 2007 feb - apr + kpno & wiyn & hydra & 400 & 3500 - 6200 & 7.1 & 9/5 & 2007 feb 20 - 21 + kpno & wiyn & hydra & 600 & 7100 - 10000 & 2.9 & 3/3 & 2007 apr 27 , may 2 + [ tab - telescopes ] cccccrrc 201 & 2452711.7580 & 3600 & wiyn - r & 0.76 & 58 & 5 & + 201 & 2452765.6208 & 1200 & smarts & 0.02 & 44 & 16 & -1.95 + 203 & 2452765.6385 & 1200 & smarts & 0.02 & 13 & 16 & -1.40 + 212 & 2452078.4665 & 900 & eso & 0.56 & 88 & 11 & -2.11 + 212 & 2452760.6437 & 300 & smarts & 0.58 & 109 & 16 & -1.99 + [ tab - observations ] ccccccccccccccc 151 & 169.625160 & -0.764498 & 14.78 & c & 0.29370 & 2451611.7002 & 7.0 & 2 & 2 & 90 & 28 & 28 & -35 & -1.32 + 157 & 171.487530 & -0.161508 & 14.39 & ab & 0.70840 & 2451571.8154 & 5.9 & 2 & 2 & 142 & 22 & 22 & 22 & -1.67 + 160 & 172.957305 & -2.240574 & 15.43 & ab & 0.67619 & 2451582.7058 & 9.4 & 2 & 1 & 217 & & 20 & 93 & -2.52 + 161 & 173.844930 & -0.895030 & 15.40 & ab & 0.60699 & 2451255.6642 & 9.3 & 2 & 2 & 18 & 7 & 15 & -100 & -1.71 + [ tab - velocities ]

we study the behavior of the chern classes of graph hypersurfaces under the operation of deletion - contraction of an edge of the corresponding graph . we obtain an explicit formula when the edge satisfies two technical conditions , and prove that both these conditions hold when the edge is multiple in the graph . this leads to recursions for the chern classes of graph hypersurfaces for graphs obtained by adding parallel edges to a given ( regular ) edge . analogous results for the case of grothendieck classes of graph hypersurfaces were obtained in . both grothendieck classes and chern classes were used to define ` algebro - geometric ' feynman rules in . the results in this paper provide further evidence that the polynomial feynman rule defined in terms of the chern - schwartz - macpherson class of a graph hypersurface reflects closely the combinatorics of the corresponding graph . the key to the proof of the main result is a more general formula for the chern - schwartz - macpherson class of a transversal intersection ( see [ generals ] ) , which may be of independent interest . we also describe a more geometric approach , using the apparatus of ` verdier specialization ' .

1106.1447

[ [ section ] ] _ graph hypersurfaces _ are hypersurfaces of projective space associated with the parametric formulation of feynman integrals in scalar quantum field theories . the study of their geometry was prompted by certain conjectures concerning the appearance of multiple zeta values in the results of computation of feynman amplitudes , and has been the object of intense investigation ( see e.g. , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and many others ) . in this paper we study chern classes of graph hypersurfaces , from the point of view of deletion - contraction and multiple - edge formulas . [ [ section-1 ] ] ` algebro - geometric feynman rules ' are invariants of graphs which only depend on the isomorphism class of the corresponding hypersurfaces , and have controlled behavior with respect to unions . ( this definition of course captures only a very small portion of the quantum field theory feynman rules ; it would be very interesting to have examples of algebro - geometric feynman rules mirroring more faithfully their physical counterparts . ) matilde marcolli and the author note in @xcite that the classes of graph hypersurfaces in the grothendieck ring of varieties satisfy this basic requirement ; grothendieck classes of graph hypersurfaces have been studied rather thoroughly , given their relevance to the conjectures mentioned above . in the same paper we produced a different example of algebro - geometric feynman rules , with values in @xmath0 $ ] , based on the _ chern classes _ of graph hypersurfaces . the definition of this polynomial invariant @xmath1 will be recalled below ; its main interest lies in the fact that it carries intersection - theoretic information about the singularities of graph hypersurfaces . for example , the ( push - forward to projective space of the ) segre class of the singularity subscheme of a hypersurface may be recovered from the polynomial feynman rules . the milnor number of the hypersurface ( in its natural generalization to arbitrary hypersurfaces as defined by parusiski , @xcite ) is but one piece of the information carried by @xmath1 . the fact that the invariant satisfies the basic requirements of algebro - geometric feynman rules is proved in @xcite , theorem 3.6 , and is substantially less straightforward than the corresponding fact for grothendieck classes . grothendieck classes of graph hypersurfaces also satisfy a ` deletion - contraction ' relation : this fact has been pointed out by several authors , see e.g. , @xcite , @xcite , @xcite . the purpose of this article is to examine deletion - contraction relations for chern classes of graph hypersurfaces , in terms of the polynomial feynman rules mentioned above . as in @xcite , it is natural to expect the situation for chern classes to be substantially subtler than for classes in the grothendieck ring , and in fact our first guess concerning a double - edge formula for chern classes , based on somewhat extensive evidence computed for small graphs , turns out to be _ incorrect _ as stated in conjecture 6.1 in @xcite . nevertheless , we will be able to show here that these invariants do satisfy the expected general structure underlying multiple - edge formulas examined in @xcite . [ [ section-2 ] ] the _ graph hypersurface _ associated with a graph @xmath2 is the zero - set of the polynomial @xmath3 where @xmath4 ranges over the maximal spanning forests of @xmath2 . this is a homogeneous polynomial in variables @xmath5 corresponding to the edges @xmath6 of @xmath2 , and its zero set may be viewed in @xmath7 or @xmath8 , depending on the context . graph hypersurfaces are singular in all but the simplest cases , and in this article ( as in @xcite ) we employ the theory of _ chern - schwartz - macpherson _ ( csm ) classes . csm classes are defined for arbitrary varieties , and agree with the ordinary ( total homology ) chern class of the tangent bundle when evaluated on nonsingular varieties . the reader may refer to 2.2 - 3 of @xcite for a quick summary of this theory , which has a long and well - documented history . csm classes can be viewed as a generalization of the topological euler characteristic : indeed , the degree of the csm class of a variety _ is _ its euler characteristic , and to some extent csm classes maintain the same additive and multiplicative behavior of the euler characteristic . in this respect they are similar in flavor to the grothendieck class . they also offer a direct measure of ` how singular ' a variety is , by comparison with other characteristic classes of singular varieties , see e.g. 2.2 in @xcite and references therein . csm classes are in fact defined for _ constructible functions _ on a variety ( @xcite ) , and what we call the csm class of @xmath9 is the class @xmath10 of the constant function @xmath11 . as our objects of study are hypersurfaces of projective space , we view csm classes as elements of the chow group of projective space , i.e. , as polynomials in the hyperplane class . the polynomial feynman rules mentioned above are closely related to the csm class of the _ complement _ of a graph hypersurface @xmath12 : if a graph @xmath2 with @xmath13 edges is not a forest , then the polynomial feynman rules @xmath1 are determined by the relation @xmath14\quad,\ ] ] where @xmath15 denotes the hyperplane class ; see prop . 3.7 in @xcite . [ [ section-3 ] ] in graph theory , _ deletion - contraction _ formulas express invariants for a graph @xmath2 directly in terms of invariants for the ` deletion ' graph @xmath16 obtained by removing an edge @xmath6 , and the ` contraction ' graph @xmath17 obtained by contracting the same edge . _ tutte - grothendieck _ invariants are the most general invariants with controlled behavior with respect to deletion - contraction ; an impressive list of important graph invariants are of this kind , ranging from chromatic polynomials to partition functions for potts models . in fact , these invariants may be viewed as ` feynman rules ' in a sense closely related to the one adopted in @xcite , see prop . 2.2 in @xcite . in @xcite we show that the invariant arising from the grothendieck class satisfies a weak form of deletion - contraction ( which involves ` non - combinatorial ' terms ) ; and that enough of this structure is preserved to trigger combinatorial ` multiple - edge ' formulas . more precisely , let @xmath18 $ ] denote the grothendieck class of the complement of the affine graph hypersurface , and denote by @xmath19 the graph obtained by doubling the edge @xmath6 in @xmath2 . if @xmath6 is neither a bridge nor a looping edge , then ( @xcite , proposition 5.2 ) @xmath20 where @xmath21 is the class of @xmath22 . note that ( * ) holds without further requirements on @xmath6 . ( simpler formulas hold in case @xmath6 is a bridge or a looping edge . ) [ [ section-4 ] ] as we will show in this paper , the situation concerning the polynomial invariant @xmath1 recalled above is somewhat different . as in the case of the grothendieck class , this invariant does not satisfy on the nose a deletion - contraction relation ( this was already observed in @xcite , prop . 3.2 ) . unlike in the case of the invariant @xmath23 , however , even a weaker non - combinatorial form of deletion - contraction only holds under special hypotheses on the edge @xmath6 . the main result of this article is the determination of conditions on a pair @xmath24 such that a sufficiently strong deletion - contraction relation ( and corresponding consequences , such as multiple - edge formulas ) holds for the edge @xmath6 of @xmath2 . these conditions are presented in [ condis ] ; somewhat surprisingly , they appear to hold for ` many ' graphs . one of them can be formulated as follows . assume that @xmath6 is neither a bridge nor a looping edge of @xmath2 . we may consider the polynomial @xmath25 for the deletion @xmath16 . the condition is then that @xmath26 _ belongs to the jacobian ideal of @xmath25 . _ the smallest counterexample to this requirement appears to be the graph with respect to the vertical edge . we find it surprising that this condition is satisfied as often as it is . the second condition is more technical . see [ condis ] for a discussion of both conditions . once a ( weak ) deletion - contraction formula is available , one should expect combinatorial multiple - edge formulas to hold . and indeed , we will prove the following analogue of ( * ) for the chern class feynman rules : [ double ] if the conditions on @xmath27 mentioned above are satisfied , then @xmath28 this is the formula that was proposed in @xcite , conjecture 6.1 , on the basis of many examples computed explicitly in @xcite . however , the additional conditions on @xmath27 had not been identified at the time ; the formula proposed in _ loc . _ for the class of a triangle with doubled edges is incorrect , and will be corrected here in example [ doubletriangex ] . [ [ section-5 ] ] as mentioned above , we do not have a sharp combinatorial characterization on @xmath27 ensuring that the technical hypotheses needed for theorem [ double ] are satisfied . however , there is one important case in which we are able to prove that these conditions are indeed satisfied : the conditions hold if @xmath6 is itself a multiple edge , i.e. , if the endpoints of @xmath6 are adjacent in @xmath16 . thus , theorem [ double ] implies that the polynomial feynman rules satisfy essentially the same recursive structure for multiple - edge formulas that is studied in general in @xcite , 6 . if @xmath6 is neither a bridge nor a looping edge of @xmath2 , then @xmath29 in a sense , this recursion is ` nicer ' than the corresponding one for grothendieck classes ( see the comments following lemma [ recurs ] ) . [ [ section-6 ] ] this paper is organized as follows . in [ condis ] we state precisely the technical conditions mentioned above , in the case of graph hypersurfaces , providing a few simple examples to illustrate them . this is also done in the hope that others may identify sharp combinatorial versions of these conditions . we prove ( lemma [ mainlemma ] ) that the conditions hold for @xmath24 if @xmath6 has parallel edges in @xmath2 , and describe one class of examples in which the conditions do not ( both ) hold . in [ generals ] we discuss a formula for the csm class of a transversal intersection , needed for the application to graph hypersurfaces presented here ; this section can be read independently of the rest of the paper . in [ particulars ] we apply these formulas to the case of graph hypersurfaces , obtaining the deletion - contraction relation ( theorem [ delconthm ] ) . in [ multedges ] we apply this relation to obtain multiple - edge formulas as mentioned above ( theorem [ goodform ] , lemma [ recurs ] ) . in [ speci ] we describe a different and more ` geometric ' ( but in practice less applicable ) approach to the main deletion - contraction formula , using _ verdier s specialization . _ _ acknowledgment . _ i thank matilde marcolli for stimulating my interest in graph hypersurfaces through our previous joint work , and for the hospitality at caltech , where this paper was written . i also thank don zagier for a conversation concerning a technical point in [ multedges ] . we work over an algebraically closed field @xmath30 of characteristic zero . [ [ condintro ] ] as in [ intro ] , @xmath2 denotes a finite graph , with @xmath13 edges ; we allow looping edges as well as parallel edges . we associate with each edge @xmath6 a variable @xmath5 , and we consider the _ graph polynomial _ @xmath31 where @xmath4 ranges over the maximal spanning forests of @xmath2 . ( note : according to this definition , the polynomial for a graph is the product of the polynomials for its connected components . ) we denote by @xmath32 the projective hypersurface defined by @xmath33 . we present in this section two conditions for a pair @xmath27 , where @xmath6 is an edge of @xmath2 , encoding some geometric features of @xmath32 . finding more transparent , combinatorial versions of these conditions is an interesting problem . we will say that an edge @xmath6 of @xmath2 is _ regular _ if @xmath6 is neither a bridge nor a looping edge , and further @xmath16 is not a forest . we will essentially always assume that @xmath6 is regular on @xmath2 ; non - regular edges are easy to treat separately ( see e.g. , [ trivca ] ) . if @xmath6 is a regular edge of @xmath2 , then @xmath34 this is well - known , and easily checked . as @xmath25 is not a forest , @xmath35 ; in this case a point @xmath36 of @xmath32 is determined by setting all variables except @xmath5 to @xmath37 . we denote by @xmath38 the blow - up of @xmath32 at @xmath36 , and by @xmath39 the exceptional divisor of this blow - up . the variety @xmath38 may be realized as a hypersurface in the blow - up of @xmath7 at @xmath36 . denoting by @xmath40 the exceptional divisor of this latter blow - up , @xmath39 is the intersection @xmath41 . heuristically , the conditions we will present below amount to requiring this intersection to be sufficiently transversal . [ [ condi ] ] assume @xmath6 is regular . both @xmath42 and @xmath43 are hypersurfaces of a projective space @xmath44 with homogeneous coordinates corresponding to the edges of @xmath2 other than @xmath6 . the first condition on the pair @xmath27 may be expressed as a relation between them : @xmath45 here , @xmath46 denotes the ideal of partial derivatives of @xmath25 , defining the _ singularity subscheme _ @xmath47 of @xmath42 . the condition essentially ( that is , up to saturating @xmath46 ) amounts to requiring @xmath47 to be a subscheme of @xmath43 . it is of course easy to verify whether this condition holds on any given graph , by employing a symbolic manipulation package such as macaulay2 ( @xcite ) . the following examples illustrate a few cases , showing in particular that the condition depends on global features of the graph . [ ciexample ] for the graph condition i is satisfied with respect to all edges . for the wheel condition i is satisfied with respect to the spokes , and it is not satisfied with respect to the rim edges . condition i is satisfied with respect to all edges for the graph and with respect to all edges _ except @xmath6 _ for the graph this is the ` smallest ' example not satisfying condition i. @xmath48 [ [ condii ] ] the second condition we will consider is more technical than condition i. let @xmath6 be a regular edge on @xmath2 , and consider the blow - up introduced in [ condintro ] . for any point @xmath49 of @xmath50 , let @xmath51 be the ideal of @xmath52 at @xmath49 , and denote by @xmath53 an equation for @xmath39 at @xmath49 . the second condition we must consider on the pair @xmath27 is @xmath54 again , checking this condition on any given graph is possible with a tool such as macaulay2 , although computing power will limit the size of graphs that can be analyzed in practice . ( note that , by artin - rees , only finitely many @xmath55 need be checked . ) [ ciiexample ] condition ii is verified for the graph with respect to all edges , while it does _ not _ hold for with respect to @xmath6 . ( both assertions may be verified with macaulay2 ; also , see [ disjoinable ] for a generalization . ) @xmath48 [ [ section-7 ] ] we would be interested in purely combinatorial interpretations in terms of @xmath2 and @xmath6 of the conditions presented in [ condi ] and [ condii ] ; it is not even clear to us that such sharp characterizations exist . however , we can provide one combinatorial situation in which both conditions are satisfied , and this situation is at the root of the application to multiple - edge formulas in [ multedges ] . [ mainlemma ] let @xmath2 be a graph , and let @xmath6 be a regular edge that has parallel edges in @xmath2 . then both conditions i and ii are verified for @xmath24 . let @xmath56 be an edge parallel to @xmath6 . we first assume that @xmath56 is not a bridge in @xmath16 . then @xmath57 where @xmath58 and @xmath59 . _ condition i. _ among the partials of @xmath25 is @xmath60 . since @xmath25 is homogeneous ( and we are in characteristic zero ) , it is in the ideal of partials @xmath46 . it follows that so is @xmath61 , and as a consequence @xmath62 as needed . _ condition ii . _ since the condition only depends on the part of @xmath38 over @xmath36 , it is unaffected by analytic changes of coordinates at @xmath36 . first , we can center an affine chart at @xmath36 by setting @xmath63 , and the equation of @xmath32 in this chart is @xmath64 next , we can set @xmath65 , i.e. , @xmath66 ; this does not affect the geometry of @xmath32 near @xmath36 ( where @xmath67 ) . in coordinates @xmath68 , the equation for @xmath32 is @xmath69 this equation is _ homogeneous , _ so @xmath32 is a cone with vertex at @xmath36 in these coordinates . it is then clear that condition ii holds : the equations of @xmath38 in the standard charts do not depend on the variable @xmath53 defining the exceptional divisor , so at each @xmath70 the ideal @xmath51 has a set of generators independent of @xmath53 , and it follows that @xmath53 is a non - zero - divisor modulo @xmath71 for all @xmath55 . this concludes the proof in the case in which @xmath56 is not a bridge in @xmath16 . if @xmath56 is a bridge in @xmath16 , then @xmath72 , and one verifies easily that @xmath73 since @xmath74 in this case , it is immediate that @xmath75 . further , @xmath32 has equation @xmath76 in the affine chart @xmath63 , and near @xmath36 . again this is a cone with vertex at @xmath36 , so condition ii holds by the same argument used above . [ [ transv ] ] we now discuss more in detail the geometric meaning of the two conditions presented above . this will also clarify the sense in which the two conditions may be interpreted as transversality statements . [ condiex ] if condition i is satisfied , then @xmath77 . recall that @xmath78 denotes the blow - up of @xmath9 at the point @xmath36 obtained by setting @xmath5 to @xmath79 and all other coordinates @xmath80 to @xmath37 ( where @xmath13 is the number of edges of @xmath2 , and @xmath6 is assumed to be a regular edge ) . working in the affine chart @xmath81 centered at @xmath36 , @xmath9 has equation @xmath82 where the summands are homogeneous polynomials of degree @xmath83 and @xmath84 respectively , with @xmath85 . we can cover the blow - up of @xmath81 at @xmath36 with standard coordinate patches ; in one of them we have coordinates @xmath86 so that the blow - up map is given by @xmath87 the equation of @xmath78 in this chart is then @xmath88 and @xmath89 is the equation of the exceptional divisor in this chart ; thus , @xmath39 has ideal @xmath90 in this chart . note that the exceptional divisor @xmath40 of the blow - up of @xmath81 is a projective space @xmath44 . the above computation ( together to the same in the other patches ) shows that @xmath91 , a hypersurface in @xmath92 . this computation also shows that @xmath93 has ideal @xmath46 . in the representative patch chosen above , this is @xmath94 on the other hand , in the same patch , @xmath95 has ideal @xmath96 and hence @xmath50 has ideal @xmath97 comparing ( @xmath98 ) and ( @xmath99 ) ( and the analogous ideals in all patches covering the blow - up ) , we see that the ideals agree if @xmath100 , that is , if condition i holds . this verifies claim [ condiex ] . the picture we have in mind is that of the nonsingular exceptional divisor @xmath101 intersecting @xmath78 along @xmath39 . according to claim [ condiex ] , condition i implies that the intersection @xmath102 is only singular along the intersection of @xmath40 with the singularity subscheme of @xmath38 , as would be expected if @xmath40 met @xmath38 transversally . @xmath48 in order to interpret condition ii , we have to introduce the blow - up @xmath103 of @xmath38 along its singularity subscheme @xmath95 . in this blow - up we may consider two subschemes : the proper transform @xmath104 ( isomorphic to the blow - up of @xmath39 along @xmath105 ) and the inverse image @xmath106 . [ condiiex ] if condition ii holds , then @xmath107 . indeed , assume condition ii holds . then letting @xmath108 , @xmath109 denote respectively the ideal sheaves of @xmath95 and @xmath39 in @xmath38 , condition ii implies @xmath110 for @xmath111 , and hence the natural morphism @xmath112 is an isomorphism for @xmath111 . it follows that the natural inclusion @xmath113 is an equality , verifying claim [ condiiex ] . by the same token , the proper transform of the exceptional divisor @xmath40 equals its inverse image in the blow - up along @xmath95 this is the behavior expected if the ( nonsingular ) hypersurface @xmath40 meets the center of a blow - up transversally , so condition ii , like condition i , appears to express a measure of transversality of the intersection of @xmath40 with @xmath78 . @xmath48 [ inotii ] the two conditions differ : for example , condition ii does not hold for the second graph displayed in example [ ciiexample ] with respect to edge @xmath6 , while condition i does hold in this case . however , we do not know of examples of graphs for which condition ii holds and condition i does not . it is conceivable that such examples exist ( cf . [ sharps ] ) . @xmath48 [ [ disjoinable ] ] finally , we discuss one case in which conditions i and ii do not both hold . if a graph is obtained by taking the union of two graphs @xmath114 , @xmath115 , joined at a vertex , then its graph polynomial equals the product @xmath116 . if neither @xmath114 nor @xmath115 is a forest , we say that the graph is ` disjoinable ' ; note that the graph hypersurface of a disjoinable graph is singular in codimension @xmath79 ( we do not know whether the converse holds . ) let @xmath2 be a graph such that @xmath32 is nonsingular in codimension @xmath79 . let @xmath6 be an edge such that @xmath16 is disjoinable . then at least one of conditions i and ii fails for @xmath27 . for example , the second graph drawn in example [ ciiexample ] is of this type , with respect to the bottom edge . ( the first is not , since while removing the bottom edge does produce the join of two graphs , one of these is a tree . ) if condition i does not hold , we are done ; so we may assume that condition i holds , and we will show that condition ii does not hold in this case . we use notation as in the discussion following the statement of claim [ condiex ] . after setting @xmath5 to @xmath79 , we have @xmath117 by assumption , where @xmath118 and @xmath61 are graph polynomials of degree @xmath119 . in the chart of @xmath120 with coordinates @xmath86 , the equation of @xmath78 is @xmath121 and @xmath39 has ideal @xmath122 in particular , the singularity subscheme @xmath123 of @xmath39 contains the locus @xmath124 with ideal @xmath125 . as we are assuming that condition i holds , we have that @xmath126 ( cf . claim [ condiex ] ) ; in particular , @xmath127 . on the other hand , note that @xmath124 has codimension @xmath79 in @xmath39 , hence codimension @xmath128 in @xmath78 ; since @xmath9 is nonsingular in codimension @xmath79 and @xmath129 , @xmath124 must consist of a collection of components of @xmath52 . but then @xmath39 contains components of @xmath52 , and it follows that the condition in claim [ condiiex ] is not verified . hence @xmath27 does not satisfy condition ii . [ [ section-8 ] ] in this section we discuss a formula expressing the chern - schwartz - macpherson class of the intersection of a variety @xmath9 with a hypersurface @xmath40 , in terms of of @xmath130 and of the class of @xmath40 . this section can be read independently of the rest of the paper . our template is the transversal intersection of nonsingular varieties . let @xmath131 be a nonsingular variety , and let @xmath40 , @xmath9 be nonsingular subvarieties of @xmath131 . assume that @xmath40 is a hypersurface , and that @xmath40 and @xmath9 meet transversally . in this case @xmath132 is a nonsingular hypersurface of @xmath9 , and @xmath133 , hence ( harmlessly abusing notation ) @xmath134 = \frac{c(tx)}{c(n_dv)}\cap [ d\cap x ] = \frac{d}{1+d } \cap c(tx)\cap [ x]\quad,\ ] ] i.e. , @xmath135 ( this equality holds in @xmath136 . ) we are interested in generalizing this formula to the case in which @xmath9 is possibly singular . [ [ section-9 ] ] the key question is , of course , what ` transversal ' should mean in the singular case . the conditions presented in [ condis ] are precisely concocted to make this requirement precise . we assume @xmath131 is a nonsingular variety , @xmath40 and @xmath9 are reduced hypersurfaces of @xmath131 , and @xmath40 is nonsingular . we denote by @xmath137 the singularity subscheme of @xmath9 , defined by the ideal of partial derivatives of local equations . we denote by @xmath138 the blow - up along @xmath137 . further , @xmath139 denotes the proper transform of @xmath40 in @xmath140 . [ adj ] assume that 1 . @xmath141 ; 2 . @xmath142 . then @xmath143 in @xmath136 . our main tool is theorem i.4 in @xcite , which relates the chern - schwartz - macpherson class of a hypersurface with the segre class of its singularity subscheme . viewing @xmath132 as a hypersurface in @xmath40 , this result yields @xmath144 ( with notation as in @xcite , 1.4 ) ; by the same token , @xmath145 it follows that @xmath146 therefore , the formula stated in theorem [ adj ] would follow from the equality @xmath147 and hence from @xmath148 and finally from @xmath149 ( in @xmath150 . this reduces the proof of theorem [ adj ] to the following claim : under hypotheses ( 1 ) and ( 2 ) from the statement of theorem [ adj ] , @xmath151 in @xmath152 . to prove this , consider the blow - up of @xmath131 along @xmath137 , with exceptional divisor @xmath153 ; the proper transform of @xmath40 may be viewed as the blow - up of the latter along @xmath154 : @xmath155^(.6)\sigma \ar@{^(->}[rr ] & & { { \widetilde{v } } } \ar[dd]^\rho \\ { { \widetilde{d}}}\cap f\ar[dd]_{\sigma ' } \ar@{^(->}[rr ] \ar@{^(->}[ur ] & \ar[d ] & { { \widetilde{d } } } \ar[dd ] \ar@{^(->}[ur ] \\ & \partial x \ar@{^(-}[r ] & \ar[r ] & v \\ d\cap \partial x \ar@{^(->}[rr ] \ar@{^(->}[ur ] & & d \ar@{^(->}[ur ] } \ ] ] by the birational invariance of segre classes , @xmath156\right)\\ s(d\cap \partial x , d ) & = \sigma ' _ * s({{\widetilde{d}}}\cap f,{{\widetilde{d } } } ) = \sigma'_*\left(\frac{f}{1+f}\cap [ { { \widetilde{d}}}]\right)\quad.\end{aligned}\ ] ] by hypothesis ( 2 ) , @xmath157 , hence @xmath158 is represented by @xmath159 . therefore , by the projection formula ( @xcite , proposition 2.3 ( c ) ) @xmath160\right ) = \sigma'_*\left(\frac{[{{\widetilde{d}}}\cap f]}{1+f}\right ) = \sigma'_*\left(\sigma^ * d\cdot \frac{f}{1+f}\cap [ { { \widetilde{v}}}]\right ) = d\cdot s(\partial x , v)\quad.\ ] ] thus , @xmath161 finally , @xmath162 by hypothesis ( 1 ) , and the claim follows . this concludes the proof of theorem [ adj ] . [ [ sharps ] ] theorem [ adj ] is sharp , in the sense that both hypotheses are necessary for the stated formula to hold . to see that the first hypothesis is needed , let @xmath9 be a nonsingular quadric in @xmath163 , and let @xmath40 be a hyperplane tangent to @xmath9 . then @xmath132 is singular , while @xmath164 ; in particular , ( 1 ) is not satisfied . working in the ambient @xmath165 $ ] for convenience , we have @xmath166 + 3[{{\mathbb{p}}}^0]\ ] ] and @xmath167 + 4[{{\mathbb{p}}}^1 ] + 4[{{\mathbb{p}}}^0 ] ) = 2[{{\mathbb{p}}}^1]+2[{{\mathbb{p}}}^0]\ ] ] ( where @xmath15 is the hyperplane class ) . therefore , the stated formula does not hold . note that the second hypothesis holds ( trivially ) in this case , since @xmath164 . to see that the second hypothesis is also necessary , let @xmath168 be a quadric cone , and let @xmath40 be a general hyperplane through the vertex . then ( 2 ) fails as @xmath40 contains @xmath137 . again working in @xmath163 , we have @xmath166 + 3[{{\mathbb{p}}}^0]\ ] ] and @xmath167 + 4[{{\mathbb{p}}}^1 ] + 3[{{\mathbb{p}}}^0 ] ) = 2[{{\mathbb{p}}}^1]+2[{{\mathbb{p}}}^0]\ ] ] as before . ( the coefficient of @xmath169 $ ] in @xmath130 is in fact irrelevant to this computation . ) note that @xmath170 the vertex of the cone , and in particular @xmath171 , so that the first hypothesis does hold in this case . [ [ section-10 ] ] now we return to the case of graph hypersurfaces . recall that @xmath2 denotes a graph with @xmath13 edges ; @xmath6 denotes a regular edge of @xmath2 ; @xmath172 is the corresponding hypersurface . as @xmath16 is not a forest , then the point @xmath36 obtained by setting all coordinates except @xmath5 to zero is a point of @xmath32 . we consider the blow - up @xmath173 at @xmath36 , and the blow - up @xmath38 of @xmath32 at @xmath36 realized as the proper transform of @xmath32 in @xmath174 . we denote @xmath40 the exceptional divisor in @xmath174 , so that @xmath102 is the exceptional divisor in @xmath38 . applying theorem [ adj ] to this situation yields : [ maincor ] assume conditions i and ii from [ condis ] hold for @xmath27 . then @xmath175 indeed , as observed in [ transv ] , conditions i and ii imply the hypotheses of theorem [ adj ] . [ [ gende ] ] we are ready to prove a deletion - contraction formula for chern classes of graph hypersurfaces , subject to the conditions presented in [ condis ] . we can view the blow - up @xmath174 as the graph of the projection @xmath176 centered at @xmath36 : @xmath177 \ar[dr ] \\ { { \mathbb{p}}}^{n-1 } \ar@{-->}[rr ] & & { { \mathbb{p}}}^{n-2 } } \ ] ] and likewise for @xmath38 : @xmath178_\nu \ar[dr]^\pi \\ x_\gamma \ar@{-->}[rr ] & & { { \mathbb{p}}}^{n-2 } } \ ] ] this situation was briefly described in @xcite , end of 4 : @xmath179 is the blow - up of @xmath32 at @xmath36 , while @xmath180 realizes @xmath38 as the blow - up of @xmath44 along @xmath181 ; the fibers of @xmath180 are points away from @xmath182 , and @xmath183 over over points of @xmath181 . the exceptional divisor @xmath39 of @xmath179 is a copy of @xmath42 in @xmath92 ( as was verified in [ transv ] ) . the restriction of @xmath184 to @xmath40 gives an isomorphism @xmath185 . we are aiming for formulas involving the polynomial ` feynman rules ' @xmath1 carrying the information of the csm class of @xmath32 . denote by @xmath15 the hyperplane class in @xmath7 . our main objective is essentially a formula for the polynomial @xmath186 encoding the degrees of the terms in @xmath187 ; @xmath1 may be computed easily from this polynomial . [ [ section-11 ] ] we let @xmath15 , resp . @xmath188 , denote the class in @xmath7 , resp . the chow group of @xmath38 is generated by @xmath189 and the class @xmath39 of the exceptional divisor . [ tech1 ] with notation as above , @xmath190 by the functoriality property of csm classes , @xmath191\quad.\end{aligned}\ ] ] using that @xmath91 , and applying the projection formula , this gives @xmath192 with the last equality due to the fact that push - forwards preserve degrees , and condensing the summation into a rational function for notational convenience . now i claim that @xmath193 : indeed , this may be verified by realizing @xmath15 as the class of a general hyperplane containing @xmath36 . also , note that @xmath194 : realize @xmath15 as the class of a hyperplane _ not _ containing @xmath36 to verify this . therefore , @xmath195 the statement follows . [ tech2 ] @xmath196 by the projection formula , and since push - forwards preserve degree , @xmath197 applying the functoriality of csm classes : @xmath198 + { { c_{\text{sm}}}}(x_{\gamma\smallsetminus e } \cap x_{\gamma / e})\quad,\end{aligned}\ ] ] where we have used the description of the fibers of @xmath180 recalled in [ gende ] . as @xmath199 , the statement follows . [ [ section-12 ] ] combining lemma [ tech1 ] and [ tech2 ] , we obtain that _ if @xmath6 is a regular edge on @xmath2 , then _ @xmath200 the more technical conditions i and ii presented in [ condis ] play no role in this statement . they become relevant in evaluating the last term , @xmath201 here we have replaced @xmath39 by @xmath40 ( since @xmath202 , and in particular @xmath40 restricts to the class of @xmath39 on @xmath38 ) , and @xmath203 with the corresponding hyperplane class @xmath188 on @xmath92 . the class @xmath204 is supported on @xmath91 . without further information on @xmath6 and @xmath2 , it does not seem possible to express this class in more intelligible terms . [ tech3 ] assume @xmath27 satisfies conditions i and ii . then @xmath205 by corollary [ maincor ] , @xmath206 here we have identified @xmath207 with @xmath208 , and used the fact that the class of the exceptional divisor @xmath40 restricts to @xmath209 . the statement is obtained by applying mindless manipulations ( and noting @xmath210 : @xmath211 with the stated result . [ [ section-13 ] ] collecting what we have proved at this point : [ delconprop ] let @xmath2 be a graph with @xmath13 edges , and let @xmath6 be a regular edge of @xmath2 . assume @xmath27 satisfies the conditions given in [ condis ] . then @xmath212 this statement improves considerably once it is expressed in terms of the ` polynomial feynman rules ' introduced in @xcite , 3 ; we take this as a further indication that the polynomial captures interesting information about @xmath2 . the polynomial essentially evaluates the csm class of the _ complement _ of @xmath32 in projective space . we will denote the polynomial corresponding to @xmath2 by @xmath213 , since it depends directly on the graph hypersurface ( this makes it an _ algebro - geometric _ feynman rule ) , and since it can be defined for any subset of projective space . [ poly ] if @xmath2 is not a forest and has @xmath13 edges , then @xmath214 this is obtained from proposition 3.7 of @xcite , by applying simple manipulations . if @xmath2 is a forest , then @xmath32 is empty , and the corresponding polynomial is a power of @xmath215 , cf . proposition 3.1 in @xcite . consistently with the expression in lemma [ poly ] , we set @xmath216 for every nonempty subscheme @xmath217 , where @xmath188 denotes the hyperplane class . note that @xmath218 depends on the dimension of the space containing @xmath124 ; this is always clear from the context . if @xmath219 is empty , we set @xmath220 . [ delconthm ] let @xmath6 be a regular edge of @xmath2 . assume @xmath27 satisfies both conditions i and ii given in [ condis ] . then @xmath221 this is the form taken by the formula in proposition [ delconprop ] , once it is written using the notation recalled above . on the right - hand side , both @xmath42 and @xmath182 are viewed as subschemes of @xmath44 . the deletion - contraction formula of theorem [ delconthm ] holds also if @xmath16 is a forest , provided that @xmath222 and @xmath223 are both taken to equal @xmath224 . [ myst ] differentiating the formula in theorem [ delconprop ] and setting @xmath225 to @xmath37 gives @xmath226 the value of the derivative at @xmath37 equals the euler characteristic of the complement ( @xcite , proposition 3.1 ) ; and as @xmath16 is not a forest , then @xmath227 . in this case , @xmath228 i.e. , @xmath229 remarkably , this formula holds as soon as @xmath6 is a regular edge on @xmath2 , as verified in @xcite , ( 3.20 ) . in fact ( _ loc . theorem 3.8 ) this formula follows from an analogous formula at the level of grothendieck classes which holds if @xmath6 is a regular edge on @xmath2 , regardless of whether conditions i and ii are verified . we find it very mysterious that these conditions should affect the csm classes involved in the deletion - contraction formula , but not affect their zero - dimensional terms . @xmath48 [ [ section-14 ] ] while the argument proving the deletion - contraction formula requires the technical conditions i and ii to hold for @xmath27 , there could be a legitimate doubt that the formula itself may hold for more general edges ; after all , deletion - contraction for grothendieck classes does hold in a more general situation ( cf . remark [ myst ] ) . the example that follows shows that the formula does not necessarily hold if the second condition fails . [ iifails ] condition ii fails for the graph with respect to the bottom edge @xmath6 ( example [ ciiexample ] ; also cf . [ disjoinable ] ) . labeling the edges by coordinates as indicated , the graph polynomial is @xmath230 the corresponding hypersurface @xmath32 is singular along two nonsingular conics meeting at the point @xmath231 . the blow - up @xmath38 is singular along the proper transforms of these two conics , and along a curve contained in the exceptional divisor . as the exceptional divisor contains a component of @xmath95 , it is clear that condition ii is not satisfied , cf . claim [ condiiex ] . it is equally straightforward to verify that condition i _ is _ satisfied in this case . since @xmath32 is nonsingular in codimension @xmath79 , its codimension-@xmath37 and @xmath79 terms must agree with the chern class of its virtual tangent bundle , i.e. , with the class for a nonsingular hypersurface of degree @xmath232 in @xmath233 : @xmath234 + 6[{{\mathbb{p}}}^2]+\dots\quad.\ ] ] this observation suffices to determine @xmath235 deletion and contraction : have polynomials @xmath236 we have @xmath237 this is easy to obtain directly , as @xmath42 consists of the union of two planes in @xmath163 ; and it also follows from the formularium in proposition 3.1 of @xcite , and theorem 3.6 of _ loc . _ as for @xmath181 , this is easily checked to consist of three lines in @xmath163 , meeting at two points . it follows that @xmath238 + 2[{{\mathbb{p}}}^0])-2[{{\mathbb{p}}}^0]=3[{{\mathbb{p}}}^1 ] + 4[{{\mathbb{p}}}^0]$ ] , and hence @xmath239 thus , we see that @xmath240 verifying that the formula in theorem [ delconthm ] need not hold if condition ii fails . a more thorough analysis shows that @xmath241 , so that the underlined coefficient is the only discrepancy . in fact , this may be checked by using theorem [ delconthm ] , using deletion - contraction with respect to a diagonal edge @xmath242 ; both conditions i and ii hold in this case by lemma [ mainlemma ] . deletion and contraction are : the reader can check that the graph hypersurface for the deletion is a quadric cone in @xmath163 , and hence @xmath243 + 4[{{\mathbb{p}}}^1]+3[{{\mathbb{p}}}^0]$ ] . the intersection @xmath244 consists of the union of a nonsingular conic and a line meeting at a point , hence @xmath245 + 3[{{\mathbb{p}}}^0]$ ] . this yields @xmath246 and hence @xmath247 as claimed , by theorem [ delconthm ] . @xmath48 as pointed out in remark [ inotii ] , we do not know an example for which the first condition fails while the second one holds . we list the relevant classes in an example for which both conditions fail . [ ifails ] condition i fails for the graph with respect to the vertical edge @xmath6 ( example [ ciexample ] ) . labeling the edges by coordinates as indicated , the graph polynomial is @xmath248 and the corresponding @xmath32 is a hypersurface of degree @xmath249 in @xmath250 . the computation of the terms needed to verify the formula in theorem [ delconthm ] for this case is more involved than in example [ iifails ] , and we omit the details . ( the macaulay2 procedure accompanying @xcite was used for these computations . ) we obtain : @xmath251 + 12[{{\mathbb{p}}}^4]+26[{{\mathbb{p}}}^3]+29[{{\mathbb{p}}}^2]+21[{{\mathbb{p}}}^1 ] + 7[{{\mathbb{p}}}^0]\quad,\ ] ] yielding @xmath252 as for deletion and contraction : @xmath253 + 9[{{\mathbb{p}}}^3]+14[{{\mathbb{p}}}^2 ] + 14[{{\mathbb{p}}}^1]+7[{{\mathbb{p}}}^0 ] \\ { { c_{\text{sm}}}}(x_{\gamma\smallsetminus e}\cap x_{\gamma / e } ) = 6[{{\mathbb{p}}}^3]+10[{{\mathbb{p}}}^2]+13[{{\mathbb{p}}}^1]+7[{{\mathbb{p}}}^0]\end{gathered}\ ] ] from which @xmath254 and therefore @xmath255 this differs from @xmath256 by @xmath257 , and shows that the formula in theorem [ delconthm ] need not hold if condition i is not satisfied . ( but note that condition ii also fails in this example . ) @xmath48 [ [ trivca ] ] if @xmath6 is _ not _ a regular edge of @xmath2 , then the corresponding deletion - contraction formulas are trivial consequences of properties of @xmath213 listed in @xcite , 3 ( especially proposition 3.1 ) . * if @xmath6 is a bridge in @xmath2 , then @xmath258 * if @xmath6 is a looping edge in @xmath2 , then @xmath259 if @xmath6 is a looping edge , then @xmath260 , and hence @xmath261 . thus , the formula given above matches the formula obtained by applying theorem [ delconthm ] . the formula in theorem [ delconthm ] is also trivially satisfied if @xmath16 is a forest . the formula for bridges has a transparent geometric explanation . if @xmath6 is a bridge , then @xmath262 ; thus @xmath32 is a cone over @xmath42 , and the formula given above follows easily from this fact . [ [ section-15 ] ] deletion - contraction formulas may be used to obtain formulas for the operation of ` multiplying edges ' , i.e. , inserting edges parallel to a given edge of a graph . this is carried out in @xcite , 5 , for the case of the grothendieck class . as in theorem [ delconthm ] , the deletion - contraction formula involves a ` non - combinatorial ' term ( the grothendieck class of the intersection @xmath263 ) . by virtue of a propitious cancellation , the resulting formula for doubling an edge only relies on combinatorial data : @xmath264 ( @xcite , proposition 5.2 ) , provided @xmath6 is a regular edge on @xmath2 , and denoting by @xmath19 the graph obtained by adding an edge parallel to @xmath6 . in this section we show that a similar situation occurs for the csm invariant . again , the deletion - contraction formula ( theorem [ delconthm ] ) involves a summand which we are not able to interpret directly in combinatorial terms ( that is , @xmath222 ) ; and again a fortunate cancellation leads to a purely combinatorial formula for doubling edges . we will prove : [ doubled ] let @xmath2 be a graph , and let @xmath6 be a regular edge of @xmath2 such that @xmath27 satisfies conditions i and ii of [ condis ] . denote by @xmath19 the graph obtained by doubling the edge @xmath6 . then @xmath265 ( here we write @xmath1 for @xmath213 , etc . ) by lemma [ mainlemma ] , this formula applies in particular if @xmath6 is already multiple in @xmath2 . this fact will be used in the proof of theorem [ doubled ] , and will lead to multiple - edge formulas in [ multiples ] . the formula also holds if @xmath6 is a looping edge , interpreting @xmath266 to be @xmath37 in this case ( cf . [ trivca ] ) . [ [ section-16 ] ] _ proof of theorem [ doubled ] . _ by theorem [ delconthm ] , under the hypothesis of the theorem , we have @xmath267 where @xmath242 denotes the edge parallel to @xmath6 in @xmath19 . indeed , the first formula holds as @xmath27 satisfies conditions i and ii by hypothesis , and the second formula holds since @xmath268 satisfies conditions i and ii by lemma [ mainlemma ] ; note that @xmath269 . the theorem will be obtained by comparing the two intersections @xmath270 and @xmath271 ( where @xmath272 number of edges of @xmath2 ) . let @xmath5 , @xmath273 be the variables corresponding to the two parallel edges @xmath6 , @xmath242 in @xmath19 , and let @xmath274 be the variables corresponding to the other edges . we have @xmath275 the graph @xmath276 may be obtained by attaching a looping edge marked @xmath6 at the vertex obtained by contracting @xmath6 in @xmath2 : as a consequence , @xmath277 and @xmath278 , @xmath279 have ideals @xmath280 respectively . the first should be viewed as an ideal in @xmath281 $ ] , and the second as an ideal in @xmath282 $ ] . denoting by @xmath283 the locus @xmath284 , we have @xmath285 this shows that @xmath279 is the union of a copy of @xmath286 and a cone in @xmath7 over @xmath287 . the intersection of these two loci is @xmath288 , that is , a copy of @xmath287 . the invariant @xmath289 satisfies an inclusion - exclusion property because so does the chern - schwartz - macpherson class , and its behavior with respect to taking a cone amounts to multiplication by @xmath215 : this follows easily from the formula for csm classes of cones , see proposition 5.2 in @xcite . therefore , @xmath290 using this fact together with the two formulas given at the beginning of the proof , we get @xmath291 which yields immediately the stated formula . [ [ multiples ] ] by lemma [ mainlemma ] , a multiple regular edge satisfies both conditions i and ii , and hence the doubling edge formula of theorem [ doubled ] applies to it . for a regular edge @xmath6 on a graph @xmath2 , and @xmath292 , denote by @xmath293 the graph obtained by replacing @xmath6 with @xmath294 edges connecting the same vertices . in particular , @xmath295 . [ goodform ] let @xmath6 be a regular edge of @xmath2 . with notation as above , @xmath296 where @xmath297 equivalently , @xmath298 the fact that the coefficient of @xmath1 determines the others by taking derivatives is a consequence of the analogous feature displayed by the coefficients of the basic recursion for @xmath299 : [ recurs ] let @xmath6 be a regular edge of @xmath2 . then for @xmath292 @xmath300 this feature reflects the fact that the characteristic polynomial for the recursion is a function of @xmath301 : @xmath302 this fact is intriguing , and we do not have a conceptual explanation for it . ( note that the corresponding fact is not verified for the recursion computing the grothendieck class . using ( 5.8 ) in @xcite , it is immediate to show that @xmath303 with notation as in loc . , if @xmath6 is regular in @xmath2 . the characteristic polynomial factors as @xmath304 . ) lemma [ recurs ] is an immediate consequence of theorem [ doubled ] : if @xmath6 is a regular edge , then @xmath6 satisfies conditions i and ii in @xmath305 for all @xmath306 , by lemma [ mainlemma ] . apply theorem [ doubled ] to obtain @xmath307 for @xmath292 . the stated recursion follows immediately , by eliminating @xmath266 from these expressions . theorem [ goodform ] follows directly from this lemma , by standard methods . [ multicor ] theorem [ goodform ] may be rewritten in the following fashion : for all regular edges @xmath6 of @xmath2 , @xmath308 that is , @xmath309 is given by the expression @xmath310 for all @xmath292 . these expressions are simpler to apply than theorem [ goodform ] , since @xmath266 is usually more immediately accessible than @xmath311 . @xmath48 [ bananaex ] the @xmath13-edge _ banana graph _ consists of @xmath13 edges connecting two distinct vertices . we have @xmath312 indeed , @xmath2 is a single bridge , and @xmath19 is a @xmath128-polygon ( see proposition 3.1 in @xcite ) . the formula in remark [ multicor ] yields the following pretty generating function : @xmath313 or equivalently @xmath314 for @xmath315 . this agrees with the formula given in example 3.8 of @xcite , which was obtained by a very different method . @xmath48 [ [ section-17 ] ] the coefficients obtained in remark [ multicor ] agree with the ones conjectured in @xcite , 6.2 , with the difference that that the formulas given here are applied to @xmath19 rather than @xmath295 ; this accounts for the extra factor of @xmath225 in the last coefficient . the point is that the hypotheses of theorem [ doubled ] are automatically satisfied for @xmath19 since @xmath6 has parallel edges in @xmath19 , while they are not necessarily satisfied for @xmath2 . accordingly , while it is tempting to interpret @xmath316 as @xmath16 , the generating function given in remark [ multicor ] can not be extended in general to provide information about this graph . [ doubletriangex ] we verify that theorem [ doubled ] does not necessarily hold if @xmath6 does not satisfy the hypotheses presented in [ condis ] . for this , we return to the graph of examples [ ciiexample ] and [ iifails ] . we have ( cf . examples [ iifails ] and [ bananaex ] ) @xmath317 the formula in theorem [ doubled ] would give @xmath318 and we can verify that this does _ not _ equal @xmath319 . indeed , the graph polynomial for @xmath19 is @xmath320 ( where the pairs of variables @xmath321 , @xmath322 , @xmath323 correspond to the three pairs of parallel edges ) . macaulay2 confirms that @xmath324 is nonsingular in codimension @xmath79 , hence @xmath325 + 8\ , [ { { \mathbb{p}}}^3]+\cdots\ ] ] from which @xmath326 differing from ( * ) . in fact , a computation using the code from @xcite shows that @xmath327 this differs from ( * ) by exactly @xmath328 . the corresponding csm class is @xmath329 + 8[{{\mathbb{p}}}^3]+18[{{\mathbb{p}}}^2]+14[{{\mathbb{p}}}]+7[{{\mathbb{p}}}^0]\quad,\ ] ] correcting the formula given at the end of 6 in @xcite . ( incidentally , the fact that in these examples the discrepancies occurring when conditions i and ii fail are pure powers of @xmath225 is also intriguing , and calls for an explanation . ) by remark [ multicor ] , we also obtain that @xmath330 for all @xmath292 . [ [ fami ] ] in this section we explain briefly a different approach to the question studied in this paper , based on the theory of specialization of chern classes ; this theory is originally due to verdier ( @xcite ) . for simplicity , we now work over @xmath331 . let @xmath332 be a family of hypersurfaces over a disk ; assume that the fibers over @xmath333 are all isomorphic . under suitable ( and mild ) hypotheses , one may define a specific constructible function @xmath334 on the central fiber @xmath335 , with the property that @xmath336 equals the specialization ( in the sense of intersection theory ) of the csm class of the general fiber . the value @xmath337 at a point @xmath338 equals the euler characteristic of the intersection of the @xmath339-ball centered at @xmath36 with nearby fibers @xmath340 , as @xmath341 and @xmath342 . see @xcite , proposition 4.1 . it may also be computed in terms of an embedded resolution of @xmath9 , see @xcite . in the situation we will consider here , the hypersurfaces @xmath340 will be elements of a pencil in projective space . it is easy to verify that @xmath343 if @xmath9 is nonsingular at @xmath36 , and @xmath344 if @xmath36 is a point of transversal intersection of two nonsingular components of @xmath9 , provided that @xmath36 does not belong to the base locus of the pencil . parusiski and pragacz ( @xcite , proposition 5.1 ) prove that @xmath343 if @xmath345 ( @xmath333 ) _ is _ a point of the base locus , if @xmath340 is _ smooth and transversal to the strata of a fixed whitney stratification of @xmath9 _ at @xmath36 . this beautiful observation will be used below . returning to graph hypersurfaces , recall that if @xmath6 is a regular edge of @xmath2 , then @xmath346 as @xmath16 is not a forest , both @xmath25 and @xmath347 are polynomials of positive degree . thus , we may view @xmath348 as an element of the pencil spanned by @xmath349 and @xmath347 : @xmath350 in @xmath351 we note that @xmath352 is isomorphic to @xmath353 for @xmath333 : indeed , with @xmath354 , @xmath355 where @xmath356 denotes the variables corresponding to edges other than @xmath6 . view ( * ) as defining a family as in [ fami ] , with central fiber @xmath357 . that is , @xmath358 where @xmath272 number of edges in the graph @xmath2 , @xmath359 is the hyperplane defined by @xmath360 , and @xmath361 denotes the cone over @xmath362 with vertex at the point @xmath363 . as a consequence of verdier s theorem ( thorme 5.1 in @xcite ) , @xmath364 where these classes are taken in @xmath365 , and @xmath334 is the specialization function on @xmath9 defined in [ fami ] . the difficulty with this approach lies in the explicit computation of @xmath334 . again we indicate two conditions under which a result may be obtained , matching the result of the more algebraic approach taken in the rest of the paper . the first condition is a ` set - theoretic ' version of condition i : @xmath366 as pointed out in [ condi ] , condition i implies this inclusion at the level of schemes ; here , we are only requiring it at the level of sets . if this condition is satisfied , then the value of @xmath334 may be determined for all points of @xmath367 . this consists of the complements of @xmath32 in @xmath368 , in @xmath369 , and in @xmath370 . thus : the first statement holds since @xmath361 is nonsingular at @xmath49 , and @xmath49 does not belong to the base locus of the pencil . the third statement likewise holds because @xmath49 is not in the base locus , and @xmath377 is nonsingular . in the second statement , @xmath49 is nonsingular on both @xmath361 and @xmath15 , and these hypersuraces intersect transversally at @xmath36 , so @xmath374 as recalled in [ fami ] . to prove the last assertion , we have to consider @xmath378 . ( we have already dealt with the other points in @xmath370 . ) at these points , both @xmath9 and @xmath32 are nonsingular ( by condition i ) . further , the intersection @xmath379 is also nonsingular at such points , again by condition i. thus the hypotheses of the result of parusiski and pragacz recalled in [ fami ] are satisfied , and it follows that @xmath372 . the locus unaccounted for in lemma [ condip ] is @xmath380 . these are points in the base locus which are contained in the component @xmath361 . _ if _ @xmath32 were nonsingular and transversal to the strata of @xmath361 at these points , then ( according to the formula of parusiski and pragacz ) we would have @xmath372 for @xmath381 ; the deviation of @xmath334 from @xmath79 at such points is a measure of non - transversality of the two hypersurfaces . also note that this locus equals the cone over @xmath181 with vertex at @xmath36 . the following condition should again be interpreted as a subtle notion of ` transversality ' of the intersection of the various loci considered here . @xmath382 applying verdier s theorem and additivity of csm classes , we get @xmath384 recalling @xmath359 , and expressing in terms of polynomial feynman rules , this gives @xmath385 with the stated result . here we also used the fact that @xmath386 , an easy consequence of proposition 5.2 in @xcite . in conclusion , we have verified that the deletion - contraction formula obtained under the ` algebraic ' conditions i and ii described in [ condis ] also holds under the conditions i and ii given in [ condipstat]-[condiipstat ] . these latter conditions have a more ` geometric ' flavor : condition i is a set - theoretic statement , and condition ii amounts to a statement on the euler characteristics of intersections of an @xmath339-ball with nearby fibers in a fibration naturally associated with the pair @xmath27 . in this sense they are perhaps easier to appreciate , although in practice the algebraic counterparts i and ii are more readily verifiable in given cases , by means of tools such as @xcite . it would be interesting to relate these two sets of conditions more precisely : does @xmath27 satisfy i and ii if and only if it satisfies i and ii ? do these conditions admit a transparent combinatorial interpretation ?

we have determined the distance to a second eclipsing binary system ( eb ) in the large magellanic cloud , hv 982 (b1 iv - v +b1 iv - v ) . the measurement of the distance among other properties of the system is based on optical photometry and spectroscopy and space - based uv / optical spectrophotometry . the analysis combines the `` classical '' eb study of light and radial velocity curves , which yields the stellar masses and radii , with a new analysis of the observed energy distribution , which yields the effective temperature , metallicity , and reddening of the system plus the distance `` attenuation factor '' , essentially . combining the results gives the distance to hv 982 , which is kpc . this distance determination consists of a detailed study of well - understood objects ( b stars ) in a well - understood evolutionary phase ( core h burning ) . the results are entirely consistent with but do not depend on stellar evolution calculations . there are no `` zeropoint '' uncertainties as , for example , with the use of cepheid variables . neither is the result subject to sampling biases , as may affect techniques which utilize whole stellar populations , such as red giant branch stars . moreover , the analysis is insensitive to stellar metallicity ( although the metallicity of the stars is explicitly determined ) and the effects of interstellar extinction are determined for each object studied . after correcting for the location of hv 982 , we find an implied distance to the optical center of the lmc s bar of kpc . this result differs by nearly 5 kpc from our earlier result for the eb hv 2274 , which implies a bar distance of 45.9 kpc . these results may reflect either marginally compatible measures of a unique lmc distance or , alternatively , suggest a significant depth to the stellar distribution in the lmc . some evidence for this latter hypothesis is discussed .

astro-ph0010526

the distance to the large magellanic cloud ( lmc ) is a key factor in determining the size scale of the universe . indeed , the uncertainty in the length of this `` cosmic meter stick '' is responsible for much of the current uncertainty in the value of the hubble constant , as noted by mould et al . the lmc distance is as controversial as it is important . existing determinations span a wide range ( see fig . 1 of mould et al . ) and are often grouped into the `` long '' distance scale results ( @xmath4 kpc and ( @xmath5 mag ) and the `` short '' distance scale results ( @xmath6 kpc and ( @xmath7 mag ) . in some cases , the same technique ( e.g. , cepheids ) can support both the long and short scales , depending on the assumptions and details of the analysis . reviews of lmc distance determinations can be found in westerlund ( 1997 ) and cole ( 1998 ) , and numerous new papers have appeared in the last several years , indicating the high activity level and interest in the field . recently , we showed that well - detached main sequence b - type eclipsing binary ( eb ) systems are ideal standard candles and have the potential to resolve the lmc distance controversy ( guinan et al . 1998a ; hereafter paper i ) . the advantages of using ebs are numerous . first , an accurate distance can be determined for each individual system and there are many systems . this is in contrast to techniques that utilize , for example , cepheids or red giant stars , where entire populations are used to derive a single distance estimate . it also contrasts with analyses of sn 1987a , which have the potential to yield a precise distance but for which there is only one object to study . the ebs can provide not only the mean lmc distance , but also can be used to probe the structure of the lmc . second , the analysis involves well - understood objects in a well - understood phase of stellar evolution ( core hydrogen burning ) and the results for each object can be verified independently by but do not depend on stellar evolution theory . third , the analysis is robust and not subject to any zeropoint uncertainties , nor are there any adjustable parameters . for example , the results are extremely insensitive to stellar metallicity , although the metallicity of the individual ebs is explicitly determined and incorporated in the analysis . likewise , the determination of interstellar extinction is an integral part of the analysis of each object . the study of the eb system hv 2274 in paper i yielded a distance of @xmath8 kpc and stellar properties consistent with stellar evolution theory ( ribas et al . 2000a ) . correcting for the position of hv 2274 relative to the lmc center yielded a lmc distance of @xmath9 kpc corresponding to @xmath10 mag . this result argues strongly in favor of the `` short distance '' to the lmc . since paper i , several partial - reanalyses of the hv 2274 system have appeared in the literature ( udalski et al . 1998 ; nelson et al . 2000 ; groenewegen & salaris 2001 ) . these authors advocate various adjustments in the results from paper i which yield lmc distance moduli in the range 18.2218.42 mag . these adjustments all stem from complications arising from the incorporation of optical photometry in the analysis . we will return to this issue later in this paper . the main goal of this paper is to apply our analysis to a second lmc eb system , hv 982 , and derive its stellar properties and distance . this eb , with @xmath11 , consists of two mildly evolved main sequence b stars each corresponding to spectral class @xmath0b1 iv a key difference between this analysis and that presented in paper i for hv 2274 is that we have here obtained space - based spectrophotometric measurements extending from 1150 to 7500 , eliminating the reliance on optical photometry . these new data preclude the ambiguities which plague the hv 2274 result and allow us to realize the full potential of our analysis technique . in 2 , we describe the data included in this study . in 3 , the analysis which incorporates the light curve , radial velocity curve , and spectral energy distribution of the system is discussed . some aspects of our results relating to the interstellar medium towards hv 982 are described in 4 , including an indication of the relative location of hv 982 within the lmc . the general stellar properties of the hv 982 system and their consistency with stellar evolution theory are described in 5 . in 6 , we show how the distance to the system is derived from our analysis , and compare this result with a reanalysis of the hv 2274 data . we discuss the distance to the lmc and summarize our conclusions from study of two binary systems in 7 . three distinct datasets are required to carry out our analyses of the lmc eb systems : high - resolution spectroscopy ( yielding radial velocity curves ) , precise differential photometry ( yielding light curves ) , and multiwavelength spectrophotometry ( yielding temperature and reddening information ) . each of these three is described briefly below . note that in this paper , the primary ( `` p '' ) and secondary ( `` s '' ) components of the hv 982 system are defined photometrically and refer to the hotter and cooler components , respectively . as we will show , the primary star is the less luminous and less massive of the pair . radial velocity curves for hv 982 , and a number of other lmc ebs , were derived from optical echelle spectra obtained by us during 6-night and 8-night observing runs in january and december 2000 , respectively , with the blanco 4-m telescope at cerro tololo inter - american observatory in chile . the seeing conditions during the two runs ranged between 0.7 and 1.8 arcsec . we secured eighteen spectra of hv 982 near orbital quadratures covering the wavelength range 36005500 , with a spectral resolution of @xmath12 , and a s / n of @xmath020:1 . the plate scale of the data is 0.08 pix@xmath13 ( 5.3 km s@xmath13 pix@xmath13 ) and there are 2.6 pixels per resolution element . identical instrumental setups were used for both observing runs . the exposure time per spectrum was 1800 sec , sufficiently short to avoid significant radial velocity shifts during the integrations . all the hv 982 observations were bracketed with thar comparison spectra for proper wavelength calibration . the raw images were reduced using standard noao / iraf tasks ( including bias subtraction , flat field correction , sky - background subtraction , cosmic ray removal , extraction of the orders , dispersion correction , merging , and continuum normalization ) . spectra of radial velocity standard stars were acquired and reduced with the same procedure . visual inspection of the hv 982 spectra revealed prominent h balmer lines and conspicuous lines of he i ( 4009 , 4026 , 4144 , 4388 , 4471 , and 4922 ) . various features due to ionized c , si , and o are also expected , but with strengths comparable to the noise level in the individual spectra . for illustration , three 200- sections of one of the observed spectra are shown in figure [ figspec ] . the strongest he i features and the h i balmer lines are labeled , with arrows marking the expected line positions for the two components of the system ( according to the radial velocity curve solution described in 3.2 ) . this spectrum was obtained at orbital phase 0.725 and illustrates the clean velocity separation of the two components of the binary . ccd differential photometric observations of hv 982 were reported by pritchard et al . ( 1998 ; hereafter p98 ) . these data were obtained between 1992 and 1995 with a 1-m telescope at mt . john university observatory ( new zealand ) . the resultant light curves in the strmgren @xmath14 , johnson @xmath15 and cousins @xmath16 passbands have very good phase coverage , with 132 , 565 , and 205 measurements , respectively . more sparsely - covered light curves were obtained in the strmgren @xmath17 passbands , with 48 , 45 , and 44 measurements for @xmath18 , @xmath19 , and @xmath20 , respectively . the precision of the individual differential photometric measurements is @xmath21 mag . we obtained spectrophotometric observations of hv 982 at uv and optical wavelengths with the faint object spectrograph ( fos ) aboard the _ hubble space telescope _ on 31 january 1997 , at binary phase 0.533 . data were obtained in four wavelength regions , using the g130h , g190h , g270h , and g400h observing modes of the fos with the 3.7x 1.3 aperture , yielding a spectral resolution of @xmath22 . the dataset names are y3fu5503 t , y3fu5506 t , y3fu5505 t , and y3fu5504 t , respectively . the data were processed and calibrated using the standard pipeline processing software for the fos . the four individual observations were merged to form a single spectrum which covers the range 1145 to 4790 . additional _ hst _ observations of hv 982 were obtained on 22 april 2001 , at binary phase 0.621 , using the space telescope imaging spectrograph ( stis ) . data were obtained in the g430l and g750l observing modes with the 52 x 0.5 aperture , yielding a spectral resolution of @xmath23 . the dataset names are o665a8030 and o665a8040 . the data were processed and calibrated using the standard pipeline processing software . cosmic ray blemishes were cleaned `` by hand '' and the spectra were trimmed to the regions 35105690 and 54107490 for the g430l and g750l data , respectively . because of concerns about photometric stability the two stis spectra were not merged ( see 3.3.3 ) . our study of hv 982 depends on three separate but interdependent analyses . these involve the radial velocity curve , the light curve , and the observed spectral energy distribution ( sed ) . the combined results provide essentially a complete description of the gross physical properties of the hv 982 system and a precise measurement of its distance . each of the three analyses is described below . to determine the radial velocities of the hv 982 components , we restricted our attention to the 40005000 wavelength region of the spectra described in 2.1 . data at higher and lower wavelengths were badly contaminated with h balmer lines , or had very poor s / n , or both . as is well known , the broad h balmer lines are generally not suitable for radial velocity work because of blending effects , which may lead to systematic underestimation of radial velocity amplitudes ( see , e.g. , andersen 1975 ) . in the 40005000 range , the h@xmath24 , h@xmath25 and h@xmath26 lines were masked by setting the normalized flux to unity in a window around their central wavelength . our initial approach for measuring radial velocities was to use the cross - correlation technique , with a very high s / n ( @xmath0250 ) spectrum of hr 1443 ( @xmath26 cae , b2 iv - v , @xmath27 km s@xmath13 ) as the velocity template . two clean cross - correlation function peaks ( one per stellar component ) were clearly visible for all the object spectra as might be anticipated from figure [ figspec ] . this allowed us to determine individual velocities with accuracies of 10 - 15 km s@xmath13 . a number of tests , however , indicated that the resulting velocities were moderately dependent on the filtering parameters used . this phenomenon adds a component of subjectivity to the measured radial velocities , and prompted us to move to an alternate , and ultimately superior , technique . `` spectral disentangling '' is an improvement over classical cross - correlation because it essentially uses information from the entire spectral dataset to derive the individual radial velocities . the basic idea of the technique is very simple : an individual ( observed ) double - lined spectrum is assumed to be a linear combination of two single - lined spectra ( one per component ) at different relative velocities ( determined by the orbital phase at the time of observation ) . the goal is to retrieve the two single - line spectra and the set of relative velocities for each observed spectrum by considering the whole dataset simultaneously . the numerical implementation is an inversion algorithm of an over - determined system of linear equations . in contrast with cross - correlation , this technique eliminates the need for a spectral template , but it does require a homogeneous dataset of spectra taken at a variety of orbital phases . the practical implementation of the disentangling method has been carried out using two independent approaches : simon & sturm ( 1994 ) based their algorithm on a singular value decomposition , and hadrava ( 1995 , 1997 ) employed a fourier transform . in principle , the two implementations are equally valid and we decided to adopt the fourier disentangling code korel developed by hadrava . we have made a number of modifications to the original fortran source , the most significant of which is an increase of the maximum number of radial velocity bins . to enhance the performance of the disentangling algorithm , we provided korel with orbital information ( period , time of periastron passage , eccentricity , longitude of periastron ) so a reasonable set of starting values for the velocity semiamplitudes could be computed ( hadrava 2001 , priv . comm . ) . several korel runs from different initial conditions were performed to ensure the uniqueness of the solution . the final heliocentric radial velocities derived from all the ctio spectra using the procedure outlined above are listed in table [ tabrv ] ( `` rv@xmath28 '' and `` rv@xmath29 '' ) , along with the date of observation and the corresponding phase . the individual errors of the velocities are not provided by korel and a reliable estimation is not straightforward . this issue is addressed in more detail in 3.2 . as noted above , the individual spectra for the two components are also products of the analysis . since they combine information from all the data , the quality of these two spectra is significantly improved with respect to the individual observations . figure [ figsecon ] shows a comparison of the disentangled spectrum of the secondary star , i.e , the more massive and luminous component , with a synthetic spectrum computed using r.l . _ atlas9 _ atmosphere models , ivan hubeny s spectral synthesis program _ synspec _ , and the appropriate stellar properties ( i.e. , @xmath30 = 23600 k , @xmath31 = 3.72 , @xmath32 = -0.3}$ ] , and @xmath33 km s@xmath13 ) . note that the products of korel are two spectra that have not been corrected for `` light dilution , '' i.e. , the continuum level contains the light contribution from the two components and thus the absorption lines are diluted to roughly half their true strengths . the spectrum in figure [ figsecon ] has had the primary s contribution removed and then been renormalized . the contribution of the primary was determined using the line strengths in the synthetic spectrum as a reference . the appropriate value of @xmath34 for the model was also determined by comparing the observed and synthetic spectra . the derivations of the basic stellar properties which serve as inputs for the synthetic spectrum calculations are described below in the rest of 3 and all the stellar properties are summarized in 5 . the significance of the derived @xmath34 value is also discussed in 5 . as can be seen in figure [ figsecon ] , the agreement between the disentangled and synthetic spectra is excellent , including not only the depths of all the features but also the profile shapes of both the he i and h i balmer lines . ( recall that the balmer lines were masked out when obtaining radial velocities , but a final run of korel with no free parameters was carried out to extract the complete spectrum that we show in figure [ figsecon ] . ) the disentangled spectrum of the primary star is nearly identical to that of the secondary and shows similarly good agreement with a synthetic spectrum computed using the appropriate stellar properties ( i.e. , @xmath30 = 24200 k , @xmath31 = 3.78 , @xmath35 = -0.3}$ ] , and @xmath36 km s@xmath13 ) . these comparisons provide a `` reality check '' for korel and , as will be shown in 5 , give valuable confirmation for some of the basic results of our overall analysis . the radial velocity data were analyzed using an improved version of the wilson - devinney program ( wilson & devinney 1971 ; hereafter wd ) that includes an atmosphere model routine developed by milone et al . ( 1992 ) for the computation of the stellar radiative parameters . the full analysis can potentially yield determinations of the component velocity semi - amplitudes @xmath37 ( or , equivalently , the mass ratio @xmath38 ) , the systemic velocity @xmath25 , the orbital semi - major axis @xmath39 , the eccentricity @xmath40 , and the longitude of the periastron @xmath41 . in our particular case , we adopted the very well - determined value of @xmath40 from the light curve solution ( 3.2 ) . it is not possible , however , to utilize the value of @xmath41 yielded by the light curve analysis without first correcting for the significant apsidal motion of the system ( an increase in @xmath41 of about 12@xmath42 between the epoch of p98 s observations and ours ) . instead of using the empirical apsidal motion rate , we treated @xmath41 as a free parameter and obtained a best fit for the mean epoch of our data ( j2000.5 ) . thus , solutions were run in which @xmath38 , @xmath39 , @xmath25 , and @xmath41 were the adjustable parameters . our best fit to the radial velocity curve is shown in figure [ figrv ] . note that the details of the curve shape ( such as its skewness and the abrupt changes during eclipse ) are not a product of the radial velocity analysis , but rather result from the adopted light curve solution and from the physical effect of partially - eclipsed rotating stars ( the `` rossiter effect '' ) . the fit residuals ( indicated as `` o c '' in the figure and in the last two columns of table [ tabrv ] ) correspond to r.m.s . errors of @xmath01.5 km s@xmath13 . this small internal error gives an indication of the good quality of the spectroscopic data and the excellent performance of the disentangling technique . the best - fitting parameters to the radial velocity curve are listed in table [ tabparms ] . the uncertainties quoted in the table bear some comment . indeed , the formal errors derived from the fit to the radial velocity curve are significantly smaller . for example , the wd program returns an estimated uncertainty of only @xmath00.5 km s@xmath13 in the velocity semiamplitudes . while formally correct , this uncertainty may be underestimated because it fails to account for any systematic effects that may be present in the data . as discussed by hensberge , pavlovski , & verschueren ( 2000 ) , more realistic estimates of the uncertainty in the velocity semiamplitudes follow from considering the scatter of the velocities derived from the analysis of separate spectral regions . thus , we divided our entire spectrum into four wavelength intervals and analyzed these separately with korel . the standard deviation of the resulting velocities turned out to be 3.2 km s@xmath13 . this is most likely an overestimate of the true error of the velocities because of the spectral coverage being significantly degraded ( and so the number of spectral lines available for radial velocity determination ) . nonetheless , we conservatively adopted 3.2 km s@xmath13 as the uncertainty of the velocity semiamplitudes and scaled the rest of the parameter errors listed in table [ tabparms ] accordingly . p98 ran simultaneous solutions for the 6 available light curves using the same version of the wd program described above . the wd program was run in an iterative mode in order to explore the full - extent of parameter space and also to make a realistic estimation of the errors . furthermore , the authors considered different mass ratios ranging between 0.9 and 1.1 ( since no spectroscopic observations were available at that time ) and found the light curve solution to be completely insensitive to changes within this range . as often occurs for eclipsing binary stars in eccentric orbits , several parameter sets four in this case were found to yield equally good fits to the observed light curves . the main distinction among the possible solutions is that p98 s cases 1 and 2 predict the primary star to be somewhat smaller and hotter than the secondary ( with relative luminosity in the @xmath15 and @xmath16 spectral regions of @xmath43_{v , i } \simeq 1.1 $ ] ) , while cases 3 and 4 yield a bigger and cooler primary ( @xmath44_{v , i } \simeq 0.9 $ ] ) . this degeneracy can be broken only by considering some external source of information , e.g. , a spectroscopically - determined luminosity ratio . without access to such information , p98 were unable to favor any of their four cases . the disentangled spectra discussed above provide such a spectroscopic luminosity ratio and allow us to distinguish between p98 s two general scenarios . as noted in 3.1 , the comparison of the disentangled spectra with synthetic spectra yields values of @xmath34 for both stars and their light dilution factors , which are related to their luminosities in the blue ( @xmath45 ) spectral region . the ratio of these dilution factors yields @xmath44_b = 1.15\pm0.05 $ ] , clearly favoring cases 1 and 2 , in which the primary star is smaller and hotter than the secondary . note that these cases are also the physically preferable ones , since our radial velocity analysis shows the secondary star to be the more massive member of this non - interacting main sequence system and , therefore , necessarily the more luminous star . in an attempt to distinguish between p98 s cases 1 and 2 , we redid the light curve analysis published by p98 using an identical computational setup ( i.e. , the iterative wd program ) . we applied the wd program to the observed light curves both individually and as a group , constraining the spectroscopically - determined mass ratio @xmath38 and trying a variety of weighting schemes for the different bandpasses . the tests revealed p98 s case 1 to be a spurious solution caused by excessive weight on the strmgren @xmath14 light curve . this solution never appeared in the analysis of the well - covered @xmath15 and @xmath16 light curves . therefore , our results clearly favor p98 s case 2 over the others . in figure [ figlc ] we illustrate this solution to the @xmath15 and @xmath16 light curves . the final orbital and stellar parameters adopted from the light curve analysis are listed in table [ tabparms ] . these were derived from a simultaneous solution to all the bandpasses , weighted by their observational errors . the parameters @xmath46 and @xmath47 represent the _ relative _ stellar radii , i.e. , the physical radii divided by the orbital semi - major axis @xmath39 . note that the fractional radii listed are those corresponding to a sphere with the same volume as the roche equipotential ( `` volume radius '' ) . the final step in the analysis of hv 982 is modeling the observed shape of the uv / optical energy distribution . this procedure is the same as that used for hv 2274 ( see paper i ) , and is based on the technique developed by fitzpatrick & massa ( 1999 ; hereafter fm99 ) . the analysis consists of a non - linear least squares determination of the optimal values of all the parameters which contribute to the right side of equation 1 . we represent the stellar surface fluxes with r. l. kurucz s _ atlas9 _ atmosphere models , which each depend on four parameters : effective temperature ( @xmath30 ) , surface gravity ( @xmath31 ) , metallicity ( [ m / h ] ) , and microturbulence velocity ( @xmath57 ) . in addition , we utilize the six - term parametrization scheme from fitzpatrick & massa ( 1990 ) and the recipe given by fitzpatrick ( 1999 ; hereafter f99 ) to construct the wavelength dependent uv - through - ir extinction curve @xmath58 . thus , in principle , the problem can involve solving for two sets of four kurucz model parameters , the ratios @xmath59 and @xmath60 , @xmath54 , six extinction curve parameters for @xmath58 , and @xmath61 . for hv 982 , several simplifications can be made which reduce the number of parameters to be determined : ( 1 ) the temperature ratio of the two stars is known from the light curve analysis ; ( 2 ) the surface gravities can be determined by combining results from the light and radial velocity curve analyses and are log g = 3.78 and 3.72 for the primary and secondary stars , respectively ( see 5 ) ; ( 3 ) the values of [ m / h ] and @xmath57 can be assumed to be identical for both components ; ( 4 ) the ratio @xmath60 is known ; and ( 5 ) the standard mean value of @xmath62 found for the milky way can reasonably be assumed given the existing lmc measurements ( e.g. , koornneef 1982 ; morgan & nandy 1982 ; see 4 ) . with these simplifications in place , we modeled the observed uv / optical energy distribution of hv 982 solving for the best - fitting values of @xmath63 , [ m / h]@xmath64 , @xmath65 , @xmath59 , @xmath54 , and six @xmath58 parameters . prior to the fitting procedure , the three spectrophotometric datasets ( one merged fos spectrum and two stis spectra ) were ( 1 ) velocity - shifted to bring the centroids of the stellar features to rest velocity ; ( 2 ) corrected for the presence of a strong interstellar h i ly@xmath66 absorption feature in the fos spectrum at 1215.7 ; and ( 3 ) binned to match the _ atlas9 _ wavelength scale . the ly@xmath66 correction was performed by dividing the spectrum by the intrinsic ly@xmath66 profile corresponding to a total h i column density of @xmath67 , distributed in an lmc component and a milky way foreground component . the determination of the column densities is discussed in more detail in 4 below . the binning was accomplished by forming simple unweighted means within the individual wavelength bins of the _ atlas9 _ models . in the wavelength range relevant to this study , the bin sizes are typically 10 ( for @xmath68 ) and 20 ( for @xmath69 ) . the statistical errors assigned to each bin were computed in the usual way from the statistical errors of the original data , i.e. , @xmath70 , where the @xmath71 are the statistical errors of the individual spectrophotometric data points within the bin . for all the spectra , these uncertainties typically lie in the range 0.5% to 1.5% of the binned fluxes . note that we do not merge the fos and stis data into a single spectrum , but rather perform the fit on the three binned spectra simultaneously and independently . this is because stis is less photometrically stable than fos and there are likely to be flux zeropoint offsets among the spectra ( see the instrument handbooks for fos and stis available online at www.stsci.edu ) . we account for this effect in the fitting procedure by assuming that the fos data represent the true flux levels and including two zeropoint corrections ( one for each stis spectrum ) to be determined by the fit . we later explicitly determine the uncertainties in the results introduced by zeropoint errors in fos . the nominal weighting factor for each bin in the least squares procedure is given by @xmath72 . we exclude a number of individual bins from the fit ( i.e. , set the weight to zero ) for the reasons discussed by fm99 ( mainly due to the presence of interstellar gas absorption features ) . the best - fitting values of the energy distribution parameters and their 1-@xmath73 uncertainties are listed in tables [ tabparms ] ( stellar properties ) , [ tabstis ] ( stis offsets ) , and [ tabext ] ( extinction curve parameters ) . a comparison between the observed spectra and the best - fitting model is shown in figure [ figsed ] . the three binned spectra are plotted separately in the figure for clarity ( small filled circles ) . the zeropoint offset corrections ( see table [ tabstis ] ) were applied to all stis spectra in figure [ figsed ] . note that we show the quantity @xmath74f@xmath75 as the ordinate in figure [ figsed ] ( rather than f@xmath75 ) strictly for plotting purposes , to `` flatten out '' the energy distributions . our final fit to the hv 982 energy distribution was computed after adjusting the weights in the fitting procedure to yield a final value of @xmath76 . although figure [ figsed ] shows that the model provides an extremely good fit to the data , nevertheless the overall r.m.s . deviation of @xmath02.4@xmath77 is significantly larger than the statistical errors of the individual data bins . with no adjustments , the formal reduced @xmath78 would be much greater than one ( 4.6 in this case ) . this is , in general , an expected result since the observational uncertainties from which @xmath78 is computed include statistical errors only and are thus certainly underestimates . in addition , it is unlikely that the absolute photometric calibration of the data , the extinction curve representation , or the atmosphere models are perfect representations of reality . the rationale for adjusting the fitting weights is to yield more realistic estimates of the parameter uncertainties , which scale as @xmath79 . the adjustment of the weights could be accomplished in a number of ways , most simply by either scaling upward the statistical errors of each bin by a single factor ( 2.1 in this case ) or by quadratically combining the statistical errors with an overall uncertainty represented as a fraction of the local binned flux ( 2% in this case ) . for hv 982 we adopted the latter technique , although both yield virtually identical results ( which are indistinguishable from the no - adjustment case ) . note that this procedure is not rigorously justifiable , since it implies that the discrepancies between the model and the data are due entirely to underestimated observational errors . however , the resultant parameter uncertainties do appear reasonable for those cases when an external check is possible . we have such comparisons for two parameters : ( 1 ) @xmath80 as we will show in 6 below , the values of t@xmath81 , with their attendant uncertainties , are completely consistent with expectations from stellar evolution calculations ; ( 2 ) @xmath31 the fitting procedure utilizes the surface gravities determined from the binary analysis . however @xmath31 can also be determined directly from the energy distributions . when we fit the hv 982 spectrum , constraining only the difference in @xmath31 between the two components , we find for the primary star a value of @xmath82 , which compares very well with the observed value of @xmath83 from the binary analysis . we conclude that the adopted procedure yields reasonable estimates of the true uncertainties of the parameters determined from the energy distribution analysis . the values of most of the parameters derived above will be discussed in the sections below . here we comment briefly only on the results for the stis offsets and e(b - v ) . the correction factors of 10.4% and 6.4% required to rectify the stis g430l and g750l spectra , respectively , are extremely well - determined and appear surprising large . nevertheless , they are consistent with the stis calibration goals ( of @xmath8410% ) for absolute photometry with the ccd cameras . the discrepancies may arise from several factors , including the absolute photometric calibration , instrument stability , and possible light loss in the 0.5-wide slit . the relative roles of these various effects are uncertain at this point . we have two other lmc binary systems with similar sets of stis and fos spectra ( eros 1044 and hv 5936 ) . in all three cases , the stis fluxes are below the fos level , with the g430l data always being the most discrepant . note that , in the region of overlap between the fos and stis g430l , the _ shapes _ of the spectra agree to within several percent , it is only the general levels that are in discord . we will examine the cross - calibration of stis and fos using these and other data in a future paper . an accurate determination of @xmath85 is one of the most critical factors in deriving accurate distances using this analysis and has been one of the most problematic . in this paper , the determination of @xmath86 is straightforward , highly precise , and unambiguous because we have an ideal dataset consisting of spectrophotometry spanning the entire wavelength range over which @xmath87 is defined . in an earlier version of this work ( fitzpatrick et al . 2000 ) , however , we performed the analysis utilizing only fos data ( due to the lack of reliable optical photometry for hv 982 ) and found a much different value of @xmath88 . it is clear in retrospect that the fos data themselves , which truncate at 4790 , do not extend far enough into the optical region to allow an accurate estimate of @xmath87 . with those data alone , @xmath89 is wholly determined by only several hundred of spectrum at the very end of the fos g400h camera and is highly subject to any systematic errors in the fos photometric calibration or in the shape of the adopted extinction curve in this small spectral region . in fact , we see a strong systematic effect in the analysis of three binaries for which stis data are now available the fos data alone always yield higher estimates of @xmath87 than when stis spectra ( or reliable optical photometry ) are incorporated in the analysis . the moral of the story is that a good determination of @xmath87 requires data which span the wavelength range of the johnson @xmath15 and @xmath45 filters , i.e. , the wavelength range over which @xmath87 is defined . our analysis provides some general information regarding interstellar gas and dust along the hv 982 sightline and also yields some insight into the relative location of the star within the lmc . this latter point will prove important in interpreting the results from our ensemble of lmc binaries . as noted in 3.3.2 above , we find a total h i absorption column density of @xmath90 toward hv 982 . this consists of an assumed galactic foreground contribution at 0 km s@xmath13 of @xmath91 @xmath92 ( see , e.g. , schwering & israel 1991 ) and an lmc contribution at 260 km s@xmath13 of @xmath93 @xmath92 determined by us from the strength of the interstellar ly@xmath66 absorption line in the fos g130h spectrum . a conservative estimate of the lmc column density uncertainty is @xmath0@xmath8420% . the fos data used to derive the h i measurement are shown in figure [ fighi ] where we plot a small section of the spectrum centered on h i ly@xmath66 with various stellar and interstellar lines labeled . the spectral resolution and s / n of the data are not sufficient to reveal the complex absorption profiles of the interstellar lines , which span nearly 300 km s@xmath13 in velocity . the dotted line shows a synthetic spectrum of the hv 982 system , constructed from two individual spectra , velocity shifted to match the stellar velocities at the time of the fos observations . the individual spectra were computed using _ model atmospheres of the appropriate physical parameters and the spectral synthesis program _ synspec_. the thick solid line shows the synthetic spectrum convolved with an interstellar ly@xmath66 profile computed using the velocities and column densities noted in the paragraph above . the adopted lmc column density is that which produces the best agreement ( based on visual estimate ) between the convolved spectrum and the data . as can be seen in figure [ fighi ] , the agreement is impressive , particularly in the broad damping wings of ly@xmath66 . note that the non - zero flux seen at line center is due to geocoronal scattering of solar ly@xmath66 photons . the h i 21-cm emission line survey of rohlfs et al . ( 1984 ) reveals that the _ total _ lmc h i column density along the hv 982 line of sight is about @xmath94 @xmath92 , centered at 260 km s@xmath13 ( this velocity was adopted for the ly@xmath66 analysis ) . this result , combined with the ly@xmath66 determination for h i _ in front of _ hv 982 , clearly demonstrates that hv 982 is actually located behind most , if not all , of the lmc h i along its line of sight . we will return to this point in 8 . the interstellar extinction curve determined for the hv 982 sightline is shown in figure [ figext ] . small symbols indicate the normalized ratio of model fluxes to observed fluxes , while the thick solid line shows the parametrized representation of the extinction , which was actually determined by the fitting process . as noted in 3.3 , the recipe for constructing such a `` custom '' extinction curve is taken from f99 and the parameters defining the curve are listed in table [ tabext ] . extinction - producing dust grains along the hv 982 line of sight lie in both the milky way and the lmc . the results of oestreicher , gochermann , & schmidt - kaler ( 1995 ) show that the milky way foreground extinction in this direction corresponds to @xmath95 mag . combined with our determination of the total value of @xmath85 ( see table [ tabparms ] ) , this indicates an lmc contribution of @xmath96 . the curve in figure [ figext ] is thus clearly a composite , but weighted more heavily towards the galactic extinction component . this prevents detailed conclusions about either extinction component , although it is useful to note that the width and position of the remarkably weak 2175 extinction bump are consistent with galactic values ( fitzpatrick & massa 1990 ) . further , it is reasonable to conclude that the very weak 2175 bump is a feature of both the milky way and lmc extinction components along this sightline . several of the stars in our lmc distance program have little or no extinction beyond the milky way foreground contribution . for these stars we will be able to derive explicitly the shape of the galactic foreground extinction and perhaps use this result to `` deconvolve '' composite curves such as that for hv 982 . the `` gas - to - dust '' ratio for the lmc interstellar medium towards hv 982 can be computed from the results above and is @xmath97 . this value is consistent with the range of values seen by fitzpatrick ( 1986 ) for a variety of lmc sightlines and is significantly higher than the mean milky way value of @xmath98 ( bohlin et al . 1978 ) , which may reflect the lower abundance of metals in the lmc . a final comment on extinction concerns the value of @xmath61 ( @xmath99 ) . in this analysis we adopt the mean galactic value of @xmath100 due , essentially , to a lack of any other option . an individual determination for a relatively lightly - reddened star like hv 982 would require higher precision near - ir photometry than is currently available ( from _ 2mass _ ) . a correlation between @xmath61 and the slope of extinction in the uv ( as parametrized by the fit coefficient _ @xmath101 _ ) has been shown for a small sample of galactic stars ( f99 ) , but it can not be assumed that such a correlation is applicable to a mixed lmc / galactic halo sightline . the safest course for our analysis is to adopt a value of @xmath61 and then incorporate a reasonable estimate of its uncertainty in the final error analysis . as will be shown in 6 , the uncertainty in hv 982 s distance due to @xmath61 is only a small component of the overall error budget . the results of the analyses described above can be combined to provide a detailed characterization of the physical properties of the hv 982 system . we summarize these properties in table [ tabstar ] ; notes to the table indicate how the individual stellar properties were derived from the analysis . it is important to realize that the results in table [ tabstar ] were derived completely independently of any stellar evolution considerations . thus , stellar structure and evolution models can be used to provide a valuable check on the self - consistency of our empirical results . to test this consistency , we considered the evolutionary models of claret ( 1995 , 1997 ) and claret & gimnez ( 1995 , 1998 ) ( altogether referred to as the cg models ) . these models cover a wide range in both metallicity ( @xmath102 ) and initial helium abundance ( @xmath103 ) , incorporate the most modern input physics , and adopt a value of 0.2 h@xmath104 as the convective overshooting parameter . the locations of the hv 982 components in the @xmath105 vs. @xmath106 diagram are shown in figure [ fighrd ] . the skewed rectangular boxes indicate the 1@xmath73 error locus ( recall that errors in @xmath30 and @xmath107 are correlated ) . if our results are consistent with stellar evolution calculations , then the evolution tracks corresponding to the masses and metallicity derived from the analysis should pass through the error boxes in the @xmath105 vs. @xmath106 diagram . indeed , this is the case . the thin solid lines show the tracks corresponding to zams masses of 11.7 and 11.4 @xmath108 and @xmath109 ( based on the value of @xmath110}$ ] resulting from the spectrophotometry analysis ) . the models predict that such stars should lose about 0.1 @xmath108 due to stellar winds by the time they reach the positions of the hv 982 stars , yielding the present - day masses of 11.6 and 11.3 @xmath108 . further , the two stars are compatible with a single isochrone , corresponding to an age of 17.4 myr ( dotted line in figure [ fighrd ] ) . note that the only adjustable parameter in the model comparison is the initial helium abundance @xmath103 , for which we find an optimal value of @xmath111 . this value is in excellent agreement with expectations from empirical chemical enrichment laws ( see ribas et al . 2000b ) . an additional , independent test of the compatibility of our results and stellar structure theory can be made because the hv 982 system has an eccentric orbit and a well - determined apsidal motion rate ( @xmath112 ) . the value of ( @xmath112 ) can be found by combining the individual results for @xmath41 determined from the light curve and the radial velocity curve analyses ( see table [ tabparms ] ) . the resultant apsidal motion rate is @xmath113 deg / yr . this is marginally consistent with the value of @xmath114 deg / yr from p98 , based on eclipse timings , although the error in p98 s result is likely to be underestimated due to large uncertainties in some of the earliest timings . the expected value of @xmath112 for a binary system can be computed as the sum of a general relativity term ( gr ) and a classical term ( cl ) . the latter , which is the most important contribution in close systems like hv 982 , depends on the internal mass distributions of the stars , which can be derived from stellar evolution models . the mass concentration parameters @xmath115 ( i.e , the ratio of the central density to mean density ) for the appropriate cg models ( i.e. , those shown in fig . [ fighrd ] ) are @xmath116 and @xmath117 . ( the error bars reflect the observational uncertainties in the stellar masses . ) these values contain a small correction for stellar rotation effects , according to claret & gimnez ( 1993 ) . using the formulae of claret & gimnez , we compute a classical term of @xmath118 deg / yr and a general relativistic term of @xmath119 deg / yr , yielding a total theoretical apsidal motion rate of @xmath120 deg / yr . this result is in excellent agreement with the observed values , once again demonstrating consistency between our results and stellar interior theory . note that the eccentric nature of hv 982 s orbit is not surprising since circularization is not expected to occur until a later evolutionary phase , when the stars have expanded to nearly fill their roche lobes . final reality checks on our results can be obtained from the disentangled spectra of the primary and secondary components of the hv 982 system , discussed in 3.1 . first , and most simply , the spectra of these stars are completely consistent with the parameters ( particularly @xmath80 and @xmath31 ) derived in our analysis , as is well - illustrated in figure [ figsecon ] . second , the values of @xmath121 ( see table [ tabstar ] ) also provide a remarkable confirmation of our results . these were derived by fitting the disentangled spectra with a grid of synthetic spectra computed with @xmath121 values ranging from 20 to 160 km s@xmath13 . the uncertainties in the best - fitting values of @xmath34 were estimated from the scatter in the results when small sections of the 40005000 disentangled spectra were considered individually . these measured @xmath121 values can be compared with theoretical expectations since the hv 982 stars are expected to have undergone `` pseudo - synchronization , '' in which the stellar rotational speeds are determined by the orbital angular velocity at periastron and the individual stellar radii . for a binary such as hv 982 , pseudo - synchronization should occur in only @xmath122 myr ( using the recipe of claret , gimnez , & cunha 1995 ) , much less than the age of the system . the pseudo - synchronized values of @xmath34 for the primary and secondary stars are 89.5 km s@xmath13 and 97.7 km s@xmath13 , respectively ( as computed from the results in kopal 1978 ) , in excellent agreement with the measurements . our analysis has shown that hv 982 is an extraordinarily well - characterized system consisting of a pair of normal , mildly - evolved , early - b stars . the results are all internally consistent and consistent with a host of external reality checks , such as the expected lmc metallicity , @xmath123 calibrations of galactic b stars , stellar evolution calculations , and binary evolution calculations . this detailed characterization and the unremarkable nature of the hv 982 stars make this system ideal for the determination of a precise distance . as in paper i , we derive the distance to the system simply by combining results from the eb analysis which yields the absolute radius of the primary star @xmath124 and from the spectrophotometry analysis which yields the distance attenuation factor @xmath59 . the result , shown in table [ tabstar ] , is @xmath125 kpc corresponding to a distance modulus of @xmath126 . the uncertainty in the hv 982 distance determination arises from three independent sources : ( 1 ) the internal measurement errors in @xmath127 and @xmath128 given in table [ tabparms ] ; ( 2 ) uncertainty in the appropriate value of the extinction parameter @xmath61 ; and ( 3 ) uncertainty in the fos flux scale zeropoint due to calibration errors and instrument stability . straightforward propagation of errors shows that these three factors yield individual uncertainties of @xmath129 kpc , @xmath130 kpc ( assuming @xmath131 ) , and @xmath132 kpc ( assuming @xmath133% ) , respectively . the overall 1@xmath73 uncertainty quoted above is the quadratic sum of these three errors . note that the only `` adjustable '' factor in the analysis is the extinction parameter @xmath61 , for which we have assumed the value 3.1 . the weak dependence of our result on this parameter is given by : @xmath134 $ ] . as noted in 1 , there is a `` mini - controversy '' over the distance to hv 2274 , the lmc eb system we analyzed in paper i. close examination of the results from the various groups involved ( including our own early result for hv 2274 reported by guinan et al . 1998b ) , reveals that the discord among the various results arises from the inclusion of ground - based optical photometry in the energy distribution portion of the analysis . the problem is twofold : first , there are significant differences among the available _ ubv _ measurements for hv 2274 differences larger than the claimed errors ( see table 2 of nelson et al . second , there is no obvious or objective way to determine how the various optical photometric indices should be weighted in the sed analysis with respect to the spectrophotometry . the need to incorporate ground - based photometry in the hv 2274 analysis is clear : @xmath85 can not be determined reliably unless the energy distribution data extend through the @xmath15 spectral region ( see the discussion at the end of 3.3.3 ) and the available fos spectrophotometry for hv 2274 truncate in the @xmath45 region at 4790 . the complications arise in determining which data to use and how to use them . since paper i , we have studied extensively the effects of optical photometry on the sed analysis and have concluded that the best way to analyze datasets such as that for hv 2274 is also the simplest way , namely , utilize the fos spectrophotometry and only the @xmath15 magnitude and force the best - fitting model to agree exactly with @xmath15 . the benefits of this are that it eliminates the need to determine weighing factors for the optical photometry and , importantly , allows the effect of the adopted @xmath15 magnitude on the resulting distance estimate to be examined explicitly . the exclusion of @xmath135 and @xmath136 from the analysis is not a loss since , even in the best of circumstances ( i.e , no observational errors ) , they are wholly redundant . we have re - run the energy distribution portion of the hv 2274 analysis using only the fos data described in paper i and a value of @xmath137 ( udalski et al . 1998 ; watson et al . 1992 ) , which we believe to be well - determined . the fos data processing and the fitting procedure was applied exactly as described here for hv 982 , with the addition of the @xmath15 magnitude constraint . synthetic @xmath15 photometry was performed on the models as described by fm99 . the parameters determined in the analysis are the same as for hv 982 , namely , @xmath63 , [ m / h]@xmath64 , @xmath65 , @xmath59 , @xmath54 , and six @xmath58 parameters . before fitting the energy distribution , we corrected the hv 2274 fos spectrum for presence of a strong interstellar h i ly@xmath66 feature . from fitting the broad line profile ( using the same technique as described in 4 above for hv 982 ) , we find a lmc h i absorption column density of @xmath138 towards hv 2274 . the lmc h i 21-cm emission column density in this same direction is @xmath139 ( rohlfs et al . . the results of this reanalysis are actually nearly identical to those reported in paper i and none of the conclusions in paper i or ribas et al . ( 2000a ) regarding the stellar properties of hv 2274 and their consistency with stellar evolution theory are altered . this agreement in hindsight is not surprising since in paper i we gave high weight to the @xmath15 magnitude of udalski et al . 1998 and low weights to the @xmath140 and @xmath45 data due to conflicting observational reports . we thus , by accident , approached what we now believe is the optimal way to combine the fos and photometric datasets . the distance derived for hv 2274 from the radius of the primary star ( @xmath141 r@xmath142 ) and the parameter @xmath143 is @xmath144 kpc corresponding to a distance modulus of @xmath145 . this distance is slightly larger than the value 46.8 kpc reported in paper i. the error analysis incorporated ( 1 ) the internal uncertainties in @xmath124 and @xmath59 , ( 2 ) an uncertainty of @xmath146 in the adopted value of @xmath62 , ( 3 ) an uncertainty of @xmath147% in the fos flux scale , and ( 4 ) an uncertainty of @xmath148 in @xmath15 . these are all independent effects and were combined quadratically to yield the quoted 1@xmath73 error in @xmath149 and @xmath150 . this result is larger than the error of @xmath151 kpc quoted in paper i , due to the more realistic treatment given to the effects of uncertainty in @xmath15 . as in the case of hv 982 , the only `` adjustable '' parameter in this result is the assumed value @xmath62 . its influence , and the explicit effect of the @xmath15 magnitude , on the result can be expressed as : @xmath152 + 3.2 \times [ v-14.16]$ ] . note that @xmath85 is not an adjustable parameter its value ( 0.12 mag ) is fully determined by the analysis . the hv 2274 distance is more sensitive to the uncertainty in @xmath61 than for hv 982 because of hv 2274 s larger reddening . determining the distance to the lmc from the individual distances to hv 982 and hv 2274 requires correcting for the spatial orientation of the lmc s disk and the stars apparent locations within it . we adopt as a reference point the optical center of the lmc s bar , at ( @xmath66 , @xmath26)@xmath153 = ( @xmath154 , @xmath155 ) according to isserstedt ( 1975 ) , assume a disk inclination of @xmath156 , and a line - of - nodes position angle of @xmath157 ( schmidt - kaler & gochermann 1992 ) . figure [ figlmc ] shows a photo of the lmc with the adopted center and the orientation of the line - of - nodes indicated by the open box and solid line , respectively . the `` near - side '' of the lmc is eastward of the line - of - nodes . the hv 982 system is located , in projection , relatively close to the line - of - nodes ( see fig . [ figlmc ] ) and , given its measured distance , should _ if it lies in the lmc s disk _ be positioned about 450 pc in front of the bar center . this would imply a distance to the lmc bar center of 50.7 ( @xmath158 ) kpc , corresponding to a distance modulus of 18.52 ( @xmath159 ) mag . as discussed in paper i , hv 2274 s location ( on the `` far - side '' of the lmc ; see fig . [ figlmc ] ) places it _ if it lies in the lmc s disk _ about 1100 pc beyond the bar center . this result implies a distance to the bar center of 45.9 ( @xmath160 ) kpc , corresponding to a distance modulus of 18.31 ( @xmath161 ) mag . these two estimates of the distance to the lmc bar differ by nearly 5 kpc . ( the magnitude of the discrepancy depends somewhat on the adopted orientation of the lmc disk ; however , this effect is only at the level of a few hundred pc given the various estimates for the lmc s geometry ; see westerlund 1997 , page 30 . ) if the uncertainties in the two results were uncorrelated , then this would amount to a @xmath02@xmath73 difference . however , the uncertainties in the two results are not completely uncorrelated . for example , errors in the fos flux zeropoint would affect both analyses in the same way , as ( probably ) would errors in the adopted value of @xmath61 due to the similar lines of sight through the milky way halo . in addition , the stars in the hv 982 and hv 2274 systems bear close resemblance to each other . any small systematic effects in the analyses would affect both results in a similar way . as a result , the discrepancy between the two measurements of the lmc distance is likely to be somewhat larger than the formal @xmath02@xmath73 . if these results represent marginally consistent , independent measures of the same quantity ( _ the _ lmc distance ) then they imply a mean distance of between 46 and 51 kpc . ( including the eros 1044 result noted below , that mean is @xmath048 kpc ) . an alternate interpretation , however , is that the hv 982 and hv 2274 results are discrepant because the systems do not lie in a common disk ( with the orientation usually ascribed to the lmc ) , and therefore do not both constrain the distance to this disk . in this scenario , hv 2274 is associated with the disk as indicated by the comparison of h i emission and absorption column densities in its direction ( i.e. , it lies behind @xmath162 of the h i in its direction ) and hv 982 is located several kpc behind the disk consistent with it s lying behind all the h i in its direction ( see 4 ) . this scenario would indicate a significant depth to the massive star distribution of the lmc and would have implications for the value of the lmc as a calibrator of the cosmic distance scale . the results for hv 982 and hv 2274 , by themselves , do not persuasively argue for the existence of a `` thick '' lmc . some support for this hypothesis is available , however , from other data . specifically : 1 ) the agreement between hv 982 s distance and that of the nearby sn 1987a ( @xmath163 kpc , panagia 1999 ) ; 2 ) the agreement between the hv 2274 result and that for the eb system eros 1044 ( ribas et al . 2001 ) which is also located in the lmc s bar and implies a lmc center distance of @xmath164 kpc ; and 3 ) the discrepancy between red clump distance estimates for the 30 doradus region ( @xmath165 kpc ; romaniello et al . 2000 ) and the lmc bar ( @xmath166 kpc ; udalski 2000 ) . in general , the existence of significant line - of - sight structure in the lmc would not be surprising , given its history of gravitational interaction with the milky way ( e.g. , weinberg 2000 ) , although this hypothesis may be difficult to reconcile with some observational indications of strong regularity within the system ( see , e.g. , the h i synthesis maps of kim et al . 1998 ) . clearly , additional results are needed to determine the extent of line - of - sight structure in the lmc and to derive a best estimate of the distance to the system . we have completed analysis of two more eb systems , eros 1044 ( ribas et al . 2001 ) and hv 5936 ( in preparation ) , and have begun work on two more , eros 1066 and macho 0537 . ( see figure [ figlmc ] for the locations . ) for all of these systems , spectrophotometry spanning the range 1150 to 7500 has been obtained and the quality of the individual distance determinations will be comparable to that for hv 982 in this paper . within the next few years we hope to expand the program to include about 20 systems . our overall ensemble of targets , in addition to nailing down the distance to the lmc , will provide a detailed probe of the structure and spatial extent of this important galaxy . this work was supported by nasa grants nag5 - 7113 , hst go-06683 , hst go-08691 , and nsf / rui ast-9315365 . we are grateful for the skilled assistance of the ctio support staff during our january and december 2000 observing runs . e.f . acknowledges support from nasa adp grant nag5 - 7117 to villanova university and thanks michael oestreicher for kindly making his lmc foreground extinction data available . i.r . acknowledges the catalan regional government ( cirit ) for financial support through a postdoctoral fulbright fellowship . we thank the referee mario mateo for a variety of helpful suggestions and comments . guinan , e.f . , ribas , i. , fitzpatrick , e.l . , and pritchard , j.d . 1998b , in _ ultraviolet astrophysics beyond the iue final archive _ , eds . r. gonzalez - riestra , r. , w. wamsteker , and r. harris ( esa sp-413 ) , p. 315 cccccc 51558.6813 & 0.7203 & 446.2 & 135.6 & 3.5 & @xmath167 + 51558.7073 & 0.7252 & 444.8 & 137.3 & 3.3 & @xmath168 + 51560.6335 & 0.0862 & 185.1 & 383.7 & @xmath169 & @xmath170 + 51560.6589 & 0.0910 & 182.5 & 389.2 & 0.0 & @xmath171 + 51560.7966 & 0.1168 & 158.0 & 409.1 & @xmath172 & @xmath169 + 51561.7346 & 0.2926 & 107.8 & 462.0 & 1.9 & @xmath173 + 51561.7569 & 0.2968 & 107.3 & 463.8 & @xmath168 & 1.8 + 51561.7823 & 0.3016 & 110.4 & 459.6 & @xmath174 & 0.2 + 51563.7379 & 0.6681 & 448.3 & 128.7 & @xmath175 & @xmath176 + 51563.7601 & 0.6723 & 450.2 & 128.8 & 0.4 & @xmath176 + 51897.5627 & 0.2381 & 97.7 & 472.8 & 2.0 & @xmath177 + 51897.5859 & 0.2425 & 98.1 & 473.7 & 2.7 & @xmath174 + 51899.5515 & 0.6109 & 439.7 & 138.1 & @xmath178 & @xmath179 + 51899.5730 & 0.6149 & 443.4 & 135.9 & 0.8 & @xmath177 + 51900.5567 & 0.7993 & 409.5 & 168.2 & @xmath180 & 0.9 + 51900.5764 & 0.8030 & 410.7 & 171.5 & 1.3 & 2.2 + 51902.5584 & 0.1745 & 119.2 & 452.7 & @xmath171 & 2.0 + 51902.5782 & 0.1782 & 118.1 & 452.1 & 0.9 & @xmath181 + [ tabrv ] lc + @xmath41 ( 2000.5 ) ( deg ) & @xmath182 + @xmath183 ( km s@xmath13 ) & @xmath184 + @xmath185 ( km s@xmath13 ) & @xmath186 + @xmath187 & @xmath188 + @xmath25 ( km s@xmath13 ) & @xmath189 + @xmath39 ( r@xmath142 ) & @xmath190 + + period ( days ) & @xmath191 + eccentricity & @xmath192 + inclination ( deg ) & @xmath193 + @xmath41 ( 1994.0 ) ( deg ) & @xmath194 + @xmath195 & @xmath196 + @xmath44_{v , i}$ ] & @xmath197 + @xmath198 & @xmath199 + @xmath200 & @xmath201 + @xmath202 & @xmath203 + @xmath204 & @xmath205 + + @xmath206 ( k ) & @xmath207 + @xmath208}_{\rm ps}$ ] & @xmath209 + @xmath210 ( km s@xmath13 ) & 0 + e(b@xmath211v ) ( mag ) & @xmath212 + @xmath213 & @xmath214 + [ tabparms ] clc @xmath217 & uv bump centroid & @xmath218 @xmath219 + @xmath25 & uv bump fwhm & @xmath220 @xmath219 + @xmath221 & linear offset & @xmath222 + @xmath101 & linear slope & @xmath223 + @xmath224 & uv bump strength & @xmath225 + @xmath226 & fuv curvature & @xmath227 + @xmath61 & @xmath228 & 3.1 ( assumed ) + [ tabext ] lcc spectral type & b1 v - iv & b1 v - iv + @xmath15 ( mag ) & 15.46 & 15.32 + mass ( m@xmath229 ) & @xmath230 & @xmath231 + radius ( r@xmath229 ) & @xmath232 & @xmath233 + @xmath31 ( cgs ) & @xmath234 & @xmath235 + t@xmath81 ( k ) & @xmath207 & @xmath236 + @xmath237 & @xmath238 & @xmath239 + @xmath240fe / h@xmath241 $ ] & @xmath242 & @xmath242 + @xmath34 ( km s@xmath13 ) & @xmath243 & @xmath244 + @xmath245 ( mag ) & @xmath246 & @xmath247 + @xmath248 ( mag ) & @xmath249 & @xmath250 + age ( myr ) & + @xmath251 ( kpc ) & + [ tabstar ]

transport properties of open chaotic ballistic systems and their statistics can be expressed in terms of the scattering matrix connecting incoming and outgoing wavefunctions . here we calculate the dependence of correlation functions of arbitrarily many pairs of scattering matrices at different energies on the ehrenfest time using trajectory based semiclassical methods . this enables us to verify the prediction from effective random matrix theory that one part of the correlation function obtains an exponential damping depending on the ehrenfest time , while also allowing us to obtain the additional contribution that arises from bands of always correlated trajectories . the resulting ehrenfest - time dependence , responsible e.g. for secondary gaps in the density of states of andreev billiards , can also be seen to have strong effects on other transport quantities such as the distribution of delay times .

1007.1595

after the conjecture by bohigas , gianonni and schmit in 1984 @xcite , that chaotic systems are well described by random matrix theory ( rmt ) @xcite , research started to demonstrate this connection on dynamical grounds by means of semiclassical methods based on analyzing energy averaged products of expressions similar to the gutzwiller trace formula @xcite for the density of states , that are asymptotically exact in the limit @xmath0 . for open systems we are particularly interested in the scattering matrix @xmath1 , which is an @xmath2 matrix if the scattering leads carry @xmath3 states or channels in total . its elements can , like the gutzwiller trace formula , be expressed @xcite in terms of sums over the classical trajectories containing the stability factors of the orbits @xmath4 and rapidly oscillating phases depending on the classical actions @xmath5 of the considered trajectories @xmath6 divided by @xmath7 @xmath8 with @xmath9 with the mean level spacing of the quantum system @xmath10 . here the sum is over the scattering trajectories that connect the two channels @xmath11 and @xmath12 . for systems with two ( or more ) leads the scattering matrix breaks up into reflecting and transmitting subblocks , so we might restrict our attention to trajectories starting and ending in certain leads . in the context of spectral statistics , i.e. for the two point correlation function of the density of states containing a double sum over periodic orbits , this dynamical understanding of the conjecture @xcite was - as for other quantities - achieved in several steps . starting with the pairing of identical ( or time reversed orbits in the presence of time reversal symmetry ) the so called diagonal contribution was evaluated in ref . using a sum rule from ref . . nondiagonal contributions consisting of pairs of long orbits differing essentially only in the place where one of the orbits possesses a self crossing and the other avoids this crossing were analyzed in ref . . this was extended @xcite and formalized for orbits differing at several places , so called encounters . in the context of transport , i.e. for example for the two - point correlator of scattering matrix elements , which if restricted to the transmission subblocks is via the landauer - bttiker formalism @xcite proportional to the conductance , the diagonal contribution was calculated in ref . . an orbit pair differing only in one crossing was analyzed in ref . and this was again extended to orbits differing at several places@xcite . these results and those for closed systems agreed with results from rmt , but besides this dynamical understanding of the rmt results , these semiclassical calculations proved very successful in determining the effect of a finite ehrenfest time @xmath13 on transport quantities , starting with the pioneering work of ref . . the ehrenfest time @xcite separates times when the time evolution of a particle follows essentially the classical dynamics from times when it is dominated by wave interference . its value is obtained as the time when two points inside a wave packet initially of quantum size @xmath14 with the fermi momentum @xmath15 evolve to points with a distance @xmath16 of the linear system size . we thus get due to the exponential separation of neighboring trajectories in the chaotic case @xmath17 with the lyapunov exponent @xmath18 . before these semiclassical calculations of the ehrenfest - time dependence , there already existed theories to describe the effect of a finite ehrenfest time on the correlators of scattering matrix elements : aleiner and larkin obtained @xcite for the correlator of two transmission matrices , i.e. the conductance , an exponential suppression with increasing ehrenfest time in agreement with semiclassics . this work was however unsatisfactory in one main aspect : a small amount of impurity scattering was introduced by hand to imitate the effects of diffraction in a ballistic system . another phenomenological theory to describe the effect of a finite ehrenfest time is effective rmt @xcite . it splits the phase space and thereby also the underlying scattering matrix of the considered system into a classical and a quantum part , where the first one is determined by all trajectories shorter than @xmath13 and the second one by all trajectories longer than @xmath13 , as well as introducing an artificial phase dependent on the ehrenfest time . the predictions of this theory are only partially correct : weak localization is predicted to be independent of the ehrenfest time , while the previously mentioned theories and also numerical simulations @xcite predict it to decay with the ehrenfest time . in contrast to the quantum correction of weak localization , effective rmt gave good predictions for effects at leading order in @xmath3 such as shot noise @xcite or the gap in the density of states of a chaotic andreev billiard @xcite . staying only at the leading order in inverse channel number we will consider the correlation function of @xmath19 scattering matrices at alternating energies defined as @xmath20^n , \ ] ] where for simplicity the energy @xmath21 is measured with respect to the ( fermi ) energy @xmath22 and in units of the so called thouless energy @xmath23 with the dwell time @xmath24 measuring the typical time a particle stays inside the system . the latter is related to the heisenberg time @xmath25 via the relation @xmath26 . the ehrenfest - time dependence is incorporated in @xmath27 . the explicit form is @xmath28 with the rmt ( i.e. @xmath29 ) part of this correlation function denoted by @xmath30 . the term in ( [ eq5 ] ) derives from effective rmt @xcite . although this theory describes certain phenomena quite well , e.g. the dependence of the andreev gap on the ehrenfest time @xcite , a dynamical justification of this result is still lacking . so far ref . calculated @xmath31 for @xmath32 while refs . showed the separation into two terms in ( [ eq4 ] ) to be a consequence of the preservation under time evolution of a phase - space volume of the system . moreover they also calculated the explicit form we give in ( [ eq6 ] ) for the second term and that the first term in ( [ eq5 ] ) is proportional to the factor @xmath33 . because of ( [ eq1 ] ) the correlation function can be written semiclassically in terms of @xmath19 scattering trajectories connecting channels along a closed cycle like in fig . this leads to trajectory sets with encounters as in fig . [ fig1]b , c which can then be moved into the leads to create the remaining diagrams in fig . [ fig1 ] . including the correct prefactors and the energy dependence , the correlation function becomes semiclassically @xmath34 @xmath35 are the times trajectories @xmath6 spend inside the system , and we identify the channels @xmath36 . note that ( [ eq6a ] ) and this identification imply that the trajectories and their partners ( traversed in reversed direction ) considered in @xmath31 form a closed cycle . ] in this paper we want to show how ( [ eq4],[eq5],[eq6 ] ) can be obtained using the trajectory based methods developed in refs . . in sec.[part1 ] we consider the first term in ( [ eq4 ] ) : we show that the prefactor @xmath30 of the exponential is indeed given by the rmt expression obtained in ref . and that this is multiplied by the exponential given in ( [ eq5 ] ) . the underlying diagrams considered here are the same as the ones occurring also in the semiclassical calculation of the rmt contribution . in sec . [ part2 ] we consider the second term in ( [ eq4 ] ) and show how this contribution arises from trajectories that are always correlated . furthermore we show in sec . [ mixing ] that there exist no mixed terms between the first and the second term in ( [ eq4 ] ) , that could result - expressed in terms of the considered diagrams - from correlations between trajectories always correlated with each other on the one side and trajectories only correlated with each other during encounters on the other side . the main idea in this section is to split our diagrams in a different way compared to the semiclassical analysis without ehrenfest time ( referred to as the rmt - treatment ) and the analysis of the ehrenfest - time dependence of the cases @xmath32 in ref . : in the semiclassical calculation one considers an arbitrary number of orbits encountering each other . it turns out in the rmt - treatment to be sufficient to consider only encounters where all orbits are linearizable up to the _ same _ point , see for example fig . [ fig2 ] . when taking into account the ehrenfest - time dependence this is no longer sufficient as was first shown in ref . ; see fig . [ fig3 ] for an example of an additional diagram analyzed in this case . the main complication arising in ref . is then to treat these encounters . to simplify the calculation we imagine these encounters being built up out of several encounters , each of which consist of two encounter stretches . we have distinguished these 2-encounters by different boxes in fig . [ fig4 ] . in this way it is much easier to consider encounter diagrams of arbitrary complexity with finite ehrenfest time , which did not appear in the formalism used in ref . . the encounter stretches are marked by a box ( shown red ) . ] we first illustrate our procedure by considering three correlated orbits with two 2-encounters as in fig . [ fig4 ] and show how the result given in ref . can be obtained in this case and then we treat the general case of @xmath37 orbits with @xmath38 independent or overlapping 2-encounters . in the treatment of the rmt - type contribution ( [ eq5 ] ) we first consider the case in which all the encounters occur inside the system . for @xmath39 we have the two semiclassical diagrams in fig . [ fig1]b , c which include a trajectory set ( of three original trajectories and three partners ) with two 2-encounters in fig . [ fig1]b and a single 3-encounter in fig . [ fig1]c . by shrinking the link connecting the two encounters in fig . [ fig1]b we can see how we deform them into the diagram in fig . [ fig1]c and we use this idea in our ehrenfest - time treatment . for the calculation of contributions resulting from diagrams differing in encounters we first need to review the notation and the important steps of the corresponding calculation in ref . . an encounter of two orbits is characterized by the difference of the stable and unstable coordinates @xmath40 and @xmath41 measured in a poincar surface of section ( pss ) put inside the encounter ; see fig . [ fig4 ] . in terms of these coordinates the duration of the encounters is given by @xmath42 derived from the condition that the coordinates @xmath43 are only allowed to grow up to a classical constant @xmath44 ( which is later related to the ehrenfest time ) . the weight function measuring the probability to find these encounters is obtained by integrating over all possible positions where the encounter stretches can be placed and dividing by the volume of available phase space ( in the corresponding closed system ) @xmath45 and further by the durations of the encounters @xmath46 to avoid overcounting the same set of correlated trajectories . the action difference @xmath47 between the orbits is in general given by a quadratic form of the coordinates @xmath43 determined by where the partner trajectories must pierce the pss s to reconnect in the right way to form a closed cycle . for example for a 3-encounter one obtains @xcite @xmath48 where the prime denotes that the coordinates are measured in _ one _ pss from the central trajectory . if we instead measure the coordinates in two _ different _ sections , we obtain @xmath49 where the time @xmath50 denotes the time the particle needs to travel between the two sections . this leads in the limit of well separated encounters to @xmath51 . from this and from ref . we can draw the following conclusions for the form of the action difference in the case of an arbitrary number of ( possibly overlapping ) 2-encounters : in the case of @xmath52 well separated 2-encounters we obtain for the action difference @xmath53 . when these encounters overlap the action difference can differ from the last expression by terms exponentially damped with the time difference between the two sections . in our treatment , the overall contribution @xmath54 of the two 2-encounters ( depicted in more detail in fig . [ fig4 ] ) is obtained by allowing the upper trajectory to possess a minimal length of the first 2-encounter and the lowest one a minimal length of the second 2-encounter . the middle trajectory , which passes through both encounters has a minimal length given by the maximum of the two encounter times as we allow the encounters to overlap . however we do not yet allow one encounter to be subsumed into the other so we also set the time @xmath55 between the start of the first encounter and the end of the second to be longer than the maximum encounter time . to write down the semiclassical contribution of the diagram in fig . [ fig4 ] we sum over the number of possible classical orbits using the open sum rule @xcite . converting the time integrals resulting from this rule to time integrals with respect to link durations , we obtain @xmath56 where the superscript refers to fig . we have summed over the possible channels , and @xmath57 with @xmath58 label the links from the channels to the encounters . in ( [ eq8 ] ) where @xmath59 and @xmath60 , @xmath40 and @xmath41 with @xmath61 are the stable and unstable coordinate differences between the two parts of the trajectories piercing through a pss placed in the @xmath11-th encounter . as explained above , the action difference is given by @xmath62 . by expanding the part of the exponential @xmath63 containing this @xmath50-dependent part into a taylor series one verifies easily that contributions from higher order terms than the leading ( time independent ) one are of higher order in @xmath64 and can be neglected . this reasoning also holds for diagrams with more than two 2-encounters . in the first line of ( [ eq8 ] ) we can see that each integral over the links is weighted by its classical probability to remain inside the system for the time @xmath57 which decays exponentially with the average dwell time @xmath24 . we only want to consider trajectory sets where the _ whole _ diagram remains inside the system , as if any parts were to hit the lead and escape the diagram would be truncated at that point . with the energy dependence in ( [ eq6a ] ) this gives the factors @xmath65 in ( [ eq8 ] ) . inside the encounters however we have trajectory stretches that are so close that the conditional survival probability of secondary traversals is 1 and we need only consider the survival probability of one stretch . if that stretch does not escape then neither will the other . the energy dependence still depends on the total time so that encounter 1 would lead to the factor @xmath66 . with the overlap , encounter 2 would then have a more complicated exponential factor , but because the time @xmath55 ( between the two outer ends of the encounter stretches on the middle trajectory shown in fig . [ fig4 ] ) passes through both encounters their survival probability ( of both stretches of both encounters ) can be expressed as the survival probability of a stretch of duration @xmath55 as in the last line of ( [ eq8 ] ) . the energy dependence instead also requires the extra traversal of the encounters as given by the exponential factor in the middle line of ( [ eq8 ] ) . performing the integrals in the first line of ( [ eq8 ] ) we have @xmath67 where we have moved all of the ehrenfest - time dependent parts into the factor @xmath68 with the superscript again referring to fig . [ fig4 ] , @xmath69 here we can also see the connection with the previous ehrenfest - time treatment of such a diagram . when @xmath70 the two encounters separate ( the integrals can then be further broken down into products ) and this is the case in which the trajectories can be considered to have two independent 2-encounters as in ref . . because we choose a different lower limit however , the contribution above also includes some of the diagrams previously treated as 3-encounters in ref . . the reason for our choice becomes clear in the following steps . we first substitute @xmath71 , @xmath72 and then substitute @xmath73 , @xmath74 and perform the @xmath75-integrals using the explicit form of the @xmath42 ( for details of this calculation , see also ref . ) . this results in @xmath76 now we substitute @xmath77 and obtain @xmath78 here we split the resulting expression into an @xmath7-independent integral ( or more exactly trivially dependent on @xmath7 ) , that exists due to the energy average that is always contained in our calculations , and an ehrenfest - time or @xmath7 dependent part with @xmath79 . this contains the ehrenfest - time dependence that is expected from ( [ eq5 ] ) , so ( [ eq13 ] ) already shows that the diagrams considered here yield the correct ehrenfest - time dependence . now we consider the case in which one of the two 2-encounters lies fully inside the other one , which we will refer to as a generalized version of a 3-encounter , as depicted in fig . [ fig5 ] . for the ehrenfest - time dependent part we have a similar contribution as in ( [ eq10 ] ) with two differences : first @xmath55 is best defined as the distance between the midpoints of the two different encounter stretches and so it can vary between @xmath80 second the survival probability of the encounters is determined by the longest encounter stretch and is independent of @xmath55 . the ehrenfest - time dependent part can then be written as @xmath81 performing the @xmath55 integral and following the same steps as for ( [ eq12],[eq13 ] ) , we find @xmath82 this part also shows an ehrenfest - time dependence as expected from ( [ eq5 ] ) . note that when performing the @xmath55-integral the result in this case is of course proportional to @xmath83 which contains , after the substitution from @xmath84 to @xmath85 , two times the same terms linear in @xmath13 with different signs that thus cancel each other . up to now we have concentrated on encounters inside the system , but apart from these diagrams we also need to consider diagrams where the encounters touch the opening as in fig . [ fig1]d - h . we will , as above , start by considering encounters built up out of two 2-encounters and we focus here on how the calculation of the contribution is changed when encounters move into the lead compared to the treatment of encounters inside the system . as can also be found in more detail in ref . when encounters touch the lead one includes in the semiclassical expressions for encounters inside the system an additional time integral running between zero and the corresponding encounter time , which characterizes the duration of the part of the encounter stretch that has not yet been moved into the lead . of it remains inside the system . ] we consider two encounters with durations @xmath86 and @xmath87 with the second encounter touching the opening as in fig . [ fig1]d and drawn in more detail in fig . [ fig6 ] . as the second encounter enters the lead we now define the time @xmath55 to be from the start of the first encounter until the lead and introduce the time @xmath88 which measures the part of the second encounter that has not yet been moved into the lead . we also separate the ehrenfest - time relevant contribution @xmath89 in this detailed calculation into two cases : in the first case ( a ) ; @xmath90 ; we have @xmath91 with the additional integral over the time @xmath88 @xmath92 where the limits on the time integrals derive from the fact that the first encounter is not allowed to touch the lead ( this would be included as a 3-encounter ) and that the second must . performing the time integrals this is @xmath93{\rm e}^{-t_{{\rm enc},1}(1 - 2{\rm i}\epsilon)/\td } , \end{aligned}\ ] ] with the first and second term in the square brackets resulting from the upper and lower limit of the @xmath88-integration . in the second case ( b ) ; @xmath94 ; we obtain @xmath95 , \nonumber\end{aligned}\ ] ] where the more complicated limits derive from not allowing the second encounter to move further left than the first . after integrating we have @xmath96{\rm e}^{-t_{{\rm enc},1}(1 - 2{\rm i}\epsilon)/\td } \right . \nonumber\\ & & { } + \frac{1}{(1 - 2{\rm i}\epsilon)}{\rm e}^{-t_{{\rm enc},1}(1 - 3{\rm i}\epsilon)/\td } \nonumber \\ & & { } - \left . \frac{1}{(1 - 2{\rm i}\epsilon)}{\rm e}^{{\rm i}\epsilon t_{{\rm enc},1}/\td}{\rm e}^{-t_{{\rm enc},2}(1 - 2{\rm i}\epsilon)/\td } \right ] . \nonumber\end{aligned}\ ] ] the last line comes from the upper limit of the second @xmath88-integral and has the same ehrenfest - time dependence as before and in line with ( [ eq5 ] ) . likewise the upper @xmath88 time limit for case a in ( [ eq17 ] ) leads to the same dependence and we can conclude that the upper limits of the @xmath88-integrations yield contributions similar to when the encounters are inside the system and with the same ehrenfest - time dependence . the remaining ( lower ) limits of the time integrations in ( [ eq17],[eq19 ] ) give contributions possessing a different ehrenfest - time dependence which however always yield zero in the semiclassical limit due to the fact that the corresponding terms contain no @xmath97 in the exponentials containing @xmath24 . apart from the action difference , the only term depending on @xmath98 is the @xmath99 . the resulting expression is rapidly oscillating as a function of the energy @xcite and thus canceled by the energy average . we can repeat this procedure for the remaining diagrams in fig . [ fig1 ] and see that the contributions are determined by the upper limits of the corresponding @xmath88 integrals . for the diagrams with a generalized 3-encounter ( fig . [ fig1]g , h ) this follows as for the 3-encounter inside the system but for fig . [ fig1]e where the two 2-encounters enter different channels ( and possibly different leads ) there is an additional subtlety . the two encounters are still allowed to overlap , so that during the time @xmath55 the stretch now connecting both channels can always be inside encounters but the individual encounters are not allowed to connect leads at both ends . these additional possibilities are considered later , where if both encounters connect to the leads at both ends we actually have a band of correlated trajectories ( treated in sec . [ part2 ] ) and if only one does we have a mixed term ( treated in sec . [ mixing ] ) . with this organization of the encounters we see that each diagram has the same ehrenfest - time dependence as when the encounters are inside the system , which is in line with ( [ eq5 ] ) . the reasoning so far in this section proves the form of ( [ eq5 ] ) for @xmath39 . first of all , we know that the resulting contribution from the diagrams analyzed contains an overall factor @xmath100 . secondly , the remaining integrals are independent of @xmath7 and thus independent of the ehrenfest time . thirdly , the diagrams we analyze are the same as the ones analyzed in the rmt - case in the first part of ref . . as in the limit @xmath101 we must recover that previous result , this implies that @xmath102 in ( [ eq5 ] ) is indeed given by the rmt - expression . before proceeding to the general case however , we first want to illustrate how our calculation can be used to obtain , apart from just the ehrenfest - time dependence , the complete dependence on @xmath24 and @xmath21 . we therefore start for the two 2-encounters from fig . [ fig4 ] from the last expression in ( [ eq13 ] ) and perform first the @xmath103-integral @xmath104 where it is simpler to rewrite the result in terms of the maximum and minimum value of @xmath105 . for calculating the @xmath106-integrals we perform partial integrations ( integrating each time the @xmath107 functions ) and then perform the resulting integrals from zero to infinity @xmath108 in the first line the additional terms due the partial integration are either zero or cancel due to the energy average . the final result in the last line of ( [ eq22 ] ) can be also obtained by replacing @xmath109 and @xmath110 and performing the integrals with respect to @xmath111 from zero to infinity . to evaluate the contribution from the generalized 3-encounter in fig.[fig5 ] we again perform two partial integrations in ( [ eq16 ] ) and obtain @xmath112 where we have also left out the terms from the partial integrations that cancel due to the energy average . with these results we can now show how they connect to the rmt - type results . for this we need to split our diagrams differently and first we need the result for an ideal 3-encounter as depicted in fig . [ fig2 ] whose contribution was calculated @xcite to be @xmath113 with the extra factors in ( [ eq9 ] ) it is clear how in the limit @xmath114 this reduces to the rmt - type result for a 3-encounter as in ref . . all the remaining contributions should be collected together as two 2-encounters , and as the ideal 3-encounter is included in our generalized 3-encounter we first subtract ( [ eq24 ] ) from ( [ eq23 ] ) @xmath115 before we add the result from our separation of two 2-encounters in ( [ eq22 ] ) we remember that in the treatment we enforce that the first encounter is to the left of the second . the result in ( [ eq25 ] ) does not have this restriction so we divide by 2 to ensure compatibility and then add the result in ( [ eq22 ] ) to obtain @xmath116 this then reduces to the rmt - type result for trajectories with two 2-encounters when @xmath114 as in ref . . the agreement of these results with the previous ehrenfest - time treatment@xcite can be seen as the result in ( [ eq26 ] ) including both the result from two independent 2-encounters as well as most of the contribution of the diagram referred to as a 3-encounter in ref . . when splitting the contribution in a different way as in ref . this also leads to terms in both classes that contain different ehrenfest - time dependencies that only cancel when summed together . although up to now we have just reproduced results from ref . , the procedure used here has the advantage that it yields a simple algorithm for determining the ehrenfest - time dependence of the corresponding contributions to @xmath117 at arbitrary order . for our example of @xmath39 we showed how it was possible to split the diagrams into two classes that _ both _ showed the ehrenfest - time dependence as expected from ( [ eq5 ] ) . we want now to show how to generalize our way of splitting considered for three trajectories to diagrams containing @xmath37 trajectories . we start again with the situation in which all of the encounters are inside the system and by considering a case analogous to fig . [ fig4 ] , but now involving @xmath37 instead of three trajectories . we first take a diagram that consists of a ladder of @xmath118 2-encounters so that the central @xmath119 trajectories each contain two encounter stretches while the two outside trajectories only contain one encounter stretch each . this situation is depicted in fig . [ fig7 ] and the encounters are thus characterized by @xmath118 @xmath120-coordinates . in this case we obtain for the ehrenfest - time relevant contribution @xmath121 that the @xmath55-integral measuring the time difference between the end points of the two encounter stretches on the middle orbit in ( [ eq10 ] ) is replaced by @xmath119 integrals over times @xmath122 with the same meaning as @xmath55 ; they measure the time difference between the end points of the two ( consecutive ) encounter stretches on the central trajectories containing two encounter stretches . these times likewise run from the maximum of the corresponding encounter times to infinity . the survival probability is determined by a single ( artificial ) stretch that runs through all the encounters so that the exponential term describing the @xmath24- and @xmath21-dependence is now given by @xmath123 where @xmath124 are the durations of the @xmath118 individual 2-encounters and the middle exponential compensates for the fact that the middle encounters are traversed by two @xmath57 and that only one traversal should contribute to the survival probability . setting @xmath125 and repeating now the steps of ( [ eq12],[eq13 ] ) we find the ehrenfest - time dependent factor in this case to be @xmath126 again confirming the ehrenfest - time dependence of ( [ eq5 ] ) . along with the case in which none of the encounters in the ladder can move completely inside another , we can look at the opposite extreme where all the encounter stretches lie inside of the encounter @xmath52 with the longest duration @xmath127 where @xmath46 are the durations of the @xmath118 individual 2-encounters with one of the two orbits containing the stretch of duration @xmath128 . this situation is like a generalization of the diagram in fig . [ fig5 ] and we similarly now define the times @xmath57 to be measured between the centers of encounter @xmath11 and the encounter @xmath52 of maximum length ( with @xmath129 ) . here the same ehrenfest - time dependence @xmath130 follows by taking into account that each time @xmath122 has a range of variation of size @xmath131 and that the @xmath24- and @xmath21-dependent exponential in this case is @xmath132 this yields for the ehrenfest - time dependent factor @xmath133\nonumber\\ & & \times { \rm e}^{-{\rm i}\epsilon\sum_{i=1}^{n-1}\ln x'_{i}/(\lambda\td)}{\rm e}^{-\tau\left(1-{\rm i } n\epsilon \right ) } , \end{aligned}\ ] ] confirming again the ehrenfest - time dependence predicted by ( [ eq5 ] ) . of course it is additionally possible to have a mixed form between these two extreme cases . this means that some 2-encounters only overlap like in the case of a ladder diagram while the others form ` single ' encounters ; see fig.[fig7a ] for a possible diagram . we then have a ladder of ` combined ' encounters that themselves can be made up of one or more 2-encounters . the treatment of such diagrams is very similar to the treatments above , and the only slight complication is in defining the appropriate times to extract the ehrenfest - time dependence . we recall that the first and last trajectories only pass through one 2-encounter while the central @xmath119 trajectories pass through two . numbering the central trajectories from @xmath134 , so that trajectory @xmath11 has encounters @xmath11 and @xmath135 along it , we divide them into two sets : those whose encounter stretches lie fully inside each other or a connected encounter , as in the case of a single encounter above , that we place in the set @xmath136 . we place the remaining orbits with two stretches separated as in ladder diagrams in the set @xmath137 . as mentioned above , we condense the overlapping encounters into combined encounters and record in the set @xmath138 the labels of the trajectories that pass through the second stretch in each combined encounter . we also include in this set combined encounters made of a single separated ( ladder ) 2-encounter . we then use @xmath139 for @xmath140 to record the number of additional consecutive trajectories involved in the same encounter , so that @xmath141 for separated 2-encounters and @xmath142 for larger encounters corresponding to the single encounter case above . if the last combined encounter is a 2-encounter its second stretch is traversed by the last trajectory in the diagram which we number by @xmath143 and include as an element of @xmath138 . for example , for the diagram in fig . [ fig7a ] we would have @xmath144 , @xmath145 , @xmath146 , @xmath147 , @xmath148 and @xmath149 . for the elements @xmath140 we also label by @xmath150 the corresponding encounter of maximum length among those from encounter @xmath11 to encounter @xmath151 . to be precise , the two stretches that stay together longest have length @xmath152 while the other encounter times are defined by how long the remaining stretches remain close to one of the two longest . for the trajectories passing through two separated condensed encounters @xmath153 we define the times @xmath122 to include the whole of the leftmost and rightmost condensed encounters , i.e. to include the encounters @xmath154 and @xmath155 , where @xmath156 is the largest element in @xmath138 that is @xmath157 . in this case the @xmath24- and @xmath21-dependent exponential can be written as @xmath158 where @xmath159 is @xmath138 with its largest element removed so that the second term accounts for the overlap between the @xmath122 s and the third term for the fact that there is also no overlap at the start of the first such stretch . this equation incorporates both ( [ ladderexponentialeqn ] ) and ( [ singleencounterexponentialeqn ] ) . if we introduce the notation @xmath160\equiv\max\left\lbrace t_{{\rm enc},i},t_{{\rm enc},j}\right\rbrace$ ] , we can then define the times @xmath161 $ ] as before ( [ lad ] ) . making the substitutions as done previously yields for the ehrenfest - time dependent factor @xmath162\nonumber\\ & & { \rm e}^{-\sum_{i\in\omega_2}\left(t'_i+\hat{m}[k_{\bar{i}},k_{i+1 } ] /\lambda\right)\left(1-{\rm i}\epsilon\right)/\td}\nonumber\\ & & \times{\rm e}^{-\sum_{i\in\bar{\lambda } } \ln x'_{k[i]}\left(1-{\rm i}\epsilon\right)/(\lambda\td)}{\rm e}^{\ln x'_{k[1]}\left(1-{\rm i}\epsilon\right)/(\lambda\td)}\nonumber\\ & & \times{\rm e}^{-{\rm i}\epsilon\sum_{i=1}^{n-1}\ln x'_{i}/(\lambda\td ) } { \rm e}^{-\tau\left(1-{\rm i } n\epsilon \right ) } , \end{aligned}\ ] ] with @xmath163=\max\left\lbrace -\ln x'_{i},-\ln x'_{j}\right\rbrace$ ] . as @xmath137 and @xmath159 must have the same number of elements , this again shows the predicted ehrenfest - time dependence . up to now we restricted our discussion to diagrams in which each trajectory is involved in one or two encounters this is however not yet the most general case where the only restriction is that each trajectory contains at least one encounter stretch , so that some trajectories can also contain more than two encounter stretches . note that the situation where two trajectories interact ( pass through the same 2-encounter block ) more than once can not occur at leading order in inverse channel number . an example of a diagram that is possible is depicted in fig . [ fig8 ] . in the most general case we define the times @xmath122 slightly differently : first we separate the @xmath164 trajectories that have one encounter stretch from the remaining @xmath165 that have more than one . then we number our encounters accordingly , first those along the trajectories with one encounter stretch with duration @xmath124 , @xmath166 then the remaining encounters with duration @xmath124 , @xmath167 . for the @xmath165 trajectories with two or more encounter stretches we now define @xmath57 , @xmath168 , to be the time difference between the outer edges of the outermost encounters along those trajectories . for any trajectories with more than two encounter stretches we will need additional time differences to fully fix the positions of the encounters . because we defined the times @xmath57 to go through the outmost encounters , importantly the exponential factor with the survival probability and the energy dependence does not depend on these additional time differences and is given by @xmath169 where the middle term ensures that the survival probability only includes one copy of each encounter and the energy dependence involves all traversals of all the encounters . in the case of more than two encounter stretches on one orbit . the encounter stretches are shown thicker ( blue ) . ] for the remaining times we notice that , starting with the ladder system with 2 trajectories containing one encounter stretch and @xmath119 trajectories containing two stretches , every time we increase the number of trajectories with one encounter stretch we simultaneously increase the number with more than two . therefore there are @xmath170 additional time differences needed to fix the positions of the central encounters along trajectories with more than two and we define times @xmath171 for @xmath172 from the left hand side of one encounter stretch to the right hand side of the next encounter stretch following on the right on those trajectories , see also fig . [ fig9 ] . as the encounters are ordered , they are not ( yet ) allowed to be subsumed by each other or pushed past the outside encounters . the ranges of the times @xmath171 are then fixed by these restrictions . using again @xmath160 $ ] defined after ( [ mix ] ) in the following to make the notation more compact , we obtain for a trajectory containing @xmath173 encounter stretches of durations @xmath124 , @xmath174 , as illustrated in fig . [ fig9 ] , the integrals @xmath175}^{t_i-\sum_{o=1}^{m-3}\left(\tilde{t}_o - m[o , o+1]\right)}{\rm d}\tilde{t}_{m-2}\nonumber\\ & = & \int_0^{t_i - m[1,2]}{\rm d}\tilde{t}_1'\ldots\int_0^{t_i-\sum_{o=1}^{m-3}\tilde{t}_o'-m[m-2,m-1]}{\rm d}\tilde{t}_{m-2}'.\nonumber\\ \label{eq32}\end{aligned}\ ] ] in the second line we substituted @xmath176 $ ] . the time differences @xmath57 , which are more important for the ehrenfest - time dependence , must instead just be longer than the maximal length of the encounter stretches lying on the considered trajectory . in general the numbering of the encounters and time differences can be more complicated than in fig . [ fig9 ] so we define @xmath177 to be a list of length @xmath139 of the encounters enclosed by the time @xmath57 ( including the outer encounters ) and @xmath178 a list of the corresponding @xmath179 times @xmath180 between the ends of those encounters . now we can make the substitution @xmath181 . after this substitution we recognize that ( [ eq32 ] ) has become independent of @xmath7 or the ehrenfest time . following then the steps in ( [ eq12],[eq13 ] ) we obtain @xmath182\right)/\lambda}{\rm d}\tilde{t}_{l_{1}}'\ldots\nonumber\\ & & \times \int_0^{t_i'-\sum_{o=1}^{m-3}\tilde{t}_{l_{o}}'-\left(\ln x'_{\max , i}+\hat{m}[l_{m-2},l_{m-1}]\right)/\lambda}\hspace{-0.5em}{\rm d}\tilde{t}_{l_{m-2 } } ' \nonumber \\ & & \times { \rm e}^{-\sum_{i=1}^{n - k}\left(t'_{i}-\ln x'_{\max , i}/\lambda\right)\left(1-{\rm i}\epsilon\right)/\td } \nonumber\\\ & & \times { \rm e}^{-\sum_{i = k+1}^{n-1}\ln x'_{i}/(\lambda\td)}{\rm e}^{-{\rm i}\epsilon\sum_{i=1}^{k}\ln x'_{i}/(\lambda\td ) } \nonumber \\ & & \times { \rm e}^{-\tau\left(1-{\rm i } n\epsilon \right ) } , \end{aligned}\ ] ] with @xmath183 linked to the duration of the longest encounter stretch contained within @xmath122 . we have also used @xmath163 $ ] defined after ( [ mix1 ] ) and dropped the explicit @xmath11 dependence of @xmath184 and @xmath173 above . again we obtain the ehrenfest - time dependence predicted by ( [ eq5 ] ) . as in the case of the ladder diagram above , we can also have the possibility of some encounter stretches being contained in larger encounter stretches and some separated from those larger encounters , see fig . [ fig9a ] encounter stretches here are allowed to be contained fully inside others . the encounters are marked by boxes ( shown in different colors ) . ] for an example of a possible diagram . this just implies that some of the @xmath122 integrals have to be treated as was done in the case of the configuration shown in fig . [ fig5 ] , and the ehrenfest - time dependence predicted by ( [ eq5 ] ) also follows in this case . when the encounters are allowed to enter the lead we again have to consider times representing how far each encounter has moved into the lead ( actually how much of the encounter remains inside the system ) . as for the case treated in detail for @xmath39 it is only the upper limit ( namely the full encounter time ) of these time integrals that have the necessary encounter time dependence to contribute in the semiclassical limit . the reasoning for @xmath39 can then be carried over directly to the more general cases as the upper limits of these integrations yield contributions that are ( up to constant factors ) the same as the ones obtained when the encounters are inside the system . we thus obtain the same ehrenfest - time dependence from encounters moved into the leads . the separate diagrams considered in the rmt - type semiclassical treatment @xcite can be created from the original collapse of trajectories onto each other and by sliding the individual encounters together or into the leads . the ehrenfest - time treatment however suggests treating all of these possibilities instead as part of continuous families . what we have shown above in this section is that , if we partition this family in a particular way , for any partition we can find a suitable set of coordinates that allows us to transform the semiclassical contribution so that we can extract the overall ehrenfest - time dependence . though the exact details of this transformation depend on the structures of the partition , the algorithmic routines described above all lead to the same ehrenfest - time dependence . each partition and hence family then has the factor @xmath185 and no other ehrenfest - time or @xmath7 dependence . as we know that we must recover the rmt - type result @xmath30 in ( [ eq5 ] ) when @xmath114 ( since we treat the same diagrams ) with no further ehrenfest - time dependence , we then obtain the full result in ( [ eq5 ] ) and hence provide a semiclassical justification of the effective rmt ansatz . correlated trajectories . the length of the orbits is marked by a box ; the duration of the encounter @xmath186 is marked by a dotted box . ] in this section we determine the so called classical contribution in ( [ eq6 ] ) . to obtain this contribution @xmath187 semiclassically we consider a band of @xmath37 trajectories that are correlated ( inside the same encounter ) for their entire duration between entering and leaving the system as in fig.[fig10 ] . this implies that all the trajectories have the same length @xmath55 and that the maximum of the differences @xmath188 between their stable and unstable coordinates lies below the constant @xmath44 ( related to the ehrenfest time ) . for the case @xmath189 this configuration was first considered in ref . and then extended to @xmath39 in ref . . for our calculation we follow ref . and place a pss at a distance @xmath190 from the left end of the trajectories while the remaining time on the right of the section is denoted by @xmath191 . the semiclassical contribution @xmath187 can be written as @xmath192 where we only include one traversal of the band in the survival probability and the restrictions on the @xmath193 and @xmath194 integrals ensure that the band always remains together under the exponential divergence of the trajectories due to the chaotic dynamics . performing an integral over @xmath195 and the @xmath41 integrals gives @xmath196 where @xmath197 . using that in the semiclassical limit @xmath198 with the heaviside theta function @xmath199 , we finally obtain @xmath200 proving the ehrenfest - time dependence of the @xmath187 in ( [ eq6 ] ) . and @xmath201 , and the durations of the band before and after the pss by @xmath190 and @xmath202 . ] finally we want to consider possible correlations between trajectory structures giving the rmt - type contribution and those giving the classical part , i.e.contributions from correlations between bands of trajectories ( that are always correlated with each other ) and trajectories that are only correlated with each other during encounters . in particular we want to show that diagrams that have a correlated band that has any encounter with other trajectory structures ( with encounters ) give no contribution in the semiclassical limit . this , once generalized , then excludes the existence of mixed terms in ( [ eq4 ] ) so that ( [ eq4 ] ) is complete . first we consider the case in which @xmath143 of the trajectories form a correlated band with the remaining trajectory meeting the band in an encounter inside the system as depicted in fig . [ fig11 ] . this contribution @xmath203 to the correlation function @xmath31 can be written by treating the correlated band as before and introducing the times @xmath204 and @xmath201 to represent the durations of the parts of the trajectory that encounter the band on the left and on the right of the pss . it reads @xmath205\nonumber \\ & & \times\int_{c { \rm e}^{-\lambda t_1}<|s|\leq c}\hspace{-1em}{\rm d}s\int_{c { \rm e}^{-\lambda t_2}<|u|\leq c}\hspace{-1em}{\rm d}u \nonumber \\ & & \times \frac{{\rm e}^{\left({\rm i}/\hbar\right)\delta s}{\rm e}^{{\rm i}\epsilon \left(t_{\rm enc}+(n-2)(t_1+t_2)\right)/\td}}{\left(t_1+t_2\right ) } , \end{aligned}\ ] ] where @xmath206 is the time during which the remaining trajectory encounters the band . in this case no overcounting factor @xmath207 occurs as the limits of the @xmath120-integrals in the third line are chosen such that the time integral accounting for all possible positions of the single encounter stretch with respect to the band cancels this factor . choosing different limits for the @xmath120-integrals we could replace the current overcounting factor of @xmath208 in ( [ eq38 ] ) again by the usual @xmath207 as in sec . [ part1 ] . the integrals over the @xmath209 @xmath120-coordinates within the band are performed in the same manner as in the last section in ( [ eq36 ] ) and lead to the same heaviside function @xmath210 . for the integrals in the third line in ( [ eq38 ] ) over the differences @xmath120 between the coordinates of a band trajectory and the trajectory encountering it we obtain @xmath211 with @xmath212 . this heaviside function is opposite to the one from the band , so that the total contribution vanishes due to the opposing restrictions of the theta functions . note that when evaluating the ehrenfest - time dependence of the rmt contributions in sec . [ part1 ] we chose a different way for performing the integrals . in sec . [ part1 ] we performed restricted time integrals first and unrestricted @xmath85 integrals afterwards . the different order of calculating these integrals leads to the theta functions which make it clear that mixed terms like in fig . [ fig11 ] vanish ; this is why we use this ordering here . if we move more trajectories from the band ( composed of at least two trajectories ) to the trajectory structure with encounters we still obtain these opposing heaviside functions and hence no contribution . a similar reasoning can be applied if the encounter of a trajectory ( or part of a trajectory structure ) with a band does not happen inside the system but enters the lead at the beginning or the end . in this case we obtain an additional time integral with respect to the time of the encounter that remains inside the system but , as the @xmath120-integrals still yield the same heaviside functions , this contribution also vanishes . note that if we move both ends of the encounter into the leads then the encountering trajectory can be considered as part of the band and treated as above or in sec . [ part2 ] . the reasoning in this section applies to an arbitrary number of bands of correlated trajectories connected by trajectories that are only correlated in encounters . therefore all such mixed terms vanish . up to now we have concentrated on energy dependent correlation functions involving the whole scattering matrix . because we use the same semiclassical diagrams our result can be applied directly to dc - transport properties of chaotic systems such as the moments of the transmission or reflection eigenvalues . assuming the system has two scattering leads and taking just the transmission subblock @xmath55 of the scattering matrix connecting the @xmath213 channels in lead 1 to the @xmath214 channels in lead 2 ( without an energy difference ) the moments of the transmission eigenvalues can be written as @xmath215^{n},\ ] ] where @xmath216 is the total number of channels . within the landauer - bttiker@xcite approach to quantum transport the moments carry information about statistical properties of the transmission in the phase coherent regime , the counting statistics@xcite . for example the first moment characterizes the average conductance @xmath217 and the second one the power of the shot noise @xmath218 . using the semiclassical results from this paper we can simply write @xmath219 where we have included the probability @xmath220 of starting in lead 1 in the moments . equation ( [ eq40 ] ) can be generalized to ac - transport considered in ref . by including in the latter equation the @xmath21-dependent factors given in ( [ eq5],[eq6 ] ) . the result in ( [ eq40 ] ) again splits into two parts with the first involving the semiclassical moments @xmath221 calculated in ref . , which in turn lead to the random matrix probability distribution @xcite . the second term in ( [ eq40 ] ) leads to the classical bernoulli distribution where the transmission amplitude @xmath222 is 1 with probability @xmath223 and 0 otherwise ( i.e. @xmath220 ) . the ehrenfest time then provides a smooth interpolation between these two distributions giving a weight @xmath224 to the rmt one and the remaining weight @xmath225 to the classical one . a similar formula and result follows for the moments and probability distribution of the reflection eigenvalues whose zero ehrenfest time contributions can be simply derived from the treatment in ref . . taking the full correlation functions @xmath31 it is possible to obtain not only the ehrenfest - time dependence of the density of states of chaotic andreev systems , covered in detail in ref . , but also the moments and distribution of the wigner delay times . we start with the wigner - smith matrix @xcite @xmath226 which can be shown to be hermitian by using the unitarity of the scattering matrix . because of this unitarity @xmath227 can also be written as @xmath228\big\vert_{\epsilon=0 } , \ ] ] where the scattering matrix energy differences are measured with respect to the energy @xmath22 . the delay times are simply the eigenvalues of @xmath227 so their moments are @xmath229^{n } .\ ] ] using the relation @xmath230^{n}\big\vert_{\epsilon=0 } = \left[f'(0)\right]^{n } , \ ] ] the moments of the delay times are @xcite @xmath231^n\big\vert_{\epsilon=0 } . \nonumber\end{aligned}\ ] ] expanding ( [ eq45 ] ) , the moments can be expressed as @xmath232 in terms of the correlation functions calculated before . as this is additive we can look at the two parts of the ehrenfest - time dependent results in ( [ eq4 ] ) separately . for the first part of the @xmath233 however , it is possible to put our ehrenfest dependence into the framework of ref . where the moments and the probability distribution @xmath234 of the wigner delay times @xmath235 were calculated ( without the ehrenfest time ) . we can actually obtain the result in a simple way . using ( [ eq44 ] ) again we can see that since @xmath236\big\vert_{\epsilon=0 } , \ ] ] we have @xmath237^{n } & = & \frac{\td^{n}}{{\rm i}^{n}n ! } \frac{{\rm d}^{n}}{{\rm d}\epsilon^{n } } \sum_{k=1}^{n } \\ & & \left(-1\right)^{n - k } \left(\begin{array}{c}n\\ k \end{array}\right ) c(\epsilon,\tau , k){\rm e}^{-{\rm i}k\epsilon\tau}\big\vert_{\epsilon=0 } . \nonumber\end{aligned}\ ] ] plugging in our result for @xmath233 from ( [ eq5 ] ) the energy dependent exponentials cancel so , apart from the damping factor @xmath238 , we just have the moments without any ehrenfest - time dependence , leading @xcite to the rmt result @xcite . of course on the left hand side of ( [ eq48 ] ) we have a simple translation by the ehrenfest time , meaning that the translated probability distribution is the same as the rmt - one ( damped ) . for the full ehrenfest - time dependent distribution we simply translate back again and have @xmath239 for the second ` classical ' contribution in ( [ eq4 ] ) we first take the simplest part of the contribution @xmath240 and substitute into ( [ eq46 ] ) obtaining @xmath241 these moments clearly come from an exponential distribution , so that in the limit @xmath242 we recover , for the probability distribution @xmath243 , the classical exponential decay of trajectories inside the system @xmath244 for the remaining contribution of the second part @xmath245 , we have the damping factor @xmath224 , and the energy dependent phase again just leads to a shift in the exponential distribution so this contribution starts at @xmath13 , thus yielding @xmath246 the minus sign however means we truncate the previous exponential at @xmath13 , so the total second contribution to the probability distribution is @xmath247 and 0 elsewhere . since this contribution to the delay time probability distribution is considered to be made up of the two distributions @xmath248 and @xmath249 and as the shifted and damped one @xmath249 has the same mean ( @xmath250 ) and weight as the shifted and damped rmt distribution @xmath251 but a minus sign , it is clear that the average time delay stays at @xmath24 ( i.e. from the untruncated exponential distribution ) and is unaffected by the ehrenfest time . in fact this is an example of the general relation @xcite , derived from the unitarity of the scattering matrix , in which the mean time delay depends just on the average spacing of the resonant levels of the scattering system and the number of scattering channels , i.e. it should have no ehrenfest time or other dependence . . ] this shape of the distribution however is significantly affected by the ehrenfest time , and as an example we plot the complete probability density of the wigner delay times for @xmath252 in fig . [ fig12 ] . there we can see the exponential decay truncated at @xmath253 before a hard gap separates it from the damped rmt distribution which is shifted to the right by @xmath13 . as the rmt distribution in ( [ eq49 ] ) starts at @xmath254 which is @xmath255 above the ehrenfest time , the hard gap is a constant @xmath256 wide . the rmt distribution is continuous ( unlike the truncation of the exponential at @xmath13 ) and peaks very quickly ( @xmath257 ) after the gap at around @xmath258 . related to the semicircle distribution , it likewise has an upper bound ( @xmath259 ) , which interestingly can be connected @xcite to the excitation gap in the density of states of andreev billiards . by expanding , for small energies , the determinantal equation that governs this excitation gap in terms of the wigner - smith matrix @xmath227 , ref . showed that the gap s width is approximately the inverse of the maximum delay time . there the maximum was increased by disorder scattering , leading to a decrease in the andreev gap , while here the maximum delay time is increased by the ehrenfest time as the rmt distribution is correspondingly shifted right . a simultaneously decreasing andreev gap fits with the effective rmt @xcite and semiclassical @xcite treatments . however the connection between the secondary gaps , which appear in the density of states of andreev billiards @xcite as a consequence of the ehrenfest dependence in ( [ eq4],[eq5],[eq6 ] ) shown in this paper , and the gap in the density of the delay times is not so clear . in this paper we have shown how to treat the effect of the ehrenfest time on correlation functions of arbitrarily many pairs of scattering matrices . in our semiclassical approach we extended and combined the zero ehrenfest time approach @xcite ( which leads to the rmt result ) and the @xmath39 ehrenfest time approach @xcite and showed how the results of the effective rmt ansatz can be obtained . the different contributions are described by simple diagrams and following an innovative way of partitioning these diagrams we implemented an algorithmic procedure that allows one to easily obtain the ehrenfest - time dependence . interestingly this always led to the same factor ( which can be traced back to the survival probability only depending on one traversal of each encounter ) so that the rmt - type expression is simply modified by the ehrenfest time by the additional factor @xmath33 . this is in line with the effective rmt result , but as our result is derived just from the underlying chaotic dynamics of the system we can justify for this situation the use of effective rmt which instead conjectures the ehrenfest - time dependence . as the semiclassical framework is based on the underlying classical dynamics we can equally well move away from the rmt arena and obtain the ` classical ' contribution to the correlation functions this can be seen to come from bands of trajectories that remain correlated with each other for the entire duration of their stay inside the system . furthermore the fact that no mixed ( between the rmt - type and classical - type ) terms arise is simply due to their opposing classical restrictions . this lack of mixed terms as well as the classical contribution were previously shown to be more generally due to the preservation of volume under the dynamical evolution and the separation of phase - space into two essentially independent subsystems @xcite . the separation of the correlation functions into two contributions , which each have a straightforward dependence on the ehrenfest time was previously shown to be responsible for secondary gaps in the density of states of andreev billiards @xcite , but has an equal effect on other transport quantities . for the transmission eigenvalues ( and their moments ) with no energy dependence we just get a straightforward interpolation between the rmt @xcite and classical values . for the distribution of the wigner delay times we further see a truncation of the classical ( exponential ) distribution and a shift to higher times of the rmt - type distribution . between the two though a hard gap remains . the method described in this paper allows for the computation of the ehrenfest - time dependence of the trace of arbitrarily many scattering matrix pairs but only to leading order in inverse channel number @xmath260 . the calculation was only doable because at this order the corresponding semiclassical diagrams involve no closed loops and have no periodic orbit encounters ( surrounded periodic orbits ) . when such surrounded periodic orbits are involved , for example for the conductance variance @xcite or the next to leading order quantum correction to the transmission , reflection and the spectral form factor @xcite , the relatively simple cancellation mechanism observed in this paper no longer holds . but by taking into account all possibilities for partial correlations within the ` fringes ' as in refs . in a systematic way , it should however also be possible to extend our ehrenfest - time results to infinite order . the authors thank cyril petitjean and robert whitney for helpful discussions and acknowledge funding through the deutsche forschungsgemeinschaft ( forschergruppe for 760 ) ( kr ) and the alexander von humboldt foundation ( jk ) . 10 o. bohigas , m. j. giannoni and c. schmit , phys . lett . * 52 * , 1 ( 1984 ) . m. l. mehta , _ random matrices , pure and applied mathematics _ , elsevier , amsterdam ( 2004 ) . m. c. gutzwiller , _ chaos in classical and quantum mechanics _ , springer , new york ( 1990 ) . k. richter , _ semiclassical theory of mesoscopic quantum systems _ , springer , berlin ( 2000 ) . m. v. berry , proc . r. soc . a * 400 * , 229 ( 1985 ) . j. h. hannay and a. m. ozorio de almeida , j. phys . a , * 17 * , 3429 ( 1984 ) . m. sieber and k. richter , phys . t * 90 * , 128 ( 2001 ) . s. mller , s. heusler , p. braun , f. haake , and a. altland , phys . , 014103 ( 2004 ) ; phys . e * 72 * , 046207 ( 2005 ) . r. landauer , ibm j. res . * 1 * , 223 ( 1957 ) ; * 32 * , 306 ( 1988 ) ; m.bttiker , phys . rev . lett . * 57 * , 1761 ( 1986 ) . r. a. jalabert , h. u. baranger , and a. d. stone , phys . lett . , 2442 ( 1990 ) ; h. u. baranger , r. a. jalabert , and a. d. stone , phys . rev.lett . * 70 * , 3876 ( 1993 ) ; chaos * 3 * , 665 ( 1993 ) . k. richter and m. sieber , phys . rev . lett . * 89 * , 206801 ( 2002 ) . s. heusler , s. mller , p. braun , and f. haake , phys . lett . * 96 * , 066804 ( 2006 ) ; s. mller , s. heusler , p. braun , and f. haake , new j. phys . * 9 * , 1 ( 2007 ) . i. adagideli , phys . b * 68 * , 233308 ( 2003 ) . b. v. chirikov , f. m. izrailev , and d. l. shepelyansky , sov . rev.sect . c * 2 * , 209 ( 1981 ) . i. l. aleiner and a. i. larkin , phys . b * 54 * , 14423 ( 1996 ) . g. silvestrov , m. c. goorden , and c. w. j. beenakker , phys . rev.lett . * 90 * , 116801 ( 2003 ) . s. rahav and p. w. brouwer , phys . * 95 * , 056806 ( 2005 ) . jacquod and r. s. whitney , phys . b * 73 * , 195115 ( 2006 ) . o. agam , i. aleiner , and a. larkin , phys . 85 * , 3153 ( 2000 ) . g. silvestrov , m. c. goorden , and c. w. j. beenakker , phys . rev . b * 67 * , 241301(r ) ( 2003 ) . j. tworzydo , a. tajic , h. schomerus , and c. w. j. beenakker , phys.rev . b * 68 * , 115313 ( 2003 ) . r. s. whitney and ph . jacquod , phys . * 96 * , 206804 ( 2006 ) . m. c. goorden , p. jacquod , and c. w. j. beenakker , phys . b * 72 * , 064526 ( 2005 ) . c. w. j. beenakker , lect . notes phys . * 667 * , 131 ( 2005 ) . p. w. brouwer and s. rahav , phys . b * 74 * , 085313 ( 2006 ) . r. s. whitney and ph . jacquod , phys . lett . * 94 * , 116801 ( 2005 ) . j. kuipers , d. waltner , c. petitjean , g. berkolaiko , and k. richter , phys.rev . * 104 * , 027001 ( 2010 ) ; j. kuipers , t. engl , g. berkolaiko , c.petitjean , d. waltner , and k. richter , phys . b * 83 * , 195316 ( 2011 ) . v. nazarov , _ quantum noise in mesoscopic physics _ kluwer , dordrecht ( 2003 ) . c. petitjean , d. waltner , j. kuipers , i. adagideli , and k. richter , phys . b * 80 * , 115310 ( 2009 ) . g. berkolaiko , j. m. harrison , and m. novaes , j. phys . a * 41 * , 365102 ( 2008 ) . p. w. brouwer , and c. w. j. beenakker , j. math . phys . * 37 * , 4904 ( 1996 ) . m. novaes , phys . b * 75 * , 073304 ( 2007 ) . e. p. wigner , phys . rev . * 98 * , 145 ( 1955 ) ; f. t. smith , phys . , 349 ( 1960 ) . g. berkolaiko and j. kuipers , j. phys . a * 43 * , 035101 ( 2010 ) . p. w. brouwer , k. m. frahm and c. w. j. beenakker , waves in random media * 9 * , 91 ( 1999 ) . v. l. lyuboshitz , phys . b * 72 * , 41 ( 1977 ) . f. libisch , j. mller , s. rotter , m. g. vavilov , and j. burgdrfer , europhys . lett . * 82 * , 47006 ( 2008 ) . p. w. brouwer and s. rahav , phys . b * 74 * , 075322 ( 2006 ) . d. waltner and j. kuipers , phys . e * 82 * , 066205 ( 2010 ) .

we consider a linear analytic ordinary differential equation with complex time having a nonresonant irregular singular point . we study it as a limit of a generic family of equations with confluenting fuchsian singularities . in 1984 v.i.arnold asked the following question : is it true that some operators from the monodromy group of the perturbed ( fuchsian ) equation tend to stokes operators of the nonperturbed irregular equation ? another version of this question was also independently proposed by j .- p.ramis in 1988 . we consider the case of poincar rank 1 only . we show ( in dimension two ) that generically no monodromy operator tends to a stokes operator ; on the other hand , in any dimension commutators of appropriate noninteger powers of the monodromy operators around singular points tend to stokes operators . _ dedicated to yu.s.ilyashenko on the occasion of his 60 birthday . _ 2000 math . subj . class . 34m35 ( 34m40 ) key words and phrases : linear equation , irregular singularity , stokes operators , fuchsian singularity , monodromy , confluence . \1 . introduction . 1.1 . brief statements of results , the plan of the paper and the history . 1.2 . analytic classification of irregular equations . canonical solutions and stokes operators . 1.3 . historical overview . \2 . main results . stokes operators and limit monodromy . 2.1 . stokes operators as limit transition operators between monodromy eigenbases . 2.2 . stokes operators as limits of commutators of appropriate powers of the monodromy operators . 2.3 . the case of higher poincar rank . \3 . convergence of the commutators to stokes operators . proof of theorem 2.16 . 3.1 . properties of the monodromy and the transition operators . the plan of the proof of theorem 2.16 . 3.2 . the upper - triangular element of the transition matrix . proof of ( 3.3 ) . 3.3 . commutators of operators with asymptotically common eigenline . proof of lemma 3.5 . 3.4 . convergence of the commutators to stokes operators : the higher - dimensional case . . generic divergence of monodromy operators along degenerating loops . 4.1 the statement of the divergence theorem . 4.2 . projectivization . the scheme of the proof of theorem 4.6 . 4.3 . divergence of word of projectivizations . proof of lemma 4.10 . references .

math0304243

consider a linear analytic ordinary differential equation @xmath1 with a nonresonant irregular singularity of order ( the poincar rank ) @xmath2 at 0 ( or briefly , an irregular equation ) . this means that @xmath3 is a holomorphic matrix function such that the matrix @xmath4 has distinct eigenvalues ( denote them by @xmath5 ) . then the matrix @xmath4 is diagonalizable , and without loss of generality we suppose that it is diagonal . two equations of type ( 1.1 ) are analytically ( formally ) equivalent , if there exists a change @xmath6 of the variable @xmath7 , where @xmath8 is a holomorphic invertible matrix function ( respectively , a formal invertible matrix power series ) , that transforms one equation into the other . the analytic classification of irregular equations ( 1.1 ) is well - known ( [ 2 ] , [ 3 ] , [ 9 ] , [ 10 ] , [ 16 ] ) : the complete system of invariants for analytic classification consists of a formal normal form ( 1.4 ) and stokes operators ( 1.6 ) defined in 1.2 ; the latters are linear operators acting in the solution space of ( 1.1 ) comparing appropriate `` sectorial canonical solution bases '' . on the other hand , an irregular equation ( 1.1 ) can be regarded as a result of _ confluence of fuchsian singular points _ ( recall that a fuchsian singular point of a linear equation is a first order pole of its right - hand side ) . namely , consider a deformation @xmath9 of equation ( 1.1 ) that splits the irregular singular point 0 of the nonperturbed equation into @xmath10 fuchsian singularities @xmath11 of the perturbed equation , i.e. , @xmath12 for @xmath13 . the family ( 1.2 ) depends on a parameter @xmath14 , @xmath15 , @xmath16 . the _ monodromy group _ of a fuchsian equation acts linearly in its solution space by analytic extensions of solutions along closed loops . the analytic equivalence class of a generic fuchsian equation is completely determined by the local types of its singularities and the action of its monodromy group . everywhere below by @xmath17 we denote the monodromy operator of the perturbed equation ( 1.2 ) along a loop going around the singular point @xmath18 ( the choice of the corresponding loops will be specified later ) . the monodromy group of the perturbed equation is generated by appropriately chosen operators @xmath17 . in 1984 v.i.arnold proposed the following question . consider a generic deformation ( 1.2 ) . is there an operator @xmath19 from the monodromy group of the perturbed equation that converges to a stokes operator of the nonperturbed equation ? a version of this question was proposed independently by j .- p.ramis in 1988 . it appears that already in the simplest case of dimension 2 and poincar rank @xmath20 generically _ each operator from the monodromy group ( except for that along a circuit ( and its powers ) around both singularities ) tends to infinity _ ( theorem 4.6 in section 4 ) , so , no one tends to a stokes operator . in other terms , generically , no word ( 1.3 ) with @xmath21 tends to a stokes operator . but if @xmath20 , then appropriate words ( 1.3 ) with _ noninteger _ exponents @xmath22 tend to stokes operators ( theorem 2.16 in 2.2 ) . the previous question and its nonlinear analogues were studied by j .- p.ramis , b.khesin , a.duval , c.zhang , j.martinet , the author and others ( see the historical overview in 1.3 ) . it was proved by the author in [ 6 ] that appropriate branches of the eigenfunctions of the monodromy operators @xmath17 of the perturbed equation tend to appropriate canonical solutions of the nonperturbed equation ( theorem 2.5 in 2.1 ) . in the case of poincar rank @xmath20 this implies ( corollary 2.6 in 2.1 ) that stokes operators of the nonperturbed equation are limits of transition operators between appropriate eigenbases of the monodromy operators @xmath17 . this corollary has a generalization for higher poincar rank [ 6 ] . the proofs of the results of the present paper are based on the previously mentioned results from [ 6 ] , which are recalled in 2.1 . in 1.2 we recall the analytic classification of irregular equations ( 1.1 ) and the definitions of sectorial canonical solution bases and stokes operators . in 2.2 we state theorem 2.16 on convergence of appropriate word ( 1.3 ) with noninteger exponents @xmath22 to a stokes operator in the case of poincar rank @xmath20 . its proof is given in section 3 . the corresponding exponents @xmath22 do not depend on the choice of deformation . in fact , in the case of the higher poincar rank @xmath23 and @xmath24 one can also prove a similar statement , but now the choice of the corresponding exponents @xmath22 will depend on the choice of deformation . the latter case will be discussed in 2.3 . in section 4 in the case , when @xmath20 , @xmath24 , for a typical nonperturbed equation ( 1.1 ) we prove the divergence of the operators from the monodromy group of the perturbed equation ( except for the monodromy along a circuit around both singularities and its powers ) . let ( 1.1 ) be an irregular equation , @xmath5 , @xmath25 , be the eigenvalues of the corresponding matrix @xmath4 . one can ask the question : is it true that the variables @xmath26 in the equation can be separated , more precisely , that ( 1.1 ) is analytically equivalent to a direct sum of one - dimensional linear equations , i.e. , a linear equation with a diagonal matrix function in the right - hand side ? generically , the answer is `` no '' . at the same time any irregular equation ( 1.1 ) is formally equivalent to a unique direct sum of the type @xmath27 where @xmath28 are polynomials of degrees at most @xmath2 , @xmath29 . the normalizing series bringing ( 1.1 ) to ( 1.4 ) is unique up to left multiplication by constant diagonal matrix . the system ( 1.4 ) is called the _ formal normal form _ of ( 1.1 ) ( @xmath30 ) . # 1#1 generically the normalizing series diverges . at the same time there exists a finite covering @xmath31 of a punctured neighborhood of zero in the @xmath32- line by radial sectors @xmath33 ( i.e. , those with the vertex at 0 ) that have the following property . there exists a unique change of variables @xmath34 over each @xmath33 that transforms ( 1.1 ) to @xmath35 , where @xmath36 is an analytic invertible matrix function on @xmath33 that can be @xmath37-smoothly extended to the closure @xmath38 of the sector so that its asymptotic taylor series at 0 coincides with the normalizing series . the previous statement on existence and uniqueness of sectorial normalization holds in any good sector ( see the two following definitions ) ; the covering consists of good sectors ( @xmath30 ) . * case @xmath20 , @xmath24 , @xmath39 . * a sector in @xmath40 with the vertex at 0 is said to be _ good _ , if it contains only one imaginary semiaxis @xmath41 , and its closure does not contain the other one ( see fig.1 ) . * general case . * ( see , e.g. , [ 9 ] ) . let @xmath42 , @xmath43 be a @xmath44- ple of distinct numbers , @xmath32 be the coordinate on @xmath40 . for a given pair @xmath45 the rays in @xmath40 starting at 0 and forming the set @xmath46 are called the @xmath47- _ imaginary dividing rays _ corresponding to the pair @xmath48 . a radial sector is said to be _ @xmath47- good _ , if for any pair @xmath48 , @xmath49 , it contains exactly one corresponding imaginary dividing ray and so does its closure . in the case , when @xmath20 , @xmath24 , @xmath39 , the imaginary dividing rays are the imaginary semiaxes , and the notions of good sector and @xmath47- good sector coincide . the ratio @xmath50 of solutions of equations from ( 1.4 ) tends to either zero or infinity , as @xmath32 tends to zero along a ray distinct from the imaginary dividing rays corresponding to the pair @xmath48 . its limit changes exactly when the ray under consideration jumps over one of the latter imaginary dividing rays . we consider a covering @xmath31 of a punctured neighborhood of zero by good ( or @xmath47- good ) sectors numerated counterclockwise and put @xmath51 . the standard splitting of the normal form ( 1.4 ) into the direct sum of one - dimensional equations defines a canonical base in its solutions space ( uniquely up to multiplication of the base functions by constants ) with a diagonal fundamental matrix . denote the latter fundamental matrix by @xmath52 together with the normalizing changes @xmath53 in @xmath33 , it defines the canonical bases @xmath54 in the solution space of ( 1.1 ) in the sectors @xmath33 with the fundamental matrices @xmath55 where for any @xmath56 the branch ( `` with the index @xmath57 '' ) of the fundamental matrix @xmath58 in @xmath59 is obtained from that in @xmath33 by the counterclockwise analytic extension for any @xmath56 . ( we put @xmath51 . the corresponding branch of @xmath60 `` with the index @xmath61 '' is obtained from that `` with the index 0 '' by the right multiplication by the monodromy matrix of the formal normal form ( 1.4 ) . ) in a connected component of the intersection @xmath62 there are two canonical solution bases coming from @xmath33 and @xmath59 . generically , they do not coincide . the transition between them is defined by a constant matrix @xmath63 : @xmath64 the transition operators ( matrices @xmath63 ) are called _ stokes operators ( matrices ) _ ( see [ 2 ] , [ 3 ] , [ 9 ] , [ 10 ] , [ 16 ] ) . the nontriviality of stokes operators yields the obstruction to analytic equivalence of ( 1.1 ) and its formal normal form ( 1.4 ) . stokes matrices ( 1.6 ) are well - defined up to simultaneous conjugation by one and the same diagonal matrix . let @xmath65 . in this case without loss of generality we assume that @xmath66 ( one can achieve this by linear change of the time variable ) . then the above covering consists of two sectors @xmath67 and @xmath68 ( figure 1 ) . the former contains the positive imaginary semiaxis and its closure does not contain the negative one ; the latter has the same properties with respect to the negative ( respectively , positive ) imaginary semiaxis . there are two components of the intersection @xmath69 . so , in this case we have a pair of stokes operators . _ the stokes matrices ( 1.6 ) are unipotent : _ the one corresponding to the left intersection component is lower - triangular ; the other one is upper - triangular ( @xmath30 ) . stokes operators of an irregular equation ( 1.1 ) with a diagonal matrix in the right - hand side are identity operators . in this case ( 1.1 ) is analytically equivalent to its formal normal form . in general , _ two irregular equations are analytically equivalent , if and only if they have the same formal normal form and the corresponding stokes matrix tuples are obtained from each other by simultaneous conjugation by one and the same diagonal matrix _ , cf . the previous remark . thus , formal normal form and stokes matrix tuple taken up to the previous conjugation present the complete system of invariants for analytic classification of irregular equations ( see @xmath30 ) . earlier in 1919 r. garnier [ 5 ] had studied some particular deformations of some class of linear equations with nonresonant irregular singularity . he obtained some analytic classification invariants for these equations by studying their deformations . the complete system of analytic classification invariants ( stokes operators and formal normal form ) for general irregular differential equations was obtained later in 1970-th years in the papers by jurkat , lutz , peyerimhoff [ 10 ] , sibuya [ 16 ] and balser , jurkat , lutz [ 3 ] . later jurkat , lutz and peyerimhoff had extended their results to some resonant cases [ 11 ] . it is well - known that the monodromy operators of a linear ordinary differential equation belong to its galois group ( see [ 9 ] , [ 14 ] ) . in 1985 j .- p.ramis have proved that the stokes operators also belong to the galois group ( [ 14 ] , see also [ 9 ] ) . in 1989 he considered the classical confluenting family of hypergeometric equations and proved convergence of appropriate branches of monodromy eigenfunctions of the perturbed equation to canonical solutions of the nonperturbed one by direct calculation [ 15 ] . in the late 1980-th years b.khesin also proved a version of this statement , but his result was not published . in 1991 a.duval [ 4 ] proved this statement for the biconfluenting family of hypergeometric equations ( where the nonperturbed equation is equivalent to bessel equation ) by direct calculation . in 1994 c. zhang [ 17 ] had obtained the expression of garnier s invariants via stokes operators ( for the class of irregular equations considered by garnier ) . the conjecture saying that stokes operators are limit transition operators between monodromy eigenbases of the perturbed equation was firstly proposed by a.a.bolibrukh in 1996 . it was proved by the author in [ 6 ] . later this result was extended to a generic resonant case [ 8 ] . nonlinear analogues of the previous statements for parabolic mappings ( i.e. , one - dimensional conformal mappings tangent to identity ) and their calle - voronin moduli , saddle - node singularities of two - dimensional holomorphic vector fields and their martinet - ramis invariants ( sectorial central manifolds in higher dimensions ) were obtained by the author in [ 7 ] . generalizations and other versions of the statement on parabolic mappings were recently obtained in the joint paper [ 12 ] by p.mardesic , r.roussarie , c.rousseau , and in two unpublished joint papers by the following authors : 1 ) x.buff and tan lei ; 2 ) a.douady , francisco estrada , p.sentenac . a particular case of the result from [ 7 ] concerning parabolic mappings ( analogous to the previously mentioned statements on linear equations ) was obtained by j.martinet [ 13 ] . everywhere below ( whenever the contrary is not specified ) we consider that the ( nonperturbed ) irregular equation under consideration has poincar rank @xmath20 . in the present section we recall the statements from [ 6 ] expressing the stokes operators as limit transition operators between monodromy eigenbases of the confluenting fuchsian equation ( theorems 2.5 , 2.11 and corollary 2.6 in subsection 2.1 ) . in 2.2 we state the results expressing the stokes operators as limits of some words ( 1.3 ) of noninteger powers of monodromy operators ( theorem 2.16 ) . in 2.3 we discuss the extension of these results to the case of higher poincar rank in dimension two . we formulate the result from the title of the subsection firstly in the case , when @xmath20 , @xmath24 , and then in the general case . * case @xmath24 , @xmath20 . * let @xmath5 , @xmath70 , be the eigenvalues of the matrix @xmath4 . without loss of generality we assume that @xmath66 : one can achieve this by linear change of the time variable . we consider a deformation of ( 1.1 ) , @xmath71 where @xmath72 and @xmath73 depend continuously on a parameter @xmath74 so that @xmath75 for @xmath76 . without loss of generality we assume that @xmath77 . we formulate the statement from the title of the subsection for a generic deformation ( 2.1 ) , see the following definition . a family of quadratic polynomials @xmath73 depending continuously on a nonnegative parameter @xmath78 , @xmath79 , with roots @xmath11 , @xmath80 , @xmath77 , is said to be _ generic _ , if @xmath75 for @xmath81 , and the line passing through @xmath82 and @xmath83 intersects the real axis by angle bounded away from 0 uniformly in @xmath78 . a family ( 2.1 ) of _ linear equations _ with @xmath24 , @xmath20 , @xmath84 is said to be _ generic _ , if so is the corresponding family of polynomials @xmath73 . ( see , e.g. , [ 2 ] ) . a singular point @xmath85 of a linear analytic ordinary differential equation @xmath86 is said to be _ fuchsian _ , if it is a first order pole of the right - hand side ( i.e. , the corresponding matrix function @xmath87 is holomorphic at @xmath85 ) . characteristic numbers _ of a fuchsian singularity are the eigenvalues of the corresponding residue matrix @xmath88 ( which are equal to the logarithms divided by @xmath89 of the eigenvalues of the corresponding monodromy operator ) . a family ( 2.1 ) of linear equations is generic , if and only if the difference of the characteristic numbers at @xmath82 ( or equivalently , at @xmath83 ) of the perturbed equation is not real for small @xmath78 and moreover has argument bounded away from @xmath90 uniformly in @xmath91 small enough . the latter condition implies that the monodromy operator of the perturbed equation around each singular point @xmath18 has distinct eigenvalues ( moreover , their modules are distinct ) , and hence , a well - defined eigenbase in the solution space ( for small @xmath78 ) . the singularities of the perturbed equation from a generic family have imaginary parts of constant ( and opposite ) signs ( by definition ) . without loss of generality everywhere below we consider that @xmath92 let ( 2.1 ) be a generic family of linear equations ( see the previous definition ) whose singularity families satisfy the previous inequalities . let @xmath33 , @xmath93 , be a pair of good sectors in the @xmath32- line ( see definition 1.2 ) such that for any @xmath78 small enough @xmath94 , @xmath93 , @xmath95 , @xmath96 ( see fig . 1 ) . the sector @xmath33 is said to be _ the sector associated to the singularity family _ @xmath97 , @xmath93 . we show that appropriate branches of the eigenfunctions of the monodromy operator @xmath17 around @xmath18 of the perturbed equation converge to canonical solutions of the nonperturbed equation in the corresponding sector @xmath98 . this will imply the statement from the title of the subsection . to formulate the latter statement precisely , consider the auxiliary domain @xmath99,\tag{2.2}\ ] ] which is simply - connected , and the canonical branches of the monodromy eigenfunctions on the domain @xmath100 . in more details , consider a small circle going around @xmath18 and take a base point on it outside the segment @xmath101 $ ] . in the space of local solutions of the perturbed equation at the base point consider the monodromy operator @xmath17 acting by the analytic extension of a solution along the circle from the base point to itself in the counterclockwise direction . the eigenfunctions of @xmath17 have well - defined branches ( up to multiplication by constants ) in the corresponding disc with the segment @xmath101 $ ] deleted . their immediate analytic extension yields their canonical branches on @xmath100 . in other terms , we identify the space of local solutions with the space of solutions on @xmath100 by immediate analytic extension , consider @xmath17 as an operator acting in the latter space and take its eigenfunctions . the canonical basic solutions of the nonperturbed equation are numerated by the indices 1 and 2 , which correspond to the eigenvalues @xmath102 of @xmath4 . to state the results previously mentioned , let us define the analogous numeration of the monodromy eigenfunctions at @xmath11 . the monodromy eigenfunctions are numerated by the characteristic numbers ( see definition 2.2 ) of the corresponding singularity . the latters are proportional to the eigenvalues of the matrix @xmath103 , which tend to @xmath104 and @xmath105 , as @xmath106 . this induces the numeration of the monodromy eigenfunctions by the indices 1 and 2 corresponding to the limit eigenvalues @xmath107 and @xmath108 . ( see [ 6 ] ) . let ( 2.1 ) be a generic family of linear ordinary differential equations ( see definition 2.1 ) , @xmath11 be its singularity family , @xmath98 be the corresponding sector ( see the previous definition ) , @xmath100 be the domain ( 2.2 ) . consider the eigenbase on @xmath100 of the monodromy operator of the perturbed equation around @xmath11 . the appropriately normalized eigenbase ( by multiplication of the basic functions by constants ) converges to the canonical solution base ( 1.5 ) on @xmath98 of the nonperturbed equation . ( see [ 6 ] ) . let ( 2.1 ) be a generic linear equation family ( see definition 2.1 ) , @xmath18 be its singularity families , @xmath98 be the corresponding sectors ( see the previous definition ) chosen to cover a punctured neighborhood of zero , @xmath100 be the corresponding domains ( 2.2 ) . let @xmath109 , @xmath110 be the corresponding stokes matrices ( 1.6 ) of the nonperturbed equation in the left ( respectively , right ) component of the intersection @xmath69 . consider the eigenbase on @xmath100 of the monodromy operator of the perturbed equation around @xmath11 . denote by @xmath111 the fundamental matrix of this eigenbase . let @xmath112 ( @xmath113 ) be the transition matrix between the monodromy eigenbases @xmath111 , @xmath80 , in the left ( respectively , right ) component of the intersection @xmath114 : @xmath115 for any @xmath93 and appropriately normalized monodromy eigenbases @xmath116 , @xmath80 ( the normalization of @xmath117 ( only ) depends on the choice of @xmath118 ) @xmath119 , as @xmath106 . . to state the analogues of theorem 2.5 and corollary 2.6 in this more general case , let us firstly extend the notions of a generic family of linear equations and a sector associated to a singularity family . let @xmath120 , @xmath121 , @xmath122 be a set of @xmath44 distinct complex numbers , @xmath45 be a pair of them . a ray in @xmath40 starting at 0 is called a @xmath47- _ real dividing ray _ associated to the pair @xmath48 , if for any @xmath32 lying in this ray @xmath123 . ( or equivalently , it is a ray bisecting an angle between two neighbor imaginary dividing rays ( see definition 1.3 ) associated to @xmath48 . ) let ( 1.1 ) be an irregular equation with @xmath20 , @xmath124 be the vector of eigenvalues of the corresponding matrix @xmath4 . let ( 2.1 ) be its deformation depending continuously on a nonnegative parameter @xmath78 , @xmath79 , @xmath77 . the family ( 2.1 ) is said to be _ generic _ , if @xmath75 for @xmath81 , and the line passing through @xmath82 and @xmath83 intersects each @xmath47- real dividing ray by angle bounded away from 0 uniformly in @xmath78 . let ( 2.1 ) be a generic family ( see the previous definition ) , @xmath124 be the corresponding eigenvalue tuple of @xmath125 . let @xmath126 , @xmath127 be the corresponding singular point families , @xmath128 be the half - plane ( depending on @xmath78 ) containing @xmath18 and bounded by the symmetry line of the segment @xmath129 $ ] . the _ sector associated to _ @xmath18 is a @xmath130- good sector ( see definition 1.3 ) independent on @xmath78 that contains @xmath128 for any @xmath78 small enough . in the previous definition the sectors @xmath67 , @xmath68 associated to @xmath126 , @xmath127 respectively cover a punctured neighborhood of zero , so , the nonperturbed equation has a pair of stokes operators ( @xmath109 , @xmath110 ) associated to this covering . ( see [ 6 ] ) . let ( 2.1 ) be a generic family of linear equations ( see definition 2.8 ) , @xmath67 , @xmath68 , @xmath131 , @xmath132 be respectively the corresponding associated sectors ( see the previous definition ) and the domains ( 2.2 ) . then the statement of the previous theorem remains valid . the same is true for corollary 2.6 . in more details , consider the eigenbases @xmath116 on @xmath100 of the monodromy operators around the singular points @xmath11 of the perturbed equation , @xmath80 . let @xmath112 , @xmath113 be the transition matrices ( 2.3 ) between them in the connected components of the intersection @xmath114 . let @xmath109 , @xmath110 be the stokes matrices of the nonperturbed equation in the corresponding limit connected components of the intersection @xmath69 . then for any @xmath93 and appropriately normalized monodromy eigenbases @xmath116 ( the normalization of @xmath133 ( only ) depends on the choice of @xmath118 ) @xmath119 , as @xmath106 . the stokes and monodromy operators act in different linear spaces : in the solution spaces of the nonperturbed ( respectively , perturbed ) equations . to formulate the statement from the title of the subsection , let us firstly identify these solution spaces and specify the loops defining the monodromy operators . let ( 2.1 ) be a generic family of linear equations ( in the sense of some of definitions 2.1 and 2.8 ) . take a `` base point '' @xmath85 in the unit disc punctured at 0 . the space of local solutions of a linear equation at a nonsingular point @xmath134 is identified with the space of initial conditions at @xmath85 ( which is common for the nonperturbed and the perturbed equations ) . this identifies the solution spaces of the latters . the space thus obtained will be denoted by @xmath135 . let ( 1.1 ) be an irregular equation with @xmath20 , @xmath124 be the eigenvalue tuple of the corresponding matrix @xmath4 . let @xmath67 , @xmath68 be @xmath130- good sectors covering a punctured neighborhood of zero in the @xmath32- line . let @xmath109 , @xmath110 be the stokes operators ( 1.6 ) corresponding to the connected components of their intersection . each operator @xmath136 is well - defined in the space @xmath137 of local solutions of ( 1.1 ) at any point @xmath85 lying in the corresponding component of the intersection @xmath69 . now let us define the monodromy operators acting in the previous space @xmath137 . let ( 2.1 ) be a generic family of linear equations ( in the sense of one of definitions 2.1 and 2.8 ) , @xmath11 , @xmath80 , be its singularity families . fix a point @xmath85 ( independent on @xmath78 ) disjoint from the line passing through @xmath82 and @xmath83 for any @xmath78 . let @xmath138 be a small circle centered at @xmath11 whose closed disc is disjoint from @xmath139 , @xmath140\cap l_i$ ] , the segment @xmath141 $ ] be oriented from @xmath85 to @xmath142 . consider the closed path @xmath143\circ l_i\circ [ t_0,a_i]^{-1}$ ] , @xmath80 , which starts and ends at @xmath85 ( in the case , when @xmath20 , @xmath24 , @xmath144 , we choose @xmath145 , see fig.3 ) . define @xmath146 to be the corresponding monodromy operator of the perturbed equation . we show that commutators of appropriate noninteger powers of the operators @xmath17 ( see the following definition ) tend to the stokes operators . let @xmath147 , @xmath148 be a linear operator in a finite - dimensional linear space having distinct eigenvalues . @xmath149-th power of _ @xmath150 is the operator having the same eigenlines , as @xmath150 , whose corresponding eigenvalues are some values of the @xmath149-th powers of those of @xmath150 . let @xmath67 , @xmath68 be sectors in @xmath40 with vertex at 0 covering a punctured neighborhood of 0 . their _ left ( right ) intersection component _ is the component of their intersection crossed while going from @xmath67 to @xmath68 in the counterclockwise ( respectively , clockwise ) direction . let ( 2.1 ) be a generic family of linear equations ( in the sense of one of definitions 2.1 and 2.8 ) , @xmath11 , @xmath80 , be its singularity families . let @xmath98 , @xmath80 , be the corresponding associated sectors ( see definitions 2.4 or 2.9 respectively ) forming a covering of a punctured neighborhood of zero , @xmath109 , @xmath110 be the stokes operators ( 1.6 ) of the nonperturbed equation corresponding to the left ( respectively , right ) component of the intersection @xmath69 ( see the previous paragraph ) . let @xmath85 be a fixed point of unit disc lying in the left component of the intersection @xmath69 , @xmath137 be the corresponding local solution space ( see remark 2.12 ) . ( then the operator @xmath109 ( @xmath110 ) acts in the space @xmath137 ( recpectively , @xmath151 , see remark 2.13 . ) ) let @xmath152 be the corresponding monodromy operators from definition 2.14 . then for any pair of numbers @xmath153 such that @xmath154 @xmath155 @xmath156 theorem 2.16 is proved in section 3 . theorem 2.5 on convergence of the monodromy eigenbases to canonical solution bases is stated and proved in [ 6 ] for arbitrary irregular equation ( for arbitrary poincar rank and dimension ) . it holds for any generic family ( 1.2 ) defined to satisfy the following conditions : 1 ) @xmath157 , @xmath158 ( then @xmath159 , where the points @xmath142 form a regular polygon centered at 0 ) ; 2 ) no one of the previous points @xmath142 lies in a real dividing ray ( see definition 2.7 ) , in other terms , no radial ray of @xmath18 tends to a real dividing ray . to each singularity family @xmath160 we put into correspondence a @xmath47- good sector @xmath161 ( similarly to definition 2.9 ) so that the canonical branches in @xmath162 $ ] of the corresponding monodromy eigenfunctions converge to the canonical solutions of the nonperturbed equation on @xmath161 . in the case of higher poincar rank for some pairs of neighbor singularities of the perturbed equation the corresponding sectors can not not be chosen intersected ; then the corresponding transition operator between the monodromy eigenbases tends to a product of stokes operators . each stokes matrix is contained in some of the previous limit products , and its elements are expressed as polynomials in the elements of the corresponding limit product . on the other hand , in dimension two there are always two pairs of neighbor singularity families such that for each singularity pair the corresponding sectors may be chosen intersected . then the transition operator between the corresponding appropriately normalized monodromy eigenbases tends to the stokes operator of the nonperturbed equation corresponding to the intersection of the sectors . consider the case , when @xmath163 . then the perturbed equation has three singularities , and the number of @xmath164- good sectors covering a punctured neighborhood of zero is equal to 4 . one can prove the following version of theorem 2.16 . _ consider a generic deformation ( 1.2 ) of an irregular equation ( 1.1 ) with @xmath163 . let @xmath126 , @xmath127 be a pair of singularity families numerated counterclockwise and corresponding to intersected sectors ( denote the latters by @xmath67 and @xmath68 respectively ) . let @xmath165 be a fixed ( base ) point lying between the radial rays of @xmath82 , @xmath83 for all @xmath78 . let @xmath166 , @xmath167 be the corresponding monodromy operators ( see definition 2.14 ) . let @xmath168 be the stokes operator corresponding to the intersection @xmath69 . then for appropriate @xmath169 ( depending on the family of equations ) @xmath170 more precisely , there exist @xmath171 , @xmath172 ( depending on the family of equations but not on @xmath78 ) such that the previous statement holds whenever @xmath173 , @xmath174 satisfy the following system of inequalities _ : @xmath175 the previous coefficients @xmath138 depend on how close the radial rays of @xmath18 , @xmath80 , approach the real dividing rays : if the minimal angle between the radial ray of @xmath18 and some real dividing ray is small , then the corresponding coefficient @xmath138 should be chosen large enough ( hence , the corresponding exponent @xmath22 should be taken small enough ) . the author believes that the previous statements extend to the general case of arbitrary poincar rank and dimension . firstly we prove theorem 2.16 in the case , when @xmath20 , @xmath24 . its proof for the case of @xmath20 and arbitrary @xmath44 is analogous : the modifications needed will be discussed in subsection 3.4 thus , from now on we consider that @xmath20 , @xmath24 , until the contrary will be specified . without loss of generality we assume that @xmath176 . let us prove the convergence of the first commutator from theorem 2.16 ; the proof of the convergence of the second commutator is analogous . thus , from now on we assume that the base point @xmath85 lies in the left component of the intersection @xmath69 , and one can put @xmath177 . consider a linear diagonalizable operator acting on @xmath178 with eigenvalues of distinct modules . its _ projective multiplier _ is the ratio of its eigenvalue with the lower module over that with the higher module . its _ projectivization _ is the mbius transformation @xmath179 induced by its action and the tautological projection @xmath180 . in the conditions of the previous definition the projectivization is a hyperbolic transformation ( see [ 1 ] and definition 4.6 in section 4 ) ; in particular , it has an attracting fixed point . the projective multiplier is well - defined and its module is always less than 1 . it is equal to the multiplier of the projectivization at its attracting fixed point . let us write down the monodromy operators in the eigenbase of @xmath166 ( which converges to the canonical solution base of the nonperturbed equation on @xmath67 ) . then the matrix of @xmath166 is diagonal : denote it @xmath181 by corollary 2.6 , the matrix of @xmath167 is @xmath182 the transition matrix @xmath183 tends to the stokes matrix @xmath109 , which is lower - triangular . thus , the upper - triangular element of @xmath183 ( denoted by @xmath184 ) tends to 0 . first of all we find the asymptotics of the eigenvalues @xmath185 of @xmath17 : let ( 2.1 ) be a generic family of linear equations ( see definition 2.1 ) , @xmath177 , @xmath17 be the monodromy operators of the perturbed equation from definition 2.14 , @xmath186 , @xmath187 be their basic eigenfunctions , @xmath188 , @xmath189 be the corresponding eigenvalues . then @xmath190 @xmath191 in the conditions of the previous proposition the projective multipliers of @xmath166 and @xmath167 are equal respectively to @xmath192 @xmath193 * of proposition 3.3 . * it follows from definition that @xmath194 . the real part of the right - hand side of the previous formula is positive and tends to infinity ( since @xmath195 by assumption , and @xmath196 ) , which implies that @xmath197 . the similar formulas written for all the @xmath198 prove the rest of the statements of the proposition . let @xmath153 , @xmath154 , @xmath199 we prove that @xmath200 by definition and ( 3.1 ) , the matrix of the previous commutator in the eigenbase of @xmath166 is @xmath201 let @xmath184 be the upper - triangular element of the transition matrix @xmath183 , @xmath202 be the projective multiplier of @xmath167 . for the proof of the convergence to @xmath203 of the previous commutator we firstly prove that @xmath204 more precisely , we show in the next subsection that @xmath205 , where @xmath206 is the upper - triangular element of the other stokes matrix @xmath110 . let @xmath207 , @xmath208 be the projective multipliers of the operators @xmath209 , @xmath210 respectively . formula ( 3.3 ) together with the previous corollary and the condition @xmath154 imply that @xmath211 using ( 3.4 ) we show ( in the next lemma proved in 3.3 ) that if we eliminate subsequently the terms @xmath212 in ( 3.2 ) ( from the right to the left ) except for the left @xmath183 , then on each step the asymptotics of the modified expression ( 3.2 ) remains the same : the modified expression can be obtained from the initial one by composing it with an operator tending to the identity . at the last step the final modified expression will be just @xmath183 , which tends to @xmath109 . this will prove the convergence of ( 3.2 ) to @xmath109 . now the convergence of the commutator ( 3.2 ) is implied by the following lemma ( modulo ( 3.3 ) ) . let @xmath209 , @xmath210 be two families of two - dimensional complex diagonalizable linear operators depending on a positive parameter @xmath78 . let @xmath213 be the ( diagonal ) matrices of @xmath214 , @xmath80 , in their eigenbases . let @xmath183 be the transition matrix between the eigenbases , more precisely , in the eigenbase of @xmath209 one have @xmath215 . let the eigenbases converge to some bases in the space so that the corresponding transition matrix @xmath183 tends to a unipotent lower - triangular matrix ( denoted by @xmath109 ) . let @xmath216 let @xmath217 , @xmath218 and the upper - triangular element @xmath184 of the matrix @xmath183 satisfy ( 3.4 ) . then @xmath219 the lemma will be proved in subsection 3.3 . * of theorem 2.16 modulo ( 3.3 ) and lemma 3.5 . * the operators @xmath220 satisfy the conditions of the previous lemma : all the conditions follow from the previous proposition , corollary and ( 3.3 ) . this together with the lemma implies the convergence of the first commutator in theorem 2.16 . this proves theorem 2.16 modulo ( 3.3 ) and lemma 3.5 . we prove the following more precise version of ( 3.3 ) . let ( 2.1 ) be a generic family of linear equations ( see definition 2.1 ) , @xmath18 be its singularity families , @xmath98 be the corresponding sectors ( see definition 2.4 ) chosen to cover a punctured neighborhood of zero , @xmath100 be the corresponding domains from ( 2.2 ) . let @xmath109 , @xmath110 be the stokes matrices ( 1.6 ) of the nonperturbed equation ( corresponding to the left ( respectively , right ) component of the intersection @xmath69 ) , @xmath221 let @xmath17 be the monodromy operator of the perturbed equation around @xmath11 acting in the space of solutions on @xmath100 . let @xmath116 be the ( fundamental matrix of ) its eigenbase . let @xmath112 be the transition matrix ( 2.3 ) between the bases @xmath222 in the left component of the intersection @xmath114 . let the previous eigenbases be normalized to converge so that @xmath223 ( see corollary 2.6 ) : @xmath224 let @xmath225 be the eigenvalues of @xmath167 , @xmath226 be the corresponding projective multiplier . then the upper - triangular element @xmath184 of the matrix @xmath112 has the asymptotics @xmath227 where @xmath206 is the upper - triangular element of the stokes matrix @xmath110 . the transition matrix @xmath112 , @xmath228 , compares the monodromy eigenbases in the left component of the intersection @xmath114 , in particular , on a real interval in @xmath229 . it is not changed , when we extend the basic solutions analytically from @xmath229 to @xmath230 along the real line . denote @xmath231 the corresponding branch on @xmath230 of the extended fundamental matrix @xmath116 , @xmath80 . it follows from definition that @xmath232 is obtained from @xmath233 by applying the inverse monodromy operator @xmath234 : @xmath235 on the other hand , we can choose a renormalization of the eigenbase @xmath236 by multiplication of the basic solutions by constants ( i.e. , changing it to @xmath237 , @xmath238 is some family of diagonal matrices ) so that in the right component of the intersection @xmath114 the transition matrix @xmath113 between @xmath237 and @xmath239 tends to the stokes matrix @xmath110 : @xmath240 by definition , @xmath241 . substituting the latter and ( 3.7 ) to the previous formula yields @xmath242 the matrices @xmath243 tend to the stokes matrices @xmath136 , which are unipotent . the matrices @xmath244 , @xmath167 are diagonal and depend on @xmath78 . this implies that @xmath245 this together with ( 3.8 ) implies ( 3.6 ) . in the proof of lemma 3.5 we use the following proposition . let @xmath183 be a family of two - dimensional matrices depending on a parameter @xmath74 and converging to a unipotent lower - triangular matrix , as @xmath106 . let @xmath246 be the upper - triangular element of @xmath183 ( thus , @xmath247 ) . let @xmath248 be a family of diagonal matrices depending on @xmath76 such that @xmath249 then @xmath250 the diagonal elements of the matrix in ( 3.10 ) are equal to those of @xmath183 , and thus , tend to 1 . its lower - triangular element tends to 0 : it is equal to that of @xmath183 ( which tends to a finite limit ) times @xmath251 ( which tends to 0 by ( 3.9 ) ) . its upper - triangular element , which is equal to @xmath252 , tends to 0 by ( 3.9 ) . this proves ( 3.10 ) . consider the commutator ( 3.2 ) : @xmath253 using the previous proposition , we firstly `` kill '' the right @xmath254 : we show that expression ( 3.11 ) is equal to @xmath255 then we kill similarly the right @xmath183 in ( 3.12 ) and the remaining @xmath254 . finally we get that the initial commutator is equal to @xmath183 times the commutator of diagonal matrices ( which is identity ) times @xmath256 . this implies that ( 3.2 ) tends to @xmath203 . the first step : killing of @xmath257 . let @xmath258 by definition , expression ( 3.11 ) is equal to @xmath259 it suffices to show that @xmath260 . this follows from the previous proposition applied to the families of matrices @xmath254 and @xmath261 : these families satisfy the conditions of the previous proposition . indeed , by ( 3.5 ) , @xmath262 . the upper - triangular element of @xmath257 ( denoted by @xmath263 ) is @xmath264 by ( 3.4 ) . this proves ( 3.9 ) for @xmath263 . the conditions of the proposition are checked . thus , by ( 3.10 ) , @xmath265 the second step : killing of the right @xmath168 in ( 3.12 ) . it repeats the previous discussions with the families @xmath183 and @xmath266 . the third step : killing of the left @xmath257 . done analogously by applying the previous proposition to the matrix families @xmath257 and @xmath267 . lemma 3.5 is proved . the proof of theorem 2.16 in higher dimensions repeats that in the two - dimensional case with some changes specified below . let ( 2.1 ) be a generic family of equations , @xmath18 be its singularity families , @xmath80 , @xmath98 , be the corresponding associated sectors . consider their `` left intersection component '' that is crossed while going counterclockwise from @xmath67 to @xmath68 . let @xmath85 be a point lying in this component . let @xmath137 be the corresponding local solution space , @xmath268 be the corresponding monodromy operators ( see definition 2.14 ) . let @xmath269 be the eigenbase of the monodromy operator @xmath166 , where the eigenfunctions are taken in the order of decreasing of the modules of the corresponding eigenvalues ( it appears that these modules are really distinct , see the next proposition ) . let @xmath270 be that of @xmath167 , and the order of the eigenfunctions coincide with the order of increasing of the modules of the eigenvalues . let @xmath183 be the transition matrix between them : @xmath271 let @xmath109 , @xmath110 be the stokes matrices of the nonperturbed equation in the left ( respectively , right ) connected component of the intersection @xmath69 . in the above conditions for any @xmath78 small enough each monodromy operator @xmath17 , @xmath80 , has distinct eigenvalues ( denote them @xmath272 ) . moreover , for any @xmath273 , @xmath274 , one has @xmath275 , @xmath276 , as @xmath106 . appropriately normalized eigenbases @xmath269 and @xmath270 converge to canonical solution bases of the nonperturbed equation in @xmath67 and @xmath68 respectively . let @xmath277 be the eigenvalues of the matrix @xmath4 numerated in the order of increasing of the values @xmath278 . the numeration of each ( converging ) monodromy eigenbase corresponds to the numeration of the limit canonical solution base by the previous eigenvalues @xmath279 . in dimension two the proposition follows from proposition 3.3 . in higher dimension its proof is analogous to that of proposition 3.3 . the stokes matrices @xmath109 and @xmath110 are lower- ( respectively , upper- ) triangular . this is implied by the last statement of the previous proposition and the following well - known fact . ( see , e.g. , [ 9 ] ) . let @xmath280 , @xmath121 , ( 1.1 ) be an irregular equation , @xmath281 be the eigenvalues of the corresponding matrix @xmath4 . let @xmath67 , @xmath68 be a pair of intersected @xmath47- good sectors . let there exist a @xmath282 such that the sequence of the values @xmath283 , @xmath284 , increases . consider the canonical sectorial solution bases of ( 1.1 ) numerated by @xmath279 . then the stokes matrix of ( 1.1 ) in the connected component containing @xmath32 of the intersection @xmath69 is lower - triangular . in the case of the inverse order of the eigenvalues it is upper - triangular . the point @xmath285 satisfies the conditions of the previous proposition with @xmath20 ( the last statement of proposition 3.8 ) . hence , by proposition 3.9 , the stokes matrix @xmath109 is lower - triangular and @xmath110 is upper - triangular . let @xmath286 formula ( 3.6 ) of lemma 3.6 extends to higher dimension as follows : @xmath287 the proof of ( 3.13 ) repeats that of ( 3.6 ) in subsection 3.2 . let @xmath153 , @xmath154 , @xmath288 , @xmath80 , @xmath284 . one has @xmath289 formula ( 3.14 ) follows from ( 3.13 ) , the inequality @xmath154 and the asymptotic formula @xmath290 , @xmath284 , which is proved analogously to proposition 3.3 . as at the end of subsection 3.1 , theorem 2.16 is implied by ( 3.14 ) and the following higher - dimensional analogue of lemma 3.5 . let @xmath209 , @xmath210 be two families of @xmath44- dimensional diagonalizable linear operators depending on a positive parameter @xmath78 . let @xmath291 be the ( diagonal ) matrices of @xmath214 in their eigenbases . let @xmath292 be the transition matrix between their eigenbases , more precisely , in the eigenbase of @xmath209 the matrix of @xmath210 is @xmath293 . let the eigenbases converge to some bases in the space so that the transition matrix @xmath183 converges to a unipotent lower - triangular matrix ( denoted by @xmath109 ) . let for any @xmath274 , @xmath273 , @xmath294 let in addition the asymptotic formula ( 3.14 ) hold . then @xmath219 the proof of the lemma repeats that of lemma 3.5 with obvious changes . in the present section we consider only _ two - dimensional irregular equations with poincar rank @xmath20 _ and their generic deformations . as before , without loss of generality we assume that @xmath295 , @xmath296 , @xmath297 . let ( 2.1 ) be a generic family of linear equations , @xmath18 , @xmath80 , be its singularity families , @xmath67 , @xmath68 be the corresponding associated sectors forming a covering of a punctured neighborhood of 0 . let @xmath298 be arbitrary fixed base point independent on @xmath78 . let @xmath299 , @xmath300 be the corresponding monodromy operators of the perturbed equation ( see definition 2.14 ) . consider the circle centered at 0 and passing through @xmath85 with the counterclockwise orientation ( it bounds a disc containing both singularities of the perturbed equation for any small @xmath78 ) . the monodromy operator along the previous circle is called the _ complete monodromy_. the complete monodromy of the perturbed equation in a generic family ( 2.1 ) converges to the monodromy of the nonperturbed equation along the counterclockwise circuit . in the previous conditions the complete monodromy is equal to @xmath301 . in the present section we state and prove the theorem saying that for any generic deformation ( 2.1 ) of a typical equation ( 1.1 ) ( see the next definition ) each word ( 1.3 ) with integer exponents tends to infinity in @xmath302 , except for the powers @xmath303 of the complete monodromy . let ( 1.1 ) be an irregular equation , as at the beginning of the paper , @xmath304 be arbitrary fixed base point , @xmath305 be the counterclockwise monodromy operator around zero . consider some branches at @xmath85 of all the sectorial canonical solutions of ( 1.1 ) as elements of @xmath137 and take the collection of the complex lines in @xmath137 generated by them . the equation is said to be _ typical _ , if for any @xmath306 no line from the previous collection is transformed by @xmath307 to another line from the same collection . the definition of typical equation does not depend on the choice of the base point and the branches of the canonical solutions . the condition that an equation ( 1.1 ) is typical is equivalent to a countable number of polynomial inequalities on the formal monodromy eigenvalues and the elements of the stokes matrices . each monodromy operator word ( 1.3 ) can be rewritten as @xmath308 where @xmath309 in the notations of ( 1.3 ) . we consider those words ( 4.1 ) that do not coincide literally with powers of the complete monodromy . let ( 1.1 ) be a typical equation ( see definition 4.2 ) , ( 2.1 ) be its generic deformation , @xmath85 , @xmath312 , be as at the beginning of the section . then any monodromy operator given by a reduced word ( 4.1 ) tends to infinity ( together with its projectivization , see definition 3.1 ) , as @xmath106 . instead of invertible linear operators @xmath313 we will consider their projectivizations , which are mbius transformations @xmath179 . denote @xmath314 , @xmath315 the projectivizations of the monodromy operators @xmath316 . we show that any reduced word @xmath317 tends to infinity in the mbius group , as @xmath106 . this will prove the theorem . ( see [ 1 ] ) . a mbius transformation @xmath318 is said to be _ hyperbolic _ , if it has one repelling fixed point ( then there is a unique attracting fixed point and each orbit except for the repeller tends to the attractor ) . a hyperbolic transformation with repeller @xmath319 and attractor @xmath320 will be presented as the picture at fig.4 let ( 2.1 ) , @xmath98 , @xmath85 , @xmath17 be as at the beginning of the section ( the nonperturbed equation is not necessarily typical ) , @xmath322 be the projectivizations of @xmath17 , @xmath80 ( see definition 3.1 ) . let @xmath323 be the eigenbases of @xmath17 , @xmath324 be the sectorial canonical solution bases on @xmath98 of the nonperturbed equation , @xmath80 . let @xmath325 be the tautological projection images of @xmath326 and @xmath327 respectively . then @xmath322 are hyperbolic transformations ( see the previous definition ) with fixed points @xmath328 : @xmath329 , @xmath330 are respectively the repelling and attracting fixed points of @xmath314 ; @xmath331 , @xmath332 are respectively the repelling and attracting fixed points of @xmath315 . let ( 2.1 ) , @xmath98 , @xmath85 , @xmath17 be as at the beginning of the section , @xmath333 , @xmath328 be the tautological projection images of the canonical basic solutions of the nonperturbed equation and the eigenfunctions of @xmath17 respectively ( see the previous proposition ) . then @xmath334 the statements on the limits in ( 4.2 ) follow from theorem 2.5 . the coincidence of @xmath335 and @xmath336 in ( 4.2 ) follows from the lower - triangularity of the stokes matrix @xmath109 ( see example 1.7 ) . let @xmath337 , @xmath338 , @xmath339 be a given triple of distinct points in @xmath340 . let @xmath341 , @xmath80 , be two families of hyperbolic mbius transformations depending on a positive parameter @xmath78 , @xmath342 , @xmath343 be the multipliers of their attractors . let @xmath330 , @xmath332 be respectively the attractors of @xmath314 and @xmath315 , @xmath329 , @xmath331 be their repellers . let @xmath344 , @xmath345 , as @xmath106 . let in addition the product @xmath346 converge to a mbius transformation @xmath347 such that for any @xmath306 , @xmath348 @xmath349 the image @xmath350 coincide with no other @xmath351 . then any reduced word @xmath317 ( see definition 4.5 ) tends to infinity . the projectivizations of the monodromy operators satisfy the conditions of the lemma . indeed , the multipliers @xmath352 tend to 0 by corollary 3.4 . the convergence @xmath344 follows from the previous proposition . the product @xmath346 tends to the projectivization ( denoted by @xmath347 ) of the monodromy of the nonperturbed equation . the inequalities @xmath353 follow from typicality ( see definition 4.2 ) . this together with the lemma proves divergence of @xmath321 . theorem 4.6 is proved modulo the lemma . in the conditions of the previous lemma let @xmath317 be a reduced word such that @xmath354 let @xmath355 be arbitrary point such that @xmath356 then the image @xmath357 converges to the limit of the attractor of @xmath358 . let us prove lemma 4.10 modulo lemma 4.11 . if the reduced word @xmath321 under consideration satisfies ( 4.3 ) , then it tends to infinity . indeed , by lemma 4.11 , the image @xmath357 of a generic @xmath359 tends to the attractor of @xmath358 , hence , @xmath360 . otherwise , @xmath361 , where @xmath306 , @xmath362 is a word satisfying ( 4.3 ) of a length less than that of @xmath321 . the new word @xmath362 tends to infinity by the previous statement . the product @xmath346 in the previous expression for @xmath321 has a finite limit . hence , @xmath321 tends to infinity as well . lemma 4.10 is proved . let @xmath363 be a family of hyperbolic mbius transformations depending on a positive parameter @xmath78 , @xmath364 , as @xmath106 , so that the attractor and the repeller of @xmath362 tend to distinct limits ( hence , the multiplier of the attractor tends to 0 ) . then for any point @xmath355 distinct from the limit of the repeller its image @xmath365 tends to the same limit , as the attractor . the same statement holds for arbitrary family @xmath366 of points bounded away from the limit of the repeller . we prove lemma 4.11 by induction in the length @xmath44 of the word . for @xmath367 its statement is obvious . suppose we have proved the lemma for the words of any length less than a given @xmath44 . let us prove it for a word @xmath317 of the length @xmath44 . without loss of generality we assume that @xmath368 : the contrary case is treated analogously . then by ( 4.3 ) and the inequality @xmath369 ( see ( 4.1 ) ) , @xmath370 consider the word @xmath371 suppose firstly that it satisfies ( 4.3 ) . then for any @xmath359 satisfying ( 4.4 ) @xmath365 tends to the limit ( denoted @xmath333 ) of the attractor of @xmath372 by the induction hypothesis . the latter attractor can be either @xmath330 , or @xmath331 , which are the attractor of @xmath314 and the repeller of @xmath315 respectively . this follows from ( 4.5 ) . hence , the limit @xmath333 is either @xmath338 , or @xmath339 ; in both cases it does not coincide with the limit @xmath335 of the repeller of @xmath314 . therefore , the image @xmath373 ( and hence , @xmath374 ) tends to the same limit , as the attractor of @xmath314 ( by the previous proposition ) . now suppose that the word @xmath362 does not satisfy ( 4.3 ) . then @xmath375 where @xmath306 , and @xmath376 is a word satisfying ( 4.3 ) of a length less than that of @xmath362 . the induction hypothesis applied to @xmath376 implies that for any @xmath359 satisfying ( 4.4 ) its image @xmath377 tends to the limit of the attractor of the last left element of the word @xmath376 , thus , to some @xmath333 . therefore , @xmath378 . the image @xmath350 coincides with no @xmath379 ( in particular , with the limit @xmath335 of the repeller of @xmath314 ) . this follows from the last condition of lemma 4.10 . therefore , @xmath380 tends to the limit of the attractor of @xmath314 ( and so does @xmath381 by the previous proposition ) . the induction step is over . lemma 4.11 is proved . the proof of lemma 4.10 is completed . i am grateful to v.i.arnold , yu.s.ilyashenko and a.a.bolibrukh for stating the problems . i wish to thank them and v.kleptsyn for their interest to the work and helpful remarks . the research was supported by part by crdf grant rm1 - 2358-mo-02 and by russian foundation for basic research ( rffi ) grant 02 - 02 - 00482 . apanasov , b.n . , gusevskii , n.a . , krushkal , s.l . , _ kleinian groups and uniformization in examples and problems_. translated from the russian by h. h. mcfaden . translation edited and with a preface by bernard maskit . translations of mathematical monographs , 62 . american mathematical society , providence , ri , 1986 . arnold , v.i . , ilyashenko , yu . s. , ordinary differential equations . - in the volume _ dynamical systems1 _ of the series `` itogi nauki i tekhniki . sovremennyie problemy matematiki . fundamentalnyie napravlenia '' , viniti publisher , moscow ( 1985 ) . - english translation in _ encyclopaedia math . , 1 , dynamical systems , i _ , 1148 , springer , berlin , 1988 . glutsyuk , a. confluence of singular points and the nonlinear stokes phenomena ( in russian ) , _ trudy moskovskogo matematicheskogo obshchestva , _ vol.62 ( 2001 ) , pp 54 - 104 . english translation in _ transactions of moscow math . soc . , _ 2001 , 49 - 95 . ilyashenko , yu . s. ; khovanskii , a. g. , galois groups , stokes operators and a theorem of ramis . ( russian ) _ funktsional . i prilozhen . _ 24 ( 1990 ) , no . 4 , 3142 , 96 ; translation in _ funct . anal . _ 24 ( 1990 ) , no . 4 , 286296 ( 1991 ) . mardesic , p. , roussarie , r. , rousseau , c. , modulus of analytic classification for unfoldings of generic parabolic diffeomorphisms , prpublication no 292 ( 2002 ) du laboratoire de topologie et du laboratoire gevrey de mathmatique physique , universit de bourgogne . , dijon , france .

in this paper we present a new approach to the solution to a generalized version of hughes models for pedestrian movements based on a follow - the - leader many particle approximation . in particular , we provide a rigorous global existence result under a smallness assumption on the initial data ensuring that the trace of the solution along the turning curve is zero for all positive times . we also focus briefly on the approximation procedure for symmetric data and riemann type data . two different numerical approaches are adopted for the simulation of the model , namely the proposed particle method and a godunov type scheme . several numerical tests are presented , which are in agreement with the theoretical prediction . marco di francescoand simone fagioli massimiliano daniele rosini giovanni russo _ dedicated to professor peter a. markowich for his 60th birthday _ ( communicated by the associate editor name )

1602.06153

the mathematical modeling of human crowds has gained significant attention in the last decades , see @xcite . in 2002 , roger l. hughes @xcite proposed a ` thinking fluid ' approach to this subject , which consists in modeling a human crowd as a continuum medium , with eulerian velocity computed via a nonlocal constitutive law of the overall distribution of pedestrians . such a nonlocal dependence is encoded in a _ weighted _ distance function computed at a quasi - equilibrium regime , obeying an eikonal equation with right - hand side depending on the density of pedestrians . in its multidimensional formulation , the model reads @xmath1 here @xmath2 denotes the position variable , @xmath3 is the time , and @xmath4 represents the averaged crowd density . the model is typically posed on a bounded domain @xmath5 , with boundary conditions depending on the position of the exits in @xmath6 , see @xcite for further details . the map @xmath7 represents the absolute value of the velocity , and is assumed to be decreasing , as lower velocities correspond to higher densities . the function @xmath8 in models the pedestrians perception of the exits , and is computed as a weighted distance encoding the overall distribution of the crowd . it may be interpreted as an estimated exit time for a given distribution of pedestrians . more precisely , pedestrians choose their direction upon evaluation of a _ running cost _ function @xmath9 . such a function is assumed to be increasing , since pedestrians choose their path toward the closest exit avoiding densely crowded regions . hughes approach addresses specifically the movement of pedestrians in highly populated domains . in @xcite , the model has been suggested in particular to provide assistance in the annual muslim hajj and , quoting from @xcite , ` to locate barriers that actually increase the flow of pedestrians above that when there are no barriers present ' . such a phenomenon , analogous to a similar one known in vehicular traffic modeling , is often referred to in the literature as _ braess paradox _ for pedestrians , see @xcite . hughes model is way far from being the only significant pde - based approach to pedestrian movements , see e.g. @xcite . compared to other existing models in the literature , in addition to its potential applications listed in @xcite , this model has the undeniable advantage of combining classical overcrowding avoidance effects with singular _ direction - switching _ phenomena , which occur typically in a densely crowded region with limited information on the location of the exits . clearly , the global perception of the domain implied by the eikonal equation in is not totally realistic in most situations . however , the model is very flexible and can be easily improved to localize the range of perception , see @xcite . another argument in favour of has been recently provided in @xcite , in which a full - transient version of this model has been justified as a mean field game . in this paper we focus on the hughes model @xcite in its one - dimensional formulation [ eq : model ] @xmath10_x = 0 , & & |\phi_x|=c(\rho ) , & x \in ( -1,1),~ t > 0,\end{aligned}\ ] ] coupled with homogeneous dirichlet boundary conditions @xmath11 and initial datum @xmath12 in this version of the model , two exits are located at the edges of the interval @xmath13 . it will be assumed that a maximal value of @xmath14 is prescribed , which we set equal to @xmath15 for simplicity . a preliminary _ formal _ maximum principle proved in @xcite shows that @xmath14 never exceeds the range @xmath16 $ ] if @xmath17 takes values in @xmath16 $ ] . we shall assume what follows . * the velocity map @xmath18 \to [ 0,+\infty)$ ] is @xmath19 , strictly decreasing , with @xmath20 and @xmath21 . moreover , there exists a @xmath22 such that @xmath23 ( \hat\rho - \rho ) > 0 $ ] for all @xmath24 . * the running cost map @xmath25 \to [ 1,+\infty]$ ] is @xmath26 , increasing , with @xmath27 , and with @xmath28 . an important example is @xmath29 see @xcite and example [ exe : c ] below . the boundary conditions are justified as follows . if we denote the efficiency @xmath30 $ ] of the exits by the ratio between the escape velocity allowed by the exit and the maximum pedestrian velocity , we have that the flux of pedestrians through the exit is proportional to @xmath31 , i.e. we obtain the robin type condition condition @xmath32 where the term @xmath33 gets cancelled when considering the outgoing normal direction of the flux . this is equivalent to imposing the exit velocity @xmath34 since we assume that the velocity is related to the density via a monotonic function , then imposing the velocity is equivalent to imposing the density . hence , at variance with what happens in diffusion problem in which robin boundary conditions become in the limit either dirichlet or neumann condition , in our case we simply obtain a dirichlet conditions on the density . for simplicity , we always considered the case of _ maximum _ efficiency @xmath35 , i.e. @xmath36 , which in turn gives zero boundary conditions on the density . note however that in the cases in which the characteristics are exiting the domain , no boundary conditions have to be assigned on the boundary . a discussion about this issue is provided later on in this section . by a simple duality argument , a similar zero dirichlet condition is obtained for @xmath8 , see the discussion in @xcite . a fully satisfactory existence theory for is still missing . a mathematical theory in this setting was first addressed in @xcite , in which the eikonal equation @xmath37 was replaced by two regularised versions involving a laplacian term . in both cases , existence of weak solutions was proved , and the uniqueness of entropy solutions was provided in one of the two cases . @xcite was the first rigorous attempt at the mathematical theory for the hughes model . the use of a regularised eikonal equation implies a @xmath19 regularity of the velocity field in the continuity equation for the density @xmath14 . such a property does not hold in the non - regularized case , and the numerical simulations performed in @xcite showed evidence of discontinuities in the velocity field , expressing sudden changes of direction in the movement of the pedestrians . a first attempt at a rigorous mathematical treatment of ( without any regularization ) was performed independently in @xcite and @xcite . in both papers , the authors provided a complete description of the behaviour of solutions for short times in the case of riemann - type initial data on an interval . in @xcite a solution semigroup was defined for short times for piecewise constant initial conditions . as a special case , solutions with direction - switching phenomena were produced analytically . in @xcite the riemann problem was analyzed more thoroughly , and the optimization of the evacuation time in that case was studied . numerical simulations were performed in @xcite and in @xcite , essentially based on godunov schemes in both papers . the results in @xcite laid the basis for tackling the convergence of a wave - front - tracking scheme @xcite for , see the numerical simulations in @xcite . such an approach was only partly successful , and led to the results in @xcite , which show existence of entropy solutions when the initial condition is _ well - separated _ , i.e. is yielding the formation of two distinct groups of pedestrians moving towards the two exits , with the emergence of a vacuum region in between , persisting until the total evacuation of the domain . a more general set of initial conditions was never covered so far , the main difficulty lying in the very large number of possible case studies in the interactions between classical and nonclassical waves in the wave - front - tracking algorithm . as already observed in @xcite , the one - dimensional hughes model can be reformulated in terms of the density @xmath14 as follows : @xmath38 with @xmath39 here , @xmath40 plays the role of a _ turning point _ at which pedestrians switch their target exit in view of the overall crowd distribution . the reduction of to is soon explained . in order to have a unique physically relevant ( semi - concave ) viscosity solution to the dirichlet problem for @xmath8 , the derivative @xmath41 can change its sign just once from positive to negative . the turning point @xmath40 is then defined as the point @xmath2 at which @xmath42 reaches its maximum . @xmath43 depends on time because of the @xmath14-depending term on the right hand side of the eikonal equation in . indeed , @xmath40 depends _ non - locally _ on @xmath44 . the second equation in is recovered after a simple integration on the eikonal equation in , upon the assumption that @xmath42 is continuous at @xmath40 . notice that the flux @xmath45 is possibly discontinuous along the _ turning curve _ @xmath46 . for a detailed study of a rankine - hugoniot condition along the turning point we refer to @xcite . the simplest choice for the cost function is @xmath47 . in this case the obvious solution to the eikonal equation in is the distance function @xmath48 for all @xmath49 . as a consequence , the ratio @xmath33 is equal to @xmath50 for @xmath51 and to @xmath15 for @xmath52 , and the model is equivalent to the scalar conservation law with discontinuous flux @xmath53_x=0.\ ] ] as a result , hughes model reduces to the lighthill - whitham - richards ( lwr ) model for traffic flow @xcite with negative velocity on @xmath54 and positive velocity on @xmath55 . roughly speaking , pedestrians are moving toward the closest exit regardless of the overall distribution . this is the typical behaviour of a crowd in a _ panic _ state , see @xcite . [ exe : c ] in the literature , the usual choice for the cost function is @xmath56 see @xcite . in this case , in order to bypass the technical issue of @xmath57 blowing up at @xmath58 in the eikonal equation of , the initial condition @xmath17 is assumed to take values in @xmath59 $ ] , for some constant @xmath60 sufficiently small . this assumption , together with the maximum principle , ensures that the cost computed along any solution of is well defined . rewriting hughes model in the form highlights the non - local mechanism by which the overall distribution of pedestrians in the domain biases the choice of the exit path for each single pedestrian . since @xmath57 is increasing w.r.t . @xmath14 , such a mechanism penalises regions with high density in the choice of the nearest exit . in the relevant case in , @xmath61 gets very large when @xmath14 is close to @xmath15 , and the weighted distance function @xmath8 gets very steep in this case . the reformulation of in the form clearly suggests that hughes model can be seen as a two - sided first order lwr model for traffic flow @xcite , with the turning point @xmath40 splitting the whole interval @xmath13 into two subintervals . hinted by the new particle approach developed in @xcite , in which nonnegative entropy solutions to nonlinear scalar conservation laws @xmath62_x = 0 $ ] are rigorously approximated by the empirical measure of the follow - the - leader ode particle system , in which particle @xmath63 satisfies an equation of the form @xmath64 where @xmath65 denotes the mass assigned to each particle , in the present paper we propose a new _ deterministic particle _ approach to , which can be roughly formulated as follows . here we just outline our particle method and refer to section [ sec : particle ] for its detailed application to approximating the solution of via atomization . for a fixed @xmath66 and a fixed mass @xmath67 , let @xmath68 be the initial positions of @xmath69 particles in @xmath13 . the initial position @xmath70 of the discrete turning point is determined uniquely by the identity @xmath71+\sum_{i=0}^{i_0 - 1 } \left [ \bar{x}_{i+1}-\bar{x}_i\right ] c\left(\frac{m}{\bar{x}_{i+1}-\bar{x}_i}\right)+[\bar\xi-\bar{x}_{i_0}]\\ = & \left [ \bar{x}_{i_0 + 1}-\bar\xi\right ] + \sum_{i = i_0 + 1}^{n-1 } \left [ \bar{x}_{i+1}-\bar{x}_i\right ] c\left(\frac{m}{\bar{x}_{i+1}-\bar{x}_i}\right)+\left [ 1-\bar{x}_n\right ] , \label{eq : initial_turning_discrete_intro } \end{split}\ ] ] which replaces the integral identity in , with the density @xmath14 replaced by a discretized particle density , and with the assumption that a vacuum region is placed around @xmath72 and near the boundary . we observe that here we assume zero density around the turning point and outside the interval @xmath73 $ ] , which yields in view of the assumption @xmath28 . the pair @xmath74 is uniquely determined by condition ( [ eq : initial_turning_discrete_intro ] ) and @xmath75 . assuming for simplicity that @xmath76 , i.e. that @xmath77 , we let the particles @xmath78 move towards the left exit via a backward follow - the - leader system , and the particles @xmath79 move towards the right exit via a forward one . more precisely , [ eq : discrete_hughes ] @xmath80 with @xmath81 for all @xmath82 . the discrete turning point @xmath40 is required to obey @xmath83 and the identity @xmath84+\sum_{i=1}^{i_0 - 1 } \left [ x_{i+1}-x_i\right ] c\left(\frac{m}{x_{i+1}-x_i}\right)+\left [ \xi - x_{i_0}\right ] \\ = & \left [ x_{i_0 + 1}-\xi\right ] + \sum_{i = i_0 + 1}^{n-1 } \left [ x_{i+1}-x_i\right ] c\left(\frac{m}{x_{i+1}-x_i}\right)+\left [ 1-x_n\right ] \ , . \end{split}\ ] ] the above formula is used as long as all points @xmath85 belong to the interval @xmath86 $ ] , and is suitably modified by removing the particles leaving the domain and adjusting the length of the first and last interval ( see the subsection [ sec : particle ] for the details ) . clearly , system must be restarted once @xmath40 crosses one of the two neighbor particles @xmath87 , @xmath88 . in this case , a detailed analysis comparing the speed of the neighbor particle with @xmath89 at the crossing time is needed in order to define the post - crossing evolution consistently . we shall refer to system as the _ ftl - hughes model_. the goal of this paper is twofold . * first of all we use the ftl - hughes model as a way to prove existence of weak solutions to the continuum model . we perform this task under analogous assumptions to those required in @xcite , namely in situations in which a vacuum region is generated around @xmath90 , but with a lighter proof and under more general assumptions on the cost function @xmath57 ( only the cost function @xmath91 is covered in @xcite ) . as a byproduct of our result , we obtain a rigorous deterministic many - particle approximation result for a class of solutions to . * as a second task , we implement the ftl - hughes scheme numerically in more complicated situations yielding particles crossing the turning point . we shall compare these simulations to existing ones in the literature @xcite based on godunov s scheme , and show that our deterministic particle approach is able to capture mass transfer phenomena around the turning point . next we provide our notion of solution to and state our main results . for notational simplicity , we introduce the maximal initial density @xmath92 $ ] such that @xmath57 is well defined on @xmath93 $ ] . then the initial datum @xmath17 is assumed to be in @xmath94 with @xmath95 . since we are dealing with zero dirichlet boundary data , assuming strict concavity of @xmath96 on @xmath16 $ ] , it is easily seen that the proper boundary condition set in @xcite is @xmath97 $ ] , where @xmath98 is uniquely determined by @xmath99 , see condition ( v ) above . such a condition expresses the fact that characteristics are pointing outward the domain ( both in @xmath100 and in @xmath101 ) if and only if @xmath97 $ ] on those two points . it is well known ( see @xcite ) that the boundary problem with dirichlet conditions for a nonlinear conservation law is solved by considering a riemann problem involving the boundary datum and the trace @xmath102 at the boundary . since the boundary datum is zero and since the flux is concave , the riemann solver at the boundary only consists of rarefaction waves , which leave the domain if @xmath103 and are centered on the value @xmath104 at the boundary points if @xmath105 . now , a solution of the cauchy problem with a general @xmath106 initial datum with support on @xmath86 $ ] and _ without boundary conditions _ within the wave - front - tracking algorithm @xcite easily shows that @xmath14 achieves only values in @xmath107 outside @xmath86 $ ] . therefore , the solution via wave - front - tracking without boundary data coincides with the one in which the boundary datum @xmath108 is imposed every time a wave interacts with the boundary . hence , by using the convergence results in @xcite , we recover that actually _ no boundary conditions have to be prescribed in our case _ , and the solution @xmath14 to can be constructed by coupling the integral relation in with the continuity equation on the _ whole _ @xmath109 restricted to the interval @xmath86 $ ] . for this reason , also due to the fact that we shall only consider solutions with a vacuum region around the turning point , our notion of entropy solution will simply be obtained by ` merging ' the two traffic equations on @xmath110 $ ] and @xmath111 $ ] respectively . observe that the strong traces of the solution at the boundary points exist due to the genuine non - linearity of the flux ( @xcite ) and must satisfy @xmath112 in this paper we only deal with boundary conditions with maximal efficiency @xmath113 in . we shall address the case @xmath114 in a future work . we also plan to impose ( local and non - local ) point constraints on the flow at the two boundary points to model the presence , for instance , of exit doors , see @xcite or ( * ? ? ? * chapter 6 ) for more details . we now state our notion of solution , which is motivated by the presence of a persistent vacuum region around the turning point for the class of solutions considered in this paper . [ def : entropy_solution ] a map @xmath115)$ ] is called a ( well - separated ) _ entropy solution _ to if * there exists @xmath116 such that @xmath14 is equal to zero on the open cone @xmath117 * @xmath118 is an entropy solution in the kruzhkov sense @xcite to the cauchy problem for @xmath119_x = 0 $ ] with initial datum @xmath120 . * @xmath121 is an entropy solution in the kruzhkov sense @xcite to the cauchy problem for @xmath122_x = 0 $ ] with initial datum @xmath123 . * the turning curve @xmath124 \colon x=\xi(t)\right \}$ ] is continuous with @xmath125 . moreover , @xmath44 and @xmath40 satisfy @xmath126 for almost every @xmath49 . finally , @xmath127 . as a first result , we provide the following theorem for the ` symmetric ' case . let us denote by @xmath128 the space of functions @xmath129\right)$ ] with @xmath130 that are _ even _ , namely @xmath131 for a.e . @xmath132 . [ teo:1 ] for any initial datum @xmath133 in @xmath128 , there exists a unique entropy solution @xmath14 to in the sense of definition [ def : entropy_solution ] , such that @xmath134 for all @xmath135 . such a solution is obtained as a strong @xmath136 limit of the discrete density @xmath137 constructed in subsection [ sec : particle ] via the ftl - hughes particle system . in the next theorem we state our main result , which deals with a class of _ small _ initial data in @xmath138 . for further use , we define the function @xmath139 which is strictly decreasing in view of assumption ( c ) above . we then set @xmath140}]=\max \left\ { c''\left(\rho\right ) \rho \colon \rho \in \left[0,\rho_{\max}\right ] \right\}\,,\label{eq : lipschitz}\\ & c \doteq c'(\rho_{\max})\ , \rho_{\max } = \max\left\{c'(\rho ) \ , \rho \colon \rho \in \left[0,\rho_{\max}\right]\right\}\,.\label{eq : cgrande}\end{aligned}\ ] ] [ teo:2 ] if @xmath141 and the initial datum @xmath142)$ ] satisfies @xmath143 < v(\rho_{\max}),\ ] ] then there exists a unique entropy solution @xmath14 to in the sense of definition [ def : entropy_solution ] defined globally in time . such a solution is obtained as a strong @xmath136 limit of the discrete density @xmath137 constructed in subsection [ sec : particle ] via the ftl - hughes particle system . we remark that the @xmath141 assumption above is essential in order to have the right - hand - side in the inequality strictly positive . the proofs of theorems [ teo:1 ] and [ teo:2 ] are carried out in subsections [ sec : mainproof1 ] and [ sec : mainproof2 ] respectively . the paper is structured as follows . in section [ sec : results ] we cover the analytical theory . in particular , in subsection [ sec : particle ] we set up the deterministic particle scheme for , namely we construct the approximation to via the ftl - hughes model . in subsections [ sec : mainproof1 ] and [ sec : mainproof2 ] we prove our analytical results , with theorem [ teo:2 ] being our main result . in section [ sec : numerics ] we show the numerical simulations of our particle methods in simple riemann - type initial conditions . in particular , we compare our tests with previous ones performed in @xcite . the numerical section shows evidence that the two methods , though conceptually different , are in very good agreement when using a large number of particles . we stress here that , although the analytical results concerning our deterministic particle method are restricted to cases in which particle separate into two distinct sets keeping the same direction for all times , the numerical simulations also cover cases with direction switching , showing that the particle approach works also in those cases . our study is restricted to the one - dimensional case . possible extensions to multidimensional cases rely on the extension of the results in @xcite to multi dimensional scalar conservation laws , which is not available in the literature . let @xmath17 be in @xmath144)$ ] . for a fixed integer @xmath145 , we set @xmath146 and @xmath147 , where @xmath148 . we set @xmath149 where @xmath150 denotes the support . we recursively define @xmath151 the above defines a set of @xmath152 particles @xmath153 with the property @xmath154 in particular , this implies that @xmath155 for future reference , we denote the local discrete initial densities @xmath156 and introduce the discretized initial density @xmath157 $ ] by @xmath158 we now define the initial approximated turning point @xmath159 via the formula @xmath160 clearly , the above formula defines @xmath161 uniquely . the next step is the definition of the evolving particle scheme . roughly speaking , @xmath159 splits the set of particles into left and right particles , the former moving according to a _ backward _ follow - the - leader scheme , the latter according to a _ forward _ one . by a slight modification of the initial condition , we may always assume that @xmath159 does not coincide with any of the particles . then , there exists @xmath162 such that @xmath163 . we then set @xmath164 we consider the corresponding discrete density @xmath165 notice that the above density has been set to equal zero outside the particle region @xmath166 and around the turning point , namely in @xmath167 . the latter in particular is simply due to a consistency with the numerical simulations , in which the computation of the turning point is made simpler in this way . this simplifying assumption will introduce a small error @xmath168 in the total mass . in view of the above notation , can be re - written as @xmath169 notice that @xmath170 does not bias the movement of any of the particles , and this is an argument in favour of the ansatz @xmath171 . for future reference , we compute for any @xmath172 @xmath173}{\left[x_{i+1}(t)-x_i(t)\right]^2 } = -r_{i+\frac{1}{2}}(t ) \ , \frac{\dot{x}_{i+1}(t)-\dot{x}_{i}(t)}{x_{i+1}(t)-x_{i}(t)}.\ ] ] in particular , the time derivative of the discrete density @xmath174 satisfies @xmath175}{[x_{i+1}(t)-x_i(t)]^2 } , & i \in \{0,\ldots , i_0 - 1\},\\[10pt ] \dot{r}_{i+\frac{1}{2}}(t ) = - \dfrac{m\left[v(r_{i+\frac{3}{2}}(t))-v(r_{i+\frac{1}{2}}(t))\right]}{\left[x_{i+1}(t)-x_i(t)\right]^2 } , & i \in \{i_0 + 1,\ldots , n-1\}. \end{cases } \label{eq : discrete_continuity_equation}\ ] ] the ( unique ) solution to the system is well defined until the turning point does not collide with a particle . similarly , the expressions for the time derivative of the discrete densities only hold until the first collision of a particle with the turning point . we note that the density @xmath176 is equal to zero until the turning point collides with a particle . in view of the discussion before definition [ def : entropy_solution ] on the solution of the continuum problem at the boundary , we shall not impose any boundary condition to the particle system , and we shall follow the movement of each particle whether or not they are in @xmath86 $ ] . hence , the discrete densities @xmath177 are ( in principle ) defined for all @xmath135 and for all @xmath178 . we remark that a careful analysis of the many particle approximation for the dirichlet boundary value problem for a scalar conservation law is performed in @xcite . the approximated turning point @xmath179 is implicitly uniquely defined by @xmath180 where @xmath181 $ ] is the discretized density defined by @xmath182 clearly @xmath179 belongs to @xmath13 for any @xmath183 . we underline that @xmath184 does not necessarily coincide with @xmath159 . we compute now the time derivative of @xmath179 . [ lem : dotxi ] fix @xmath135 , and let @xmath185 be such that @xmath186.\end{aligned}\ ] ] at time @xmath3 we have then @xmath187 \ , \frac{x_{i_-}+1}{x_{i_-}-x_{i_--1 } } + v(r_{i_--\frac{1}{2 } } ) \right\ } c'(r_{i_--\frac{1}{2 } } ) \ , r_{i_--\frac{1}{2 } } \\ & -v(r_{i_0-\frac{1}{2 } } ) \left[1-\upsilon(r_{i_0-\frac{1}{2}})\right ] + \sum_{i = i_-}^{i_0 - 1 } v(r_{i-\frac{1}{2 } } ) \left [ \upsilon(r_{i-\frac{1}{2 } } ) - \upsilon(r_{i+\frac{1}{2}})\right]\\ & + v(r_{i_0+\frac{3}{2 } } ) \left[1-\upsilon(r_{i_0+\frac{3}{2}})\right ] + \sum_{i = i_0 + 2}^{i_+ } v(r_{i+\frac{1}{2 } } ) \left[\upsilon(r_{i-\frac{1}{2 } } ) - \upsilon(r_{i+\frac{1}{2}})\right]\\ & -\left\ { v(r_{i_++\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } v(r_{i_++\frac{3}{2}})-v(r_{i_++\frac{1}{2}})\right ] \ , \frac{1-x_{i_+}}{x_{i_++1}-x_{i_+ } } \right\ } c'(r_{i_++\frac{1}{2 } } ) \ , r_{i_++\frac{1}{2 } } , \end{aligned}\ ] ] where @xmath188 is defined in . by we have that @xmath189 c(r_{i_--\frac{1}{2 } } ) + \sum_{i = i_-}^{i_0 - 1 } \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } \xi^n - x_{i_0}\right ] \\ = & \left[\vphantom{x_{i_- } } x_{i_0 + 1}-\xi^n\right ] + \sum_{i = i_0 + 1}^{i_+-1 } \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } 1-x_{i_+}\right ] c(r_{i_++\frac{1}{2 } } ) . \end{split}\ ] ] take the time derivative of the above equation @xmath190 c'(r_{i_--\frac{1}{2 } } ) \ , \dot{r}_{i_--\frac{1}{2}}\\ & + \sum_{i = i_-}^{i_0 - 1 } \left\ { \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c'(r_{i+\frac{1}{2 } } ) \ , \dot{r}_{i+\frac{1}{2 } } \right\ } + \left[\vphantom{x_{i_- } } \dot{\xi}^n-\dot{x}_{i_0}\right ] \\ = & \left[\vphantom{x_{i_- } } \dot{x}_{i_0 + 1}-\dot{\xi}^n\right ] + \sum_{i = i_0 + 1}^{i_+-1 } \left\ { \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c'(r_{i+\frac{1}{2 } } ) \ , \dot{r}_{i+\frac{1}{2 } } \right\ } \\ & -\dot{x}_{i_+ } \ , c(r_{i_++\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } 1-x_{i_+}\right ] c'(r_{i_++\frac{1}{2 } } ) \ , \dot{r}_{i_++\frac{1}{2}},\end{aligned}\ ] ] namely @xmath191 c'(r_{i_--\frac{1}{2 } } ) \ , \dot{r}_{i_--\frac{1}{2}}\\ & -\sum_{i = i_-}^{i_0 - 1 } \left\ { \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c'(r_{i+\frac{1}{2 } } ) \ , \dot{r}_{i+\frac{1}{2 } } \right\ } + \dot{x}_{i_0 } \\ & + \dot{x}_{i_0 + 1 } + \sum_{i = i_0 + 1}^{i_+-1 } \left\ { \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] c(r_{i+\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } x_{i+1}-x_i\right ] c'(r_{i+\frac{1}{2 } } ) \ , \dot{r}_{i+\frac{1}{2 } } \right\ } \\ & -\dot{x}_{i_+ } \ , c(r_{i_++\frac{1}{2 } } ) + \left[\vphantom{x_{i_- } } 1-x_{i_+}\right ] c'(r_{i_++\frac{1}{2 } } ) \ , \dot{r}_{i_++\frac{1}{2}}.\end{aligned}\ ] ] hence , from and we obtain @xmath192 c'(r_{i_--\frac{1}{2 } } ) \ , r_{i_--\frac{1}{2 } } \ , \frac{\dot{x}_{i_-}-\dot{x}_{i_--1}}{x_{i_-}-x_{i_--1}}\\ & -\sum_{i = i_-}^{i_0 - 1 } \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] \upsilon(r_{i+\frac{1}{2 } } ) + \dot{x}_{i_0 } + \dot{x}_{i_0 + 1 } + \sum_{i = i_0 + 1}^{i_+-1 } \left[\vphantom{x_{i_- } } \dot{x}_{i+1}-\dot{x}_i\right ] \upsilon(r_{i+\frac{1}{2 } } ) \\ & -\dot{x}_{i_+ } \ , c(r_{i_++\frac{1}{2 } } ) -\left[\vphantom{x_{i_- } } 1-x_{i_+}\right ] c'(r_{i_++\frac{1}{2 } } ) \ , r_{i_++\frac{1}{2 } } \ , \frac{\dot{x}_{i_++1}-\dot{x}_{i_+}}{x_{i_++1}-x_{i_+ } } \\ = & \left\ { \left[x_{i_-}+1\right ] \ , \frac{\dot{x}_{i_-}-\dot{x}_{i_--1}}{x_{i_-}-x_{i_--1 } } - \dot{x}_{i_- } \right\ } c'(r_{i_--\frac{1}{2 } } ) \ , r_{i_--\frac{1}{2 } } \\ & -\sum_{i = i_-}^{i_0 } \dot{x}_{i } \ , \upsilon(r_{i-\frac{1}{2 } } ) + \sum_{i = i_-}^{i_0 - 1 } \dot{x}_{i } \ , \upsilon(r_{i+\frac{1}{2 } } ) + \dot{x}_{i_0}\\ & + \dot{x}_{i_0 + 1 } + \sum_{i = i_0 + 2}^{i_+ } \dot{x}_{i } \ , \upsilon(r_{i-\frac{1}{2 } } ) -\sum_{i = i_0 + 1}^{i_+ } \dot{x}_{i } \ , \upsilon(r_{i+\frac{1}{2}})\\ & -\left\ { \dot{x}_{i_+ } + \left[\vphantom{x_{i_- } } 1-x_{i_+}\right ] \ , \frac{\dot{x}_{i_++1}-\dot{x}_{i_+}}{x_{i_++1}-x_{i_+ } } \right\ } c'(r_{i_++\frac{1}{2 } } ) \ , r_{i_++\frac{1}{2}}.\end{aligned}\ ] ] finally , by applying we conclude the proof . fix an initial datum @xmath133 in @xmath128 . in @xcite it is proven that the function @xmath193 defined for @xmath194 as the entropy solution to the initial - boundary value problem @xmath195 is an entropy solution of in the sense of definition [ def : entropy_solution ] , with @xmath196 . moreover , as already pointed out in the introduction , the entropy solution to the above initial - boundary problem coincides with the entropy solution to the cauchy problem @xmath197 where @xmath198 is obtained from @xmath17 by extending it to zero outside @xmath13 . now , adopting the atomization procedure described in subsection [ sec : particle ] , it is clear that the discrete turning point will be located at zero for all times due to the symmetry of the particle system . therefore , the particles will split into two sets which will not change in time , with the two particles nearest the turning point getting further and further away from each other . hence , by the results in @xcite , we obtain that the ftl - hughes model converges to the entropy solution of the above cauchy problem as @xmath65 goes to zero and @xmath199 to infinity , and this concludes the proof . observe that the function @xmath200 defined in is strictly decreasing , indeed @xmath201 ' = - c''\left(\rho\right ) \rho < 0 & \text{for all } \rho \in \left(0,\rho_{\max}\right],\end{aligned}\ ] ] because @xmath57 is assumed to be convex , and @xmath202 . hence @xmath203 , where @xmath204 and @xmath205 is defined in . [ prop:1 ] let @xmath205 be as defined in and @xmath206 be as defined in . if @xmath207 < v(\rho_{\max}),\ ] ] then no particles change direction and the problem admits a global - in - time solution . directly from lemma [ lem : dotxi ] we deduce that @xmath208\,\ ] ] with @xmath209\doteq \frac{v_{\max}}{2 } \left[\vphantom{\int } l \,{{\mathrm{tv}}}(\bar{\rho } ) + 3 \ , c\right].\ ] ] indeed , it is easy to obtain from the estimate proved in lemma [ lem : dotxi ] that @xmath210 \\ & \le v_{\max } \left[\vphantom{\sum } { \mathrm{tv}}_c(r)+3 \ , c\right]\end{aligned}\ ] ] and @xmath211 \\ & \ge -v_{\max } \left[\vphantom{\sum } { \mathrm{tv}}_c(r)+3 \ , c\right].\end{aligned}\ ] ] moreover , by the discrete maximum principle proven in ( * ? ? ? * lemma 1 ) , as long as no collisions occur between a particle and the turning point , we have that @xmath212 in conclusion we proved that the particles do not belong to the cone where @xmath213 evolves . as a consequence , no particle crosses the turning curve , namely no particle changes direction . hence problem can be split in two follow - the - leader problems , for which global existence follows from ( * ? ? ? * lemma 1 ) . moreover , the @xmath138 contraction estimate proven in ( * ? ? ? * proposition 5 ) implies that the @xmath214 , and this concludes the proof . we remark that in the special case @xmath91 and @xmath215 , condition is satisfied only if @xmath216 . this condition is analogous to the one obtained in @xcite . in the @xmath217 limit it is clear that @xmath218 converges to the unique @xmath219 such that @xmath220 a crucial consequence of the result in proposition [ prop:1 ] is that condition is independent of @xmath221 . let @xmath222 $ ] be the constant defined in , and consider the cone @xmath223 \ , first of all we get @xmath224 $ ] , which shows that the curve @xmath213 converges up to a subsequence strongly in @xmath225;{{\mathbb{r}}})$ ] for all @xmath226 to some @xmath227;{{\mathbb{r}}})$ ] , and the limit turning curve @xmath228 is entirely contained in @xmath229 . since no particles are placed in @xmath229 , the discrete density @xmath137 converges to zero strongly in @xmath230 . by taking @xmath60 arbitrarily small , we can apply the same procedure of ( * ? ? ? * theorem 3 ) to prove that the discretized density @xmath137 defined in converges strongly in @xmath231 towards a function @xmath232 satisfying kruzhkov s entropy condition @xcite for the conservation law @xmath233_x = 0 $ ] on @xmath234 . similarly , @xmath137 converges strongly in @xmath231 towards a function @xmath235 satisfying kruzhkov s entropy condition @xcite for the conservation law @xmath236_x = 0 $ ] on @xmath237)$ ] . by setting @xmath238 we can integrate in time on @xmath239 , multiply by an arbitrary test function @xmath240 , and get in the @xmath217 limit ( up to a subsequence ) @xmath241\varphi(t ) \ , { { \mathrm{d } } } t = 0\,,\ ] ] which shows that is satisfied for almost every @xmath49 . in order to prove that @xmath14 is the unique entropy solution to according to definition [ def : entropy_solution ] , we only need to prove that the initial condition @xmath133 is achieved as @xmath242 at least in a weak measure sense . now , it is immediately seen that @xmath243 only differs from the initial atomization @xmath244 on the interval @xmath245 . moreover , recalling the definition of @xmath246 in and of @xmath247 in ( [ eq : density ] ) , we have @xmath248 and the last term above clearly tends to zero as @xmath217 . since @xmath244 converges to @xmath133 in the sense of measures as @xmath217 , we have that also @xmath243 approaches @xmath133 as @xmath217 . this concludes the proof of theorem [ teo:2 ] . in this subsection we consider the particle approximation of an initial condition @xmath249 , \\ \rho_+ & \text{if } x\in ( 0,1 ] , \end{cases}\ ] ] with @xmath250 . in order to simplify the computations , we first fix the number of particles @xmath251 to be equal to @xmath252 , and the number of particles on @xmath253 $ ] equal to @xmath254 , with @xmath255 . we call @xmath256 . we then fix the particle mass @xmath67 such that @xmath257 , and obtain the two riemann values @xmath258 we set the initial @xmath69 particle positions as follows @xmath259 the initial position of the discrete turning point is computed as @xmath260 in order to avoid the situation in which @xmath72 coincides with the initial position of a particle , we remove that particle from the system in such a case . lemma [ lem : dotxi ] implies @xmath261+v(\rho_-)\left[\upsilon(\rho_-)-\upsilon(\rho_+)\right]\right\}\\ & = \frac{1}{2}\left\{\left[v_{\max}-v(\rho_-)\right]\left[\upsilon(\rho_+)-\upsilon(\rho_-)\right ] + v_{\max}\left[c(\rho_-)-c(\rho_+)\right]\right\}.\end{aligned}\ ] ] the assumptions on @xmath57 and @xmath262 clearly imply that @xmath263 . therefore , the avoidance of a collision between the turning point @xmath40 and a neighbor particle is guaranteed under the assumption @xmath264+v(\rho_-)\left[\upsilon(\rho_-)-\upsilon(\rho_+)\right]+2 \ , v(\rho_+)>0\,.\ ] ] it is immediately seen that condition is satisfied when @xmath265 and @xmath266 . on the other hand , if @xmath267 may not hold . in the special case @xmath268 treated in @xcite , the function @xmath45 in reads @xmath269 let us test condition on @xmath267 . we get @xmath270 and the above is satisfied on @xmath271 $ ] only if @xmath272 we now study the monotonicity of @xmath45 with respect to @xmath273 . we get @xmath274 hence , @xmath275 if and only if @xmath276 with @xmath277 now , a quick inspection shows that @xmath278 for all @xmath279 . moreover , since @xmath280 in this case , we infer the existence of a curve @xmath281 with @xmath282 , @xmath283 , and @xmath284 on @xmath285 , such that holds if and only if @xmath286 . such a condition is automatically satisfied if @xmath287 and @xmath273 are both less than @xmath288 . in this section we present some numerical simulations of the particle method described above , and compare the tests with classical methods available in the literature . more precisely , we solve the particle system using the runge - kutta matlab solver ode23 with initial mesh sizes automatically determined by the total number of particles @xmath199 and the initial density values . we then compare the particle scheme simulations with solutions of obtained via godunov scheme . an important remark has to be stated about the boundary conditions . as described in section [ sec : particle ] , we do not impose any boundary condition on the particle method . the two leading particles @xmath289 and @xmath290 move with maximal velocity towards the opposite directions . for the godunov method the boundary conditions are assigned as follows . we create two extra ghost cells , one just at the left of @xmath50 and one juts at the right of @xmath15 . in all our numerical computation we set the value of @xmath14 in those cells to zero , to mimic ` perfect exits ' . in practice , any value that results in characteristic speeds pointing out of the domain will provide the solution , since the upwind scheme will not make use of such values ( except to check that the characteristic actually point out of the domain ) . this procedure is used for convenience , since it simplifies the implementation of boundary conditions a lot . particular attention is devoted to the turning point evolution in the particle simulations , obtained by discretizing . since no boundary conditions are imposed for the particle method , particles are free to exit the domain following the evolution of the two leaders , whereas only the particles still inside the domain bias the evolution of the turning point . we test the atomization algorithm described in section [ sec : particle ] considering some examples introduced in the literature for various initial data @xmath17 , see @xcite and @xcite . in each of the example reported , the choice for the cost function is @xmath268 and we show time evolution of the discrete density @xmath137 in the domain @xmath291 . in [ fig : test 1 ] and [ fig : test 2 ] we consider two constant initial conditions @xmath292 and @xmath293 respectively . in the latter case , as we expect , the simulation shows the formation of two rarefaction waves centered at @xmath100 and @xmath101 , with trace @xmath294 . in the former case the rarefaction waves exit the boundary , hence we see three constant states inside @xmath295 . with initial data @xmath292 at times @xmath296 , @xmath297 and @xmath298 . in the particle simulations the blu dots represent particles positions , whereas the red line is the discretized density . the magenta vertical line describes the turning point evolution . [ fig : test 1],width=226 ] with initial data @xmath292 at times @xmath296 , @xmath297 and @xmath298 . in the particle simulations the blu dots represent particles positions , whereas the red line is the discretized density . the magenta vertical line describes the turning point evolution . [ fig : test 1],width=226 ] with initial data @xmath292 at times @xmath296 , @xmath297 and @xmath298 . in the particle simulations the blu dots represent particles positions , whereas the red line is the discretized density . the magenta vertical line describes the turning point evolution . [ fig : test 1],width=226 ] with initial data @xmath293 at times @xmath296 , @xmath297 and @xmath298 . [ fig : test 2],width=226 ] with initial data @xmath293 at times @xmath296 , @xmath297 and @xmath298 . [ fig : test 2],width=226 ] with initial data @xmath293 at times @xmath296 , @xmath297 and @xmath298 . [ fig : test 2],width=226 ] in [ fig : test 3 ] we consider the riemann problem with initial condition @xmath299 for which the condition in section [ sec : riemann ] is satisfied . in this case , a vacuum region is formed around the turning point . in [ fig : test 4 ] we consider the initial datum @xmath300 for which condition is _ not _ satisfied . in this case , the turning point collides with its left neighbour particle . in order to compare our method with the tests performed in @xcite , in [ fig : test 5 ] we consider the three - step initial condition @xmath301 as shown in [ fig : test 4 ] and [ fig : masstr ] , examples and , exhibit the typical _ mass transfer _ phenomenon occurring when the turning point is not surrounded by a vacuum region . in such a case , particles are crossing the turning point @xmath40 , and a _ non - classical shock _ is formed around @xmath40 , see ( * ? ? ? * remark 5 ) . these examples show that the particle method introduced in section [ sec : particle ] has good chances to rigorously approximate the continuum solution to hughes model under less restrictive assumptions than the ones considered in theorems [ teo:1 ] and [ teo:2 ] . indeed , a key requirement in order to have the particle scheme in section [ sec : particle ] well posed and consistent in the limit is the _ avoidance of wild oscillations of the particle trajectories around the turning point_. such a behaviour is ensured in the examples and , and we believe that it can be obtained in a less restrictive setting for the initial conditions than the one we imposed in theorems [ teo:1 ] and [ teo:2 ] . in support of this conjecture we performed several simulations with various initial conditions , and there is no evidence of such strange behavior . of course a more detailed analysis is needed to rigorously prove convergence of the scheme . in all the examples we set @xmath302 particles and plot the particle positions and the discrete densities . we highlight the interaction with the boundary and the evolution of the turning point . with initial data @xmath303 given in . [ fig : test 3],width=226 ] with initial data @xmath303 given in . [ fig : test 3],width=226 ] + with initial data @xmath303 given in . [ fig : test 3],width=226 ] with initial data @xmath303 given in . [ fig : test 4],width=226 ] with initial data @xmath303 given in . [ fig : test 4],width=226 ] + with initial data @xmath303 given in . [ fig : test 4],width=226 ] with initial data @xmath303 given in . [ fig : test 5],width=226 ] with initial data @xmath303 given in . [ fig : test 5],width=226 ] with initial data @xmath303 given in . [ fig : test 5],width=226 ] given in . [ fig : masstr],width=226 ] given in . [ fig : masstr],width=226 ] in the literature , the model is usually solved in an iterative manner at each time step : 1 . given @xmath14 , solve the eikonal equation . 2 . given the solution to the eikonal equation , solve the non - linear scalar conservation law using the godunov , lax - friedrichs or eno schemes ( see @xcite and @xcite ) . a comparison betweeen the first - order particle method and a classical godunov scheme is showed in [ fig : comp ] , where we plot the solution of the previous simulations ( at time @xmath298 ) both with the ftl - hughes particle method and with the godunov scheme . we set the spatial discretization according to the number of particles @xmath302 and the time discretization with @xmath304 . note that the time step is the same for both methods and is selected so as to respect the cfl condition for the godunov method . empirically , the observed time step restriction for the ftl - hughes method is much less severe than for the godunov method applied to the eulerian descriprion of the flow . however a detailed stability analysis of the method is beyond the scope of the present paper , and will be subject of future investigation . it is evident from [ fig : comp2 ] that the two methods , though conceptually different , produce solutions which differ by a very small error . a more refined analysis of the error will be produced in a future work . , width=226 ] , width=226 ] + , width=226 ] , width=226 ] + , width=226 ] } + 0.7\times \,\mathbf{1}_{(0,1]}$ ] and we set @xmath305 with @xmath306 time iterations . [ fig : comp2],width=302 ] the ftl - hughes particle scheme introduced in section [ sec : particle ] has been proven to rigorously converge to weak entropy solutions to hughes model in @xmath307 with zero dirichlet conditions in cases for which a vacuum region is generated around the turning curve @xmath90 for all times , i.e. _ no mass transfer occurs _ through the turning point @xmath40 . at the particle level , this means that particles initially split into two sets moving towards distinct directions , with _ no change of direction _ for all times . the numerical simulations in section [ sec : numerics ] supported this result , and show evidence that the proposed particle method is consistent also in cases in which particles do cross the turning point thus switching direction . the observation of such a phenomenon motivates further studies on the rigorous analytical study of the particle method , in particular to show ( for the first time ) an existence result for with _ large initial data_. even the global - in - time well - posedness of the particle scheme is difficult in such cases . indeed , even in some simple symmetric cases the continuation of the particle scheme after collision with the turning point can not be prescribed by actually crossing the turning point itself . the authors will continue the study of this problem in a future work . mdf is grateful to peter a. markowich ( to whom this paper is dedicated ) for having introduced him to the hughes model for pedestrian movements in 2009 . mdf and mdr are supported by the gnampa ( italian group of analysis , probability , and applications ) project _ geometric and qualitative properties of solutions to elliptic and parabolic equations_. sf is supported by the gnampa ( italian group of analysis , probability , and applications ) project _ analisi e stabilit per modelli di equazioni alle derivate parziali nella matematica applicata_. gr was partially supported by itn - etn marie curie actions modcompshock - ` modelling and computation of shocks and interfaces ' . mdr acknowledges hospitality by the gran sasso science institute in laquila in a period in which this manuscript was being written . d. amadori and m. di francesco , the one - dimensional hughes model for pedestrian flow : riemann - type solutions , _ acta math . _ , * 32 * ( 2012 ) , 259280 , http://dx.doi.org/10.1016/s0252-9602(12)60016-2 . b. andreianov , c. donadello , u. razafison and m. d. rosini , riemann problems with non local point constraints and capacity drop , _ mathematical biosciences and engineering _ , * 12 * ( 2015 ) , 259278 , http://aimsciences.org/journals/displayarticlesnew.jsp?paperid=10696 . b. andreianov , c. donadello and m. d. rosini , crowd dynamics and conservation laws with nonlocal constraints and capacity drop , _ math . models methods appl . _ , * 24 * ( 2014 ) , 26852722 , http://dx.doi.org/10.1142/s0218202514500341 . , donadello , carlotta , razafison , ulrich and rosini , massimiliano d. , qualitative behaviour and numerical approximation of solutions to conservation laws with non - local point constraints on the flux and modeling of crowd dynamics at the bottlenecks , _ esaim : m2an _ , http://dx.doi.org/10.1051/m2an/2015078 . c. bardos , a. y. le roux and j .- c . ndlec , first order quasilinear equations with boundary conditions , _ comm . partial differential equations _ , * 4 * ( 1979 ) , 10171034 , http://dx.doi.org/10.1080/03605307908820117 . m. burger , m. di francesco , p. a. markowich and m .- wolfram , mean field games with nonlinear mobilities in pedestrian dynamics , _ discrete contin . b _ , * 19 * ( 2014 ) , 13111333 , http://dx.doi.org/10.3934/dcdsb.2014.19.1311 . j. a. carrillo , s. martin and m .- t . wolfram , a local version of the hughes model for pedestrian flow , to appear in mathematical models and methods in the applied sciences , 2015 , url : http://arxiv.org/abs/1501.07054 . r. m. colombo , p. goatin and m. d. rosini , a macroscopic model for pedestrian flows in panic situations , in _ current advances in nonlinear analysis and related topics _ , vol . 32 of gakuto internat . appl . , gakktosho , tokyo , 2010 , 255272 . m. di francesco and m. d. rosini , rigorous derivation of nonlinear scalar conservation laws from follow - the - leader type models via many particle limit , _ arch . _ , * 217 * ( 2015 ) , 831871 . m. di francesco , p. a. markowich , j .- f . pietschmann and m .- t . wolfram , on the hughes model for pedestrian flow : the one - dimensional case , _ j. differential equations _ , * 250 * ( 2011 ) , 13341362 , http://dx.doi.org/10.1016/j.jde.2010.10.015 . r. l. hughes , a continuum theory for the flow of pedestrians , _ transportation research part b : methodological _ , * 36 * ( 2002 ) , 507 535 , http://www.sciencedirect.com/science/article/pii/s0191261501000157 . r. l. hughes , the flow of human crowds , in _ annual review of fluid mechanics , vol . 35 _ , vol . 35 of annu . fluid mech . , annual reviews , palo alto , ca , 2003 , 169182 , http://dx.doi.org/10.1146/annurev.fluid.35.101101.161136 . k. h. karlsen , k .- a . lie and n. h. risebro , a front tracking method for conservation laws with boundary conditions , in _ hyperbolic problems : theory , numerics , applications , vol . i ( zrich , 1998 ) _ , vol . 129 of internat . numer . math . , birkhuser , basel , 1999 , 493502 . m. d. rosini , _ macroscopic models for vehicular flows and crowd dynamics : theory and applications _ , understanding complex systems , springer , heidelberg , 2013 , classical and non - classical advanced mathematics for real life applications .

in the past perfect surfaces have been shown to yield a local critical behaviour that differs from the bulk critical behaviour . on the other hand surface defects , whether they are of natural origin or created artificially , are known to modify local quantities . it is therefore important to clarify whether these defects are relevant or irrelevant for the surface critical behaviour . the purpose of this review is two - fold . in the first part we summarise some of the important results on surface criticality at perfect surfaces . special attention is thereby paid to new developments as for example the study of surface critical behaviour in systems with competing interactions or of surface critical dynamics . in the second part the effect of surface defects ( presence of edges , steps , quenched randomness , lines of adatoms , regular geometric patterns ) on local critical behaviour in semi - infinite systems and in thin films is discussed in detail . whereas most of the defects commonly encountered are shown to be irrelevant , some notable exceptions are highlighted . it is shown furthermore that under certain circumstances non - universal local critical behaviour may be observed at surfaces .

cond-mat0402574

our current understanding of critical phenomena results from an intensive interplay between experimental studies of a large variety of physical systems , ground - breaking theoretical developments ( including renormalization group methods , finite - size scaling theory and conformal invariance ) and extensive numerical investigations of model systems . in many cases the systems under investigation are treated as bulk systems , thus neglecting the existence of surfaces which are unavoidable in real physical systems . discarding surfaces in systems with short - range interactions is justifiable when studying bulk critical properties , as the contribution of the surface to extensive quantities is vanishing in the thermodynamic limit . however , a surface breaks the translation symmetry of a system and changes local quantities . thirty years ago , it has been realized that this leads to surface critical behaviour which differs from the bulk critical behaviour . since that time numerous theoretical and experimental studies have been undertaken in order to determine local critical quantities in systems with perfect surfaces . real surfaces , however , are usually not perfectly smooth but display some degree of roughness due to the presence of surface defects as for example steps , islands , or vacancies . impurities are also often encountered at crystalline surfaces and may be viewed as the source of some disorder at the surface . furthermore , experimentalists nowadays create thin films which do not appear in nature , by growing artificial structures on the film surface . all these defects have some impact on magnetic surface quantities . the present work reviews the recent progress achieved in the study of critical phenomena in systems with boundaries . besides semi - infinite systems and films with perfect surfaces , more complex geometries with wedges and corners as well as more realistic surfaces with defects are discussed . the review hereby focuses on the question whether the different types of geometries and/or of surface defects have an impact on the surface critical behaviour . it is therefore complementary to earlier reviews on surface criticality @xcite that exclusively considered flat , perfect surfaces . the thermodynamics of a surface is completely described by the surface free energy per area @xmath0 . singularities occurring in @xmath0 determine the phase diagram of semi - infinite systems . at some phase boundaries of this phase diagram surface and bulk free energies both exhibit singularities , whereas at other boundaries only @xmath0 becomes singular . an example for the former case is the * ordinary transition * where bulk and surface ordering occur at the same temperature , whereas the latter case is encountered at the so - called * surface transition * where the surface layer alone orders , while the bulk remains disordered . this surface transition is encountered at temperatures higher than the bulk critical temperature , the critical fluctuations of the @xmath1-dimensional semi - infinite system then being essentially ( @xmath2)-dimensional , corresponding to a phase transition in @xmath2 dimensions . in section 2 the critical behaviour at perfect surfaces is discussed . at the bulk critical point different surface universality classes are obtained for every bulk universality class . these universality classes are discussed and their differences emphasized . furthermore , recent progress in our understanding of surface critical behaviour in systems with competing interactions is reviewed . surface critical dynamics and the effect of symmetry breaking surface fields are also briefly discussed . section 3 is devoted to more complex geometries with a wedge . wedge - shaped models are very interesting systems where local critical exponents , which change continuously with the wedge opening angle , arise because of the particular geometrical properties of the wedge . however , at a given opening angle , surface critical behaviour at the ordinary transition is still universal and does not depend on microscopic details of the model as for example the lattice type or the strengths of the local interactions . this is completely different at the surface transition , where under certain circumstances non - universal local critical behaviour can be observed . at a fixed opening angle , edge critical exponents then not only depend continuously on the values of the local couplings but also reflect the existence of the disordered bulk . this intriguing behaviour results from the fact that at the surface transition the edge acts like a defect line in a two - dimensional critical system . corner critical behaviour is discussed in this section , too . in the last years a great deal of activity focused on the critical behaviour of wedges in presence of external surface fields . this rapidly developing field is also briefly summarized . section 4 deals with the important issue of thin films and semi - infinite systems with non - perfect surfaces . surfaces are very often naturally rough , due to the growth mechanism or because of erosion effects . adatom islands , vacancy islands or steps are typical defects encountered at real surfaces . on the other hand , specific surface structures , as for example lines of adatoms or regular geometrical patterns , can be created on purpose by using advanced experimental methods . as these surface defects have an impact on local surface quantities , one has to ask the question whether they do change the surface critical behaviour . for semi - infinite systems we must again distinguish between the ordinary transition and the surface transition . at the ordinary transition , common surface defects ( presence of a step , surfaces with uncorrelated roughness , amorphous surface ) are usually irrelevant for the surface critical behaviour , but there are some notable exceptions . at the surface transition , defect structures like steps and additional lines of atoms located at the surface may yield non - universal local critical exponents , as observed in semi - infinite ising models with additional surface structures . interestingly , additional lines located at the surface of thin ising films always lead to a non - universal local critical behaviour . section 5 finally contains concluding remarks . the aim of this section is to review the main results on surface criticality in systems with flat , perfect surfaces . in sections 2.1 and 2.2 we pay special attention to the phase diagrams of semi - infinite systems as well as to the sets of critical exponents characterizing the different surface phase transitions . as the amount of papers published on this topic is rather impressive , only selected results are presented . the interested reader is referred to the reviews of binder @xcite , abraham @xcite , and diehl @xcite for further readings . the main experimental techniques available for the investigation of surface critical behaviour are reviewed in the book of dosch @xcite . section 2.3 is devoted to critical phenomena in thin films , whereas in section 2.4 the recent studies of surface criticality in systems with competing interactions are discussed . brief accounts of works on surface critical dynamics and on surface critical behaviour in presence of external fields close this section . when comparing systems with and without surfaces , one remarks that extensive quantities are altered by the presence of surfaces thus , the free energy is given by @xcite @xmath3 with the bulk free energy per volume @xmath4 and the surface free energy per surface area @xmath0 . @xmath5 and @xmath6 are the volume and the surface of the system , respectively . further terms in equation ( [ gl:3_2 ] ) may result from the presence of edges , as discussed in section 3 . similar corrections also show up in other extensive quantities . the surface free energy @xmath0 completely describes the thermodynamics of the surface . similar to @xmath4 , singularities occur in @xmath0 . the values of the variables of @xmath0 , where these singularities appear , determine the boundaries separating the different phases of the surface phase diagram . as we shall see later in the course of this section , at some phase boundaries both @xmath4 and @xmath0 exhibit singularities whereas at other ones only @xmath0 becomes singular . local quantities as the local magnetisation density ( often called local magnetisation for brevity ) or the local energy density are also modified by the presence of a surface . examples of order parameter profiles obtained by monte carlo simulations of three - dimensional ising films with @xmath7 layers at temperatures below the bulk critical temperature @xmath8 are shown in figure [ fig_surf_1 ] @xcite . here , all the couplings involved have been chosen to have equal strength @xmath9 and no magnetic fields have been retained . the surface magnetisation @xmath10 , related to @xmath0 by the equation @xmath11 is smaller than the bulk magnetisation due to the reduced coordination number at the surface . the field @xmath12 acts exclusively on the surface layer . the magnetisation increases from its surface value to the bulk value at distances exceeding the bulk correlation length @xmath13 . in case the thickness of the film is not large compared to @xmath13 ( as for example at the temperature @xmath14 in figure [ fig_surf_1 ] ) the bulk magnetisation is never reached . for surface couplings exceeding the bulk couplings by a large amount , magnetisation profiles decreasing monotonically towards the center of the system may also be observed . for ising models in film geometry we have the general hamiltonian @xmath15 where the spins @xmath16 can take on the values @xmath17 . the first sum runs over bonds connecting neighbouring spins where at least one of the spins is a bulk spin , whereas the second sum runs over all surface links . bulk couplings , @xmath9 , and surface couplings , @xmath18 , are ferromagnetic . the bulk field @xmath19 acts on all the spins , the surface field @xmath12 only on spins located at the surface , i.e. a surface spin sees the field @xmath20 . sometimes a different coupling constant is used for the bonds connecting surface and bulk spins @xcite . in general , the perturbations due to the presence of a surface are supposed to be of short range . analytical results on surface critical behaviour are commonly obtained in the framework of continuum field theory , see @xcite for comprehensive reviews . the standard @xmath21 ginzburg - landau model , appropriate for the universality class of the ising model , is thereby augmented in order to include contributions coming from the surface : @xmath22 with @xmath23 , whereas @xmath24 is related to the surface enhancement of the spin - spin coupling constant in the corresponding lattice model . the resulting ginzburg - landau free energy density is readily generalized to cases where the order parameter exhibits a different symmetry . layer - dependent quantities are very useful when investigating systems with surfaces . examples are the magnetisation per layer @xmath25 and the susceptibility per layer @xmath26 where @xmath27 labels the layers parallel to the surface . the surface magnetisation is @xmath28 and the local susceptibility at the surface , i.e. the response of the surface magnetisation to a surface field , @xmath29 , is given by @xmath30 . from the profiles @xmath25 and @xmath26 one also obtains the surface excess quantities @xmath31 and @xmath32 where @xmath33 and @xmath34 are the bulk magnetisation and the bulk susceptibility . of further importance is the surface layer susceptibility , i.e.the response of the local surface magnetisation to a bulk field , usually denoted by @xmath35 . the critical surface pair correlation function behaves as @xmath36 where @xmath37 is the scaling dimension of the surface order parameter . for a @xmath1-dimensional model @xmath38 is a @xmath39-dimensional vector parallel to the surface . this behaviour of the surface correlation function differs from that of the bulk correlation function : @xmath40 where the value of the bulk scaling dimension @xmath41 is usually smaller than that of @xmath37 . here @xmath42 is a @xmath1-dimensional vector . critical correlations between a surface point and a bulk point exhibit a power law behaviour with the exponent @xmath43 . the surface critical exponents associated with the different critical quantities are listed in table [ table_surf_1 ] . .[table_surf_1 ] surface critical exponents and their definitions . [ cols="^,^",options="header " , ] to explain these findings , note that the corners are intersection points of edges and recall that at the surface transition the critical fluctuations are essentially two - dimensional . the critical exponent of the corner magnetisation , @xmath44 , has been related to that of the magnetisation at the intersection of three semi - infinite defect lines in the two - dimensional ising model . the critical exponent @xmath45 for star defects formed by three intersecting ladder defects has been calculated by henkel _ one may assign to the edges an effective ladder coupling @xmath46 by setting @xmath47 . inserting this effective coupling into the equation derived for a star defect formed by three ladder defects @xcite then gives @xmath45 . the comparison of @xmath44 with @xmath45 yields a satisfactory agreement , see table [ table_edge_3 ] . of course , a more refined analysis would focus on the fact that the correct geometry of the present problem is not that of a plane but that of a cone with three defect lines meeting at an angle @xmath48 @xcite ( see @xcite for a recent discussion of ising models with conical singularities ) . furthermore , the rather complicated nature of the edge as a simultaneously ladder- and chain - type defect line as well as the effect of the bulk spins on the corner magnetisation should then also be taken into account . the critical behaviour near edges has also been studied recently at the normal transition @xcite . the authors investigated critical adsorption near edges due to the presence of symmetry - breaking surface fields in a wedge . besides studying the problem in mean - field approximation , they also presented some exact results for the two - dimensional case . especially , the critical edge exponent @xmath49 was discussed . it was shown that at the normal transition local critical exponents also depend continuously on the opening angle . critical adsorption near edges is , however , only one of the many intriguing phenomena encountered in systems with both wedges and surface fields . in the case of adsorption of a fluid on a solid substrate with wedge geometry @xcite it is expected from thermodynamic arguments that the liquid fills the wedge completely at temperatures @xmath50 , where the filling temperature @xmath51 is lower than the wetting temperature @xmath52 of a planar , but otherwise identical , wall @xcite . in a series of recent papers parry _ et al . _ @xcite studied the corner filling transition in detail , focusing especially on fluctuation effects and on the universality classes of filling transitions . one of their prediction was that critical filling , i.e. the filling transition for @xmath53 , could be continuous even in cases where the related wetting transition is of first order . they also studied the divergence of various length scales associated with this phase transition and predicted , basing themselves on a fluctuation theory , that the interfacial height @xmath54 from the bottom of the wedge should diverge as @xmath55 @xcite . the wedge filling transition can be studied in ising models with wedge geometries by applying surface fields which favor one of the two phases in the wedge @xcite . exact results in the two - dimensional case @xcite permit to establish the existence of this kind of transition . recent numerical investigations @xcite revealed a very good agreement with the theoretical predictions of parry _ et al_. in addition , a new type of interface localisation - delocalisation transition was revealed in the three - dimensional double wedge forming a pore with a square cross section @xcite . the critical exponents of this transition can be related to the critical exponents of the filling transition in a simple wedge . the influence of wedges on critical behaviour has not only been studied in equilibrium systems but also in non - equilibrium systems . @xcite introduced an edge into a directed percolation process and analysed the impact this edge has on the non - equilibrium phase transition observed in this system . this work generalized earlier studies of directed percolation in a semi - infinite geometry @xcite to the wedge geometry . angle dependent edge critical exponents were observed in this non - equilibrium case , too . directed percolation processes have also been studied in two - dimensional parabolic - like systems with a free surface at @xmath56 @xcite . for @xmath57 , @xmath27 being the dynamical exponent , the surface shape is a relevant perturbation and the tip order parameter displays a stretched exponential behaviour . in the marginal case , @xmath58 , a non - universal local critical behaviour is again observed . other non - equilibrium absorbing phase transitions have also been studied in semi - infinite systems recently @xcite , but these studies have not been extended to wedge - shaped geometries . in the preceding sections we discussed critical phenomena in systems with various geometries : semi - infinite systems , wedge - shaped systems , and cubes . all the results presented so far have in common that only idealized , perfect surfaces were considered . however , real surfaces are often naturally rough , as steps or islands occur during growth processes or result from the effect of erosion . furthermore , methods from nanoscience permit to create artificial structures on top of films . examples include lines of adatoms or regular arrangements of geometric structures . all these defects have some impact on magnetic surface quantities . in the following i review theoretical studies where surface critical phenomena in systems with various surface defects have been investigated . section 4.1 deals with semi - infinite ( i.e. bulk terminated ) systems , whereas section 4.2 is devoted to the influence of surface defects on the critical behaviour of ( ultra)thin films . geometric surface imperfections ( e.g. , islands or vacancies ) and impurities may be stable on the time scale of magnetic phenomena and thus lead to quenched surface disorder @xcite . some experiments indicate an enhancement of disorder near surfaces , thus pointing to the possible realisation of quenched surface disorder in systems where bulk disorder is negligible . in a theoretical description , this kind of surface disorder is usually mimicked by random surface fields or by random surface couplings . early studies mainly focused on the global phase diagram observed in systems with random surface couplings or random surface fields . in @xcite dilute semi - infinite ising models with bond and site dilution both in the bulk and at the surface were considered . mean - field results obtained for semi - infinite ising models with random surface and bulk fields were presented in @xcite . the possible types of phase diagrams were established and the phase transitions as well as multicritical points were discussed . phase diagrams obtained for semi - infinite transverse ising models with random surface and bulk fields were analysed recently in @xcite . in a series of papers , kaneyoshi ( see , e.g. , @xcite ) studied semi - infinite ising systems with an amorphous surface layer in detail . the amorphous surface can be mimicked by choosing randomly weak or strong nearest - neighbour ferromagnetic couplings between surface spins . sometimes , he also considered random couplings between surface spins and the underlying bulk spins . using different analytical approaches , surface magnetic properties as , for example , the magnetisation of the surface layer , were analysed . the corresponding phase diagrams were derived from the results of an effective field theory which includes correlations . one interesting result obtained in these studies concerns the location of the special transition point . kaneyoshi observed that the critical coupling ratio @xmath59 is shifted to higher values in systems with a diluted surface and he conjectured that this is a common effect of surface amorphisation . this was confirmed by a monte carlo study @xcite where the strength of surface bonds was chosen randomly between two different values @xmath60 and @xmath61 , the ratio @xmath62 measuring the degree of dilution . for @xmath63 the special transition point is located at the mean coupling ratio @xmath64 for the simple cubic lattice , noticeably larger than the value obtained for the perfect surface @xmath65 , see section 2.1 . in order to understand this shift we recall that in the limit @xmath66 the surface effectively decouples from the underlying bulk and can be regarded as a two - dimensional system . however , it is well known that at a given mean coupling @xmath67 the critical temperature of the two - dimensional ising model is reduced by randomness @xcite . therefore , for fixed @xmath68 , increasing randomness shifts the line of the surface transition to lower temperatures which in turn shifts the location of the special transition point to larger values of the coupling ratio . sometimes , alloy surfaces are mimicked by semi - infinite surfaces with randomly decorated magnetic and nonmagnetic atoms located on the surface @xcite . again , the location of the special transition point is observed to shift as a function of the model parameters . quantities having an impact on this shift are for example the concentration of the nonmagnetic surface atoms or the spin of the magnetic atoms . the investigations reviewed so far almost exclusively focused on global phase diagrams of semi - infinite systems with non - perfect surfaces . the different authors did not try to study in detail the influence of these imperfections on the surface critical behaviour . however , as a certain degree of imperfections is unavoidable when studying surfaces experimentally , it is very important to clarify whether the surface critical exponents are robust against surface imperfections . a first step in this direction was taken by mon and nightingale in their study of the influence of a random surface field on the surface critical behaviour of the semi - infinite ising model @xcite . this work was motivated by the determination of the surface order parameter critical exponent @xmath69 in a study of wetting phenomena of binary mixtures consisting of a polar and a nonpolar liquid @xcite . the value obtained for @xmath69 in that study was compatible with the value encountered in the semi - infinite ising model with a perfect surface , even so @xmath69 described the vanishing of the surface order parameter in the presence of a random surface field with zero average strength . using a harris type criterion @xcite , mon and nightingale conjectured that , at the ordinary transition , a random surface field is irrelevant for the surface critical behaviour of the ising model . they verified their prediction with the aid of monte carlo simulations . using renormalization group techniques it was shown in @xcite that surface bond - dilution is irrelevant for the semi - infinite ising model , the universality classes of the different transitions being those of the pure system . in an extensive study , diehl and nsser @xcite derived harris type criteria for various types of quenched surface disorder with the aim to assess the relevance or irrelevance of these random imperfections on the pure system surface critical behaviour . in his original work , harris @xcite studied the stability of critical bulk systems in the presence of randomness . the well - known harris criterion states that , in the weak - disorder limit , bond disorder is irrelevant for the bulk critical behaviour provided the specific heat critical exponent @xmath70 is negative . disorder is relevant for @xmath71 , whereas the case @xmath72 is marginal . looking at the ising model , one concludes , based on the harris criterion , that in three dimensions disorder is relevant as @xmath73 , whereas in two dimensions one encounters the marginal case @xmath72 which has attracted much interest @xcite . one should note that the condition for irrelevance of disorder resulting from a harris type criterion is a necessary but not a sufficient condition for the stability of the pure system s critical behaviour . nevertheless , a harris type criterion is usually a reliable indicator for the irrelevance of randomness . the harris criterion generalized in @xcite to surface critical behaviour states that short - range correlated disorder is relevant or irrelevant depending on whether some ( surface ) susceptibility , which depends on the kind of randomness under investigation , diverges or is finite at the critical point . for random surface fields the quantity of interest is the surface susceptibility @xmath74 of the pure system , whereas for random surface couplings the relevant susceptibility is the local specific heat @xmath75 @xcite . as these generalized susceptibilities have a singularity of the form @xmath76 at the critical point , the criterion indicates that the disorder is relevant for positive @xmath77 ( i.e. @xmath78 or @xmath79 ) , but that it is irrelevant for negative @xmath77 . this criterion has been applied by diehl and nsser in various cases involving surfaces . thus , the presence of random surface fields is expected to be irrelevant at the ordinary transition for all @xmath80 models with @xmath81 if the dimension @xmath82 , in accordance with the results obtained by mon and nightingale @xcite for the special case of the semi - infinite ising model ( @xmath83 ) . interestingly , feldman and vinokur recently showed @xcite that weak quenched surface disorder destroys bulk ordering in the case of a system with continuous symmetry ( as the @xmath84 model ) , leading to a power law decay of correlation functions and therefore to the appearance of quasi - long - range order . in the two - dimensional semi - infinite ising model the perturbation caused by the random surface field is marginal so that the criterion does not yield a definite answer . this case was studied subsequently in @xcite where it was shown that the surface critical behaviour of the 2d ising model is described by ising critical exponents with logarithmic corrections to scaling . for the special transition point , the harris like criterion of @xcite indicates that random surface fields are relevant in dimensions @xmath85 . however , the three - dimensional semi - infinite ising model with quenched random surface fields does not present a special transition point nor an extraordinary transition , as the surface transition line , where the surface orders alone , does not exist anymore . this follows from the fact that at finite temperatures random bulk fields destroy long - range order in the two - dimensional bulk ising system @xcite . in the cases where the extraordinary transition still exists in the presence of random surface fields , the criterion predicts that the disorder is then irrelevant . for random surface couplings , the perturbation is found to be irrelevant both at the ordinary and at the extraordinary transitions . more interesting is the situation at the special transition point where early estimates of the critical exponent @xmath79 in three dimensions yielded small negative @xcite _ or _ small positive values @xcite , implying that short - range enhancement disorder is close to being relevant in three dimensions . recently , new field - theoretical estimates also resulted in negative values for @xmath79 @xcite . the monte carlo results obtained for three - dimensional semi - infinite ising models with random surface - bond disorder @xcite are also compatibel with the irrelevance of random surface enhancement for the special transition point critical behaviour . however , as short - range random surface enhancement is close to being relevant , one might expect long - range correlated enhancement disorder to be relevant at the special transition point @xcite . diehl and nsser also studied the impact of surface - enhancement disorder at the special transition of a bulk tricritical system . as in this case the harris criterion did not permit a definite prediction , renormalization group techniques were used in a second paper @xcite to clarify the situation . the robustness of the surface critical exponents at the ordinary transition against two types of surface imperfections was established in a monte carlo study @xcite . the studied imperfections correspond to an amorphous surface , mimicked by random strong and weak couplings in the surface , and to a simple case of corrugation , due to the presence of a step of monoatomic height superimposed on a perfect surface . for the amorphous surface , the monte carlo results showed that the dilution , at a fixed temperature @xmath86 , decreases the magnetisation at and near the surface . the effective critical exponent , derived from the magnetisation of the diluted surface , follows in the ising case closely that of the perfect case , yielding the same asymptotic value @xmath87 . randomness in the surface couplings is therefore irrelevant for the asymptotic behaviour of the surface magnetisation , and even of minor importance for the corrections to scaling @xcite . similarly , the estimate for surface susceptibility critical exponent @xmath88 is compatibel with the one in the perfect case . these numerical findings provide support for the conjecture of @xcite that surface enhancement disorder is irrelevant at the ordinary transition . the robustness of the critical exponent @xmath69 against dilution observed in the numerical study was later proven by diehl @xcite who derived upper and lower bounds on the surface magnetisation of the diluted surface and showed in a rigorous way that @xmath69 takes the same value in the diluted system as in the perfect system . in @xcite the effect of corrugation on surface critical behaviour was studied in ising systems where an extra half layer in the form of a strip - like terrace of monoatomic height was superimposed on a perfect surface , as shown in figure [ fig_dirt_1 ] . for this geometry the magnetisation at the step - edge deviates most significantly from the magnetisation of the perfect surface . the local critical exponent describing the behaviour of the step - edge magnetisation was found to have the value 0.800(15 ) , in agreement with that of the magnetisation of the flat surface . however , in contrast to the case of random couplings , corrections to scaling are here distinctively different from those of the perfect surface . diehl also considered this type of imperfections in @xcite and showed in a rigorous way that the critical exponent of the step - edge magnetisation is identical to that of the magnetisation of a perfect surface . note that at the surface transition local critical magnetic exponents are expected to be non - universal close to the step - edge . indeed , the step - edge should then act like a defect line in a system governed by two - dimensional critical fluctuations . this is supported by monte carlo simulations which yield for @xmath89 the value @xmath90 for the step - edge magnetisation critical exponent @xcite in the ising model . the study of magnetic properties of rough , corrugated surfaces near criticality is a rather new and promising field . the case of one additional terrace on an otherwise perfect surface may be generalized to the more complex situation of vicinal surfaces with terraces separated by equally spaced monoatomic steps @xcite . in nature , rough surfaces often result from a growth process and strongly affect the surface magnetisation . diffusion - limited growth results in a rough growth front following a poisson distribution . this may be realized in simulations by considering a surface formed of columns of random heights @xcite , see also @xcite . layer - by - layer growth , however , may result in films with a finite number of complete layers and one partially filled layer , thus yielding a different type of roughness @xcite . whereas the different kinds of roughnesses discussed so far have been shown to be ( or are supposed to be ) irrelevant for the surface critical behaviour at the ordinary transition , this is not always the case . in a recent interesting work , hanke and kardar @xcite showed that self - affine rough surfaces may give rise to novel surface critical behaviour with surface critical exponents different from those of the perfect case . similar results have previously been obtained in two - dimensional systems with fractal boundaries @xcite yielding new multifractal boundary exponents . these results in two dimensions have been interpreted in terms of a scale - dependent distribution of opening angles of the fractal boundary @xcite . the same analogy has been evoked in the discussion of the surface critical behaviour of self - affine surfaces in @xcite . hanke and kardar also studied critical correlations in the vicinity of a regularly patterned surface ( as might for example result from a lithographic preparation of the surface ) . using a perturbative calculation in the deformations in height from a flat surface they showed that the leading power law decay of the correlations is the same as for a flat surface , but with a modulated amplitude reflecting the shape of the surface @xcite . extended surface defects have been studied intensively in two dimensions . in the hilhorst - van leeuwen model @xcite one considers a semi - infinite square lattice with inhomogenous couplings ( alternatively , the triangular lattice has also been investigated ) . in the direction parallel to the surface one has couplings with a constant strength @xmath91 , whereas the strength of the couplings varies perpendicular to the surface as a function of the distance @xmath92 to the surface : @xmath93 this extended perturbation is irrelevant for @xmath94 , yielding the same critical behaviour as the homogeneous semi - infinite system . for @xmath95 , however , the perturbation is relevant and a spontaneous surface magnetisation is observed at the bulk critical point . the most interesting case is @xmath96 , where the perturbation is marginal . here a spontaneous surface magnetisation is again observed for values of @xmath97 larger than some threshold @xmath98 , whereas for @xmath99 the scaling dimension @xmath37 varies continuously as a function of @xmath97 : @xmath100 . this intriguing behaviour has attracted much interest in the past @xcite and has recently led to the discovery of an up to then unnoticed effect in the short - time non - equilibrium critical dynamics at surfaces @xcite . finally , before turning to thin films with surface imperfections , let us briefly mention some recent work concerning surface critical behaviour in an ising model with quenched random defects in the bulk . in the three - dimensional bulk ising model , quenched random defects are a relevant perturbation leading to modified critical bulk exponents @xcite . the surface critical behaviour of the three - dimensional ising model with quenched bulk disorder has been investigated both at the ordinary @xcite and at the special transition point @xcite , using different renormalization group techniques . the crossover behaviour between these two transitions was studied recently in @xcite . interestingly , modified surface critical behaviour is encountered independently whether quenched surface disorder is present or not . the corresponding marginal case @xmath101 has also been studied in detail recently @xcite , yielding ising surface critical exponents with logarithmic corrections to scaling . boundary critical behaviour of @xmath102-state random potts models hase been studied in @xcite for @xmath103 . it should be noted that the problem of bond percolation has also been investigated in semi - infinite geometries @xcite . in this case distinct surface percolation transitions have been identified and the values of the local critical exponents have been computed . in a series of papers aaro reis studied the dependence of physical quantities in two - dimensional ising stripes @xcite and three - dimensional ising thin films @xcite on the surface roughness . using transfer matrix methods , reis investigated stripes where the column heights were chosen according to a gaussian distribution with mean @xmath104 ( @xmath104 : integer ) and variance @xmath105 @xcite . two different cases were considered : @xmath106 constant and @xmath107 with some constant @xmath108 . whereas the roughness becomes unimportant for large @xmath104 in the former case , in the latter case the roughness increases with the mean thickness . reis paid special attention to the finite - size scaling and to the finite - size corrections due to randomness . in @xcite he also considered the case of noninteger mean @xmath104 and showed that the free energy displays an interesting oscillating behaviour as function of continuously changing @xmath104 due to oscillations in the mean coordination number . recently @xcite , he completed his investigation of stripes of random width by studying roughness of the more general form @xmath109 with @xmath110 . he also considered correlated roughness by imposing the maximal height difference between neighbouring columns to be not larger than one lattice constant . interestingly , these computations showed that the correlations had no systematic effect on the corrections to scaling . more relevant to the understanding of the critical behaviour of real films are reis studies of films with rough surfaces @xcite . using monte carlo techniques , he studied the same kind of uncorrelated roughness as for the stripes : gaussian distribution of thicknesses with integer mean @xmath104 and @xmath106 constant for all @xmath104 or @xmath107 @xcite as well as noninteger mean @xmath104 @xcite . these studies were partly motivated by the observation that some film quantities depend on the growth condition , the different growth mechanism leading to different types of roughness . reis studied the magnetic susceptibility , the specific heat , and the total magnetisation of films with different thicknesses . especially , he showed that the simple equation ( [ gl:5_1 ] ) connecting the critical temperature of finite films with the bulk critical temperature does not hold for noninteger mean thicknesses @xmath104 , as @xmath111 shows a convex behaviour between two consecutive integer values of @xmath104 @xcite . he also studied the critical behaviour of rough films and thereby observed that the considered types of disorder are irrelevant : as for flat films , the critical exponents of the two - dimensional ising model are recovered in rough thin films . up to now only rough films with uncorrelated roughness have been studied . it would be very interesting to investigate the influence of spatial correlated roughness on the critical thin film behaviour as well . the impact of additional regular structures , located on the surface , on the critical behaviour of ising films is studied numerically in @xcite . the additional structures consist of one or two adjacent lines formed by adatoms as well as straight steps of unit height . one may introduce local couplings with different strengths , similar to what has been done in section 3.1 for the edges . for example , the strength of the bonds connecting two neighbouring adatoms in the defect line or the strength of the coupling between the adatoms and the underlying magnetic film may be varied . the main finding of @xcite is that non - universal critical behaviour is encountered in ising films with additional surface defects : the local magnetic critical exponents close to the defects depend continuously on the local couplings as well as on the layer thickness . interestingly , the presence of an additional line is already enough to change the local critical exponent of a one - layer system by a large amount . indeed , in the case where all the couplings have equal strength one obtains the value @xmath112 for the critical exponent of the local order parameter @xcite . it is worthwhile mentioning that related experiments on films with additional adatom lines have been published @xcite , even if there the lines of adatoms were not connected magnetically to the film . the magnetism of lines of @xmath113 adatoms on ag surfaces has been the subject of some theoretical studies @xcite . the magnetic behaviour of thin films with large terraces has also been analysed experimentally @xcite . increasing the film thickness further , the semi - infinite system with surface imperfections is reached at the end . the following two typical scenarios must be distinguished : ( i ) for @xmath114 , bulk and surface order at the same temperature ( ordinary transition ) , whereas ( ii ) for @xmath115 the surface orders at a higher temperature than the bulk ( surface transition ) . @xmath59 is the critical coupling ratio of the multicritical point , see section 2.1 . therefore , a completely different behaviour of local quantities near the surface imperfections is expected in both cases when varying the number of layers @xmath104 . numerical simulations of non - perfect films are in complete agreement with this expectation @xcite . for @xmath114 the surface critical exponent @xmath116 is obtained in the limit @xmath117 everywhere on the surface , independently of any additional surface structure , as discussed already in section 4.1 . in a film the local magnetisation near the defect closely follows the local magnetisation of the corresponding semi - infinite system for low temperatures . at temperatures where the correlation length is comparable with the thickness of the film a crossover from a regime with isotropic three - dimensional fluctuations to a regime dominated by two - dimensional critical fluctuations takes place . increasing the film thickness the crossover temperatures approaches the bulk critical temperature , yielding in the limit of the semi - infinite system , @xmath117 , the value @xmath118 for the critical exponent of the local magnetisation . choosing @xmath115 , non - universal local critical behaviour close to the adchain is expected at the surface transition even in the limit @xmath117 . this situation is indeed comparable to that of an edge at the surface transition , the edge corresponding to an extended defect line as discussed in section 3.2 . the local magnetic critical exponents should therefore depend continuously on the local couplings at the additional line . this has been studied in @xcite by examining the case @xmath119 . the results obtained show that near an additional line @xmath120 is affected both by the values of the local couplings and by the presence of additional bulk - like layers leading to non - universal local critical behaviour . further results have been obtained for systems with a straight step on top of film . at the ordinary transition , a straight step on top of a semi - infinite system does not change the local critical exponents @xcite . in thin films the critical behaviour is however quite different @xcite . introducing a straight step by adding half a layer of magnetic adatoms to the surface of the magnetic film one observes two sharp peaks in the specific heat @xcite , see figure [ fig_dirt_2 ] , pointing to the existence of two different phase transitions . in fact , when considering a system with one half layer on top of a film one is dealing with a composite system displaying in the thermodynamic limit two distinct phase transitions at two different temperatures . consider for simplicity the case of a single layer plus half a layer . one phase transition then takes place at the critical temperature of the 2d ising model , @xmath121 and one at the critical temperature of the double layer , @xmath122 . the value of the critical exponent of the step magnetisation at the higher temperature phase transition is 1/2 , i.e. is identical to the value of the surface critical exponent of the two - dimensional ising model . at the lower critical temperature the single semi - infinite layer then orders in presence of ordered surface spins , the ordered double layer acting at that transition as a surface field . one is therefore dealing with the normal transition . the same scenario is expected to hold for finite films with an additional half layer with @xmath123 , yielding at @xmath124 the critical exponent 1/2 near the step edge , in accordance with the numerical findings of @xcite and @xcite . note that very recently @xcite , two sharp susceptibilities peaks observed experimentally in thin films with large terraces have been interpreted as the signatures of two distinct phase transitions . thin films with an additional terrace covering half of the surface are only one example of composite systems displaying two distinct phase transitions . further examples include two - dimensional ising models with a defect column where the bonds differ on the two sides of the column @xcite , ising quantum chains where couplings and transverse fields differ in the two half chains @xcite , or layered magnetic systems consisting of thin layers of coupled ising spins with @xmath125 and @xmath126 @xcite . in @xcite , wetting phenomena were studied in models consisting of two semi - infinite systems with unequal critical temperatures connected by a defect plane . finally , let us briefly mention that in the context of spin systems with continuous spin symmetry thin films with selected surface defects ( amorphisation of the surface layer @xcite or steps @xcite ) have been considered in some cases , but no systematic study of the critical behaviour of this kind of systems has been done up to now . surface criticality has been the subject of intensive studies in the last thirty years . thereby a large variety of different methods ( analytical , numerical , and experimental ) has been used , yielding a host of interesting results , as reviewed in this work . critical phenomena at perfect surfaces are now in general well understood , at least when dealing with static critical quantities . this is not really the case for dynamic critical behaviour at surfaces for which a coherent picture has not yet emerged . a major problem in this context is the total lack of experimental studies on surface dynamic properties at criticality . the situation is also not very satisfactory from the theoretical point of view , as only selected results on equilibrium and non - equilibrium surface dynamic behaviour at criticality have been published . clearly , this is one of the most important aspects of surface criticality that warrants more attention in the future . wedge - shaped geometries , which can be viewed as generalisations of semi - infinite systems , have also been discussed in detail in this review . it is encouraging that the effect of curved surfaces on the local critical behaviour has been observed in simulations of liquid - vapour transitions near a weakly attractive surface . this may point to possible experimental systems where this kind of problems can be studied . there are indeed a vast number of theoretical predictions for this kind of geometry , and experimental investigations are therefore welcomed . in the last years the focus of research on surface criticality has somehow shifted , as the main emphasis has been on more realistic surfaces . the facts that real surfaces are usually rough , displaying a variety of different surface defects , and that experimental physicists can create artificial structures on top of a surface directly lead to the question whether these quantities have an impact on local critical behaviour . we have presented a comprehensive overview of the field , thereby discussing in detail critical behaviour in semi - infinite systems with surface defects as well as in thin films with additional surface structures . some common surface defects have been shown to be irrelevant for the surface critical behaviour at the ordinary transition . there are however some interesting exceptions , as for example the case of self - affine rough surfaces . the situation is even more complex at the surface transition where in a three - dimensional system the critical fluctuations are of two - dimensional nature . indeed , additional structures as steps or lines of adatoms have been shown in numerical studies to lead to non - universal local critical behaviour where the values of the local critical exponents reflect the strengths of the coupling constants as well as the presence of the disordered bulk . it is worth noting that in thin films this kind of additional surface structures in general leads to non - universal critical behaviour . it is obvious from our overview that a large number of recent studies of the effects of surface defects on the local critical behaviour are either of purely numerical nature or are using rather crude approximations . this is especially the case when dealing with non - perfect surfaces at the surface transition . there is a need for more elaborate analytical approaches , and it is one of the intention of this review to encourage further theoretical ( and also experimental ) investigations of critical phenomena at non - perfect surfaces . it is a pleasure to thank all my collaborators who worked with me on various aspects of surface criticality : f. . bagamry , d. catrein , m .- c . chung , f. igli , m. kaulke , i. peschel , w. selke , and l. turban . i am indebted to w. selke for introducing me to the field of surface critical behaviour and for many years of fruitful collaboration . i also thank h.w . diehl , m. henkel , and a. hller for many inspiring discussions . 999 binder k 1983 _ phase transitions and critical phenomena _ vol 8 ( london / new york : academic press ) p. 1 abraham a b 1986 _ phase transitions and critical phenomena _ vol 10 ( london / new york : academic press ) p. 1 diehl h w 1986 _ phase transitions and critical phenomena _ vol 10 ( london / new york : academic press ) p. 75 diehl h w 1997 _ int . j. mod . _ b * 11 * 3503 dosch h 1992 _ critical phenomena at surfaces and interfaces _ ( berlin , heidelberg , and new york : springer ) fisher m e and caginalp g 1977 _ commun . math . phys . _ * 56 * 11 caginalp g and fisher m e 1979 _ commun . math . _ * 65 * 247 pleimling m and selke w 1998 _ eur . j. _ b * 1 * 385 lubensky t c and rubin m h 1975 b * 12 * 3885 nakanishi h and fisher m e 1982 1565 mills d l 1971 b * 3 * 3887 binder k and hohenberg p c 1972 b * 6 * 3461 binder k and hohenberg p c 1974 b * 9 * 2194 burkhardt t w and eisenriegler e 1977 b * 16 * 3213 lipowsky r and wagner h 1981 _ z. phys . _ b * 42 * 355 binder k and landau d p 1984 318 ruge c , dunkelmann a and wagner f 1992 2465 ruge c , dunkelmann a , wagner f and wulf j 1993 _ j. stat . * 73 * 293 au - yang h 1973 _ j. math . * 14 * 937 mermin n d and wagner h 1966 1133 kosterlitz j m and thouless d j 1973 1181 kosterlitz j m 1974 1046 frhlich j and pfister c e 1986 _ commun . math . phys . _ * 107 * 337 landau d p , panday r and binder k 1989 b * 39 * 12302 peczak o and landau d p 1991 b * 43 * 1048 diehl h w and eisenriegler e 1984 b * 30 * 300 affleck i 2000 6473 bakchich a and el bouziani m 2001 b * 63 * 064408 igli f and carlon e 1999 b * 59 * 3783 benayad n and dakhama a 1997 b * 55 * 12276 karevski d and henkel m 1997 b * 55 * 6429 theumann a 1979 b * 19 * 6295 debell k and essam j w 1980 4811 carton j p 1980 _ j. phys . ( paris ) _ * 41 * , l175 christou a and stinchcombe r b 1986 757 diehl h w and lam p m 1989 _ z. phys . _ b * 74 * 395 lipowsky r 1982 1575 lajzerowicz j 1981 _ ferroelectrics _ * 35 * 219 turban l and igli f 2002 b * 66 * 014440 watson g m , gibbs d , lander g h , gaulin b d , berman l e , matzke h and ellis w 2000 b * 61 * 8966 lipowsky r 1982 _ z. phys . _ b * 45 * 229 dobrovolny c , laanait l and ruiz j 2004 _ j. stat . _ * 114 * 1269 diehl h w and shpot m 1994 3431 diehl h w and shpot m 1998 595 debell k , lookman t and whittington s g 1990 a * 41 * 682 hegger r and grassberger p 1994 4069 landau d p and binder k 1990 b * 41 * 4633 ruge c and wagner f 1995 b * 52 * 4209 nightingale m p and blte h w j 1988 1562 krech m 2000 b * 62 * 6360 berche b 2003 585 sigl l and fenzl w 1986 2191 mailnder l , dosch h , peisl j and johnson r l 1990 2527 burandt b , press w and hausshl s 1993 1188 alvarado s , campagna m and hopster h 1982 51 diehl h w and dietrich s 1981 _ z. phys . _ b * 42 * 65 cardy j 1984 514 res i and straley j p 2000 b * 61 * 14425 deng y and blte h w j 2003 e * 67 * 066116 diehl h w and dietrich s 1981 b * 24 * 2878 rau c and robert m 1987 2714 tang h , weller d , walker t g , scott j c , chappert c , hopster h , pang a w , dessau d s and pappas d p 1993 444 tober e d , palomares f j , ynzunza r x , denecke r , morais j , wang z , bino g , liesegang j , hussain z and fadley c s 1998 2360 shick a b , pickett w e and fadley c s 2000 b * 61 * r9213 rau c 1983 _ j. magn . magn . mater . _ * 31 - 34 * 874 rau c , jin c , and roberts m 1988 _ j. appl . phys . _ * 63 * 3667 mamaev y a , petrov u v and starovoitov s a 1987 _ sov . phys . tech . * 13 * 642 marynowski m , franzen w , el - batanouny m and staemmler v 1999 b * 60 * 6053 el - batanouny m 2002 6281 murphy b m , stettner j , traving m , sprung m , grotkopp i , mller m , oglesby c s , tolan m and press w 2003 _ physica _ b * 336 * 103 polak m , rubinovich l and deng j 1995 4059 arnold c s and pappas d p 2000 5202 bray a j and moore m a 1977 1927 burkhardt t w and diehl h w 1994 b * 50 * 3894 drewitz a , leidl r , burkhardt t w and diehl h w 1997 1090 abraham d b and upton p j 2000 2541 poulopoulos p and baberschke k 1999 9495 ou j t , wang f and lin d l 1997 b * 56 * 2805 henkel m , andrieu s , bauer p and piecuch m 1998 4783 saber a , ainane a , dujardin f , el aouad n , saber m and stb b 2000 43 zhang r and wills r f 2001 2665 cabral neta j , ricardo de sousa j and plascak j a 2002 b * 66 * 064417 barber m n 1983 _ phase transitions and critical phenomena _ vol 8 ( london / new york : academic press ) p. 145 krech m and dietrich s 1991 345 krech m and dietrich s 1992 a * 46 * 1886 krech m and dietrich s 1992 a * 46 * 1922 li y and baberschke k 1992 1208 nickel b , donner w , dosch h , detlefs c and grbel g 2000 134 schilbe p , siebentritt s and rieder k .- h . 1996 _ phys . lett . _ a * 216 * 20 chung m - c , kaulke m , peschel i , pleimling m , and selke w 2000 _ eur . j. _ b * 18 * 655 marqus m i and gonzalo j a 2000 _ eur . j. _ b * 14 * 317 moussa n and bekhechi s 2003 _ physica _ a * 320 * 435 schulz b and baberschke k 1994 b * 50 * 13467 guo w , shi l p and lin d l b * 62 * 14259 halilov s v , perlov a y , oppeneer p m , yaresko a n and antonov v n 1998 b * 57 * 9557 qiu z q , pearson j and bader s d 1994 b * 49 * 8797 huang f , kief m t , mankey g j and willis r f 1994 b * 49 * 3962 popov a p and pappas d p 2001 b * 64 * 184401 bak p 1982 _ rep . prog . _ * 45 * 587 selke w 1988 _ phys . rep . _ * 170 * 213 yeomans j m 1988 _ solid state physics _ * 41 * 151 selke w 1992 _ phase transitions and critical phenomena _ vol 15 ( london / new york : academic press ) p. 1 neubert b , pleimling m and siems r 1998 _ ferroelectrics _ * 208 - 209 * 141 hornreich r m , luban m and shtrikman s 1975 1678 elliott r j 1961 346 fisher m e and selke w 1980 1502 selke w and duxbury p m 1984 _ z. phys . _ b * 57 * 49 selke w , catrein d and pleimling m 2000 l459 selke w , pleimling m , peschel i , kaulke m , chung m - c and catrein d 2002 _ j. magn . _ * 240 * 349 selke w , pleimling m and catrein d 2002 _ eur . j. _ b * 27 * 321 mello v d , chianca c v , dantas a l and carrio a s 2003 b * 67 * 012401 charnaja e v , ktitorov s a and pogorelova o s 2003 _ cond - mat/0303294 _ diehl h w 2002 _ acta physica slovaca _ * 52 * 271 gumbs g 1986 b * 33 * 6500 binder k and frisch h l 1999 _ eur . j. _ b * 10 * 71 frisch h l , kimball j c and binder k 2000 29 jacobs a e , mukamel d and allender d w 2001 e * 63 * 021704 pleimling m 2002 b * 65 * 184406 diehl h w , rutkevich s and gerwinski a 2003 l243 diehl h w , gerwinski a and rutkevich s 2003 b * 68 * 224428 garel t and pfeuty p 1976 l245 kumar p and maki k 1976 b * 13 * 2011 dietrich s and diehl h w 1983 _ z. phys . _ b * 51 * 343 kikuchi m and okabe y 1985 1220 riecke h , dietrich s and wagner h 1985 3010 xiong g m and gong c d 1989 8673 xiong g m and gong c d 1989 _ z. phys . _ b * 74 * 379 frank d and dohm v 1989 1864 binder k and frisch h l 1991 _ z. phys . _ b * 84 * 403 diehl h w and janssen h k 1992 a * 45 * 7145 diehl h w 1994 b * 49 * 2846 wichmann f and diehl h w 1995 _ z. phys . _ b * 97 * 251 ritschel u and czerner p 1995 3882 majumdar s n and sengupta a m 1996 2394 mukherji s 1999 _ eur . phys . j. _ b * 8 * 423 krech m , karl h and diehl h w 2001 _ physica _ a * 297 * 64 diehl h w , krech m and karl h 2002 b * 66 * 024408 pleimling m and igli f 2004 [ _ cond - mat/0312583 _ ] hohenberg p c and halperin b i 1977 _ rev . phys . _ * 49 * 435 janssen h k , schaub b , schmittmann b 1989 _ z. phys . _ b * 73 * 539 dietrich s and wagner h 1985 _ z. phys . _ b * 59 * 35 dietrich s 1990 _ magnetic properties of low - dimensional systems ii _ ( berlin / heidelberg : springer ) p. 150 dietrich s 1988 _ phase transitions and critical phenomena _ vol 12 ( london / new york : academic press ) p. 1 bonn d and ross d 2001 _ rep . phys . _ * 64 * 1085 binder k , landau d and mller m 2003 _ j. stat . phys . _ * 110 * 1411 symanzik k 1981 1 mcavity d m and osborn h 1993 655 ritschel u and czerner p 1996 3645 czerner p and ritschel u 1997 _ int . j. mod . _ b * 11 * 2075 czerner p and ritschel u 1997 _ physica _ a * 237 * 240 ciach a and ritschel u 1997 653 maciolek a , ciach a and drzewiski 1999 e * 60 * 2887 maciolek a , evans r and wilding n b 2003 _ j. chem . phys . _ * 119 * 8663 cho j - h j and law b m 2001 2070 brzin e and leibler s 1983 b * 27 * 594 ritschel u r693 krimmel s , donner w , nickel b , dosch h , sutter c and grbel g 1997 3880 schmid f 1993 _ z. phys . _ b * 91 * 77 leidl r and diehl h w 1998 b * 57 * 1908 cardy j 1983 3617 igli f , peschel i and turban l 1993 _ adv . phys . _ * 42 * 683 barber m n , peschel i and pearce p a 1984 _ j. stat . phys . _ * 37 * 497 peschel i 1985 313 kaiser c and peschel i 1989 _ j. stat . phys . _ * 54 * 567 davies b and peschel i 1991 1293 davies b and peschel i 1998 _ ann . physik _ * 6 * 187 abraham d b and latrmolire f 1994 e * 50 * r9 abraham d b and latrmolire f 1995 _ j. stat . phys . _ * 81 * 539 abraham d b and latrmolire f 1996 4813 guttmann a j and torrie g m 1984 3539 duplantier b and saleur h 1986 3179 karevski d , lajk p and turban l 1997 _ j. stat _ * 86 * 1153 peschel i , turban l and igli f 1991 l1229 landau d p 1976 b * 14 * 255 merikoski j , timonen j , manninen m and jena p 1991 938 aguilera - granja f , morn - lpez j l and montejano - carrizales j m 1995 _ surf . sci . _ * 326 * 150 saxena v k 1987 6623 grassberger p 1992 5867 gaunt d s and colby s a 1990 _ j. stat . * 58 * 539 wang z g , nemirovsky a m , freed k f and myers k r 1990 2575 saxena v k 1987 b * 35 * 3612 larsson t a 1986 1691 mon k k and valls j l 1989 b * 40 * 2419 lai p y and mon k k 1989 5167 privman v 1988 b * 38 * 9261 lai p y and mon k k 1989 b * 39 * 12407 cardy j l and peschel i 1988 377 kachkachi h and garanin d a 2001 _ physica _ a * 300 * 487 pleimling m and selke w 1998 _ eur . j. _ b * 5 * 805 pleimling m 2002 _ comp . commun . _ * 147 * 101 pleimling m and selke w 1999 b * 59 * 65 pleimling m and selke w 2000 e * 61 * 933 brovchenko i , geiger a and oleinikova a 2004 _ cond - mat/0402030 _ fisher m e and ferdinand f e 1967 169 bariev r z 1979 _ sov . jetp _ * 50 * 613 mccoy b m and perk j h h 1980 840 ko l f , au - yang h and perk j h h 1985 1091 turban l 1985 l325 delfino g , mussardo g and simonetti p 1994 518 oshikawa m and affleck i 1996 2604 simes c s and drugowich de felcio j r 1998 7265 szalma f and igli f 1999 _ j. stat . phys . _ * 95 * 795 henkel m and patks a 1988 l231 henkel m , patks a and schlottmann m 1989 609 turban l _ private communication _ costa - santos r 2003 b * 68 * 224423 frjdh p , howard m and lauritsen k b 1998 2311 janssen h k , schaub b and schmittmann b 1988 _ z. phys . _ b * 72 * 111 kaiser c and turban l 1994 l579 kaiser c and turban l 1995 351 lauritsen k b , frjdh p and howard m 1998 2104 howard m , frjdh p and lauritsen k b 2000 e * 61 * 167 hanke a , krech m , schlesener f and dietrich s 1999 e * 60 * 5163 hauge e h 1992 a * 46 * 4994 napirkowski m , koch w and dietrich s 1992 a * 45 * 5760 rejmer k , dietrich s and napirkowski m 1999 e * 60 * 4027 parry a o , rascn c and wood a j 1999 5535 parry a o , rascn c and wood a j 2000 345 parry a o , wood a j and rascn c 2000 7671 parry a o , wood a j , carlon e and drzewiski a 2001 196103 parry a o , wood a j and rascn c 2001 4591 bednorz a and napirkowski m 2001 e * 63 * 031602 lipowski a 1998 e * 58 * , r1 abraham d b and maciolek a 2002 286101 albano e v , de virgiliis a , mller m and binder k 2003 333 milchev a , mller m , binder k and landau d p 2003 136101 diehl h w and nsser a 1990 _ z. phys . _ b * 79 * 69 benyoussef a , boccara n and saber m 1985 4275 saber m 1987 2749 dakhama a , fahti a and benayad n 2001 _ eur . j. _ b * 21 * 393 kaneyoshi t 1988 _ phys . ( b ) * 150 * 297 kaneyoshi t 1989 b * 39 * 557 kaneyoshi t 1991 _ introduction to surface magnetism _ , ( boca raton , ann arbor , and boston : crc press ) wang j - s , selke w , dotsenko vl s and andreichenko v b 1990 _ physica _ a * 164 * 221 kaneyoshi t and shin s 1998 _ physica _ a * 260 * 455 mon k k and nightingale m p 1988 b * 37 * 3815 durian d j and franck c 1987 555 harris a b 1974 1671 da silva l r , tsallis c and sarmento e f 1988 b * 37 * 7832 shalaev b n 1994 _ phys . rep . _ * 237 * 129 feldman d e and vinokur v m 2002 227204 igli f , turban l and berche b 1991 l1031 cardy j 1991 l1315 de martino a , moriconi m and mussardo g 1998 615 pleimling m , bagamry f , turban l and igli f 2004 _ cond - mat/0402188 _ nattermann t and villain j 1988 _ phase transitions _ * 11 * 5 eisenriegler e , kremer k and binder k 1982 _ j. chem . phys . _ * 77 * 6296 diehl h w and nsser a 1990 _ z. phys . _ b * 79 * 79 diehl h w 1998 _ eur . j. _ b * 1 * 401 bengrine m , benyoussef a , ez - zahraouy h , and mhirech f 1999 _ physica _ a * 268 * 149 zhao d , feng l , huber d l and lagally m g 2000 b * 62 * 11316 mokrani a and vega a 2001 b * 63 * 094403 bahmad l , benyoussef a and ez - zahraouy h 2002 _ physica _ * 303 * 525 aaro reis f d a 1998 b * 58 * 394 aaro reis f d a 2000 b * 62 * 6565 hanke a and kardar m 2001 4596 hanke a and kardar m 2002 e * 65 * 046121 duplantier b 1998 5489 cardy j 1999 l177 hilhorst h j and van leeuwen j m 1981 1188 harris a b and lubensky t c 1974 1540 grinstein g and luther a 1976 b * 13 * 1329 ohna k and okabe y 1992 b * 46 * 5917 usatenko z , shpot m and hu c - k 2001 e * 63 * 056102 usatenko z and hu c - k 2002 e * 65 * 066103 usatenko z and hu c - k 2003 e * 68 * 066115 igli f , lajk p , selke w and szalma f 1998 2801 lajk p and igli f 2000 e * 61 * 147 palgyi g , chatelain c , berche b and f. igli f 2000 _ eur . j. _ b * 13 * 357 aaro reis f d a 1997 b * 55 * 11084 aaro reis f d a 1998 _ physica _ a * 257 * 495 aaro reis f d a 1998 9105 aaro reis f d a 2001 _ physica _ a * 291 * 375 shen j , skomski r , klaua m , jenniches h , manoharan s s and kirschner j 1997 b * 56 * 2340 shen j , pierce j p , plummer e w and kirschner j 2003 r1 skomski r 2003 r841 bazhanov d i , hergert w , stepanyuk v s , katsnelson a a , rennert p , kokko k and demangeat c 2000 b * 62 * 6415 bellini v , papanikolaou n , zeller r and dederichs p h 2001 b * 64 * 094403 andrieu s , chatelain c , lemine m , berche b and bauer p 2001 3883 bariev r z , malov o a and barieva n a 1990 _ physica _ a * 169 * 281 berche b and turban l 1991 245 igli f and indekeu j o 1990 b * 41 * 6836 hamedoun m , bakrim h , hourmatallah a and benzakour n 2003 _ surface science _ * 539 * 155 zhao d , liu f , huber d l and lagally m g 2002 _ j. appl . phys . _ * 91 * 3150

various formation channels for the puzzling ultra - compact dwarf galaxies ( ucds ) have been proposed in the last few years . to better judge on some of the competing scenarios , we present spectroscopic [ fe / h ] estimates for a sample of 26 compact objects in the central region of the fornax cluster , covering the magnitude range of ucds and bright globular clusters ( or mag ) . we find a break in the metallicity distribution of compact objects at mag ( 3 m ) : for mag the mean metallicity is [ fe / h].62 0.05 dex , 0.56 0.15 dex higher than the value of.18 0.15 dex found for mag . this metallicity break is accompanied by a change in the size - luminosity relation for compact objects , as deduced from hst - imaging : for mag , scales with luminosity , while for mag , is almost luminosity - independent . in our study we therefore assume a limiting absolute magnitude of mag between ucds and globular clusters . the mean metallicity of five fornax de , n nuclei included in our study is about 0.8 dex lower than that of the ucds , a difference significant at the 4.5 level . this difference is marginally higher than expected from a comparison of their colors , indicating that ucds are younger than or at most coeval to de , n nuclei . because of the large metallicity discrepancy between ucds and nuclei we disfavor the hypothesis that most of the fornax ucds are the remnant nuclei of tidally stripped de , ns . our metallicity estimates for ucds are closer to but slightly below those derived for young massive clusters ( ymcs ) of comparable masses . we therefore favor a scenario where most ucds in fornax are successors of merged ymcs produced in the course of violent galaxy - galaxy mergers . it is noted that in contrast to that , the properties of virgo ucds are more consistent with the stripping scenario , suggesting that _ different _ ucd formation channels may dominate in either cluster .

astro-ph0512474

in their spectroscopic studies of the fornax region , hilker et al . @xcite and later drinkwater et al . ( @xcite and @xcite ) reported on the discovery of six isolated compact stellar systems in the fornax cluster , having @xmath15 mag and half - light radii @xmath11 between 15 and 30 pc . these objects had been interpreted as galactic stars in previous imaging surveys because of their star - like appearance on photographic plates . due to their compactness at luminosities comparable to average dwarf galaxies , they were dubbed ultra - compact dwarf galaxies ( ucds ) ( phillips et al . @xcite ) . high resolution imaging and spectroscopy ( drinkwater et al . @xcite ) place the fornax ucds between the sequence of globular clusters ( gcs ) and dwarf elliptical galaxies ( des ) in the fundamental plane of stellar systems . their m / l ratios are in the range 2 - 4 , at the upper end of values obtained for globular clusters . the luminosities and sizes of ucds are comparable to those of bright nuclei of nucleated dwarf ellipticals ( de , ns ) ( lotz et al . @xcite ) . it is not yet clear how objects as compact and luminous as the ucds may have formed . one much discussed hypothesis is the stripping scenario ( bekki et al . @xcite ) , which proposes that ucds are remnant nuclei of dwarf galaxies stripped in the potential well of their host cluster . it has also been proposed that most ucds are simply very bright globular clusters ( mieske et al . @xcite ) , although we note that the sizes of the brightest fornax ucds are much larger than those of globular clusters ( drinkwater et al . @xcite , jordn et al . ucds could also be explained by stellar super - clusters created in gas - rich galaxy galaxy mergers ( fellhauer & kroupa @xcite and @xcite ) . related to the latter scenario , bastian et al . @xcite propose that young massive clusters ( ymcs ) created in recent galaxy mergers may resemble ucd properties after a hubble time . maraston et al . @xcite show that the young massive cluster w3 in the merger remnant galaxy ngc 7252 will resemble ucd fundamental plane properties if aged to several gyrs . recently , ucds have also been detected in other environments : ucd candidates brighter than those in fornax have been found in the massive galaxy cluster abell 1689 ( mieske et al . @xcite ) , consistent with the idea that tidal forces play a major role in the creation of ucds . hasegan et al . @xcite report on the discovery of several ucds in the virgo cluster , some of which have high m / l ratios between 6 and 9 . those authors suggest a high m / l ratio as a criterion to tell ucds from globular clusters . also jones et al . @xcite have discovered 9 ucd candidates in virgo , in a survey analogous to the one performed by drinkwater et al . @xcite in fornax . like the investigation of hasegan et al . , they find that ucds in virgo are close to the blue peak of the gc color distribution , in contrast to fornax where they coincide with the red peak ( mieske et al . @xcite ) . the data of jones et al . @xcite give a @xmath16 colour difference of 0.33 @xmath8 0.08 mag between fornax and virgo ucds , significant at the 4@xmath13 level . this color difference hints at possible different formation scenarios , which will be discussed in the course of this paper . de propris et al . @xcite show from hst imaging that the surface brightness profile of five fornax ucds is only matched by the brightest de , n nuclei in virgo , while the ucds are significantly more extended than fainter nuclei . that is not necessarily inconsistent with the stripping scenario , because ucds have luminosities comparable to the brightest nuclei . fainter ucds might have different scale sizes . furthermore , stripping processes can alter the structural parameters of embedded nuclei ( bekki et al . therefore , a direct comparison of surface brightness profiles might be a biased way of judging on differences between ucds and embedded nuclei . a note regarding nomenclature : the term `` ultra - compact dwarf galaxies '' ( phillipps et al . @xcite ) should be understood as a purely _ morphological _ description , referring to astronomical objects of luminosities comparable to average dwarf galaxies and sizes much smaller than those . this term does not imply a specific ucd _ formation scenario _ , be it as remnant of more extended dwarf galaxies or as merged super - clusters . furthermore , in order to call a source ucd , it should clearly be distinguishable from the bulk of `` normal '' gcs , based on properties like mass , size or stellar populations . in our studies , whenever a separation between gcs and ucds is not clear , we therefore use the term `` compact objects '' . in the fornax compact objects survey fcos ( mieske et al . @xcite and @xcite ) we have analyzed the photometric and kinematic properties of compact objects in the fornax cluster . our sample of compact objects was defined via radial velocity cluster membership , spanning a range of @xmath0 ( @xmath1 ) mag . at the bright end , these are the ucds , while at the faint end the sample is dominated by gcs . a main finding of the fcos was that compact objects with @xmath17 mag have somewhat different kinematics , spatial distribution and photometric properties than compact objects with @xmath18 mag . furthermore , the latter sample was more consistent with the gc system of ngc 1399 than the former one . this was interpreted such that ucds dominate the sample for @xmath17 mag , while gcs dominate for @xmath18 mag . specifically , ucds ( i.e. compact objects with @xmath17 mag ) are found to be redder by about 0.10 mag than the overall gc population of ngc 1399 . their colors are shifted about 0.2 mag red - wards of the well known color magnitude relation for des ( hilker et al . @xcite , karick et al . @xcite ) and consistent with a color magnitude trend of comparable slope . the nuclei of present - day de , ns are bluer than ucds by 0.10 to 0.15 mag ( lotz et al . @xcite ) , suggesting that ucds either have higher metallicities or older integrated stellar populations than the present - day nuclei . the magnitude distribution of compact objects observed in the fcos shows a soft transition between ucds and gcs . there is a 2@xmath13 overpopulation with respect to the extrapolated bright end of ngc 1399 s gc luminosity function . the findings obtained from the fcos are consistent both with the stripping scenario ( bekki et al . @xcite ) and the stellar super - cluster scenario ( fellhauer & kroupa @xcite ) as a source of ucds . the major uncertainty of the data interpretation for the fcos is the origin of the color difference between fornax ucds , the nuclei of de , ns and the main body of gcs . the redder color of ucds as compared to nuclei either means older integrated ages or higher metallicities . within the scenario of bekki et al . @xcite , the m / l ratio of the stripping remnant ( nucleus + remainder of envelope ) is lower by a factor of two compared to the progenitor m / l , while the nucleus itself loses about 20% of its mass . this mass loss and the associated weakening of the gravitational potential will lead to expulsion of a substantial gas fraction out of the stripping remnant , a process conceivably supported by ram - pressure stripping . this would lower the efficiency of or completely halt subsequent star formation events in the naked nucleus . therefore one would expect the ucds to have older integrated ages and lower or at most equal metallicity as compared to de , n nuclei ( see also jones et al . one would _ not _ expect in the stripping scenario that the ucds are on average more metal - rich than still existing de , n nuclei . the aim of this paper is to lift the age - metallicity degeneracy inherent in the broad - band color measurements for fornax ucds , gcs and de , n nuclei , in order to better assess the validity of the various ucd formation scenarios . to this end we present new spectroscopic data for compact objects in fornax , obtained with the 6.5 m magellan telescope at las campanas observatory , chile . the paper is structured as follows : in sect . [ data ] we present the data and their reduction . [ cmrel ] re - assesses the color - magnitude relation for fornax compact objects proposed in mieske et al . @xcite . in sect . [ results ] we present the results of the spectroscopic metallicity measurements for ucds , gcs and de , n nuclei . the results are discussed in sect . [ discussion ] . we finish this paper with the conclusions in sect . [ conclusions ] . the data we present in this paper were obtained in the night of 5th to 6th december 2004 at las campanas observatory ( lco ) , chile . the instrument used was the inamori magellan areal camera and spectrograph `` imacs '' in spectroscopy mode with the `` short '' f/2 camera , mounted at the 6.5 m baade telescope . the double - asphere , glass - and - oil - lens f/2 camera produces an image of 27.4@xmath19 field diameter at 0.20 arcsec per pixel . the imacs detector is an 8192@xmath208192 ccd mosaic camera which uses 8 thinned 2k@xmath204k detectors . gaps of about 50 pixels separate the chips . we used the intermediate resolution 300 grism , which has a dispersion of 1.34 per 0.2@xmath21 pixel and covers a wavelength range of 3900 - 10000 . at a typical seeing of 0.8@xmath21 , the instrumental resolution was of the order of 5 . we observed one mos mask centered on ngc 1399 , targeting 35 compact fornax members in the magnitude range @xmath0 ( @xmath1 ) mag . those members were chosen from the fcos and the investigation of ngc 1399 s gcs by richtler et al . the total integration time was 2.5 hours , split up into three exposures of 30 minutes and three exposures of 20 minutes . see figs . [ map ] and [ cmdall1 ] for a map and color - magnitude diagram of all compact fornax members in the magnitude range @xmath22 mag . the 35 sources observed with imacs are especially indicated . the ucds originally discovered by drinkwater et al . @xcite are those brighter than @xmath23 mag . [ data ] the reduction procedure from raw images to two - dimensional , sky subtracted , wavelength calibrated spectra was performed with the imacs version of the cosmos - package . cosmos performs the following reduction steps , based on the observation definition file used to generate the slit mask : construct spectral maps , use a comparison arc to adjust spectral maps , calculate positions of spectral lines using a spectral map , construct a spectroscopic flat field frame , do bias and flat field corrections of frames , subtract sky from spectral frame , extract 2d - spectra , co - add spectral exposures , using a cosmic - ray rejection algorithm . the extraction of the one - dimensional spectrum from the two - dimensional output of cosmos was performed with the apall task within the iraf twodspec package . to test the absolute accuracy of the wavelength calibration , these 1d - spectra were cross - correlated against the same template spectra as used for the radial velocity measurement in the fcos . for the 35 compact objects observed with imacs , the mean radial velocity was insignificantly higher by only 13 @xmath8 20 km / s than measured in the fcos . in order to properly derive lick line indices , we slightly downgraded our spectra to the lick system resolution of 9 . for the brightest object observed , namely ucd 3 with @xmath24 mag , the s / n at the peak of the spectral throughput was around 100 . we derived lick indices for our 35 observed sources using the latest lick pass - band definitions of trager et al . @xcite , appropriately red - shifted according to the radial velocity of each investigated object . for converting the line index measurements to [ fe / h ] , we used the empirical calibration of puzia et al . @xcite between [ fe / h ] and several metal line indices for a large sample of metal - poor and metal - rich galactic globular clusters , based on the zinn & west @xcite scale . specifically , we employed the four most strongly correlated indices mg2 , mgb , @xmath25 , @xmath26 $ ] . for each of those indices we derived [ fe / h ] according to puzia s calibration relation and we attributed a weight corresponding to the inverse square of the rms of the calibration relation . then , we adopted the weighted mean of the four measurements as [ fe / h ] . the scatter of the four values around their mean is adopted as error estimate for [ fe / h ] . the results of these measurements are given in table [ fehtable ] and plotted in figs . [ fe_v_nucl ] and [ vi_fe_v ] . [ hb_mg2 ] shows the age indicator h@xmath27 vs. [ fe / h ] . see sect . [ results ] for further details . before proceeding to the metallicity distribution of compact objects in fornax , we revisit the possibility of a color - magnitude relation as suggested by mieske et al . @xcite . [ cmdall1 ] shows a @xmath28 cmd of compact fornax members with @xmath29 ( @xmath30 ) mag , i.e. about two magnitudes brighter than the turn - over of the globular cluster luminosity function ( gclf ) . the difference to the cmd shown in mieske et al . @xcite is that now we include additional confirmed cluster members from richtler et al . @xcite and dirsch et al . @xcite in the magnitude range @xmath31 mag . photometry is from hilker et al . @xcite for all sources . over - plotted is a linear fit to all data points , applying a 2.5@xmath13 rejection algorithm . for this fitting , the iraf task nfit1d within the stsdas package was used . we have slightly modified the error computation for the fitted coefficients as compared to mieske et al . @xcite . in the latter paper the coefficient errors were calculated in the following way : after fitting the function , the independent variable vector @xmath14 is replicated 15 times , each data point being replaced by the fitted function value plus noise with dispersion given by _ the error bar of each data point _ ( as determined by artificial star experiments in mieske et al . this does not take into account the fact that there are _ physical _ variations in color that may go beyond the level of the photometric accuracy . a more conservative error estimate is achieved when the dispersion for the re - sampling is adopted as the _ dispersion of the original data around the fit . _ the significances quoted in the following are based on this error estimate . as can be seen in fig . [ cmdall1 ] , for the sample as a whole there is a trend of redder color with increasing luminosity , significant at the 2.5@xmath13 level ( compared to 3.0@xmath13 in mieske et al . the significance of this trend remains unchanged when rejecting the brightest data point from the sample . however , the slope becomes insignificant when restricting the sample to the sources with @xmath17 mag , due to low number statistics . interestingly , the slope remains almost unchanged at even slightly higher significance ( 2.8@xmath13 ) when excluding all sources with @xmath17 mag . the difference in color between the faint ( @xmath18 mag ) and bright ( @xmath17 mag ) subsample is 0.080 @xmath8 0.024 mag , significant at the 3.4@xmath13 level . all this is consistent with a gradual color - magnitude trend over the entire investigated magnitude range @xmath22 mag . however , the significance levels are still not very high , and spectroscopic information is indispensable to clarify whether the color shift is due to age or metallicity . in fig . [ fe_v_nucl ] we plot [ fe / h ] vs. apparent @xmath32 magnitude for 26 out of the 35 observed sources . the 9 objects not plotted had too low s / n for line index measurements . over - plotted are values for 5 nuclei of de , ns in the fornax cluster . two of those measurements are from the data set in this paper , three from the first part of fcos ( mieske et al . @xcite ) . one striking feature of fig . [ fe_v_nucl ] is a metallicity break at @xmath2 mag : among the brightest compact objects in fornax there are no metal - poor ones . when separating the sample into bright and faint with a limit of @xmath33 mag inspired by the results of the fcos , the mean [ fe / h ] of the bright sample is @xmath90.63 @xmath8 0.07 dex , that of the faint sample @xmath91.14 @xmath8 0.15 dex . the difference then is 0.51 @xmath8 0.16 dex , significant at the 3.2@xmath13 level . if instead of @xmath33 mag adopting @xmath34 mag ( @xmath12 mag ) as the limit , the values for the bright and faint sub - samples are @xmath90.62 @xmath8 0.05 dex and @xmath91.18 @xmath8 0.15 dex , respectively . the difference rises to 0.56 @xmath8 0.15 dex , somewhat more significant at the 3.7@xmath13 level . these confidence levels are of the same order than the difference in _ color _ @xmath14 between bright and faint subsamples . note that while we can trace back the color _ shift _ to a metallicity _ shift _ , the size of our imacs sample is not large enough to judge on the existence of a _ continuous _ luminosity - metallicity relation . to get some feeling for possible systematic biases in our derivations of [ fe / h ] i.e. offsets of our line index measurements with respect to the lick system consider the cmd in fig . [ vi_fe_v ] . here , we show the _ measured _ @xmath14 values plus the measured [ fe / h ] values _ transformed _ to @xmath14 , using the transformation relation by kissler - patig et al . @xcite , which was calculated for gcs in ngc 1399 with metallicities above [ fe / h]@xmath35 dex . for the eight sources brighter than @xmath12 mag with @xmath28 photometry available ( all of which have [ fe / h]@xmath36 dex ) there is an offset between measured and transformed @xmath14 of 0.018 @xmath8 0.016 mag with an rms scatter of 0.045 mag . under the assumption that the lick calibration itself between line indices and [ fe / h ] is correct , this indicates an upper limit to a potential systematic bias in [ fe / h ] of about 0.10 dex . the next question to ask is : does the color difference in @xmath14 between bright and faint sub - sample ( @xmath12 mag as limit ) correspond to the measured metallicity - difference ? the mean color for the bright sub - sample , taking only objects with @xmath28 photometry available , is 1.15 @xmath8 0.02 mag . for the faint sub - sample it is 1.01 @xmath8 0.04 mag . to convert this difference of @xmath37 0.04 mag into an expected metallicity difference ( under the assumption of two coeval populations ) , we again apply the transformation of kissler - patig et al . @xcite , specifically their calibration slope @xmath38}{d(v - i)}= $ ] 3.27 @xmath8 0.32 . we would then expect a [ fe / h ] difference of 0.46 @xmath8 0.15 dex , in very good agreement with the actually found difference of 0.56 @xmath8 0.15 dex . note that the mean metallicity of the faint sub - sample is at about the limit of the linear approximation by kissler - patig et al . @xcite . for lower metallicities , the slope becomes steeper . kissler - patig et al . @xcite found a slope twice as high for the metal - poor milky way globular clusters ( see also couture et al . @xcite , kundu & whitmore @xcite ) . taking this steepening of the slope into account , the expected metallicity difference is higher by about 0.1 to 0.2 dex , bringing it to an even better match with the observations . thus , we can state that our data are consistent with the assumption that the only difference between bright and faint sub - sample is metallicity . are there also significant differences in age sensitive line indices between the bright and faint sub - sample ( fig . [ hb_mg2 ] ) ? the mean @xmath39 equivalent width for sources with @xmath40 mag is 2.30 @xmath8 0.10 with a dispersion of 0.3 , compared to 2.69 @xmath8 0.20 and a dispersion of 0.6 for @xmath41 mag , mg2@xmath42 and @xmath43 . for a proper age comparison , one also has to take into account the slope of isochrones in the [ fe / h]@xmath44 plane , which is approximately @xmath45}=-0.5 $ ] ( see fig . [ hb_mg2 ] ) . with a metallicity difference of 0.55 dex as estimated above , one would expect a difference in @xmath39 of 0.25 to 0.3 between the bright and faint sample in the case of coeval populations . the actual difference is 0.39 @xmath8 0.22 . the data are therefore consistent with coeval populations , but also allow for age differences of a few gyrs in both directions . the ucds appear most consistent with intermediate age populations . concluding this section : we have found a @xmath3 0.55 dex higher metallicity for compact objects in fornax with @xmath6 mag than for those with @xmath10 mag , significant at the 3.7@xmath13 level and in good agreement with their color difference . our findings hence support the hypothesis that for @xmath6 mag , the population of compact objects has a different formation history than the main body of ngc 1399 s gcs . in the following , we will refer as ucds to compact objects with @xmath6 mag and as gcs to those with @xmath10 mag . with respect to this separation it is worth noting that the metallicity range of gcs includes that of the ucds , and therefore we can not exclude that a low - luminosity extension of the ucds somewhat contaminates the gc sample . in sect . [ discussion ] the separation at @xmath12 mag will be confirmed by a size comparison from hst imaging , indicating that ucds are probably not a _ major _ fraction of fornax compact objects with @xmath10 mag . one much discussed formation scenario for the ucds is that they are the compact remnants of tidally stripped de , ns ( e.g. bekki et al . @xcite , drinkwater et al . @xcite , see also bassino et al . [ fe_v_nucl ] compares the metallicities of bright gcs plus ucds with those of the central regions of five fornax cluster de , ns . the central regions of the investigated de , ns have a mean metallicity of [ fe / h]@xmath71.38 @xmath8 0.15 dex , 0.75 @xmath8 0.17 dex lower than the mean ucd metallicity . this is a 4.5@xmath13 difference . literature estimates of [ fe / h ] for fornax des and de , ns ( held & mould @xcite , brodie & huchra @xcite , rakos et al . @xcite ) also show lower average metallicities than for the ucds , but are about 0.1 to 0.2 mag higher than our values ( see also mieske et al . @xcite ) . before proceeding further , we draw the attention to two effects that potentially bias our result : \1 . in addition to the light from the nucleus itself , a certain fraction of the stellar envelope of the investigated de , ns also contributes to the measured signal . that fraction was on average about 40% for the five de , ns investigated ( scattering between 10% and 85% ) , as judged from the slit size and width ( this paper ) , total galaxy luminosity ( hilker et al . @xcite ) and nuclear luminosity ( this paper and lotz et al . this contribution becomes important in the case of significant differences between the stellar populations of the de , n envelope and the nucleus . lotz et al . @xcite have shown that on average , de , n envelopes are redder by 0.1 to 0.2 mag than their nuclei . if this is due to a metallicity difference ( as argued by lotz et al . ) , our measurements will overestimate the nuclear metallicity . to estimate the extent of this overestimation , we assume the mean of the two different slopes derived in the metal - poor ( [ fe / h]@xmath46 ) and metal - rich regimes by kissler - patig et al . @xcite and @xcite , which is @xmath38}{d(v - i)}~\simeq$]5 . the over - estimation of the nuclear metallicity is then about 0.3 dex . this effect can explain why we obtain a slightly lower de , n metallicity than previous estimates : due to the better spatial resolution of our data , the contribution of the presumably more metal - rich stellar envelope is lower than in the data of the authors in the previous paragraph . \2 . our sample of de , n nuclei is biased towards lower luminosities compared to the ucds : the average luminosity of the five nuclei plotted in fig . [ fe_v_nucl ] is v=20.2 mag , while the ucds have a mean apparent magnitude of v=19.6 . that magnitude difference is relevant because of the color - magnitude relation for nuclei ( lotz et al . @xcite ) , which is commonly interpreted as the representation of a mass - metallicity relation . due to this relation , our sample of five lower luminosity nuclei statistically exhibits a lower metallicity than nuclei of ucd luminosity . with a slope of @xmath470.04 ( lotz et al . @xcite ) and @xmath38}{d(v - i)}\simeq$]5 ( kissler - patig et al . @xcite and @xcite ) , one would expect a higher metallicity by about 0.12 dex for nuclei of ucd luminosity . the net effect of the two biases is a slight increase of the metallicity difference by about 0.1 - 0.2 dex . we therefore confirm a metallicity difference between ucds and nuclei of at least 0.80 dex . already in mieske @xcite and mieske et al . @xcite it was noted that de , n nuclei from the hst photometry of lotz et al . @xcite are bluer than the fornax ucds by 0.10 to 0.15 mag , consistent with a lower metallicity as found here . assuming again @xmath38}{d(v - i)}~\simeq $ ] 5 ( kissler - patig et al . @xcite and @xcite ) one would expect a metallicity difference of the order 0.5 to 0.75 dex . this is consistent with , but at the low end of our measured range . . [ discussion ] for further discussion . we finally note that the [ fe / h]@xmath44 plot in fig . [ hb_mg2 ] shows no indication for an age difference between the five de , n nuclei and the ucds . however , the large scatter of data points also allows for substantial age differences of several gyrs . in the light of the new metallicity estimates outlined in the previous section , we discuss three possible ucd formation channels : 1 . unusually bright gcs . nuclei of de , ns . 3 . young massive star clusters created in violent star - forming events . + 1 . bright gcs : the brightest fornax ucd ( ucd 3 ) can not be defined as a gc , given its very high luminosity ( @xmath48 -13.4 mag ) and large half - light radius @xmath11 of 30 pc ( drinkwater et al . @xcite ) . surface brightness profile fitting even revealed a second exponential component of @xmath49 60 pc ( drinkwater et al . indeed , ucd 3 is the only ucd that is marginally resolved on ground - based images ( see hilker et al . @xcite and mieske et al . furthermore , fornax ucds are more extended around the cluster center , more numerous than expected from the extrapolation of a gaussian gclf and slightly offset in radial velocity with respect to the gcs ( mieske et al . the significances of these differences are all of the order 2@xmath13 . the strongest distinguishing feature is the metallicity difference between ucds ( @xmath6 ) and gcs ( @xmath10 mag ) found in the present paper , significant at 3.7@xmath13 . note , however , that this only implies that fornax ucds do not belong to the _ metal - poor _ , i.e. blue population of gcs . the mean metallicity of the fornax ucds is close to ( yet slightly lower than ) the typical value found for the metal - rich , red peak of the gc distribution ( e.g. peng et al . @xcite , kissler - patig et al . @xcite ) of about -0.6 to -0.3 dex . also their mean color of @xmath50 @xmath8 0.02 is identical to the red peak of ngc 1399 s bimodal gc color distribution ( gebhardt & kissler - patig ( @xcite ) . that is : the question `` can one explain ( some ) ucds as bright gcs ? '' must in the fornax cluster case be specified as : `` can the red bulge gcs extend to much higher masses / luminosities than the blue gcs ? '' . a possibility is that the most massive metal - rich star clusters preferentially form by the merging of several single , already massive clusters in the course of violent gas - rich merger events ( fellhauer & kroupa @xcite ) . this is an attractive scenario if one assumes that indeed galaxy - galaxy mergers were the main creation channel for the entire metal - rich gc bulge population of gcs ( e.g. rhode & zepf @xcite , li et al . this is opposed to the idea that both metal - poor and metal - rich gcs formed in single , monolithic collapse - like , subsequent episodes ( e.g. forbes et al . @xcite , harris & pudritz @xcite ) . in the latter case one might expect a less violent , more homogeneous star formation , not necessarily leading to the creation of super - massive gcs . kravtsov & gnedin @xcite find in their simulations on gcs formation that the maximum globular cluster mass in a given region strongly correlates with the local average star formation rate . on the observational side , larsen @xcite correspondingly found that the maximum mass of young massive clusters in spiral galaxies scales with local star formation rate . this implies that the most massive gcs are created in the most violent merger events . kravtsov & gnedin also show that the mass of the most massive gc built in a dm halo correlates with this dm halo s mass , although with a large dispersion , due to the varying star formation conditions within the halo . according to formula ( 8) of their paper , gcs as massive as the ucds ( 1 - 5@xmath2010@xmath51m@xmath5 ) must have progenitor haloes of @xmath3 5@xmath2010@xmath52m@xmath5 . this is only a few percent of ngc 1399 s dm halo mass ( @xmath53 , see richtler et al . therefore one may at face value not even require to merge _ several _ massive gcs to end up with _ one single _ . however , a strong argument against the identity ucds = bright single gcs comes from a size comparison : the five brightest fornax ucds have king half - light radii @xmath11 between 15 and 30 pc ( drinkwater et al . the typical @xmath11 of globular clusters is only @xmath543 pc ( jordn et al . @xcite ) over several orders of magnitude in mass , and even the most massive gcs are not expected to exceed radii of about 5 - 7 pc ( kravtsov & gnedin @xcite ) . the size distribution of ucds vs. gcs deserves some more detailed investigation : in fig . [ size ] we plot @xmath11 vs. luminosity for those compact objects that were imaged in the acs fornax cluster survey ( jordn @xcite and jordn et al . in preparation ) . all of them are imaged in two acs pointings . we include the data points for the five brightest ucds from drinkwater et al . @xcite and also plot the data points for all objects detected in the two acs pointings . note that the bulk of sources with @xmath55 mag are gcs , having the typical @xmath11 values around 3 pc ( jordn et al . @xcite , mclaughlin @xcite ) . the fact that they have a luminosity independent mean size is a physical effect , and not due to the resolution limit of the camera . at the distance of fornax , 3 pc subtend 0.03@xmath21 , or 0.6 wfc pixels . as shown in jordn et al . @xcite , the resolution limit regarding @xmath11 measurements is 0.25 wfc pixels ( @xmath31pc at virgo / fornax ) , which is substantially lower ( see also fig . [ size ] ) . it is very remarkable that the luminosity of the metallicity break corresponding to @xmath2 mag coincides exactly with the luminosity brighter than which there is a clear up - turn in the size - luminosity distribution of fornax compact objects . brighter than @xmath12 mag one enters the regime where size scales with mass / luminosity , while fainter than that , size is independent of mass / luminosity . this strengthens the separation occurring at @xmath12 mag between a ) metal - rich objects at brighter magnitudes whose size scales with luminosity the ucds and b ) more metal - poor objects at fainter luminosities with luminosity independent sizes the gcs . this luminosity limit corresponds to about 3@xmath56 , assuming an average m / l ratio of 3 ( drinkwater et al . note that @xmath57 centauri , the largest star cluster of the milky way , has a mass of about this value ( meylan & mayor @xcite , pryor & meylan @xcite ) . furthermore , this is also the mass which hasegan et al . @xcite propose to mark the transition between gcs and ucds . a possible clue to the up - turn in the size - luminosity distribution are the findings of fellhauer & kroupa @xcite and bekki et al . @xcite : super - clusters consisting of several single clusters that merged together will be significantly more extended than the average progenitor cluster . that is , they will deviate towards larger sizes from the constant mass - size relation of globular clusters . the corresponding effect in the fundamental plane ( drinkwater et al . @xcite , kissler - patig et al . @xcite , bastian et al . @xcite , maraston et al . @xcite ) has been found for the brightest and most massive young massive clusters and ucds , which do not follow the steep m-@xmath13 relation defined by gcs , but rather the extrapolation of the faber - jackson relation . this supports the hypothesis that ucds are successors of merged super - clusters , as opposed to being single super - massive gcs . the possibility of young massive clusters in the present universe as examples of `` early '' ucds will be discussed further in point 3 . nuclei : ucds in fornax are significantly more metal - rich and redder than nuclei of existing de , ns , with no indications for a significant age difference ( see previous section ) . if ucds are stripped nuclei , they were stripped of their gas and parts of their dark matter halo a long time ago ( see bekki et al . consequently they have had no opportunity of self - enrichment and star formation since the time of their stripping . in comparison to still existing nuclei that are embedded in a deep potential well with the principal possibility of further star formation events , one would therefore expect equal or lower metallicities and higher integrated ages for `` naked '' nuclei ( see also sect . [ intro ] , mieske et al . @xcite and jones et al . we find significantly higher metallicities , but can not tightly constrain the age difference . the large metallicity difference is not consistent with the stripping scenario as the major creation channel for ucds in fornax . we have outlined in sect . [ compnuclei ] that the metallicity difference between nuclei and ucds is slightly larger than predicted from their color difference of 0.10 to 0.15 mag . if one allows for age differences between ucds and nuclei to account for this `` deficit '' in color difference , one would require that ucds be younger than nuclei . this goes into the opposite direction of what would be expected in the stripping scenario . one conceivable way of explaining the metallicity difference between surviving `` embedded '' nuclei and ucds is that tidal stripping selects the most metal - rich de , ns . that is not pure speculation , since in some galaxy clusters , galaxy metallicities have been shown to increase towards the cluster center ( e.g. carter et al . @xcite for coma , ezawa et al . @xcite for awm 7 ) . this can lead to a selection bias because tidal stripping should affect only those de , ns that orbit close to the cluster center ( bekki et al . @xcite ) . however , the amplitude of the metallicity gradients found in the cited studies is only of the order of 0.2 - 0.3 dex over the entire radial cluster range . that is not sufficient to explain the difference of @xmath3 0.80 dex between ucds and nuclei found by us . furthermore , lotz et al . @xcite do not find any radial color gradient for fornax cluster dwarf elliptical galaxies , making a strong stripping selection bias unlikely . note that the ucds discovered in the virgo cluster exhibit slightly different properties , more consistent with the stripping scenario : the ucd candidates detected in the virgo cluster survey ( ct et al . @xcite ) based on their larger sizes than average gcs ( hasegan et al . @xcite ) are significantly bluer than the fornax ucds ( see also jones et al . @xcite ) . they are almost as blue as the blue peak of m87 s gc color distribution ( peng et al . @xcite ) and are consistent with the colors of virgo de , n nuclei ( ct et al . @xcite ) . in agreement with this finding , evstigneeva et al . @xcite report line index estimates for virgo ucds in which they are found to be old and metal _ poor_. for some of the virgo ucds , hasegan et al . @xcite report high m / l ratios between 6 and 9 , which in case of virial equilibrium requires some component of dark matter . those authors suggested a high m / l ratio as a bona - fide criterion to tell ucds from ordinary gcs . hasegan et al . @xcite find that the two ucd candidates with highest m / l ratios are the most metal - poor ones , strengthening the need for dark matter . in contrast , the fornax ucds have lower m / l ratios between between 2 and 4 ( drinkwater et al . @xcite ) and higher metallicities . this makes the scenario of ucds as successors of nuclei of dark matter dominated de , ns more likely for virgo than for fornax . the concluding remark we draw from this is : for the fornax ucds , the de , n stripping seems a less important formation channel , while for the virgo ucds the stripping scenario may be the preferred channel . young massive clusters ( ymcs ) : motivated by the difficulties of the stripping scenario to explain the fornax ucd metallicities , we discuss the hypothesis that ucds are successors of young massive clusters created in violent star - forming events . this hypothesis is closely related to the scenario of fellhauer & kroupa ( @xcite and @xcite ) that ucds may be merged super - clusters created in such mergers ( see also point 1 ) . one example for a possible future ucd is the ymc w3 in the star forming galaxy ngc 7252 ( maraston et al . @xcite ) with a mass of @xmath3 10@xmath58 m@xmath5 whose fundamental plane parameters place it in the region of ucds / m32 when passively aged several gyrs . fellhauer & kroupa @xcite show that w3 may have formed through the merging of several dozen massive single star clusters . bastian et al . @xcite and kissler - patig et al . @xcite find that also other ymcs populate the same fundamental plane region as the ucds . [ fe_v_nucl ] shows the location of w3 in the luminosity - metallicity diagram if aged passively to 10 and 13 gyrs , respectively . starting points are its present observed absolute magnitude @xmath5916.2 mag , an assumed age of 300 myr and assumed metallicities of z=0.4z@xmath60 and z=1.0z@xmath60 ( hilker et al . in preparation , maraston et al . w3 is located close to the regime of the brightest fornax cluster ucds , shifted slightly to higher metallicities . note that bastian et al . @xcite also found metallicities of [email protected]@xmath60 for several other young and massive star cluster complexes in the antenna galaxies ( ngc 4038/4039 ) . that is an indication that some of the fornax ucds may have been created in an early violent merger event similar to those occurring with ngc 7252 or the antennas , but in a lower metallicity environment . the very small number of ucds can be explained by the fact that only a small fraction of the cluster complexes observed today appears to be gravitationally bound ( e.g. fall et al . @xcite , bastian et al . that is , only very few of the complexes will survive over gyr time scales . fall et al . @xcite argue that while between 20% and 100% of all stars form in clusters , the fraction of surviving clusters over a hubble time is between 10@xmath62 and 10@xmath63 of the total stellar mass . for comparison , the total absolute brightness of all compact objects in fornax with @xmath6 mag is @xmath64 mag , about 6 mag or a factor of 250 fainter than ngc 1399 s absolute brightness of @xmath65 mag ( hilker et al . this is consistent with the fractional stellar mass range for surviving clusters as estimated by fall et al . and still leaves sufficient space for the main body of the fainter globular clusters . it is important to stress that in order to create a consistent link between ymcs and ucds in terms of metallicity , one would have to know the properties of an w3-like object created about 10 gyrs ago and then interpolate up to the present . star formation in the early universe was certainly not less intense than it is now , and mergers also happened more often than now . if the molecular clouds out of which early ymcs formed consisted of pre - processed material , one could have formed moderately metal - rich ymcs with fornax ucd metallicity , i.e. about 1/2 to 1/4 of the metallicities derived for present ymcs , relatively early in the universe . note that the cm trend amongst the fornax ucds / bright gcs ( see sect . [ cmrel ] ) may be explained with the ymc scenario : statistically , the more massive ymcs originate from more massive progenitor galaxies and more intense star - bursts ( see e.g. elmegreen et al . @xcite ) and might therefore also be more metal - enriched . + to finish the discussion , we consider in how far global differences between the fornax and virgo clusters are expected to influence the occurence of the two competing ucd formation channels . this may help to explain why properties of virgo ucds fit better the stripping scenario , while fornax ucds are more consistent with aged ymcs . the fornax cluster has a two times higher central galaxy density and a two times lower velocity dispersion than the virgo cluster ( drinkwater et al . @xcite , mieske et al . @xcite , richtler et al . @xcite , ferguson @xcite ) . since the galaxy merger rate in cluster environments has the approximate dependence @xmath66 ( makino & hut @xcite ) , one expects a much higher rate of galaxy mergers in fornax than in virgo . this qualitative difference should also hold at the time of the most violent galaxy mergers and hence ymc formation , even if both clusters were still in their early phase of evolution and had not yet settled into virial equilibrium . if the tidal stripping scenario and the fading of ymcs created in violent galaxy mergers are competing ucd formation channels , the probability of the ymc channel is therefore higher for fornax than for virgo . this fits to the observed properties of fornax and virgo ucds . to what extent is the stripping scenario influenced by the different cluster properties ? because of the higher mass of virgo and its correspondingly deeper potential well as compared to fornax , bekki et al . @xcite predict the number of ucds in virgo created by stripping de , ns to be about a factor of three higher than in fornax . given that the numbers of ucds in virgo and fornax ( jones et al . @xcite and drinkwater et al . @xcite ) do not differ by such a large factor , a plausible interpretation that fits to the observed ucd properties is that the population of fornax ucds is fueled by an additional formation channel . [ discussion ] we have presented spectroscopic [ fe / h ] estimates for a sample of 26 compact objects in the central fornax cluster , spanning the luminosity range @xmath0 ( @xmath1 ) mag . our sample includes ucds at the bright end and is dominated by gcs at the faint end . we furthermore present [ fe / h ] measurements for the nuclear regions of five fornax de , ns . a. the mean metallicity of bright compact objects ( @xmath6 mag ) is [ fe / h]@xmath70.62 @xmath8 0.05 dex , which is 0.56 @xmath8 0.15 dex higher than the mean metallicity of @xmath91.18 @xmath8 0.15 dex for faint compact objects ( @xmath10 mag ) . the difference is significant at the 3.7@xmath13 level . this metallicity break at @xmath12 mag ( @xmath67 ) is accompanied by a change in the size - luminosity relation for compact objects , as deduced from acs - imaging : for @xmath6 mag , @xmath11 scales with luminosity , while for @xmath10 mag , @xmath11 is luminosity - independent and of typical value for gcs . due to these characteristic features in the metallicity and size distribution , we identify compact objects with @xmath6 mag as ucds and those with @xmath10 mag as gcs . the metallicity difference between ucds and gcs is consistent with their color difference in @xmath14 under the assumption of coeval populations . b. the mean metallicity of the five investigated de , n nuclear regions is [ fe / h]@xmath71.38 @xmath8 0.15 dex , which is 0.75 @xmath8 0.17 dex lower than that of the ucds , a difference significant at 4.5@xmath13 . from the balance of two counteracting observational biases , we expect the mean metallicity of the actual nuclei to be about 0.1 - 0.2 dex lower , increasing the disagreement with the ucds to at least 0.8 dex . the metallicity difference between ucds and nuclei is slightly higher but still marginally consistent with that expected from their color difference . this implies that ucds are coeval with or slightly younger than the nuclei . we find our metallicity estimates for ucds close to but slightly below those derived for young massive clusters ( ymcs ) . although the fornax ucds match the metallicity and color of the red , metal - rich peak of the gc distribution , they are unlikely to be single , super - massive gcs due to their larger average sizes . their sizes are more consistent with those predicted for merged super - clusters . \3 . we propose that most fornax ucds are successors of merged ymcs produced in earlier star formation bursts in the course of gas rich galaxy - galaxy mergers . in contrast , for the virgo cluster we suggest that the stripping scenario is the more important ucd formation channel . four facts favor this distinction of formation channels : a ) the virgo ucds are - in contrast to the fornax ucds - on average metal - poor and of comparable color to de , n nuclei . b ) some virgo ucds possess high m / l ratios that may require dark matter , while fornax ucds have lower m / l ratios . c ) the fornax cluster has a higher current galaxy merger rate than virgo , favoring the ymc scenario . d ) the virgo cluster has a higher mass and deeper potential well , favoring the tidal stripping scenario . + in the future , increasing the available sample of fornax compact objects by a factor of two to three may allow us to check whether there is a _ continuous trend _ in metallicity with luminosity , as suggested from the color distribution . this will give important clues on the detailled formation histories of the fornax ucds . furthermore , one must increase the baseline for environmental comparisons of ucd properties by surveying galaxy clusters with a broad variety of intrinsic properties such as merger history , total mass and density . we owe thanks to the staff at las campanas observatory for their assistance in carrying out the observations and gus oemler for making the cosmos package publicly available . we would like to thank the acsfcs team for granting use of the data shown in fig . [ size ] in advance of publication . we also thank pat ct for useful discussions . sm acknowledges support by dfg project hi 855/1 . li was supported by fondap `` center for astrophysics '' . bassino , l. p. , muzzio , j. c. , & rabolli , m. 1994 , apj , 431 , 634 bastian , n. , gieles , m. , efremov , y. n. , lamers , h.j.g.l.m . 2005a , a&a in press , astro - ph/0508110 bastian , n. , emsellem , e. , kissler - patig , m. , maraston , c. 2005b , a&a in press , astro - ph/0509249 bastian , n. , saglia , r. , goudfrooij , p. et al . 2005c , submitted to a&a bekki , k. , couch , w.j . , drinkwater , m.j . , shioya , y. , 2003 , mnras , 344 , 399 bekki , k. , couch , w.j . , drinkwater , m.j . , shioya , y. , 2004 , apjl , 610 , 13 brodie j.p . , huchra j.p . , 1991 , apj 379 , 157 bruzual , g. , charlot , s. 2003 , mnras , 344 , 1000 carter , d. , mobasher , b. , bridges , t. et al . 2002 , apj , 567 , 772 ct , p. , blakeslee , j. p. , ferrarese , l. et al . ( vcs team ) 2004 , apjs , 153 , 223 ct , p. , piatek , s. , ferrarese , l. et al . ( vcs team ) 2005 , submitted to apj couture , j. , harris , w. e. , & allwright , j. w. b. 1990 , apjs , 73 , 671 de propris , r. , phillipps , s. , drinkwater , m. j. , gregg , m. d. , jones , j. b. , evstigneeva , e. , bekki , k. 2005 , apj , 623l , 105 dirsch , b. , richtler , t. , geisler , d. et al . 2003 , aj , 125 . , 1908 dirsch , b. , richtler , t. , geisler , d. et al . 2004 , aj , 127 , 2114 drinkwater m.j . , jones j.b . , gregg m.d . , phillipps s. , 2000 , pasa 17 , 227 drinkwater m.j . , gregg m.d . , colless , m. , 2001 , apj , 548l , 139 drinkwater , m.j . , gregg , m.d . , hilker , m. et al . , 2003 , nature , 423 , 519 elmegreen , b.g . , kaufman , m. , & thomasson , m. 1993 , apj , 412 , 90 evstigneeva , e.a . , gregg , m.d . , drinkwater , m. j. 2005 , to appear in the proceedings of the iauc198 `` near - field cosmology with dwarf elliptical galaxies '' , h. jerjen & b. binggeli ( eds . ) , astro - ph/0504289 ezawa , h. , fukazawa , y. , makishima , k. , ohashi , t. , takahara , f. , xu , h. , yamasaki , n. y. 1997 , apj , 490l , 33 fall , m.s . , chandar , r. , whitmore , b.c . 2005 , apj letters in press , astro - ph/0509293 fellhauer , m. , kroupa , p. , 2002 , mnras , 330 , 642 fellhauer , m. ; kroupa , p. , 2005 , mnras , 359 , 223 ferguson h.c . , sandage a. , 1989 , apj 346 , 53 forbes , d. a. , brodie , j. p. , grillmair , c. j. 1997 , aj , 113 , 1652 gebhardt , k. , kissler - patig , m. , 1999 , aj , 118 , 1526 harris , w. e. , pudritz , r. e. 1994 , apj , 429 , 177 hasegan , m. , jordn , a. , ct , p. et al . ( vcs team ) 2005 , apj , 627 , 203 held e.v . , mould j.r . , 1994 , aj 107 , 1307 hilker , m. , infante , l. , vieira , g. , kissler - patig , m. , & richtler , t. 1999 , a&as , 134 , 75 hilker , m. , mieske , s. , & infante , l. 2003 , a&al , 397 , l9 jones , j.b . , drinkwater , m.j . , jurek , r. et al . 2005 , aj in press , astro - ph/0509120 jordn , a. , ct , p. , blakeslee , j. et al . 2005a , apj in press , astro - ph/0508219 jordn et al . 2005b , in iau colloq . 198 , near - field cosmology with dwarf elliptical galaxies , ed . h. jerjen and b. binggeli ( cambridge : cambridge university press ) , in press . karick , a. , drinkwater , m.j . , gregg , m.d . 2003 , mnras , 344 , 188 kissler - patig , m. , kohle , s. , hilker , m. , richtler , t. , infante , l. , & quintana , h. 1997 , a&a , 319 , 470 kissler - patig , m. , brodie , j. p. , schroder , l. l. , forbes , d. a. , grillmair , c. j. , huchra , j. p. 1998 , aj , 115 , 105 kissler - patig , m. , jordn , a. , bastian , n. 2005 , submitted to a&a kravtsov , a. v. , gnedin , o. y. 2005 , apj , 623 , 650 kundu , a. , & whitmore , b. c. 1998 , aj , 116 , 2841larsen , s.s . 2002 , aj , 124 , 1393 li , y. , mac low , m. , klessen , r. s. 2004 , apj , 614l , 29 lotz , j. m. , miller , b. w. , ferguson , h. c. 2004 , apj , 613 , 262 makino , j. , hut , p. 1997 , apj , 481 , 83 maraston , c. , bastian , n. , saglia , r.p . , kissler - patig , m. , schweizer , f. , goudfrooij , p. 2004 , a&a , 416 , 467 mclaughlin , d.e . 2000 , apj , 539 , 618 meylan , g. , mayor , m. 1986 , a&a , 166 , 122 mieske , s. , hilker , m. , infante , l. , 2002 , a&a , 383 , 832 mieske , s. , hilker , m. , infante , l. 2004a , a&a , 418 , 445 mieske , s. , infante , l. , bentez , n. , et al . ( acs gto team ) , 2004b , aj , 128 , 1529 mieske , s. , infante , l. , hilker , m. et al . 2005a , a&a , 430l , 25 mieske , s. , hilker , m. , infante , l. 2005b , to appear in the proceedings of the iauc198 `` near - field cosmology with dwarf elliptical galaxies '' , h. jerjen & b. binggeli ( eds . ) , astro - ph/0505187 peng , e. , jordn , a. , ct , p. et al . ( vcs team ) 2005 , apj in press , astro - ph/0509654 phillipps s. , drinkwater m.j . , gregg m.d . , jones j.b . , 2001 , apj 560 , 201 pryor , c. , meylan , g. 1993 , aspc , 50 , 357 puzia , t. h. , saglia , r. p. , kissler - patig , m. , maraston , c. , greggio , l. , renzini , a. , ortolani , s. 2002 , a&a , 395 , 45 rakos k. , schombert j. , maitzen h.m . , prugovecki s. , odell a. , 2001 , aj 121 , 1974 richtler , t. , dirsch , b. , gebhardt , k. et al . 2004 , aj , 127 , 2094 rhode , k. l. , zepf , s. e. 2004 , aj , 127 , 302 schlegel , d.j . , finkbeiner , d.p . , & davis , m. 1998 , apj , 500 , 525 trager , s.c . , worthey , g. , faber , s.m . , burstein , d. , gonzalez , j.j . 1998 , apjs , 116 , 1 zinn , r. , west , m.j . 1984 , apjs , 55 , 45 lccccccc fcos 1 - 2053 & 3:38:54.05 & -35:33:33.8 & 18.06 & 1.18 & -0.52 @xmath8 0.11 & 2.46 @xmath8 0.09 & 0.136 @xmath8 0.022 + fcos 2 - 2143 & 3:38:05.04 & -35:24:09.7 & 18.93 & 1.14 & -0.69 @xmath8 0.21 & 2.32 @xmath8 0.05 & 0.084 @xmath8 0.014 + fcos 1 - 2083 & 3:39:35.90 & -35:28:24.2 & 19.12 & 1.12 & -0.85 @xmath8 0.29 & 2.86 @xmath8 0.12 & 0.051 @xmath8 0.023 + fcos 2 - 2111 & 3:38:06.29 & -35:28:58.8 & 19.23 & 1.12 & -0.90 @xmath8 0.33 & & 0.511 @xmath8 0.016 + fcos 1 - 021 & 3:38:41.96 & -35:33:12.9 & 19.70 & 1.16 & -0.58 @xmath8 0.15 & 2.05 @xmath8 0.07 & 0.108 @xmath8 0.020 + fcos 3 - 2004 & 3:39:20.50 & -35:19:14.2 & 19.73 & 1.21 & -0.49 @xmath8 0.04 & 2.02 @xmath8 0.13 & 0.178 @xmath8 0.036 + fcos 0 - 2031 & 3:38:28.97 & -35:22:55.9 & 19.87 & 1.11 & -0.38 @xmath8 0.05 & 2.43 @xmath8 0.09 & 0.210 @xmath8 0.022 + fcos 2 - 2153 & 3:38:06.48 & -35:23:03.8 & 20.04 & 1.06 & -0.70 @xmath8 0.21 & 2.46 @xmath8 0.10 & 0.079 @xmath8 0.023 + fcos 0 - 2066 & 3:38:23.23 & -35:20:00.6 & 20.09 & 1.10 & -0.61 @xmath8 0.14 & 2.61 @xmath8 0.15 & 0.108 @xmath8 0.031 + fcos 1 - 060 & 3:39:17.66 & -35:25:30.0 & 20.19 & 1.25 & -0.59 @xmath8 0.12 & 1.86 @xmath8 0.10 & 0.136 @xmath8 0.029 + fcos 0 - 2024 & 3:38:16.51 & -35:26:19.3 & 20.32 & 0.72 & -0.52 @xmath8 0.10 & 1.93 @xmath8 0.11 & 0.266 @xmath8 0.034 + fcos 0 - 2007 & 3:38:54.67 & -35:29:44.2 & 20.54 & 0.89 & -1.00 @xmath8 0.21 & 2.81 @xmath8 0.21 & 0.174 @xmath8 0.042 + fcos 2 - 2165 & 3:37:28.22 & -35:21:23.0 & 20.57 & 0.95 & -0.40 @xmath8 0.05 & & 0.176 @xmath8 0.031 + fcos 0 - 2023 & 3:38:12.70 & -35:28:57.0 & 20.66 & 1.09 & -1.13 @xmath8 0.14 & 3.49 @xmath8 0.15 & 0.067 @xmath8 0.025 + fcos 0 - 2069 & 3:38:26.71 & -35:30:07.2 & 20.67 & 0.97 & -1.50 @xmath8 0.20 & 1.56 @xmath8 0.14 & 0.027 @xmath8 0.046 + fcos 1 - 058 & 3:38:39.39 & -35:27:06.4 & 20.67 & 1.02 & -0.74 @xmath8 0.25 & 2.07 @xmath8 0.13 & 0.065 @xmath8 0.034 + fcos 0 - 2072 & 3:38:32.06 & -35:28:12.7 & 20.71 & 1.14 & -0.50 @xmath8 0.08 & 2.88 @xmath8 0.27 & 0.148 @xmath8 0.047 + fcos 2 - 2072 & 3:38:14.69 & -35:33:40.7 & 20.72 & 0.89 & -0.60 @xmath8 0.10 & 2.69 @xmath8 0.19 & 0.127 @xmath8 0.040 + fcos 0 - 2041 & 3:38:44.11 & -35:19:01.6 & 20.73 & 1.30 & -1.32 @xmath8 0.25 & 3.33 @xmath8 0.33 & 0.052 @xmath8 0.051 + fcos 0 - 2032 & 3:38:30.22 & -35:21:31.0 & 20.80 & 0.84 & -1.75 @xmath8 0.30 & 3.14 @xmath8 0.68 & 0.055 @xmath8 0.129 + fcos 2 - 086 & 3:37:46.77 & -35:34:41.7 & 20.81 & 0.90 & -0.66 @xmath8 0.17 & 3.05 @xmath8 0.23 & 0.242 @xmath8 0.041 + fcos 1 - 2115 & 3:38:49.18 & -35:21:42.1 & 20.85 & 1.20 & -1.26 @xmath8 0.25 & 2.50 @xmath8 0.18 & 0.491 @xmath8 0.039 + fcos 2 - 089 & 3:38:14.02 & -35:29:43.0 & 20.87 & 1.06 & -2.00 @xmath8 0.36 & & 0.015 @xmath8 0.044 + fcos 0 - 2074 & 3:38:35.66 & -35:27:15.5 & 20.91 & 1.04 & -2.25 @xmath8 0.27 & 2.17 @xmath8 0.12 & -0.006 @xmath8 0.033 + fcos 2 - 2100 & 3:38:00.17 & -35:30:08.3 & 20.94 & 1.05 & -0.91 @xmath8 0.24 & 1.91 @xmath8 0.16 & 0.069 @xmath8 0.046 + fcos 0 - 2092 & 3:39:05.02 & -35:26:53.9 & 20.96 & 1.13 & -1.66 @xmath8 0.38 & 1.59 @xmath8 0.14 & -0.006 @xmath8 0.047 +

in the context of measurement - based quantum computation a way of maintaining the coherence of a _ graph state _ is to measure its stabilizer operators . aside from performing quantum error correction , it is possible to exploit the information gained from these measurements to characterize and then counteract a coherent source of errors ; that is , to determine all the parameters of an _ error channel _ that applies a fixed but unknown unitary operation to the physical qubits . such a channel is generated , e.g. , by local stray fields that act on the qubits . we study the case in which each qubit of a given _ graph state _ may see a different error channel and we focus on channels given by a rotation on the bloch sphere around either the , or axis , for which analytical results can be given in a compact form . the possibility of reconstructing the channels at all qubits depends non - trivially on the topology of the graph state . we prove via perturbation methods that the reconstruction process is robust and supplement the analytic results with numerical evidence .

1512.07083

in order to construct devices that reliably process quantum information it is necessary to characterize , measure and eventually remove all the known sources of detrimental noise @xcite . it is however inevitable that a certain amount of disturbance will continue to affect the apparatus . hence it usually will be indispensable to perform quantum error correction ( qec ) @xcite to reduce the error rate to a tolerable amount . robust and scalable architectures for quantum information processing require that their elementary building blocks are readily calibrated and easily operated . better scalability with less manual intervention can be achieved by engineering devices which are capable of autonomously characterizing and correcting the error sources . also , since these systems will likely become increasingly more integrated and miniaturized @xcite , it will be difficult to have physical access to their building parts once the apparatus has been assembled . therefore , it is desirable that these devices are designed and built so that it is possible to gather data while the machine is running , and thus , on - the - fly , achieve good control and feedback on the internal functioning . in this paper we introduce a method that allows for an automated analysis of the presence of error sources and envision an _ agent _ , connected to the quantum device , which performs classical statistical analysis of the data in order to learn which noise processes are occurring on all qubits . a related case study of adapting to a unitary channel acting on an isolated qubit is given in ref . @xcite . furthermore , as we argued before , the device will necessarily be equipped with a qec code to protect the logical subspace from noise . hence we set out to investigate the following natural question : _ to what extent is it possible to characterize the error sources affecting the apparatus , having access to the syndrome measurements alone ? _ more specifically , we examine the scenario in which the agent collects the syndrome measurements and performs statistical analysis over them to detect and characterize a unitary component in the noise . therefore we avoid the necessity of adding additional sensors to monitor the performance of the quantum processing device . this approach is independent of the physical implementation of the apparatus and it can be studied through a multitude of different methods , see e.g. refs . @xcite . once the unitary component of the error has been characterized it is possible , at least in principle , to completely correct it through the application of a counter - unitary operation , which undoes the effect of this component of the error channel . as an alternative , we might allow the quantum computation to take place in a `` floating frame of reference '' , in which the logical quantum basis change in time in order to compensate for the rotation created by the unitary error channel . hereafter we will study the information that can be recovered from the most restrictive type of qec . we consider the extreme case of a qec code in which only one state is protected , i.e. , the logical subspace has dimension one . this qec method can be adopted to produce the resource state for measurement - based quantum computation ( mbqc ) @xcite . in mbqc particular classes of quantum states , e.g. the _ cluster states _ @xcite , are _ universal _ , meaning that they can be employed to perform arbitrary quantum computations just by applying local measurements . the cluster states are a subset of a more general family of quantum states associated to undirected graphs , which are called _ graph states_. a method for obtaining any graph state consists in measuring a complete set of stabilizer operators ( a.k.a . _ correlators _ ) and in applying the required corrections whenever an `` error '' is detected . the output of this procedure is a unique quantum state , i.e. the required graph state . another setting in which this qec can be employed is for measurement - based quantum repeaters @xcite , in which specific graph states are used as resources to perform entanglement purification and entanglement swapping to transmit quantum information over long distances . the outline of the paper is the following . in sec . [ sec : notation ] we give a brief introduction to graph states and set the notation ; in sec . [ sec : error_model ] we describe and motivate the error model we analyse in the paper ; in sec . [ sec : reconstruction ] we detail the effect of the unitary error channel over the stabilizer measurements statistics ; then in sec . [ sec : promise_settings ] we describe a special _ promise setting _ in which the problem becomes solvable even with limited resources . the central part of the work begins in sec . [ sec : solving_methods ] , where we provide a general method for reconstructing the error channels in the promise settings ; then in sec . [ sec : particular_classes ] we apply the general methods given in sec . [ sec : solving_methods ] to some specific classes of graph states ; finally in sec . [ sec : error_propagation ] we show that our reconstruction methods are resilient to small deviations from the model . we conclude by giving in sec . [ sec : extension ] a completely different approach which allows us to reconstruct the unitary error in complete generality , at the cost of requiring some extra resources . a _ graph state _ @xmath3 is quantum state associated with an _ undirected graph _ @xmath4 having _ vertices _ @xmath5 and _ edges _ @xmath6 , which connect pairs of vertices ; @xmath3 is defined as @xmath7 where @xmath8 , @xmath9 is the number of qubits of the graph state ( each vertex represents a qubit ) and @xmath10 denotes the ( symmetric ) controlled-@xmath11 operation between the qubits at vertex @xmath12 and @xmath13 . the _ neighbours _ of any vertex @xmath14 are given by the vertices in the set @xmath15 . we also use the notation @xmath16 . we say that two graph states are _ locally unitarily equivalent _ ( @xmath17-equivalent ) if it is possible to obtain one graph state from the other by applying unitary operations acting on individual qubits . a standard method that allows us to preserve the coherence of a graph state @xmath3 consists in repeatedly measuring the _ correlator _ operators @xmath18 @xmath19 . here @xmath20 , @xmath21 and @xmath22 denote the pauli matrices acting on the qubit in vertex @xmath12 ; we also use the shorthand @xmath23 . all these correlators commute pairwise thus all the measurements can be performed simultaneously and form a complete set of observables . we explicitly distinguish the _ correlator _ operators @xmath24 from the random variables @xmath25 which represent the _ realized measurement outcomes _ of @xmath24 . the variables @xmath25 assume values @xmath26 , corresponding respectively to the @xmath27 eigenvalues of @xmath24 . the graph state @xmath3 is the unique common eigenvector with eigenvalue equal to @xmath28 for all the correlators ; equivalently , this is the state one obtains when @xmath29 , @xmath19 . hence , when the measurements outcomes of the correlators are given by @xmath30 , the system is projected into the state @xmath31 where , for any quantum gate @xmath32 , @xmath33 when @xmath29 and @xmath34 when @xmath35 . the graph state @xmath36 is equivalent to @xmath3 for the purpose of performing mbqc . the pauli errors @xmath22 can be accounted for by a redefinition of the logical encoding of the qubits : i.e. , when @xmath35 , the sign of the state @xmath37 has to be reversed ( that is , we make the association @xmath38 and @xmath39 ) . thus we can assume , without loss of generality , that the graph state is re - prepared after the measurement of all the correlators in the standard graph state @xmath3 . as stated in the introduction , from a quantum error correction perspective these correlators provide a stabilizer code which protects only one word , the state @xmath3 . for a thorough introduction to these concepts , see ref . we consider the following model for a unitary error source that acts on a graph state . suppose that our graph state is physically implemented with spin-@xmath40 particles , each representing a qubit with the spin degree of freedom . particles might be effectively implemented as neutral atoms trapped in optical lattices , or as ion in a confining potential . however these real systems are more complex than the model we study in this paper . for example , the qubits are usually encoded within two energy levels of the atom or ion , and not directly in the angular momentum degrees of freedom of the particle . ] these particles carry an intrinsic magnetic dipole moment ; hence they are susceptible to undergo larmor precession when immersed in a magnetic field , with angular frequency equal to @xmath41 , where @xmath42 is the magnetic field intensity and @xmath43 is the gyro - magnetic ratio of the particle . however , as long as these stray fields are not subject to rapid and stochastic variations in time , it is possible to measure them and then apply a counter - unitary operation which undoes the unwanted precession . equivalently , if the precession rate of each particle is known , we may let the logical encoding of the qubits become time - dependent , in order to compensate at the logical level the physical precession of the spins of the system . notice that we focus on unitary processes since , if the error channel is given by a general master equation , the decoherence process can be characterized and mitigated , but not completely corrected . in summary , we assume that our graph state @xmath3 is affected by a space - varying static _ stray field _ which has a different strength at each qubit of the state . we further assume that all measurements are performed at discrete time steps @xmath44 , so that the action of the field on each qubit is given by a @xmath45 operation of the form @xmath46 . we remark that the results we will obtain in this paper will not depend on the specific implementation of the physical qubits , nor on the physical model of the error source ; our methods will be applicable whenever the error process acting on the qubits is dominated by a unitary local operation acting on the qubits in the system . nonetheless , for concreteness we will always refer to the above specified physical error model . consequently the word _ field _ can be assumed to designate any channel acting on single qubits and which is given by a fixed , but unknown , unitary mapping . notice that in the special case where the stray field is almost uniform and acts equally on all the qubits in the graph state , then an efficient way to achieve noise resilience is given by encoding the information in decoherence - free subspaces . the properties of these subspaces have been extensively studied elsewhere ( see for a review refs . @xcite ) and thus we will not discuss nor employ them in our work . in conclusion , at each time step and on each vertex of the graph the qubits are affected by the following local operations : @xmath47 where @xmath14 denotes the site at which the field acts , @xmath48 is proportional to the strengths of the field ( in our physical model @xmath49 ) and @xmath50 ( with @xmath51 ) is the normalized vector in @xmath52 that gives the axis around which the rotation is performed . observe that the relation @xmath51 allows us to recover @xmath53 from @xmath54 , assuming that its sign is always @xmath28 , since this can be enforced by the transformation @xmath55 , which is the identity on @xmath56 . naturally , in any realistic scenario there will be also other uncharted error sources that act on the physical system , causing some decoherence in the system . this further noise will in general render the reconstruction of the fields more difficult , but it will not qualitatively affect our results . in our setting , we are considering a qec scheme which aims to preserve the coherence of the graph state by performing the necessary correlator measurements and reading out the corresponding results . thus , the whole process can be schematized as follows : the graph state is prepared in the standard reference state @xmath3 ; we wait for a fixed time @xmath44 , during which the error channels act on the physical qubits ; then we measure the correlators , which project the graph state on some state @xmath36 , which depends on the measurement outcomes of the correlators ( i.e. on the _ measured syndromes _ ) ; we finally use the syndromes to bring back @xmath36 into the reference state @xmath3 , either physically or by changing the logical encoding of the physical qubits . implicitly , we are assuming that the state preparation , correlator readout and the final correction can be done very fast , w.r.t . the waiting time @xmath44 , and almost free of errors . moreover we will assume for the moment that the stray field is the only error source present , i.e. , no other error channel or decoherence process is acting on the qubits . we will lift this assumption in sec . [ sec : full_error_analysis ] , where we will consider the effect of having a decoherence channel acting alongside the stray field . as a first step in our analysis , we need to find the probability that , for each vertex @xmath14 , the measurement outcome @xmath25 of the correlator @xmath24 is equal to @xmath57 or @xmath58 , given that the graph state was initialized in a state @xmath3 corresponding to a graph @xmath59 , and assuming that a unitary @xmath60 has acted on each vertex . the result can be found via a straightforward computation , given in appendix [ app : error_probability ] . we define the _ probability difference _ @xmath61 $ ] as the difference between the probability of getting a `` correct '' outcome ( @xmath29 ) and the probability of getting an `` error detected '' outcome ( @xmath35 ) @xmath62 then we obtain the following formula for @xmath63 : @xmath64 where @xmath65 and @xmath66 are the parameters that define the action of the field on vertex @xmath12 as given by eq . . notice that from the probability differences alone it is impossible to reconstruct the correct sign of @xmath67 and @xmath68 , as eq . depends on these only through @xmath69 , nor the correct value of @xmath70 in the set @xmath71 , since we can only access @xmath72 . in principle , these ambiguities can be easily resolved . for example , it is sufficient to vary the delay time @xmath44 to linearly vary the parameters @xmath70 , which can then be discriminated through the functions @xmath73 . however , these methods rely on performing some extra operations , not included in the basic stabilizer code , and therefore we will not consider them in the following . in sec . [ sec : repeaters ] we will exploit another resource , which is automatically included in mbqc , in order to reconstruct the fields . that is , we will use that some graph states are @xmath17-equivalent even if they correspond to different graphs . @xmath17-equivalent graph states have the same power for performing mbqc , assuming that arbitrary local measurements can be performed on the qubits ; nonetheless the information that can be recovered from stabilizer measurements is very different . hence arises the possibility of exploiting different graph states with equivalent computational power in order to better characterize the error sources . the problem as specified so far is underdetermined , since we need to determine @xmath74 parameters , i.e. the field directions and intensities , from just @xmath75 parameters , i.e. from the measured probability differences . thus we need to modify the problem to make it possible to reconstruct the stray fields . we will therefore first consider simplified settings , in which the fields fulfill a condition ( a _ promise _ ) that reduce the number of parameters to be estimated from @xmath74 to @xmath75 thus permitting to estimate them through the measurement of the correlators stabilizing the graph state . we will then consider a different scenario in sec . [ sec : extension ] , where we will extend the set of stabilizer measurements in order to collect information on all the @xmath74 parameters . from now on , we will assume that the field is always globally pointing in a known and well - determined direction , but has unknown site - dependent intensity . specifically , we investigate the cases in which the field are parallel to one of the cartesian axes ; under these assumptions , eq . simplifies considerably : 1 . * z - field * @xmath76 $ ] + the field is always aligned in the @xmath2 direction @xmath77 2 . * x - field * @xmath78 $ ] + the field is always aligned in the @xmath0 direction @xmath79 3 . * y - field * @xmath80 $ ] + the field is always aligned in the @xmath1 direction @xmath81 here we will present a method that allows us to infer the stray fields , knowing the probability differences @xmath82 , in the promise settings specified above . we will show that not in all circumstances it is possible to determine all field intensities @xmath83 , and this impossibility depends non - trivially on the connectivity structure of the graph state . to determine the fields we need to solve a system of @xmath75 polynomial equations ( one for each @xmath63 ) in @xmath75 variables ( one for each field intensity @xmath83 ) of degree at most @xmath84 , where @xmath84 is the maximum number of neighbours of any vertex in the graph . in general , a system of polynomial equations can be solved via computation of the grbner basis associated to the polynomials . there exist well - known algorithms that compute grbner bases @xcite , but the time complexity in the worst case scenario grows doubly - exponentially in @xmath75 @xcite . however , with the method that we will introduce the particular polynomial equations arising from eqs . can be solved efficiently . notice in particular that the case of eq . is already trivially solved , as one reads out directly the ( cosine of the ) intensity of the field from the measured rates . the other two cases , eqs . , will be discussed in the following sections . consider the settings in which the stray field is parallel to either the @xmath2 , @xmath0 or @xmath1 axis . we can write eqs . as @xmath85 with @xmath86 , and setting @xmath87 for the @xmath11-field case , @xmath88 for the @xmath89-field case , and @xmath90 for the @xmath91-field case . they are polynomial equations in the variables @xmath83 , in which each polynomial has only two terms , i.e. , the monomials appearing on the left hand and on the right hand side . to solve these we take the logarithms of both sides of the equations . in general the cosines of the fields ( @xmath83 ) and the probability differences ( @xmath63 ) have support in the interval @xmath92 $ ] . thus we have to use the logarithm as a complex - valued function ; more precisely , we define the logarithm as taking values on the riemann surface @xmath93 : @xmath94 where @xmath95 is the real logarithm ( mapping @xmath96 in @xmath97 ) , @xmath98 and the riemann surface @xmath93 is the complex plane modulo identification of numbers that differ by an integer multiple of @xmath99 ; we also denote by @xmath100[0pt]{$\scriptstyle 2\pi i$}}}{=}}$ ] the equality on this riemann surface . if we restrict @xmath101 to be a real number , then @xmath102 is in the set @xmath103 . with this definition the logarithm is a single - valued function and the multiplication rule applies : @xmath104[0pt]{$\scriptstyle 2\pi i$}}}{=}}\ln(z_1 ) + \ln(z_2 ) \qquad \forall \ z_1,z_2 \in \mathds{c}^\ast \,.\ ] ] notice that on the riemann surface @xmath93 the division is not uniquely defined . in particular , when performing a division by @xmath105 we get @xmath106 different valid results : @xmath107[0pt]{$\scriptstyle 2\pi i$}}}{=}}\left\lbrace \ , \left . \frac{z}{m } + \frac{2k\pi}{m}i \ \right| \ , k \in \{0,\ldots,|m|-1\ } \ , \right\rbrace \,.\ ] ] thus , after taking the logarithm of both sides of eqs . , we obtain the following linear equations : @xmath108[0pt]{$\scriptstyle 2\pi i$}}}{=}}\left ( \begin{array}{c } \ln { \delta \text{p}}_1 \\ \ln { \delta \text{p}}_2 \\ \vdots \\ \ln { \delta \text{p}}_n \end{array } \right ) \,.\ ] ] the adjacency matrix @xmath109 of a graph @xmath110 is a symmetric matrix defined as @xmath111_{a , b } : = \begin{cases } 1 & \text{if } \ ( a , b ) \in e \\ 0 & \text{if } \ ( a , b ) \notin e \end{cases } .\ ] ] we also define @xmath112 for @xmath86 as @xmath113 where @xmath114 denotes the @xmath115 identity matrix . the problem thus has been reduced to solving a linear system of the form @xmath116[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w}$ ] , with @xmath117 and @xmath118 . from here , provided that the matrix @xmath112 is non - singular , the ( logarithm of the cosines of the ) fields can be reconstructed via linear algebra . however , in this context even when @xmath112 is non - singular it is not formally correct to write the solution as @xmath119 , since the system @xmath116[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w}$ ] can have multiple solutions . this happens because the entries of @xmath112 are taken as elements of @xmath93 , thus divisions by integer number have to be performed according to eq . . therefore , in order recover all the valid solutions to the linear system we need to employ eq . whenever performing a division during the solving algorithm . we adopt a modified version of gauss elimination algorithm which allows us to recover these multiple solutions . since the entries of @xmath112 are all integers , it is possible to bring @xmath112 into the form of an upper triangular matrix @xmath120 , applying only a sequence of the following _ elementary gauss operations _ on @xmath112 : rule 1 : : multiply a row of @xmath112 by @xmath121 ; rule 2 : : permute two rows of @xmath112 ; rule 3 : : add to the @xmath122-th row of @xmath112 an integer multiple of the @xmath123-th row , with @xmath124 . these operations can in fact be used to compute the _ hermite normal form _ @xcite of @xmath112 . that is , it is always possible to write @xmath112 in the form : @xmath125 where @xmath126 and @xmath120 are matrices with integer entries , @xmath126 unimodular ( @xmath127 ) and @xmath120 upper - triangular ; for more details see , e.g. , ref . notice that , @xmath126 being unimodular , @xmath128 is also unimodular and has integer entries ; moreover , @xmath128 is obtained via multiplication of the matrices that represent the elementary gauss operations . hence the solutions of @xmath116[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w}$ ] are in one - to - one correspondence with the solutions of @xmath129[0pt]{$\scriptstyle 2\pi i$}}}{=}}p^{-1 } \boldsymbol{w } \,.\ ] ] thus eqs . have been brought into the form @xmath130[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w}'$ ] , with @xmath131 , and @xmath120 upper - triangular . if @xmath112 is non - singular , then all the diagonal entries @xmath132 of @xmath120 are different from @xmath57 . therefore the solutions can be recovered finding the @xmath133 solutions ( see eq . ) to the equation @xmath134[0pt]{$\scriptstyle 2\pi i$}}}{=}}w'_n$ ] , then substituting the obtained values for @xmath135 in eq . , then solving for @xmath136 and so forth . that is , having computed the possible values for @xmath137 we obtain @xmath138 by solving the equation @xmath139[0pt]{$\scriptstyle 2\pi i$}}}{=}}w'_l - \sum_{j = l+1}^n q_{lj } v_j$ ] . in total , the number of complex solutions is given by @xmath140 . an equivalent way of looking at this operation , which will become useful in the error analysis in sec . [ sec : error_propagation ] , is to consider the vector @xmath141 as taking @xmath142 different values , and then computing the corresponding solutions for @xmath143 with standard linear algebra over the complex numbers . explicitly , we need to solve @xmath142 different linear equations in the form @xmath144 over @xmath145 , in which the known terms are given by @xmath146 where @xmath147 is the unit vector with a @xmath58 in position @xmath122 , and @xmath148 , @xmath149 . in conclusion , we face two possible scenarios : 1 . the matrix @xmath112 is singular ; the system is either inconsistent or underdetermined . the matrix @xmath112 is non - singular ; the number of complex solutions is given by @xmath150 . finally , once the solutions @xmath143 to the system have been computed , the corresponding values of the parameters @xmath83 can be obtained via element - wise exponentiation of @xmath143 . now consider @xmath151 ; the components of @xmath152 are logarithms of real numbers , thus the imaginary part must be an integer multiple of @xmath153 ( i.e. the imaginary component is either @xmath57 or @xmath154 on @xmath155 ) . after the gauss elimination algorithm , we obtain another vector @xmath156 whose entries are integer linear combinations of those of @xmath152 ; thus , the entries of @xmath141 still have an imaginary part which is an integer multiple of @xmath153 . take @xmath138 as a complex solution of an equation of the form @xmath139[0pt]{$\scriptstyle 2\pi i$}}}{=}}w'_l - \sum_{j = l+1}^n q_{lj } v_j$ ] ; if @xmath157 is odd , then among these solutions only one has an imaginary part which is an integer multiple of @xmath153 ; if @xmath157 is even , then among these solutions only two have imaginary parts which are integer multiples of @xmath153 . repeating this argument for all entries of @xmath143 we get the number of allowed solutions is @xmath158 , where @xmath159 is the number of even diagonal entries in @xmath120 . the requirement that all the probabilities and the ( cosines of the ) field strengths must be real thus restricts the allowed solutions to just a small subset of all possible complex solutions . there is still another requirement that should be enforced : the parameters @xmath83 and @xmath63 have to lie in the range @xmath160 $ ] ; this condition further restricts the number of legitimate solutions . notice that , when we enforce this requirement , the number of acceptable solutions can not be expressed in terms of properties of the adjacency matrix alone ( @xmath112 ) , since this number does depend on the data actually collected ( @xmath152 ) . in this section , we will explicitly determine whether a graph state allows a complete reconstruction of the fields acting at each site assuming that it is parallel to either the @xmath0 , @xmath1 or @xmath2 direction for some relevant classes of graphs . we will consider at the beginning graphs which represent one - dimensional lattices that is , _ linear chains_. in these graphs each vertex is connected to at most two neighbours . afterwards , we will show how to extend those results to two - dimensional ( and higher dimensional ) square lattices . finally , we will give an application for states used in quantum repeaters and measurement - based approaches to quantum error correction . a graph represents a _ closed linear chain _ if all vertices are connected to other two vertices in the graph ; it represents a _ open - ended linear chain _ if there are two endpoints which are connected to only one other vertex . the adjacency matrix of an open - ended linear chain with @xmath161 vertices is a @xmath162 matrix given by : @xmath163_{a , b } = \begin{cases } 1 & \text{if } |a - b|= 1 \\ 0 & \text{otherwise } \end{cases } \qquad \forall \ , a , b \in [ m]\ ] ] in which we introduce the notation @xmath164 : = \{1,\ldots , m\}$ ] and we have numbered the vertices from @xmath58 to @xmath123 in the natural way ( from one endpoint to the other ) . thus @xmath165 and @xmath166 are both tridiagonal toeplitz matrices and where a toeplitz matrix @xmath167 is a constant - diagonal matrix , i.e. @xmath168_{a , b } = t_{(a - b)}$ ] , @xmath169 $ ] . their determinants can be given as closed - form expressions @xcite : @xmath170 therefore , given the solution methods presented in sec . [ sec : solving_methods ] , it is evident that the system is underdetermined when @xmath123 is even in the case of @xmath89-fields , and when @xmath171 for @xmath91-fields . in all other cases there is exactly one ( complex ) solution to the equations . for a closed linear chain with @xmath123 vertices the adjacency matrix is given by : @xmath172_{a , b } = & \begin{cases } 1 & \text{if } |a - b| \in \{1,m-1\ } \\ 0 & \text{otherwise } \end{cases } \\ & \qquad \ \forall \ , a , b \in [ m ] \ , , \nonumber\end{aligned}\ ] ] and thus @xmath173 and @xmath174 are circulant matrices where a circulant matrix @xmath175 is defined by the relation @xmath176_{a , b } = c_{(a - b ) \text { mod } m}$ ] , @xmath169 $ ] . also in this case it is possible to give closed - form expressions for their determinants @xcite , but we have to take into account that @xmath177 and @xmath178 are special cases : @xmath179 therefore , the problem is underdetermined when @xmath177 or @xmath180 in the case of @xmath89-fields , and when @xmath178 or @xmath181 in the case of @xmath91-fields . in the other cases , there are a number of complex solutions equal to @xmath182 . for the @xmath91-field case , since the non - zero determinant of @xmath174 is equal to @xmath183 or @xmath184 , there is only one real solution for the fields . for the @xmath89-field case , when @xmath185 , @xmath186 , so there are two real solutions ; and when @xmath185 , @xmath187 , and it can be verified that the hermite normal form @xmath120 has two diagonal entries equal to @xmath188 , so there are four admissible solutions . a summary of these results can be found in table [ table1 ] . for reference , we also give the eigenvalues of the adjacency matrices @xmath189 and @xmath190 for @xmath191 : @xmath192 in which we have introduced the function @xmath193 to deal with the special cases arising when @xmath177 or @xmath178 : @xmath194 the results above will turn out useful in the next section . ll * open - end linear chain * & * closed linear chain * + & + * adjacency matrix : * & * adjacency matrix : * + tridiagonal toeplitz matrix & tribanded circulant matrix + @xmath195 & @xmath196 + * eigenvalues : * & * eigenvalues : * + @xmath197 & @xmath198 + @xmath199 & @xmath200 + + * x - field : * & * x - field : * + @xmath201 & @xmath202 + * y - field : * & * y - field : * + @xmath203 & @xmath204 here we describe the method for solving the problem for two dimensional square lattices . the same constructions can be applied straightforwardly in an arbitrary number of dimensions , so in the end we will present the formulas also for the general multidimensional case . as a first step , we give the expression for the adjacency matrix of a graph representing a two - dimensional square lattice , having @xmath205 rows in one direction and @xmath206 columns in the other direction , thus having in total @xmath207 vertices . we number the vertices according to a scan - line pattern , i.e. we assign to the vertex in position @xmath208 , with @xmath209 $ ] a row index and @xmath210 $ ] a column index , the value @xmath211 . with this numbering convention , the adjacency matrices @xmath109 of planar lattices take the form of block - toeplitz matrices . that is , there are @xmath212 blocks , each one consisting of @xmath213 elements , and the blocks which lie on the same diagonal are all equal ; the blocks themselves are toeplitz matrices . for a square lattice with open boundaries in both directions , i.e. for a _ planar graph _ , the adjacency matrix takes the following tridiagonal block - matrix form : @xmath214_{a , b } = & \begin{cases } a^{open}_{m_2 } & \text{if } |a - b| = 0 \\ { \mathds{1}}_{m_2 } & \text{if } |a - b| = 1 \\ 0_{m_2 } & \text{otherwise } \end{cases } \\ & \qquad \qquad \quad \forall \ a , b \in [ m_1 ] \nonumber\end{aligned}\ ] ] where here @xmath215 are block indices , the blocks are of size @xmath213 , and @xmath216 is given by eq . . now we consider another family of graphs , which are obtained from the above planar square lattice by connecting the vertices of the @xmath206-th column to the corresponding vertices in the first column ; so we obtain a graph which is topologically a cylinder . then the adjacency matrix takes the following block - matrix form : @xmath217_{a , b } = & \begin{cases } a^{close}_{m_2 } & \text{if } |a - b| = 0 \\ { \mathds{1}}_{m_2 } & \text{if } |a - b| = 1 \\ 0_{m_2 } & \text{otherwise } \end{cases } \\ & \qquad \qquad \quad \forall \ a , b \in [ m_1 ] \nonumber\end{aligned}\ ] ] where @xmath218 is given by eq . . finally , we consider the family of graphs obtained from the cylindrical graph by connecting the vertices of the @xmath205-th row to the corresponding vertices of the first row ; so we obtain a graph which is topologically a torus . the adjacency matrix takes the following block - matrix form : @xmath219_{a , b } = & \begin{cases } a^{close}_{m_2 } & \text{if } |a - b| = 0 \\ { \mathds{1}}_{m_2 } & \text{if } |a - b| \in \{1,m_1 - 1\ } \\ 0_{m_2 } & \text{otherwise } \end{cases } \\ & \qquad \qquad \quad \forall \ a , b \in [ m_1 ] \nonumber\end{aligned}\ ] ] with the same conventions as above . as usual , we have @xmath220 and @xmath221 for all the cases above , where @xmath114 is the @xmath222 identity matrix . now we will show how to obtain the eigenvalues of the above matrices in closed form . once the eigenvalues are given , it is straightforward to tell whether all information about the fields can be recovered or not , by searching for eigenvalues equal to zero . if these can be found , it means that the adjacency matrix is singular and not all the information can be recovered . the number of zero eigenvalues is the _ rank defect _ of the adjacency matrix , and is a measure of the amount of information that one can not reconstruct . .eigenvalues of tridiagonal toeplitz matrices and of tribanded symmetric circulant matrices . [ cols= " < , < " , ] we will now deal with the full error analysis , in which the stray fields satisfy the constraints only approximately . that is , we analyse the case in which the fields are _ approximately _ aligned to one of the three cartesian axes @xmath0 , @xmath1 or @xmath2 . in all three situations we have to solve an approximate linear systems of the form @xmath223[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w } + \delta \boldsymbol{w}\ ] ] where @xmath224 is a vanishing function for @xmath225 , with @xmath226 a global parameter which bounds the maximum misalignment of the rotation axis from one of the cartesian axes . we call @xmath227 a solution of this perturbed system , and with @xmath143 a solution of the original unperturbed system . in appendix [ app : continuity ] we show that , under some regularity assumptions , the magnitude of the perturbation in the solution is bounded by a continuous function of @xmath228 and of the rotation axis misalignments . in principle , one should consider that in any realistic implementation there will be some incoherent noise acting alongside the local unitary rotations induced by the stray fields . equivalently , we might consider the case in which the stray fields @xmath229 are not fixed vectors , but they vary in time . if these temporal variations are fast and stochastic , then it can be imagined that @xmath229 are random variables drawn according to some probability distribution . the simplest model , but also the worst case scenario , for this source of incoherent noise is given by the uniformly depolarizing channel @xmath230 where @xmath231 is a single qubit density matrix and @xmath232 . it is not necessary to specify in which order the depolarizing channel and the local unitary rotations are applied , since they commute . from eq . one immediately obtains the difference in probability between the measurement outcomes when both local unitary error sources and depolarizing noise @xmath233 act on the system : @xmath234 where @xmath63 is given by eq . . this noise affects the results , favouring smaller values of @xmath83 ; for example , in the @xmath11-field case all @xmath83 are just rescaled by a factor @xmath235 . we point out that , for the @xmath89 or @xmath91-field cases , for some of the qubits in the graph state the reconstruction algorithm may exhibit a natural resilience to the inaccuracy introduced by the uniformly depolarizing noise . in fact , from the expression in eq . it can be seen that , to account for the action of the depolarizing noise , the differences in probabilities have to be multiplied by @xmath235 for all qubits . thus the vector @xmath236 has to be added to the known term in the linear equation , whose entries are the logarithm of the differences in probabilities . by linearity , the effect of the depolarizing noise on @xmath152 ( the vector of the estimators in logarithmic variables ) is simply given by the addition of @xmath237 . some of the entries of the vector @xmath238 might be zero , implying that the reconstruction of the fields for those entries is unaffected by the depolarizing noise . we have tested through numerical simulation the performance of our reconstruction method when all the three noise factors considered above are present , i.e. , the effect of the _ finiteness _ of the number of measurements of the correlators , the _ misalignment _ of the field from the promised direction , and the independent action of a _ depolarizing channel _ on the qubits . these simulations provide evidence that , indeed , when all the noise factors are small ( and the numbers of measurements of each correlator is large ) then the accuracy in the reconstruction improves and , in principle , the inaccuracy can be brought arbitrarily close to zero . we define the _ reconstruction error _ @xmath239 on any vertex @xmath14 as @xmath240 in which @xmath241 is the true value of parameter and @xmath242 is a regularization of the output @xmath243 of the reconstruction algorithm . the regularization consists in setting @xmath244 when @xmath245 and @xmath246 when @xmath247 , so that the regularized value always lies in the interval @xmath160 $ ] . an excerpt of the _ reconstruction errors _ we have obtained , in the case of an open - ended chain with ten vertices , is given in fig . [ fig2 ] . a general trend that can be inferred from the numerical data is that since the correlation among the parameters is not taken into account the reconstruction error is minimal in the @xmath11-field case , is larger in the @xmath89-field case , and is still larger in the @xmath91-field case . this feature is expected , since the degree of the polynomial equations to be solved is minimal in the @xmath11-field case , larger in the @xmath89-field case and still larger in the @xmath91-field case . this implies that the cross - correlations between the estimators of the fields are maximal in the @xmath91-field case and they are minimal in the @xmath11-field case . another feature that can be inferred from the data is that the precision of the estimation of the fields varies across the vertices of the graph and depends on the global connectivity of the graph itself . notice that to produce the data shown in fig . [ fig2 ] we have assumed that the model of the disturbance is the same for all ten vertices in the chain . thus for symmetry reasons we expect that , averaging over all possible intensities of the stray field , for a open - ended chain with ten qubits the inaccuracy in the reconstruction of the fields on vertex @xmath12 and @xmath248 are the same . also , notice that the reconstruction error decreases approximately as @xmath249 for small values of @xmath250 , while for larger values of @xmath250 the reconstruction error approaches a non - zero asymptotic value , which is due to residual uncertainties introduced by the depolarizing noise and stray field misalignment . finally , we remark that at some of the vertices the reconstruction of the fields shows resilience to the depolarizing noise , for the reason given earlier . explicitly , this happens for qubits on the vertices @xmath251 for the @xmath89-field case and for qubits on the vertices @xmath252 for the @xmath91-field case , as can be easily checked computing @xmath238 for these two cases . consequently , it can be seen in fig . [ fig2 ] that these qubits are almost insensitive to the magnitude @xmath253 of the depolarizing noise , provided that @xmath254 . when @xmath255 , then the effect of the finiteness of the number of measurement @xmath250 renders unreliable the reconstruction of the fields on those qubits . a completely different approach to estimating the stray fields consists in adopting different sets of stabilizer measurements in order to collect more information . in this way we do not need to assume that the fields are always aligned in a known direction . using a different set of stabilizer measurements will produce a different graph state ; but this is not an issue , as long as the functionality of this family of graph states is the same of that of the reference graph state . explicitly , we consider the following scenario . in the defining equation , the correlators are defined in terms of the product of pauli-@xmath89 and pauli-@xmath11 matrices . actually , many other correlators will yield graph states which have the same computational capabilities as the reference graph state . for example , we could as well use correlators in the form @xmath256 ( or in the form @xmath257 ) . to see that these stabilizers define a graph state which is computationally equivalent to the original one it is enough to observe that these are obtained by a redefinition of the pauli matrices , @xmath258 ( or @xmath259 ) , which is a transformation that preserves the commutation relations between the matrices and thus all the associated algebraic properties . correspondingly , the local measurements necessary to perform mbqc have to be modified according to the new pauli basis . these considerations apply to any transformation @xmath260 , in which @xmath261 is any right - oriented orthonormal basis of @xmath52 . we call these _ logical pauli matrices_. correspondingly , eq . becomes : @xmath262 thus , using three different graph states , associated to three different logical pauli bases it is in principle possible to reconstruct all the parameters of the stray fields . notice that in general there will be many physically acceptable solutions to a system of @xmath74 polynomial equations with @xmath74 unknowns . but then using @xmath263 logical pauli bases , from the corresponding graph states one obtains an overdetermined system of @xmath264 equations with @xmath74 unknowns , which almost certainly admits only one solution ; moreover , having more independent equations usually provides a reconstruction which is more robust in presence of small uncertainties . explicitly , we look for an approximate solution to the equations through numerical minimization of the following cost function : @xmath265 here @xmath266 is computed from eq . and it is a function of the field intensities @xmath267 and directions @xmath268 at the vertices in @xmath269 , and of the choice of the logical pauli basis @xmath270 ( which can be identified with an orthonormal basis in @xmath52 ) . with @xmath271 we denote the value for the probability difference estimated from syndrome measurements for vertex @xmath12 using the logical pauli basis @xmath270 . an example of the result that can be obtained via this method is reported in fig . in this paper we have shown how to use the information collected from a set of stabilizer measurements in order to extract information about a unitary error process acting on a graph state . since the stabilizer measurements are non - local operations , each of them depends on the errors that act on several different sites of the graph state . therefore the measurement outcomes of these stabilizers are correlated . we have studied in depth the case in which the local unitary errors are given by rotations aligned to one of the cartesian axes . this condition is needed to make the problem solvable , i.e. , we match the number of parameters to be reconstructed to the number of measured parameters ; moreover this restriction allows us to find efficient reconstruction methods . a surprising result arising from the fact that the stabilizer measurement outcomes are correlated is that in some circumstances it is not possible to reconstruct all the parameters of the unitary error even under the assumption that we can measure the stabilizers an arbitrary number of times . the possibility of completely reconstructing the error turns out to depend non - trivially on the connectivity structure of the graph state . specifically , in order to determine the intensity of local unitary rotations around the @xmath0 axis , the adjacency matrix @xmath109 associated to a graph state @xmath110 has to be a full - rank matrix , whereas to reconstruct rotations around the @xmath1 axis the matrix @xmath272 has to be full - rank , while rotations around the @xmath2 axis can always be reconstructed . we have ascertained for some classes of graph states whether it is possible to reconstruct all the errors . hence we have studied the properties of the adjacency matrices of _ linear chains _ ( with open and closed boundary conditions ) , _ square lattices _ in arbitrary number of dimensions , and ghz states ( these represented either as _ fully - connected _ graphs or as _ star _ graphs ) . finally we have shown that the performed reconstruction is resistant to various kinds of imperfections . in particular we have investigated the effect of a finite number of measurements of stabilizers , of the misalignment of the rotation axis and of the action of a depolarizing noise alongside the action of the unitary channel . in each of these cases , as long as the imperfections are small , the precision in the determination of the unitary channel is only mildly affected . we acknowledge support from the austrian science fund ( fwf ) through the sfb foqus : f4012 , and the templeton world charity foundation grant twcf0078/ab46 . in this appendix we derive eq . . this equation gives the difference between the probability of getting @xmath273 and @xmath274 when measuring the correlator @xmath275 on a graph state which is subject to local noise of the form @xmath276 . we introduce the notation @xmath277 ; observe that @xmath24 has spectrum @xmath27 , therefore there is a unitary mapping @xmath278 such that @xmath279 , where @xmath280 with @xmath281 is the projector on the @xmath282-eigenspace of @xmath24 . hence we get that the difference in the probabilities is given by the expectation value of @xmath24 on @xmath283 : @xmath284 now we can compute : @xmath285 \left(x_a \bigotimes_{b \in \mathcal{n}_a } z_b \right ) { | g \rangle } \nonumber\\ & = \left[\prod_{c \in ( v \setminus \mathcal{n}_a ' ) } e^{-i \lambda_c \hat{\boldsymbol{n}}_c \cdot \boldsymbol{s}_c/2 } \right ] \cdot \nonumber\\ & \ \cdot \left ( e^{-i \frac{\lambda_a}{2 } \hat{\boldsymbol{n}}_a \cdot ( x_a \boldsymbol{s}_a x_a ) } \prod_{b \in \mathcal{n}_a } e^{-i \frac{\lambda_b}{2 } \hat{\boldsymbol{n}}_b \cdot ( z_b \boldsymbol{s}_b z_b ) } \right ) { | g \rangle}\end{aligned}\ ] ] where @xmath286 , and in which we have used that @xmath287 and that @xmath288 when @xmath32 is a unitary operator . hence : @xmath289 actually , we only need to compute the part proportional to the identity in the operator in the last line of eq . , since @xmath290 for any non - trivial graph @xmath110 , i.e. for graphs in which there is no vertex which is completely disconnected from all other vertices ( no isolated vertices ) . indeed , for the pauli @xmath11 operation on the vertex @xmath12 we have : @xmath291 in which we have used twice @xmath287 and the anti - commutation relation between @xmath22 and @xmath24 . an analogous procedure shows that @xmath292 , since also @xmath21 and @xmath24 anti - commute . then , for the pauli @xmath89 operation on vertex @xmath12 we have ( provided that @xmath12 has at least one neighbour ) : @xmath293 finally , we can evaluate the part proportional to the identity in eq . applying the exponentiation formula for pauli matrices : @xmath294 together with the identity : @xmath295 for @xmath296 , obtaining : @xmath297\nonumber\\ & \ \ \,\prod_{b \in \mathcal{n}_a } { \mathds{1}}_b \left [ \cos^2 \frac{\lambda_b}{2 } + \left ( ( n_{z}^b)^2 - ( n_{x}^b)^2 - ( n_{y}^b)^2 \right ) \sin^2 \frac{\lambda_b}{2 } \right ] { | g \rangle } \nonumber\\ & = \ \left [ ( n_{x}^a)^2 + \left ( 1 - ( n_{x}^a)^2 \right ) \cos \lambda_a \right]\nonumber\\ & \ \ \prod_{b \in \mathcal{n}_a } \left [ ( n_{z}^b)^2 + \left ( 1 - ( n_{z}^b)^2 \right ) \cos \lambda_b \right ] \,.\end{aligned}\ ] ] this is exactly the expression given in eq . , once one writes @xmath65 . suppose that the rotation axes of the rotation generated by the stray field is almost parallel to @xmath0 , @xmath1 or @xmath2 , but is slightly misaligned . in these three cases we can rewrite the parameters @xmath298 and @xmath299 ( which together define the direction in which the field points for each @xmath14 ) as follows : @xmath300 where each @xmath301 and @xmath302 is assumed to be small . then we write eq . for these three cases with the probabilities @xmath303 replaced by the measured rate differences @xmath304 , we take the logarithm of both sides , we expand @xmath305 assuming that @xmath306 and @xmath307 , finally obtaining : @xmath308[0pt]{$\scriptstyle 2\pi i$}}}{=}}\ \sum_{b \in \mathcal{n}_a } \ln \beta_b \ + \nonumber \\ & + \ \left[\epsilon_a(\beta_a-1 ) + \sum_{b \in n_a } \tilde{\epsilon}_b\frac{1-\beta_b}{\beta_b } \right ] + \mathcal{o}(\epsilon^2)\,,\nonumber\\ \ln \delta r_a & { \underset{{\raisebox{2pt}[0pt][0pt]{$\scriptstyle 2\pi i$}}}{=}}\ \ln \beta_a + \sum_{b \in \mathcal{n}_a } \ln \beta_b \ + \nonumber \\ & + \ \left[\epsilon_a \frac{1-\beta_a}{\beta_a } + \sum_{b \in n_a } \tilde{\epsilon}_b\frac{1-\beta_b}{\beta_b } \right ] + \mathcal{o}(\epsilon^2)\,,\nonumber\\ \ln \delta r_a & { \underset{{\raisebox{2pt}[0pt][0pt]{$\scriptstyle 2\pi i$}}}{=}}\ \ln \beta_a \ + \nonumber \\ & + \ \left[\epsilon_a\frac{1-\beta_a}{\beta_a } + \sum_{b \in n_a } \tilde{\epsilon}_b ( \beta_b-1 ) \right ] + \mathcal{o}(\epsilon^2)\,,\end{aligned}\ ] ] for the @xmath89-field , @xmath91-field and @xmath11-field case respectively . in all three cases we need to solve an approximate linear systems in the form @xmath309[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w } + \delta \boldsymbol{w}\ ] ] where we use the notation @xmath310 , @xmath311 and where @xmath312 is the solution of the perturbed system of equations , being @xmath143 a solution of @xmath116[0pt]{$\scriptstyle 2\pi i$}}}{=}}\boldsymbol{w}$ ] , i.e. a solution of the original unperturbed system . furthermore , we have introduced a global scale parameter @xmath226 given by @xmath313 . this perturbed ( nonlinear ) system can in principle be solved recursively , through the following relations : @xmath314 where we use the equality symbol @xmath315 rather then @xmath100[0pt]{$\scriptstyle 2\pi i$}}}{=}}$ ] as we will focus on a particular solution of the system . we have to show that the solution of the perturbed system varies continuously with respect to the input data and field directions , i.e. @xmath316 when @xmath317 and @xmath225 . initially we have to check that the vector @xmath143 , obtained from solving the unperturbed linear system , satisfies the constraints @xmath318 where @xmath319 . then it is straightforward to verify that for @xmath320 given by any of the three expressions in square brackets in eq . , the following bound holds : @xmath321 where @xmath322 , i.e. @xmath84 is the maximum number of connections for any vertex in the graph . also , using the triangle inequality , it can be verified that : @xmath323 and , provided that @xmath324 , we also have @xmath325 . then the recursive relation allows us to bound the norm of the @xmath326-th iterative solution @xmath327 , provided that @xmath328 where @xmath329 is the lambert @xmath330-function , i.e. @xmath331 is the solution of @xmath332 . this can be shown bounding @xmath333 with @xmath334 , in which again @xmath335 is given through a recurrence relation : @xmath336 and we obtain @xmath337 through the relation @xmath338 ; this can be solved when eq . is satisfied , and if this is the case we obtain : @xmath339 with @xmath340 . notice that @xmath341 when @xmath342 , which can be enforced if @xmath225 and @xmath317 . in conclusion we have shown that the perturbed solution @xmath227 ( which has to exist for physical consistency ) has a distance from the unperturbed solution @xmath143 which is at most @xmath337 , and that @xmath337 goes to zero when both the magnitude of the misalignment and the inaccuracy in measuring the rates go to zero . t. moroder , m. kleinmann , p. schindler , t. monz , o. ghne , and r. blatt , `` certifying systematic errors in quantum experiments , '' http://dx.doi.org/10.1103/physrevlett.110.180401[phys . rev . * 110 * , 180401 ( 2013 ) ] . d. nigg , m. mller , e. a. martinez , p. schindler , m. hennrich , t. monz , m. a. martin - delgado , and r. blatt , `` quantum computations on a topologically encoded qubit , '' http://dx.doi.org/10.1126/science.1253742[science * 345 * , 302305 ( 2014 ) ] . j. kelly , r. barends , a. g. fowler , a. megrant , e. jeffrey , t. c. white , d. sank , j. y. mutus , b. campbell , yu chen , z. chen , b. chiaro , a. dunsworth , i .- c . hoi , c. neill , p. j. j. omalley , c. quintana , p. roushan , a. vainsencher , j. wenner , a. n. cleland , and john m. martinis , `` state preservation by repetitive error detection in a superconducting quantum circuit , '' http://dx.doi.org/10.1038/nature14270[nature * 519 * , 6669 ( 2015 ) ] . b. brandsttter , a. mcclung , k. schppert , b. casabone , k. friebe , a. stute , p. o. schmidt , c. deutsch , j. reichel , r. blatt , and t. e. northup , `` integrated fiber - mirror ion trap for strong ion - cavity coupling , '' http://dx.doi.org/10.1063/1.4838696[rev . instrum . * 84 * , 123104 ( 2013 ) ] . j. m. hornibrook , j. i. colless , i. d. conway lamb , s. j. pauka , h. lu , a. c. gossard , j. d. watson , g. c. gardner , s. fallahi , m. j. manfra , and d. j. reilly , `` cryogenic control architecture for large - scale quantum computing , '' http://dx.doi.org/10.1103/physrevapplied.3.024010[phys . rev . applied * 3 * , 024010 ( 2015 ) ] . j. carolan , c. harrold , c.and sparrow , e.e martn - lpez , n. j. russell , j. w. silverstone , p. j. shadbolt , n. matsuda , m. oguma , m. itoh , g. d. marshall , m. g. thompson , j. c. f. matthews , t. hashimoto , j. l. obrien , and a. laing , `` universal linear optics , '' http://dx.doi.org/10.1126/science.aab3642[science * 349 * , 711716 ( 2015 ) ] . yang wang , xianli zhang , t. a. corcovilos , a. kumar , and d. s. weiss , `` coherent addressing of individual neutral atoms in a 3d optical lattice , '' http://dx.doi.org/10.1103/physrevlett.115.043003[phys . lett . * 115 * , 043003 ( 2015 ) ] . j. combes , c. ferrie , c. cesare , m. tiersch , g. j. milburn , h. j. briegel , and c. m. caves , `` in - situ characterization of quantum devices with error correction , '' http://arxiv.org/abs/1405.5656[preprint arxiv:1405.5656 [ quant - ph ] ( 2014 ) ] . a. g. fowler , d. sank , j. kelly , r. barends , and j. m. martinis , `` scalable extraction of error models from the output of error detection circuits , '' http://arxiv.org/abs/1405.1454[preprint arxiv:1405.1454 [ quant - ph ] ( 2014 ) ] . m. hein , w. dr , j. eisert , r. raussendorf , m. van den nest , and h. j. briegel , `` entanglement in graph states and its applications , '' link:\doibase 10.3254/978 - 1 - 61499 - 018 - 5 - 115[ios press proceedings of the international school of physics `` enrico fermi '' , * 162 * , 115218 ( 2006 ) ] . b. buchberger , _ ein algorithmus zum auffinden der ba- siselemente des restklassenringes nach einem nulldimen- sionalen polynomideal _ , http://www.risc.jku.at/groebner-bases-bibliography/details.php?details_id=706[ph.d . thesis ( 1965 ) ] . s. noschese , l. pasquini , and l. reichel , `` tridiagonal toeplitz matrices : properties and novel applications , '' http://www.math.kent.edu/~reichel/publications/toep3.pdf[numerical linear algebra with applications 20 , 302326 ( 2013 ) ] .

in this work the dynamics of a chain consisting of a set of beads attached to the ends of segments of fixed lengths is investigated . the chain fluctuates at constant temperature in a viscous medium . for simplicity , all interactions among the beads have been switched off and the number of spatial dimensions has been limited to two . in the limit in which the chain becomes a continuous system , its behavior may be described by a path integral , in which the rigid constraints coming from the infinitesimally small segments are imposed by means of a functional delta function . in this way a model of the dynamics of the chain is obtained , which closely resembles a two - dimensional nonlinear sigma model . the partition function of this generalized nonlinear sigma model is computed explicitly for a ring - shaped chain in the semiclassical approximation . the behavior of the chain at both long and short scales of time and distances is investigated . the connection between the generalized nonlinear sigma model presented here and the rouse model is discussed .

0705.4182

subject of this work is a study of the dynamics of a continuous chain which is subjected to thermodynamic fluctuations at constant temperature @xmath0 . the chain is represented as the limit of a system of beads and links of fixed length , in which the number @xmath1 of beads becomes infinite , while the length @xmath2 of the links goes to zero in such a way that the total length of the chain @xmath3 remains a finite constant . problems of this kind are encountered for example in polymer physics , because with some approximation long flexible polymers may be regarded as continuous chains @xcite . it is thus spontaneous to consider an isolated polymer fluctuating in some viscous environment as a concrete realization of the system investigated here . a few applications in which the dynamics of a chain turns out to be relevant are mentioned in ref . in particular , in ref . @xcite six regimes of the chain dynamics are distinguished , which apply not only to the well known case of a polymer in a solution , but also to other cases , like for instance those of an isolated cold chain and of a hot polymer in the vapor phase . from polymer physics , one may borrow the standard approach to the dynamics of a chain . it consists in considering the fluctuations of the chain as a stochastic process , which is usually described with the help of the langevin equations or , alternatively , of the fokker planck smoluchowski equations @xcite . this approach leads to the well known models of rouse @xcite and zimm @xcite which allow a satisfactory understanding of the main properties of polymers in solutions . one major drawback of these coarse grained models is that they do not take into account the rigid constraints which are necessary in order to keep constant the length of the links connecting the beads . the rouse and rouse zimm equation consider instead beads joined together by springs , where the local spring is infinitely extensible . in this way the length of the chain is not fixed and is allowed to become infinite . moreover , in the continuous limit the rouse equation is nothing but the stochastic equation ( langevin equation ) for the classical wiener measure , which yields paths without well defined tangent vectors @xcite . these problems have been tackled by various attempts , see e.g. @xcite . however , the correct use of rigid constraints in ( stochastic ) dynamics requires some mathematical effort @xcite , in contrast to the static cases where rigid constraints can be implemented by dirac delta functions in the partition functions . for instance , the probability distribution of @xmath1 ideal closed chains topologically linked together may be represented as a path integral of @xmath1 noninteracting particles with the insertion of dirac delta functions which enforce the topological constraints on their trajectories @xcite . here a strategy similar to that used for static chains will be applied to dynamics . we consider the distribution @xmath4 of the probability that a fluctuating chain passes from a initial discrete spatial conformation @xmath5 to a final one @xmath6 in a given interval of time @xmath7 . the idea behind our approach is based on the fact that the beads of the chain may be regarded as a set of @xmath1 brownian particles with constrained trajectories . the constraints arise due to the presence of the @xmath8 links of fixed lengths connecting the beads . as a consequence , it is possible to write the probability distribution @xmath4 in the form of a path integral describing the fluctuations of @xmath1 brownian particles with the insertion of dirac delta functions . the latter are needed in order to impose the necessary conditions on the trajectories of the particles . for simplicity , possible interactions among the beads have been switched off , including hydrodynamic interactions and only the two - dimensional case has been discussed . the limit in which the chain becomes continuous is not entirely trivial . it involves the vanishing of three crucial quantities , the mass of the beads , their mobility and their size . after performing this limit very carefully , we obtain as a final result the probability distribution @xmath9 of the continuous chain . it is found that @xmath9 consists of a path integral which closely resembles the partition function of a two - dimensional nonlinear sigma model @xcite . for this reason the obtained model will be called the generalized nonlinear sigma model ( gnlsm ) . the main difference from the nonlinear sigma model is that the holonomic constraint is replaced in the gnlsm by an nonholonomic constraint , which requires that the tangent at every point of the trajectory of the chain is an unit vector . the lagrange multiplier that imposes this nonholonomic constraint plays the same role of pressure in the hydrodynamics of incompressible fluids , a fact already noted in ref . the `` incompressibility '' is related to the fact that it is not possible to `` compress '' the lengths of the links joining the beads . the gnlsm is well suited to study all situations in which it is possible to assume that the thermodynamic fluctuations and the conformational changes are small and slow . systems satisfying these requirements are for instance cold chains or chains in a very viscous medium . in principle , the gnlsm is exactly solvable after performing two gaussian integrations , but the presence of non - trivial boundary conditions complicates the calculation of the probability distribution @xmath9 . even the method of the effective potential , which is useful to investigate phase transitions in nonlinear sigma models , can not be easily applied . as a matter of fact , in the nonlinear sigma model the effective potential is computed assuming that the field configurations which minimize the action are constants . however , in the gnlsm configurations of this kind correspond to the situation in which the chain has collapsed to a point and are thus unphysical . despite these difficulties , it is possible to compute the probability distribution @xmath9 and the associated generating functional of the correlation functions of the bond vectors @xmath10 $ ] using a background field approximation , in which small gaussian fluctuations are considered in the background of a dominating classical conformation . the initial and final conformations of the chain are picked up by choosing in a suitable way the background classical solution and by tuning the boundary conditions of the gaussian fluctuations . the semiclassical approximation is valid in the case of low temperatures or of highly viscous media , exactly the regimes in which the gnlsm can be applied . the explicit formulas of @xmath9 and @xmath10 $ ] derived here show that the fluctuations which deform the chain along directions which are tangent to the trajectory of the classical background conformation propagate differently from the normal fluctuations . this fact is a direct consequence of the presence of rigid constraints and could be relevant for example in the theory of formation of single polymer crystals @xcite . the material presented in this paper is organized as follows . in section ii the form of the lagrangian of a classical discrete chain in two dimensions has been derived in polar and in cartesian coordinates . no restrictions are posed to the motion of the discrete chain . the limit to a continuous chain is however performed assuming that one of the ends of the chain is fixed . the calculation in polar coordinates shows that one crucial term in the lagrangian disappears in the continuous limit . this fact simplifies the classical equations of motion of the chain . the probability distribution @xmath9 of the fluctuating chain is computed in section iii using a path integral approach . some subtleties arising when taking the continuous limit in the probability distribution of the discrete chain are discussed . section iv is dedicated to the study of the classical solutions of the gnlsm . it is shown that the only possible classical solutions are time independent , apart from rigid translations with constant velocity of the whole chain . the computation of the probability distribution @xmath9 and of the generating functional @xmath10 $ ] for a ring - shaped chain is performed in the semiclassical approximation in section v. the physical interpretation of the results obtained in the previous sections is presented in section vi . the equilibrium limit of the gnlsm and its connection with the rouse model are studied in section vii . finally , in section viii the conclusions are drawn and possible future developments are discussed . let us consider a discrete chain of @xmath8 segments of fixed lengths @xmath11 in the two - dimensional plane . each segment @xmath12 is completely specified by the positions of its end points @xmath13 and @xmath14 . in cartesian coordinates @xmath15 these positions are given by the radius vectors : @xmath16 the segments are joined together at the points @xmath14 , where @xmath17 , see fig . [ system2y ] , while @xmath18 and @xmath19 are the ends of the chain . moreover , at each point @xmath14 , with @xmath20 , a mass @xmath21 is attached . in the following we restrict ourselves to the case of a free chain . we will see below that the addition of interactions is straightforward . at this point we compute the kinetic energy of the above system : @xmath22 the subscript @xmath23 in the left hand side of eq . ( [ totenechadef ] ) is to recall that we are considering at the moment a discrete chain with @xmath8 segments . for future purposes , it will be convenient to introduce the kinetic energy of point @xmath14 : @xmath24 of course , to eq . ( [ totenechadef ] ) one should also add the constraints @xmath25 in order to enforce the requirement that the segments have fixed length @xmath26 . it is possible to eliminate these constraints passing to polar coordinates : @xmath27 @xmath28 is the angle formed by segment @xmath29 with the @xmath30axis , see fig . [ system2y ] . according to our settings , the radial coordinates @xmath31 are constants for @xmath32 . the length @xmath33 , which denotes the distance of the end point @xmath34 from the origin , is not fixed , so that @xmath35 is allowed to vary with the time @xmath36 . from eq . ( [ carpol2 ] ) the velocity components of the @xmath37th segment may be written as follows : @xmath38 segments . let us note that the end point @xmath18 is not bound to stay at a fixed distance with respect to the origin of the cartesian reference system . ] separating the contribution coming from the first @xmath39 variables as shown in eqs . ( [ velcom1][velcom2 ] ) , the kinetic energy @xmath40 of the @xmath37th point can be expressed in terms of the kinetic energy @xmath41 of the @xmath42th segment : @xmath43 it is possible to solve the above recursion relation and to find a closed expression of @xmath40 in polar coordinates . if we do that , at the end the total kinetic energy of the discrete chain becomes : @xmath44 where @xmath45 is the total mass of the chain . we wish now to perform the limit in which the chain of @xmath8 segments becomes a continuous system . to this purpose , it is convenient to consider the indices @xmath46 appearing in eq . ( [ tdiscr ] ) as discrete variables taking values in a one - dimensional lattice with @xmath1 points . quantities @xmath47 carrying the index @xmath48 may be interpreted as discrete functions of @xmath48 . their variations @xmath49 are given by : @xmath50 . clearly , @xmath51 , i. e. the spacing between two neighboring points in the lattice is 1 . in order to proceed , we rescale the distances in the lattice in such a way that the spacing in the new lattice will be @xmath2 . to this purpose , we perform the transformations @xmath52 , @xmath53 where the new variable @xmath54 has variation @xmath55 . the next step is to compute the kinetic energy of eq . ( [ tdiscr ] ) in the limit @xmath56 in which the product @xmath57 remains finite and is equal to the total length of the chain @xmath58 . clearly , in the limit ( [ contlim ] ) the right hand side of eq . ( [ tdiscr ] ) will diverge unless we suppose that the masses @xmath21 and the lengths @xmath26 of the segments are going to zero in a suitable way . reasonable assumptions are : @xmath59 where @xmath60 and @xmath61 are respectively the distribution of length and of mass along the chain . to be consistent with our settings , the distributions @xmath60 and @xmath61 must be normalized as follows : @xmath62 while it would be interesting to study chains in which the segments are allowed to have different lengths and the points have different masses , for simplicity we will suppose from now on that the length and mass distributions in the chains are uniform , i. e. : @xmath63 in the discrete case ( compare with eq . ( [ ansaone ] ) ) this implies that all segments of the chain and the masses @xmath21 are equal : @xmath64 at this point we are ready to pass to the continuous limit . functions of discrete variables will be substituted with functions of continuous variables , while sums will be replaced with integrals according to the following rules : @xmath65 after a few calculations one finds : @xmath66 where @xmath67 let us note that the right hand side of eq . ( [ tcont ] ) contains four terms , while the original discrete version of the kinetic energy in eq . ( [ tdiscr ] ) contained five terms . in fact , the contributions proportional to @xmath68 of eq . ( [ tdiscr ] ) disappear in the continuous limit . ( [ tcont ] ) may be simplified by performing the following change of variables : @xmath69 using also the formula : @xmath70 which is valid for any integrable function @xmath71 , we obtain : @xmath72 as a further simplification , one could fix the point @xmath18 in some location , so that @xmath73 exploiting the above assumptions in eq . ( [ tcontsimp ] ) , we find that the lagrangian @xmath74 of the ideal chain is given by : @xmath75 what happens if , instead of polar coordinates , we choose cartesian coordinates in order to compute the continuous limit of the kinetic energy ( [ totenechadef ] ) ? with the help of the prescriptions given in eqs . ( [ contlim][disccontrul ] ) and related comments , it is easy to show that the lagrangian of the ideal chain @xmath76 becomes in cartesian coordinates : @xmath77 of course , the fields @xmath78 and @xmath79 are not independent , because they satisfy the relationship : @xmath80 where @xmath81 and @xmath82 . ( [ constrxy ] ) is the continuous version of the constraints ( [ constrdiscln ] ) . as we see , the form of the kinetic energy is much simpler than that of its analogue in polar coordinates , but the price to be paid is the cumbersome presence of the constraints ( [ constrxy ] ) . if @xmath33 and @xmath83 are constants , according to the assumption of eq . ( [ ponefixass ] ) , one should also add to eqs . ( [ hzeroreal ] ) and ( [ constrxy ] ) the boundary conditions : @xmath84 in this way one end of the chain is fixed at the given point @xmath85 . it is possible to implement other boundary conditions . for instance , one could ask that the chain forms a closed loop : @xmath86 to show that the lagrangian in cartesian coordinates ( [ hzeroreal ] ) and the lagrangian in polar coordinates ( [ hamfrecha ] ) are equivalent , it is possible to perform in eq . ( [ hzeroreal ] ) the field transformations : @xmath87 these are analogous to the discrete changes of variables of eqs . ( [ velcom1 ] ) and ( [ velcom2 ] ) in the case in which one end of the chain is kept fixed . it is easy to check that , after the substitutions ( [ realvar ] ) in eq . ( [ hzeroreal ] ) , one obtains exactly eq . ( [ hamfrecha ] ) as desired . for future convenience we introduce also the vector notation : @xmath88 in this way we get for the functional @xmath89 and the constraint ( [ constrxy ] ) the more compact expressions : @xmath90 and @xmath91 finally , eqs . ( [ initcond ] ) and ( [ cloloocond ] ) become respectively : @xmath92 @xmath93 it is now easy to add the interactions . for example , let us suppose that the segments of the chain are immersed in an potential @xmath94 and that there are also internal interactions associated to a two - body potential @xmath95 . in this case , the lagrangian @xmath89 of eq . ( [ dffdf ] ) generalizes to : @xmath96 where @xmath97 and @xmath98 in this section it will be addressed the problem of describing the dynamics of a random chain subjected to thermodynamic fluctuations and immersed in an environment held at constant temperature @xmath0 . the strategy in order to treat this problem is to consider the discrete chain as a set of @xmath1 particles of mass @xmath99 performing a random walk while subjected to the discrete constraints ( [ constrdiscln ] ) , which we rewrite here for convenience as follows : @xmath100 it is additionally required that at the instants @xmath101 the @xmath102th particle is located respectively at the initial point @xmath103 and at the final point @xmath104 for @xmath105 . for simplicity , the interactions among the particles are switched off including hydrodynamic forces . mediated by the motion of the fluid are called hydrodynamic interactions , according to the definition of hydrodynamic interactions given in @xcite , section 3.6 , page 66 . ] if one could ignore the constraints , the probability distribution @xmath106 of the system of @xmath1 particles would be : @xmath107 where @xmath108 is the probability distribution describing the free random walk of the @xmath102th particle . as it is well known , @xmath108 satisfies the fokker - planck - smoluchowski equation @xmath109 @xmath110 being the diffusion constant . ( [ diffusioneq ] ) is completed by the boundary condition : @xmath111 the solution of eq . ( [ diffusioneq ] ) can be written up to an irrelevant normalization factor @xmath112 in the form of a path integral : @xmath113 substituting eq . ( [ psinpi ] ) in eq . ( [ fullnpart ] ) , the probability distribution @xmath106 becomes : @xmath114 now we have to add to the above free random walks the constraints ( [ constaconst ] ) . this will be done by inserting in the probability distribution of eq . ( [ postnpi ] ) a product of dirac @xmath115functions which enforce exactly these constraints : @xmath116 in the above equation @xmath117 denotes a normalization constant . moreover , we have introduced the functional @xmath118 defined as follows : @xmath119 the distribution @xmath4 measures the probability that a chain starting from the initial configuration @xmath120 of its segments ends up after a time @xmath121 in the configuration @xmath122 . we note that the diffusion constant @xmath110 appearing in ( [ actiondisc ] ) satisfies the relation : @xmath123 where @xmath124 is the mobility of the particle , @xmath125 is the boltzmann constant and @xmath0 is the temperature . this fact allows to rewrite the quantity @xmath126 in a form which reflects the analogy of the present problem with a quantum mechanical problem : @xmath127 in the above equation we have put @xmath128 the parameter @xmath129 has the dimension of a time . indeed , @xmath129 is the relaxation time that characterizes the rate of the decay of the drift velocity of the particles composing the chain . the quantity @xmath118 looks now like a real action of a set of @xmath1 quantum particles of mass @xmath99 in complex time , with the constant factor @xmath130 replacing the planck constant @xmath131 . this is not a surprise , because the connections between quantum mechanics and brownian motion are well known . indeed , one may show that the uncertainties in the position and momentum of a brownian particle are related to the constant @xmath132 @xcite . at this point we are ready to take in the probability distribution ( [ partfunpp ] ) the continuous limit ( [ contlim ] ) . by introducing the rescaled variables @xmath133 as we did in section [ sec2 ] , the probability distribution ( [ partfunpp ] ) becomes : @xmath134 in the limit @xmath135 , @xmath136 , @xmath137 we obtain from @xmath4 the probability distribution @xmath9 of the continuous chain : @xmath138 where @xmath139 this is the desired result . formally , the normalization constant @xmath117 may be written as a path integral over the initial and final configurations : @xmath140 the model described by eqs . ( [ contparttd][contacttd ] ) will be called here the generalized nonlinear sigma model due to its close resemblance to a two - dimensional nonlinear sigma model . the most striking difference is that the constraint @xmath141 of the nonlinear sigma model has been replaced by the condition ( [ constbfr ] ) , which contains the derivatives of the bond vectors @xmath142 and it is thus nonholonomic . before concluding this section , we would like to complete the derivation of the probability distribution of eqs . ( [ contparttd][contacttd ] ) by discussing the continuous limit of the relaxation time @xmath129 . this parameter has been defined in eq . ( [ reltimedef ] ) as the product of the mobility @xmath124 with the mass @xmath99 of the beads . when the distance @xmath2 between the beads goes to zero , @xmath99 goes to zero as well according to eq . ( [ liallequal ] ) . on the other side , with decreasing values of @xmath2 , two contiguous beads will become closer and closer until they will eventually merge one into another . to avoid this unphysical situation , one should add to the continuous limit ( [ contlim ] ) the requirement that the dimensions of the beads vanish together with @xmath2 . supposing for instance that the beads are circles of radius @xmath143 , for dimensional reasons one is lead to put @xmath144 , where @xmath145 is a dimensionless proportionality factor . as it is intuitive , when the size of a bead decreases , its mobility @xmath124 increases . the increasing of @xmath124 compensates the vanishing of @xmath99 , so that the product @xmath146 remains finite . this fact can be verified rigorously in three dimensions using the well known stokes formula of the mobility . to get rid of the delta function appearing in the formulation of the generalized sigma model of eqs . ( [ contparttd ] ) and ( [ contacttd ] ) , it will be convenient to introduce a lagrange multiplier @xmath147 and to use the fourier representation of the dirac delta function . moreover , we add a coupling of the bond vectors @xmath148 with an external source @xmath149 . in this way the generating functional @xmath10 $ ] of the correlation functions of the bond vectors may be written in the form : @xmath150=\int{\cal d}\mathbf r{\cal d}\mathbf \lambda \exp\left\{- \int_{t_i}^{t_f}dt\int_0^lds\left [ \frac{m}{2l\kappa}\dot\mathbf r^2+i\lambda\left ( { \mathbf r'}^2 - 1 \right ) + \frac 1\kappa\mathbf j\cdot \mathbf r\right ] \right\ } \label{pathlagmult}\ ] ] let us note that , for the sake of generality , in eq . ( [ pathlagmult ] ) no boundary conditions for the relevant fields have been specified . as it stands , eq . ( [ pathlagmult ] ) could be applied both to open or closed chains . moreover , in the case of open chains both possibilities of free or fixed end points are allowed . it turns out that the degree of complexity of the computation of @xmath10 $ ] strongly depends on the choice of boundary conditions . let s now derive the solutions @xmath151 and @xmath152 of the classical equations of motion associated with the generating functional @xmath10 $ ] of eq . ( [ pathlagmult ] ) : @xmath153 it is easy to see that , due to the constraint ( [ cleqthree ] ) , the current @xmath154 must depend only on the variable @xmath36 , i. e. @xmath155 . always for the same reason , it is possible to check that eq . ( [ cleqtwo ] ) is inconsistent unless @xmath156 or @xmath157 . depending on the fact that @xmath152 is zero or not , one finds that the relevant solutions of eqs . ( [ cleqone][cleqthree ] ) may be divided into two groups , which we call here solutions of type a and of type b. the solutions of type a are characterized by the condition @xmath157 . they are displayed in table [ tableone ] . * type a solutions * @xmath158 where @xmath159 and @xmath160 are constant vectors , @xmath161 and @xmath162 here @xmath163 is an arbitrary function of @xmath164 , while in eq . ( [ erreone ] ) @xmath165 denotes the green function which solves the differential equation : @xmath166 type a solutions admit closed chain configurations . in that case , the functions @xmath167 must satisfy the additional periodicity condition : @xmath168 besides the classical solutions of type a , there are also the solutions of type b listed in table [ tabletwo ] . * type b solutions * @xmath169 where @xmath170 and @xmath160 are constants vectors , @xmath171 and @xmath172 in the absence of the external source @xmath173 , this kind of solutions corresponds to a configuration in which the chain is stretched along the @xmath30axis with one end in the point @xmath170 and the other end in the point @xmath174 . no closed loop configuration is allowed . of course , the stretched chain may be oriented in a different way by means of a rotation . + we would like to stress the fact that , if there are no external currents , both type a and b of classical solutions are static , i. e. they do not depend on time apart from the rigid translations of the whole chain with constant velocity @xmath160 . this result is confirmed if one studies the equations of motion corresponding to the lagrangian ( [ hamfrecha ] ) obtained using polar coordinates . the only allowed classical solutions for the field @xmath175 are in fact time independent . the exact computation of @xmath10 $ ] is a formidable problem despite the simplicity of the action of the generalized nonlinear sigma model . one of the main difficulty is the presence of the nonholonomic constraint @xmath176 in the path integral ( [ contparttd ] ) . in principle , this cumbersome condition may be easily eliminated by introducing a scalar field @xmath177 and performing the formal substitutions of eq . ( [ realvar ] ) @xmath178 here we have assumed for simplicity that the chain has one fixed end in the origin of the coordinates , so that @xmath179 in agreement with eq . ( [ formsubst ] ) . to show that after the field transformation ( [ formsubst ] ) the constraint disappears from the path integral ( [ contparttd ] ) , we use the following relation which will be proved in appendix a for a generic functional @xmath180 : @xmath181 where @xmath182 is an irrelevant constant . in our particular case , in which : @xmath183 \label{arbfundef}\ ] ] one obtains from eq . ( [ funcid ] ) : @xmath184 with @xmath89 being the lagrangian defined in eq . ( [ hamfrecha ] ) . as we see , the dirac delta function with the constraint is no longer present in the path integral ( [ dfhjsfd ] ) , but the lagrangian of eq . ( [ hamfrecha ] ) is both nonlocal and nonlinear . in the following , we will stick to cartesian coordinates limiting ourselves to study small gaussian fluctuations of the field @xmath148 around the classical solutions derived in the previous section . to this purpose , in eq . ( [ pathlagmult ] ) we split both fields @xmath148 and @xmath185 into classical contributions @xmath186 and statistical corrections @xmath187 : @xmath188 moreover , it will also be convenient to split the external source @xmath154 in an analogous way : @xmath189 where @xmath190 denotes the current depending only on the time @xmath36 appearing in eqs . ( [ erreone ] ) and ( [ errebone ] ) . due to the fact that the lagrange multiplier @xmath191 is just an auxiliary field , it is possible to choose for its variation @xmath192 trivial boundary conditions at the initial and final instants : @xmath193 the boundary conditions of @xmath194 will be fixed later . at this point , we expand the action @xmath195 \label{actionbeq}\ ] ] appearing in the path integral ( [ pathlagmult ] ) with respect to the quantities @xmath196 and @xmath197 . since the latter are supposed to be small corrections of the dominating classical solutions , we stop the expansion at the second order : @xmath198 at the zeroth order we have : @xmath199 \label{acldef}\ ] ] this is just the action @xmath200 in which the fields have been replaced by their classical configurations , which may be either of type a or of type b. in both cases , the term @xmath201 , which in principle should be present in eq . ( [ acldef ] ) , has been omitted because eq . ( [ cleqthree ] ) forces it to vanish identically . let s now compute the first order contribution @xmath202 . usually , first order contributions vanish after exploiting the classical equations of motion . in our case this is in general not true . the reason is that , due to the nontrivial boundary conditions satisfied by @xmath203 , nonzero boundary terms may still appear in @xmath204 . despite this fact , it is possible to show that @xmath205 vanishes at least in the following two situations : 1 . closed chains satisfying the boundary conditions ( [ cloloopcondr ] ) . 2 . open chains in which both ends are fixed , so that besides the condition ( [ begbb ] ) also the following one is valid : @xmath206 , where @xmath207 and @xmath208 are constants . assuming that one of the above two conditions is verified , it is possible to put : @xmath209 thus , we are left only with the computation of the second order corrections @xmath210 . after simple calculations , one finds : @xmath211 \label{dstwodef}\ ] ] putting together all the above results , it is possible to conclude that , within the present gaussian approximation , the expression of @xmath10 $ ] reduces to : @xmath150=e^{{\cal a}_{cl}}z[j ] \label{probfungauss}\ ] ] where @xmath212=\int{\cal d\delta \mathbf r}{\cal d}\delta\lambda e^{-\delta { \cal a}^{(2 ) } } \label{partfungauss}\ ] ] and @xmath213 and @xmath214 are respectively given in eqs . ( [ acldef ] ) and ( [ dstwodef ] ) . from this point on we will consider only chain configurations which are closed , so that both classical configurations @xmath151 and their statistical corrections @xmath215 must satisfy the boundary conditions in @xmath216 of eq . ( [ cloloopcondr ] ) . in the lagrange multiplier sector , besides the trivial boundary conditions in time of eq . ( [ tribouconlam ] ) , we require also the following ones with respect to the variable @xmath216 : @xmath217 the case of closed chains is particularly interesting because under this assumption the path integral appearing in the right hand side of eq . ( [ partfungauss ] ) may be rewritten in such a way that it closely resembles the generating functional of a field theory of a one - dimensional system at finite temperature . in this field theory the coordinate @xmath216 plays the role of time while the real time @xmath36 becomes the spatial coordinate of the one - dimensional space . we still need to specify the boundary conditions with respect to the time for the fields @xmath218 . we fix them in such a way that the gaussian path integration over these fields in the generating functional of eq . ( [ partfungauss ] ) will be as simple as possible . to this purpose , reasonable choices are the following : @xmath219 @xmath220 @xmath221 @xmath222 at this point we are ready to perform the gaussian integrations over the fields @xmath215 in the generating functional @xmath223 $ ] given in eq . ( [ partfungauss ] ) . after some integrations by parts , which do not produce boundary terms thanks to the boundary conditions ( [ cloloopcondr ] ) and ( [ lambdatriv ] ) , one finds : @xmath212=c_1\int{\cal d}\delta\lambda e^{-s(\delta\lambda ) } \label{zc1dl}\ ] ] with @xmath224 \label{actslam}\end{aligned}\ ] ] and @xmath225 the symbol @xmath165 denotes the propagator ( [ gttpdef ] ) computed taking into account one of the boundary conditions defined in eqs . ( [ dirdir][neuneu ] ) @xcite . we have thus four possibilities : @xmath226 @xmath227 @xmath228 @xmath229 + \frac lm \frac { \left[(t - t_f)^2+(t'-t_f)^2 \right]}{2(t_f - t_i ) } \label{propneuneu}\ ] ] in eqs . ( [ propdirdir][propneuneu ] ) the function @xmath230 is the usual @xmath231function of heaviside . let us note that the function @xmath165 of eq . ( [ propneuneu ] ) is actually a pseudo green function , which satisfies the equation : @xmath232 instead of eq . ( [ gttpdef ] ) . this is due to the fact that , in the case of neumann neumann boundary conditions , one should project out the constant solution of the eigenvalue equation associated to the operator @xmath233 . we remark that in the action ( [ actslam ] ) the classical conformations appear only in the derivatives @xmath234 , which coincide with the derivatives of the vectors @xmath235 defined in eq . ( [ erretwo ] ) . in components : @xmath236 from eq . ( [ univec ] ) it is clear that the vector @xmath234 has the meaning of the unitary vector which is tangent to the classical trajectories . it is therefore convenient to decompose all vectors appearing in the action @xmath237 in components which are normal or tangent to @xmath238 . after some algebra , one obtains in this way an expression of @xmath223 $ ] in which now @xmath237 takes the simplified form : @xmath239\label{simpactsdl}\end{aligned}\ ] ] here we have introduced the _ tangential _ component @xmath240 of the current @xmath241 : @xmath242 at this point we are ready to eliminate the auxiliary field @xmath192 from the functional of eq . ( [ zc1dl ] ) , where now the action @xmath243 is defined in eq . ( [ simpactsdl ] ) . to this purpose , one needs to perform a gaussian integration , which produces the result : @xmath212=c_1c_2 e^ { -\frac 12\int_0^l ds ds'\int_{t_i}^{t_f}dtdt'g(t , t')\left [ k(s , s')\delta j_t(t , s)\delta j_t(t',s')+\frac 1 l\delta\mathbf j(t , s ) \cdot\mathbf j(t',s ) \right ] } \label{zfin}\ ] ] in the above equation @xmath244 denotes the green function which satisfies the relation : @xmath245k(s , s')=-\delta(s , s ' ) \label{gretwoeqdef}\ ] ] while @xmath246 } \label{ctwodef}\ ] ] putting all together we obtain the expression of the generating functional @xmath10 $ ] in its final form : @xmath247&=&e^{{\cal a}_{cl}}c_1c_2 \exp\left[-\frac 1l \int_0^l ds\int_{t_i}^{t_f}dtdt ' g(t , t')\delta \mathbf j(t , s)\cdot\delta\mathbf j(t',s ) \right ] \nonumber\\ & \times&\exp\left [ -\frac 12\int_0^l dsds'\int_{t_i}^{t_f}dtdt'g(t , t ' ) k(s , s')\delta j_t(t',s'))\delta j_t(t , s ) \right]\label{finforgenfun}\end{aligned}\ ] ] the right hand side of eq . ( [ finforgenfun ] ) displays the asymmetry in the propagation of transverse and longitudinal modes . the differential equation satisfied by the green function @xmath244 of eq . ( [ gretwoeqdef ] ) can not be solved analytically for any given function @xmath167 . here we discuss just the case in which the background classical solution corresponds to a chain with the configuration of a circle , i. e. : @xmath248 the radius of the circle is @xmath249 , so that the total length of the chain is @xmath58 as desired . comparing eq . ( [ rclcirc ] ) with eq . ( [ erretwo ] ) , it is clear that for this conformation @xmath250 . substituting this expression of @xmath167 in eq . ( [ gretwoeqdef ] ) , it turns out that the green function @xmath244 satisfies the relation : @xmath251 the solution of the above equation corresponding to the boundary conditions ( [ lambdatriv ] ) is : @xmath252\sinh\frac{2\pi}ls}{\frac { 2\pi}l\sinh 2\pi}\theta(s'-s)+ \frac{\sinh\left [ \frac{2\pi}l(l - s ) \right]\sinh\frac{2\pi}ls'}{\frac { 2\pi}l\sinh 2\pi}\theta(s - s ' ) \label{kcircleder}\ ] ] in the limit @xmath253 , we obtain from the generating functional of eq . ( [ finforgenfun ] ) the expression of the probability distribution @xmath9 of eqs . ( [ contparttd][contacttd ] ) in the semiclassical approximation : @xmath254 remembering the respective definitions of the constants @xmath255 and @xmath256 of eqs . ( [ conedef ] ) and ( [ ctwodef ] ) , together with the form ( [ acldef ] ) of the classical action , @xmath9 may be explicitly written as follows : @xmath257 \right\ } \int{\cal d}\delta\mathbf r \exp\left[-{\displaystyle \frac m{2l}\int_{t_i}^{t_f}dt \int_0^lds\left(\delta\dot\mathbf r\right)^2 } \right]\nonumber\\ \!\!\!\!\!\ ! & \!\!\!\!\!\ ! \times&\!\!\!\!\!\!\int{\cal d}\delta\lambda\exp\left\ { \int_{t_i}^{t_f}dtdt'\int_0^ldsg(t , t')\left [ 2\kappa^2\delta\lambda(t , s)\left({\textstyle \frac{\partial^2}{\partial s^2}-(\varphi'(s))^2 } \right)\delta\lambda(t',s ) \right ] \right\}\label{finprofun}\end{aligned}\ ] ] the classical equations of motion ( [ cleqone][cleqthree ] ) admit only static solutions in which the conformation of the chain is fixed . only rigid translations of the chain as a whole with constant velocity @xmath160 are allowed . apart from these rigid translations , the time dependence of the classical solutions @xmath258 in eqs . ( [ aerre ] ) and ( [ erreb ] ) is just an artifact of the presence of the classical current @xmath259 which is fictitious and may thus be set to zero without any loss of generality . in the following discussion , it will be assumed that both @xmath190 and @xmath160 are zero . the absence of any relevant dynamics in the classical solutions is somewhat surprising . we can only suggest that this absence could be related to the fact that , after performing the continuous limit , the fourth term appearing in the left hand side of eq . ( [ tdiscr ] ) vanishes identically . this term is important for the chain dynamics since it contains second time derivatives of the angles @xmath175 . in some sense , the continuous chain is simpler that its discrete counterpart , in which this term is present . less trivial is the treatment of the fluctuations of the chain at constant temperature @xmath0 . the nonlinear sigma model given in eq . ( [ pathlagmult ] ) or , alternatively , its formulation without the lagrange multiplier of eqs . ( [ contparttd][contacttd ] ) , describe the fluctuations of a chain of @xmath1 segments of constant length @xmath2 in the continuous limit ( [ contlim ] ) . during its motion in the time interval @xmath260 $ ] the chain spans a two - dimensional surface which , in the case of a closed conformation , has the topology of a cylinder in the @xmath261 space , see fig . [ cylclocon ] . a surface which has the topology of a cylinder.,width=226 ] if the chain is open , instead , the topology of the cylinder should be replaced with that of a strip . we recall that in deriving the gnlsm of eqs . ( [ contparttd][contacttd ] ) the contribution of the hydrodynamic interactions has been neglected . this limits the validity of this model to the following cases : the viscosity of the fluid is large , so that the motion of the particles composing the chain is slow . thus the velocity field created by each particle is too weak to influence the motion of other particles . the temperature is low , so that once again the motion of the chain is slow . the conformation of the chain is relatively straight because there is some energy cost when the chain is being bent . to this purpose , however , one should introduce the stiffness at the joints between the segments . on the other side , the semiclassical approximation used in order to derive the generating functional @xmath10 $ ] of eq . ( [ finprofun ] ) is valid in the case in which the parameter @xmath262 defined in eq . ( [ hpplanck ] ) is small . this parameter depends essentially on the temperature @xmath0 and on the relaxation time @xmath129 . since @xmath129 is inversely proportional to the viscosity in the limit of low reynolds number , it is reasonable to assume that the semiclassical approach can be applied to a cold isolated chain or to an isolated chain in a very viscous solution . these situations correspond respectively to the points b ) and a ) mentioned above . both the generating functional @xmath10 $ ] of eq . ( [ finforgenfun ] ) and the probability distribution of eq . ( [ finprofun ] ) have been computed in the case of closed chains , whose conformations are subjected to the boundary conditions ( [ dirdir][neuneu ] ) . the physical meaning of these boundary conditions may be summarized as follows . in this case @xmath263 , where @xmath238 is a given static solution of the classical equations of motion . the probability that , starting from the conformation @xmath264 , the fluctuating chain ends up at the time @xmath265 in the same conformation , is proportional up to a normalization constant to the probability distribution @xmath9 of eq . ( [ finprofun ] ) , in which the green function @xmath165 is that of eq . ( [ propdirdir ] ) . in this case the probability distribution @xmath9 of eq . ( [ finprofun ] ) is proportional to the probability that a closed chain starting from a classical static conformation @xmath266 at the time @xmath267 ends up at the instant @xmath265 in an arbitrary conformation characterized by the fact that the velocities of each segment composing the chain is zero . @xmath9 must be computed choosing the green function @xmath165 defined in eq . ( [ propdirneu ] ) . here the segments of the chain have zero velocity at the beginning , but the conformation of the chain is otherwise arbitrary . at the time @xmath265 the chain is found in a given static classical conformation , i. e. @xmath268 . the probability for this to happen is obtained after substituting in eq . ( [ finprofun ] ) the green function @xmath165 of eq . ( [ propneudir ] ) . this is the situation in which the conformation of the chain at the initial and final times @xmath267 and @xmath265 are not specified , but the velocities of all the segments composing the chains must be zero . the relevant green function @xmath165 to be inserted in the probability distribution @xmath9 is that of eq . ( [ propneuneu ] ) . by choosing neumann dirichlet boundary conditions one may check for instance if , starting from any static conformation , there is a particular conformation in which it is very likely that the chain will be found after a certain time @xmath121 . the stability of a given conformation with respect to the thermodynamic fluctuations which attempt to reshape the chain can be tested by choosing dirichlet dirichlet boundary conditions . in principle , it is also possible to study other types of boundary conditions than those considered here , provided they do not give rise to unwanted boundary terms in the action of the generating functional @xmath10 $ ] . to conclude this section , it is interesting to see how the original constraint ( [ constbfr ] ) is realized in the semiclassical approximation . in eq . ( [ probfungauss ] ) , which gives the second order corrections to the classical action @xmath213 , there are two lagrange multipliers , @xmath192 and @xmath152 . the most relevant condition is that imposed by @xmath192 : @xmath269 let us note that the above relation is at the origin of the asymmetry in the generating functional ( [ finforgenfun ] ) between modes which are tangent or normal to the classical background conformation @xmath151 . ( [ secondconstr ] ) is just the approximated version of the full constraint ( [ constbfr ] ) . as a matter of fact , remembering the splitting into classical solutions and statistical corrections of eqs . ( [ splitr ] ) and ( [ splitl ] ) , we may rewrite eq . ( [ constbfr ] ) as follows : @xmath270 due to the fact that @xmath271 and neglecting the second order term @xmath272 , we obtain from eq . ( [ ccc ] ) : @xmath273 which coincides exactly with eq . ( [ secondconstr ] ) . in the case of solutions of type b there is an additional constraint , which is associated to the nonzero constant lagrange multiplier @xmath152 . this constraint requires that the average over the time @xmath36 and over the chain length @xmath216 of the quadratic term @xmath274 is zero : @xmath275 in the solutions of type a the above condition is not present because in that case @xmath157 . first of all , we study the equilibrium limit of the gnlsm . we use the formulation of the model given in eqs . ( [ contparttd][contacttd ] ) . for simplicity , we set @xmath276 . thus @xmath277 at this point , we rescale the time variable by putting @xmath278 , so that the above equation becomes : @xmath279 in the equilibrium limit @xmath280 we obtain : @xmath281 where @xmath282 . ( [ eqdist ] ) is exactly what one should expect in the case of the statistical mechanics of a discrete chain subjected to the constraints : @xmath283 the discrete probability function of the conformation of such a chain is given by : @xmath284 in the continuous limit this becomes exactly the distribution of eq . ( [ eqdist ] ) . this result is in agreement with the analogous probability function given in @xcite . while the purpose of this work is to provide a path integral formulation of the dynamics of a freely jointed chain without having in mind concrete applications to polymer physics , it is interesting to explore possible connections between the gnlsm of eqs . ( [ contparttd][contacttd ] ) and the rouse model . a direct attempt to put the rouse model in the path integral form via the martin - siggia - rose formalism leads to a probability distribution for the rouse chain which differs profoundly from the gnlsm obtained in this work . indeed , let us start from the langevin equation : @xmath285 here we have used instead of the arc - length @xmath216 the dimensionless variable @xmath286 defined as follows : @xmath287 . moreover , @xmath288 is a stochastic force with a gaussian distribution of width @xmath289 given by : @xmath290 . @xmath291 and @xmath292 are constant parameters which will be specified later . after the application of the martin - siggia - rose method , one finds the rouse probability distribution : @xmath293 } \label{rousepd}\ ] ] in principle , in the exponent of the above equation there should be the additional term @xmath294 however , due to the fact that @xmath295 and remembering the identity @xmath296 $ ] , it is easy to realize that @xmath297 amounts to total derivative terms , which can be neglected . coming back to eq . ( [ rousepd ] ) , we see that , while the gnlsm is nonlinear and contains just second derivatives of the bond vector @xmath142 , the rouse model is linear and contains derivatives of @xmath142 up to the fourth order . as an upshot , while it is possible to investigate the rouse model by decomposing @xmath148 into normal coordinates as explained in @xcite , that kind of fourier analysis can not be easily applied to the nonlinear gnlsm . we show at this point that , indeed , the two models are quite different and that it is not possible starting from one of them to recover the probability function of the other and viceversa , because they correspond to regimes which do not overlap . to this purpose , instead of the constraint @xmath298 of the gnlsm , we introduce the more general condition @xmath299 where @xmath300 is a new length scale such that @xmath301 and @xmath302 is a function of @xmath303 normalized in a such a way that : @xmath304 let us note that @xmath2 is the smallest length at our disposal : any segment of the chain of contour length shorter than @xmath2 may be regarded as rigid . at this point , following the same strategy used in section [ sec : three ] , we build the new distribution function @xmath305 with @xmath306 . the index _ int _ means that the distribution probability @xmath307 describes a model which , as we will see , interpolates between the gnlsm and the rouse model . we remark that the insertion of the @xmath308-function @xmath309 in the path integral ( [ xtl ] ) has a double meaning . on one side , it may be seen as a condition on the length of the averaged vector : @xmath310 in the above equation the distance between points of the chain has been averaged over arc - segments of length @xmath311 . on the other side , the introduction of the @xmath308-function @xmath309 may also be related to the internal forces among the beads composing the chain , which appear due to the presence of constraints . for example , in the case of the gnlsm the presence of these forces is evident in the formulation of eq . ( [ pathlagmult ] ) , in which the free action is corrected by the addition of the interacting term @xmath312 . it is easy to realize that the gnlsm is a special case of the model described by eq . ( [ xtl ] ) . we have just to remember that in the gnlsm the motion of the chain is observed at the smallest available scale of distances , i. e. the segment length @xmath2 . thus , we choose the form of the function @xmath302 as follows : @xmath313 as a consequence , the constraint ( [ genconstr ] ) becomes @xmath314 . dividing both members of the above equation by @xmath315 and supposing that @xmath2 is very small , we get up to higher order terms in @xmath2 the relation : @xmath316 in the limit @xmath317 , this is exactly the condition which has been imposed to the chain in the gnlsm , see eq . ( [ constbfr ] ) . using the property of the @xmath308-function @xmath318 , it is possible to check that also the probability distribution @xmath307 becomes that of the gnlsm . to obtain the rouse model from the interpolating probability distribution @xmath307 , we need first of all to decrease the resolution with which the segments of the chain are observed . accordingly , we require that the function @xmath319 appearing in the constraint ( [ genconstr ] ) is constant over the whole interval @xmath320 $ ] : @xmath321 in this way , the finest details of the chain are not taken into account , because the chain conformations are averaged over the scale of distance @xmath311 , which is by hypothesis much larger than the smallest available scale @xmath2 . to pass to the rouse model , we have also to restrict ourselves to the long time - scale behavior of the chain . this is achieved by assuming following @xcite , section 4.1 , p. 93 , that , for long - time scales , @xmath148 varies slowly with @xmath216 . this hypothesis allows to stop the expansion of @xmath322 with respect to @xmath303 at the first few orders : @xmath323 substituting the above truncated expansion in eq . ( [ genconstr ] ) and performing the trivial integrations over @xmath303 we obtain the condition : @xmath324 plugging in the above constraint in eq . ( [ xtl ] ) , we get the following approximated expression of @xmath307 @xmath325 where the factor @xmath326 has been absorbed by a rescaling of the length @xmath300 . at this point we use the fact that , apart from an irrelevant infinite constant , the functional @xmath308-function present in eq . ( [ crst ] ) may be simplified as follows : @xmath327 a proof of this identity , which is valid up to an irrelevant constant , will be given in appendix b. exploiting eq . ( [ eqwwtp ] ) , the probability distribution @xmath307 of eq . ( [ crst ] ) becomes : @xmath328 we may still simplify the above equation by applying the following slightly modified version of the gaussian approximation of the @xmath308-function : @xmath329 where we have supposed that the parameter @xmath330 is very large while the ratio @xmath331 is very small . clearly , the usual fourier representation of the functional dirac @xmath308-function is recovered in the limit @xmath332 and @xmath333 . up to now @xmath334 is an arbitrary parameter . we choose it in a such a way that @xmath335 this choice is compatible with our requirement for @xmath334 , since @xmath2 is the smallest scale of lengths at our disposal , so that @xmath336 is a very small quantity . using eq . ( [ gcalc ] ) in order to eliminate @xmath330 from eq . ( [ modfrep ] ) , we obtain the relation : @xmath337 after performing the gaussian integration over @xmath191 in eq . ( [ rclst ] ) , we obtain @xmath338 we may now plug in the above expression of the @xmath308-function in the distribution probability @xmath307 of eq . ( [ atceq ] ) . the result is : @xmath339 this approximated probability distribution has the same structure of that coming from the rouse model given in eq . ( [ rousepd ] ) . to make the comparison with the rouse model more explicit , we perform in eq . ( [ psiintrel ] ) the substitution @xmath340 : @xmath341 let s now identify the coefficients appearing in eq . ( [ psint ] ) with those of eq . ( [ rousepd ] ) . we recall the fact that in the case of the rouse model : @xmath342 on the other side , the parameter @xmath145 is in the exponent of eq . ( [ psint ] ) may be written as follows : @xmath343 . it is now easy to verify that the probability function @xmath344 of eq . ( [ rousepd ] ) and that of eq . ( [ psint ] ) coincide if we make the following choice for @xmath334 : @xmath345 . this work may be considered as the ideal continuation of the seminal paper of edwards and goodyear of ref . @xcite , in which the problem of a chain subjected to the constraints ( [ constaconst ] ) has been investigated using an approach based on the langevin equation . with respect to ref . @xcite , our approach based on the fokker planck smoluchowski equation provides a path integral and field theoretical formulation of the dynamics of a freely jointed chain in the continuous limit . the gnlsm obtained here makes possible the application of field theoretical techniques to the study of the fluctuations of a freely jointed chain . as an example , we have derived in the semiclassical approximation the probability function of the chain and the associated generating functional . the approximation used in the computation of the generating functional is valid for instance in the cases of a cold isolated chain or of a chain fluctuating in a very viscous medium . most of the results obtained in this paper have been discussed in the previous two sections , so that we provide only a short summary of them : 1 . derivation of the gnlsm , which provides a path integral formalism to the freely jointed chain . computation of the partition function and of the generating functional of the gnlsm in the semiclassical approximation . the behavior of a chain at scales of length and time which are very long has been compared with the behavior at short scales of length and time in section vii . it is shown in this way that it is not possible to compare directly the rouse model and models describing a freely jointed chain like the gnlsm , because the regimes and the assumptions of these two models do not overlap . in particular , the rouse model considers only the long - time behavior of the chain and long scales of distances , while in the case of the freely jointed chain the short - time behavior is taken into account and the chain is observed at a short scale of distance . a chain model which encompasses both the regimes of the rouse model and of the gnlsm has been proposed in eq . ( [ xtl ] ) . the equilibrium limit of the gnlsm has been recovered . it gives the expected result in agreement with ref . last but not last , the dynamics of a random chain has been investigated also from the classical point of view . the equivalence of the expressions of the classical lagrangian of the chain computed starting from cartesian and polar coordinates has been verified . to conclude , we would like to mention the problems which are still open and possible further developments of our work . for simplicity , we have restricted our analysis to a two - dimensional chain . however , its extension to any dimensions is not difficult . preliminary work in three dimensions can be found in ref . it turns out that more dimensions allow the possibility of performing the continuous limit in different ways , so that one could end up with a flexible chain or with a rigid chain which privileges only certain directions along a fixed axis . there should be also no problem in switching on the interactions among the beads composing the chain . to this purpose , one may use the path integral methods applied to stochastic differential equations explained in ref . only the inclusion of the hydrodynamic interactions requires still some work . it is not simple to provide for these interactions a lagrangian based formulation as that developed here for a continuous chain . however , hydrodynamic interactions have been already implemented in the path integral formalism in ref . @xcite using the martin - siggia - rose formalism . work is in progress in order to extend the results of @xcite to the freely jointed chain discussed in this work . another open question is how the chain behaves in the short time regime when it is stretching under the action of a force . this could be interesting in the biophysics of dna @xcite . finally , work is in progress in order to linearize the gnlsm applying the same approach used in the case of the standard nonlinear sigma model . in this way it would be possible to study the gnlsm as the strong coupling limit of its linear version with the help of the techniques of refs.@xcite . this work has been financed in part by the polish ministry of science , scientific project n202 156 31/2933 . f. ferrari gratefully acknowledges also the support of the action cost p12 financed by the european union . the authors would also like to thank v. g. rostiashvili for fruitful discussions and the anonymous referees of ref . @xcite and of this work for their precious remarks and suggestions . finally , f. ferrari and j. paturej are grateful for the nice hospitality to t. a. vilgis and to the max planck institute for polymer research , mainz , germany . in this appendix we wish to prove eq . ( [ funcid ] ) . to this purpose , we start from the path integral : @xmath346 upon the transformation : @xmath347 we obtain : @xmath348 where we have made use of the fact that : @xmath349 now it is possible to eliminate the variable @xmath350 by performing in @xmath297 the substitution : @xmath351 as a consequence , the path integration over @xmath350 appearing in ( [ idel ] ) may be replaced by a path integration over the new variable @xmath352 : @xmath353 in the above equation the determinant @xmath354 is the functional determinant giving the jacobian of the transformation ( [ substfd ] ) , so that : @xmath355 applying eqs . ( [ ideaaa ] ) and ( [ idetwo ] ) in ( [ idef ] ) we get : @xmath356 finally , we perform in eq . ( [ lastbeforesincos ] ) the substitution @xmath357 . in this way @xmath297 may be rewritten as a path integral over @xmath358 : @xmath359 noting that @xmath360 we obtain , apart from an irrelevant constant @xmath361 , the final result : @xmath362 to prove the identity ( [ eqwwtp ] ) , we write the functional @xmath115function in the right hand side of eq . ( [ eqwwtp ] ) with the help of its fourier representation : @xmath363 since @xmath364 , eq . ( [ frep ] ) becomes : @xmath365 due to the fact that @xmath366 is always different from zero , one may perform the change of variables : @xmath367 applying the substitution ( [ substllt ] ) in ( [ freppol ] ) we obtain : @xmath368 which coincides with eq . ( [ eqwwtp ] ) up to the irrelevant constant @xmath369 . m. muthukumar , _ shifting paradigms in polymer crystallization _ , in _ progress in understanding of polymer crystallization _ , g. reiter and g. r. strobl ( eds . ) , lecture notes in physics * 714 * , ( springer , berlin , heidelberg 2007 ) , p. 1 sommer , _ theoretical aspects of the equilibrium state of chain crystals _ , _ ibid . _ , p. 19 ; l. jarecki , _ kinetic theory of crystal nucleation under transient molecular orientation _ _ , p. 66 ; g. strobl , _ multiphase model describing polymer crystallization and melting _ , _ ibid._. p. 481 . w. tomaszewski and p. pieraski , _ new jour . phys . _ * 7 * ( 2005 ) , 45 . h. kleinert , _ gauge fields in condensed matter _ , vol 1 , ( world scientific , 1990 ) . s. stepanow , _ j. phys . a : math . gen . _ * 17 * ( 1984 ) , 3041 ; _ j. phys . * 17 * ( 1984 ) , l507 . o. hallatschek , e. frey and k. kroy , _ phys . lett . _ * 94 * ( 2005 ) , 077804 ; _ phys . * 75 * ( 2007 ) , 031905 . h. kleinert , _ phys . rev . _ * d57 * ( 1998 ) , 2264 ; h. kleinert , _ phys . lett . _ * b434 * ( 1998 ) , 74 , [ arxiv : cond - mat/9801167 ] .

as part of an effort to understand the origin of open clusters , we present a statistical analysis of the currently observed pleiades . starting with a photometric catalog of the cluster , we employ a maximum likelihood technique to determine the mass distribution of its members , including single stars and both components of binary systems . we find that the overall binary fraction for unresolved pairs is 68% . extrapolating to include resolved systems , this fraction climbs to about 76% , significantly higher than the accepted field - star result . both figures are sensitive to the cluster age , for which we have used the currently favored value of 125 myr . the primary and secondary masses within binaries are correlated , in the sense that their ratios are closer to unity than under the hypothesis of random pairing . we map out the spatial variation of the cluster s projected and three - dimensional mass and number densities . finally , we revisit the issue of mass segregation in the pleiades . we find unambiguous evidence of segregation , and introduce a new method for quantifying it .

0801.1329

open clusters , with their dense central concentrations of stars , are relatively easy to identify . over a thousand systems are known , and the census is thought to be complete out to 2 kpc @xcite . because the clusters are no longer buried within interstellar gas and dust , their internal structure and dynamics is also more accessible than for younger groups . despite these favorable circumstances , many basic questions remain unanswered . most fundamentally , how do open clusters form ? all observed systems have undergone some degree of dynamical relaxation . thus , the present - day distribution of stellar mass differs from the one just after disruption of the parent cloud . recovering this initial configuration will clearly be of value in addressing the formation issue . but such reconstruction presupposes , and indeed requires , that we gauge accurately the stellar content of _ present - day _ clusters . even this simpler issue is a non - trivial one , as we show here . we consider one of the most intensively studied open clusters , the pleiades . again , our ultimate goal is to trace its evolution from the earliest , embedded phase to the present epoch . focusing here on the latter , we ask the following questions : what are the actual masses of the member stars ? do they follow the field - star initial mass function ? how many of the members are single , and how many are in binary pairs ? are the primary and secondary masses of binaries correlated ? what is the overall density distribution in the cluster ? what is the evidence for mass segregation , and how can this phenomenon be quantified ? all of these questions have been addressed previously by others . @xcite constructed a global mass function for the pleiades . their method was to assign masses based on the observed distribution of @xmath0-magnitudes . a more accurate assessment should account for the photometric influence of binaries . several studies directed specifically at binaries have probed selected regions for spectroscopic pairs @xcite . however , the overall binary fraction has not been carefully assessed , despite some preliminary attempts @xcite . a fuller investigation of these and the other issues raised requires statistical methods ; these should prove generally useful in characterizing stellar populations . section 2 describes our approach , which employs a regularized maximum likelihood technique ( e.g. * ? ? ? a similar method has been applied to other astronomical problems , including the reconstruction of cloud shapes @xcite and the investigation of binarity within globular clusters @xcite . our study is the first to apply this versatile tool to young stellar groups . in doing so , we also relax many of the restrictive assumptions adopted by previous researchers . section 3 presents our derived mass function for the pleiades , along with our results for binarity . the density distribution , as well our quantification of mass segregation , are the topics of section 4 . finally , section 5 summarizes our findings , critically reexamines the binarity issue , and indicates the next steps in this continuing study . the basic problem is how to assign stellar masses to all the point - like sources believed to be cluster members . in many cases , the source is actually a spatially unresolved binary pair . more rarely , it is a triple or higher - order system ; for simplicity , we ignore this possibility . the available observational data consists of photometry in several wavebands . given the inevitable , random error associated with each photometric measurement , it is not possible to identify a unique primary and secondary mass for each source . instead we adopt a statistical approach that finds the relative probability for each source to contain specific combinations of primary and secondary masses . we introduce a stellar mass probability density , to be denoted @xmath1 . this two - dimensional function is defined so that @xmath2 is the probability that a binary system exists with primary mass ( hereafter expressed in solar units ) in the interval @xmath3 to @xmath4 and secondary mass from @xmath5 to @xmath6 . single stars are viewed here as binaries with . we normalize the function over the full mass range : @xmath7 note that we integrate the secondary mass @xmath5 only to @xmath3 , its maximum value . furthermore , we set the lower limit of @xmath5 to 0 , in order to account for single stars . it is assumed that for . here , the global minimum mass @xmath8 is taken to be 0.08 , the brown dwarf limit . we consider @xmath9 to be the highest mass still on the main sequence . the age of the pleiades has been established from lithium dating as @xmath10 yr @xcite . this figure represents the main - sequence lifetime for a star of @xmath11 @xcite , which we adopt as @xmath9 . we examine separately the handful of stars that are ostensibly more massive than @xmath9 , and hence on post - main - sequence tracks . for these 11 sources , we assigned approximate masses from the observed spectral types , and obtained data on their known unresolved binary companions ; this information was taken from the bright star catalogue @xcite . these systems were then added by hand to our mass functions . finally , we ignore the brown dwarf population , thought to comprise from 10 to 16% of the total system mass @xcite . the most direct method for evaluating @xmath12 would be to guess its values over a discrete grid of @xmath3- and @xmath5-values . each @xmath13 pair makes a certain contribution to the received flux in various wavebands . thus , any guessed @xmath12 yields a predicted distribution of fluxes , which will generally differ from that observed . one changes the guessed function until the observed flux distribution is most likely to be a statistical realization of the predicted one . unfortunately , this straightforward approach is impractical . the basic difficulty is the mass - sensitivity of stellar luminosities . for secondary masses @xmath5 only modestly less than @xmath3 , the binary is indistinguishable photometrically from a single star having the primary mass . a @xmath14 main - sequence star , for example , has a @xmath15-band flux of 10.81 mag at the 133 pc distance of the pleiades @xcite . pairing this star with a @xmath16 secondary ( which is not yet on the main sequence ) only changes the flux to 10.68 mag . in summary , the function @xmath12 evaluated in this way is unconstrained throughout much of the @xmath17 plane . since binaries of even modestly large primary to secondary mass ratio are difficult to recognize observationally , we need to infer their contribution indirectly , within the context of a larger theoretical framework . the physical origin of binaries is far from settled @xcite . there is a growing consensus , however , that most pairs form together , perhaps within a single dense core . the accumulating observations of protobinaries , i.e. , tight pairs of class 0 or class i infrared sources , have bolstered this view ( e.g. , * ? ? ? if binaries indeed form simultaneously , but independently , within a single dense core , there is no reason to expect a strong correlation between the component masses . ( such a correlation would be expected if the secondaries formed within the primaries circumstellar disks , for example . ) a credible starting hypothesis , then , is that each component mass is selected randomly from its own probability distribution . if the formation mechanism of each star is identical , then these distributions are also the same . that is , we postulate that the true binary contribution to @xmath12 is @xmath18 , where the single - star probability density @xmath19 is properly normalized : @xmath20 of course , not all sources are unresolved binaries . let @xmath21 represent the fraction of sources that are . we suppose that this fraction is independent of stellar mass , provided the mass in question can represent either the primary or secondary of a pair . while this hypothesis is reasonable for low - mass stars , it surely fails for o- and early b - type stars , which have an especially high multiplicity ( e.g. , * ? ? ? * ; * ? ? ? * ) . such massive objects , however , are not present in clusters of the pleiades age . accepting the assumption of a global binary fraction , we have a tentative expression for the full stellar mass probability : @xmath22 here , the first term represents true binaries , and the second single stars of mass @xmath3 . the factor of 2 multiplying the first term is necessary because of the restricted range of integration for @xmath5 in equation ( 1 ) . that is , this integration effectively covers only half of the @xmath3-@xmath5 plane . on the other hand , the normalization condition of equation ( 2 ) applies to both the primary and secondary star , and covers the full range of mass , from @xmath8 to @xmath9 , for each component . we shall see below that the strict random pairing hypothesis , as expressed in equation ( 3 ) , does not yield the optimal match between the predicted and observed distribution of magnitudes . the match can be improved , in the statistical sense outlined previously , if one allows for a limited degree of correlation between the primary and secondary masses within binaries . in other words , there is an apparent tendency for more massive primaries to be accompanied by secondaries that have a greater mass than would result from random selection . a simple way to quantify this effect is to consider the extreme cases . if there were _ perfect _ correlation between primary and secondary masses , then the contribution to @xmath12 from binaries would be @xmath23 . with no correlation at all , @xmath12 is given by equation ( 3 ) . we accordingly define a correlation coefficient @xmath24 , whose value lies between 0 and 1 . our final expression for @xmath1 uses @xmath24 to define a weighted average of the two extreme cases : @xmath25 note that the last righthand term , representing the probability of the source being a single star , is unaffected by the degree of mass correlation within binaries . reconstructing the stellar mass probability @xmath26 requires that we evaluate the constants @xmath21 and @xmath24 , as well as the single - star probability @xmath27 . to deal with this continuous function , we divide the full mass range into discrete bins of width @xmath28 . integrating over each bin , we find @xmath29 , the probability of a star s mass being in that narrow interval : @xmath30 we symbolize the full array of @xmath29-values by the vector @xmath31 , and similarly denote other arrays below . our task , then , is to find optimal values not only for @xmath21 and @xmath24 , but also for all but one element of @xmath31 . the normalization of @xmath19 is now expressed by the constraint @xmath32 which sets the last @xmath31-value . for each choice of @xmath21 , @xmath24 , and @xmath31 , equation ( 4 ) tells us the relative probability of binaries being at any location in the @xmath17 plane . after dividing the plane into discrete bins , each labeled by an index @xmath33 , we define @xmath34 as the predicted number of systems associated with a small bin centered on an @xmath13 pair . if @xmath35 is the total number of systems , i.e. , of unresolved sources in all magnitude bins , then our chosen @xmath21 , @xmath24 , and @xmath29-values yield the relative fractions @xmath36 . as an example , consider a bin @xmath33 in which @xmath3 and @xmath5 have different values lying between @xmath8 and @xmath9 . then the system is an unequal - mass binary , for which equation ( 4 ) gives @xmath37 here , @xmath38 is the element of @xmath31 corresponding to the selected @xmath3 , while @xmath39 is similarly associated with @xmath5 . for a bin where , the corresponding relation is @xmath40 note the additional term accounting for correlated binaries . note also that a factor of 2 has been dropped from equation ( 4 ) , since we are integrating only over that portion of the mass bin with . finally , if the system is a single star , so that , we have @xmath41 our observational data consists of a catalog of @xmath42 sources , each of which has an apparent magnitude in at least two broadband filters . ( in practice , these will be the @xmath43- and @xmath15-bands ; see below . ) as before , we divide this two - dimensional magnitude space into small bins . our choice of @xmath21 , @xmath24 , and @xmath31 leads not only to a predicted distribution in mass space , but also in magnitude space . let @xmath44 be the predicted number of sources in each magnitude bin , now labeled by the index @xmath45 . then we may write the transformation from the mass to the magnitude distribution as @xmath46 which may be recast in the abbreviated form @xmath47 here , @xmath48 is the _ response matrix _ , whose elements @xmath49 give the probability that a source in a mass bin @xmath33 is observed in a magnitude bin @xmath45 . in detail , this probability utilizes a theoretical isochrone in the color - magnitude diagram ( see section 3.1 ) . we must also account for random errors in the measured photometry . in other words , a given magnitude pair can have contributions from a range of mass pairs . it is for this reason that each element of @xmath50 involves a _ sum _ over all @xmath33-values . we previously showed how to obtain the relative mass distribution @xmath36 , not the actual @xmath51 itself . however , it is the latter that we need for equation ( 11 ) . to find @xmath35 , we sum equation ( 10 ) over all @xmath45-values , and demand that this sum be @xmath42 , the total number of observed sources : @xmath52 so that @xmath53 in summary , choosing @xmath21 , @xmath24 , and @xmath31 gives us @xmath36 through through equations ( 7 ) , ( 8) and ( 9 ) we then solve equation ( 12 ) for @xmath35 . supplied with knowledge of @xmath51 , we finally use equation ( 11 ) to compute @xmath50 . having chosen @xmath21 , @xmath24 , and @xmath31 , how do we adjust these so that the predicted and observed magnitude distributions best match ? our technical treatment here closely follows that in ( * ? ? ? * chapter 11 ) , but specialized to our particular application . let @xmath54 be the number of sources actually observed in each two - dimensional magnitude bin . we first seek the probability that the full array @xmath55 is a statistical realization of the predicted @xmath50 . the supposition is that each element @xmath44 represents the _ average _ number of sources in the appropriate bin . this average would be attained after a large number of random realizations of the underlying distribution . if individual observed values follow a poisson distribution about the mean , then the probability of observing @xmath54 sources is @xmath56 this probability is highest when @xmath54 is close to @xmath44 . the likelihood function @xmath57 is the total probability of observing the full set of @xmath54-values : @xmath58 we will find it more convenient to deal with a sum rather than a product . thus , we use @xmath59 the strategy is then to find , for a given @xmath55 , that @xmath50 which maximizes @xmath60 . for this purpose , we may neglect the third term in the sum , which does not depend on @xmath50 . we thus maximize a slightly modified function : @xmath61 since , for a given @xmath55 , each @xmath62 peaks at , maximizing @xmath63 is equivalent to setting @xmath50 equal to @xmath55 in equation ( 11 ) , and then inverting the response matrix to obtain @xmath51 . such a direct inversion procedure typically yields a very noisy @xmath51 , including unphysical ( negative ) elements . the solution is to _ regularize _ our result by employing an entropy term @xmath64 : @xmath65 the function @xmath64 is largest when the elements of @xmath31 are evenly spread out . adding @xmath64 to @xmath66 and maximizing the total guarantees that the @xmath31-values are smoothly distributed , i.e. , that @xmath27 is also a smooth function . in practice , we also want to vary the relative weighting of @xmath64 and @xmath66 . we do this by defining a _ regularization parameter _ @xmath67 , and then maximizing the function @xmath68 , where @xmath69 for any given value of @xmath67 , maximizing @xmath68 yields an acceptably smooth @xmath31 that reproduces well the observed data . for the optimal solution , we find that value of @xmath67 which gives the best balance between smoothness of the derived @xmath27 and accuracy of fit . we do this by considering another statistical measure , the bias . our observational dataset , @xmath55 , is an imperfect representation of the unknown probability density @xmath27 in two senses . as already noted , @xmath55 may be regarded as only one particular realization of the underlying distribution . even this single realization would directly reveal @xmath27 ( or , equivalently , @xmath31 ) if the sample size were infinite , which of course it is not . imagine that there were other realizations of @xmath27 . for each , we employ our maximum likelihood technique to obtain @xmath31 . averaging each @xmath29 over many trials yields its expectation value , @xmath70 . however , because of the finite sample size , @xmath70 does not necessarily converge to the true value , @xmath71 . their difference is the bias , @xmath72 : @xmath73 the values of the biases , collectively denoted @xmath74 , reflect the sensitivity of the estimated @xmath31 to changes in @xmath55 . following ( * ? ? ? * section 11.6 ) , we define a matrix @xmath75 with elements @xmath76 the bias is then given by @xmath77 to evaluate the derivatives in @xmath75 , we consider variations of the function @xmath68 about its maximum . in matrix notation , @xmath78 where the matrix @xmath79 has elements @xmath80 and @xmath81 is given by @xmath82 since @xmath68 is a known function of both the @xmath31 and @xmath55 , the derivatives appearing in both @xmath79 and @xmath81 may be evaluated analytically . another matrix that will be useful shortly is @xmath83 , whose elements @xmath84 are also known analytically from equations ( 7)-(9 ) . to determine the regularization parameter @xmath67 appearing in @xmath68 , we seek to minimize the biases . in practice , we consider the weighted sum of their squared values : @xmath85 and vary @xmath67 to reduce this quantity . be reduced to @xmath86 , the number of free parameters in our fit . as noted by ( * section 11.7 ) , the average @xmath72-value at this point is about equal to its standard deviation , and so is statistically indistinguishable from zero . ] here , the @xmath87 are diagonal elements of @xmath88 , the covariance of the biases . recall that the elements of @xmath74 are here considered to be random variables that change with different realizations . we find the covariance matrix @xmath88 by repeated application of the rule for error propagation ( * ? ? ? * section 1.6 ) . we begin with @xmath89 , the covariance of @xmath55 . since these values are assumed to be independently , poisson - distributed variables , @xmath89 has elements @xmath90 here we have used the fact that the variance of the poisson distribution equals its mean , @xmath44 . in this equation only , both @xmath33 and @xmath45 range over the @xmath91-values , i.e. , @xmath89 is a square matrix . we next consider @xmath92 , the covariance of @xmath31 . this is given by @xmath93 finally , we obtain the desired @xmath88 by @xmath94 the matrix @xmath95 in this last equation has elements @xmath96 differentiating equation ( 21 ) and applying the chain rule , we find @xmath95 to be @xmath97 where @xmath98 is the identity matrix . thus far , we have focused on determining global properties of the cluster , especially the mass function @xmath27 . we also want to investigate the spatial distribution of stellar masses . for this purpose , we need not perform another maximum likelihood analysis . the reason is that we can treat the mass distribution at each radius as a modification of the global result . we divide the ( projected ) cluster into circular annuli , each centered on a radius @xmath99 . what is @xmath100 , the number of sources in an annulus that are within mass bin @xmath33 ? ( as before , each bin is labeled by the masses of both binary components . ) the quantity we seek is @xmath101 here , @xmath102 is the probability that a source observed within magnitude bin @xmath45 has component masses within bin @xmath33 . in principle , this probability depends on radius . for example , the source could have a high probability of having a certain mass if there exist many such stars in that annulus , even stars whose real magnitude is far from the observed one . if we discount such extreme variations in the underlying stellar population , then we may approximate @xmath103 by its global average . the factor @xmath104 in equation ( 32 ) is the estimated number of sources at radius @xmath99 in magnitude bin @xmath45 . we only compute , via the maximum likelihood analysis , @xmath44 , the estimated number of sources in the entire cluster . but we also know @xmath105 , the _ observed _ number of sources in the annulus . if the total number of observed sources , @xmath54 , is non - zero , then we take @xmath106 in case , then @xmath105 vanishes at all radii . we then assume @xmath107 where @xmath108 is the observed source number of all magnitudes in the annulus . in the end , equation ( 32 ) attributes the radial mass variation to changes in the local magnitude distribution , rather than to improbable observations of special objects . it is clear that the global @xmath103 must be closely related to the response matrix @xmath48 , which is the probability that a source with a given mass has a certain magnitude . the precise relation between the two follows from bayes theorem : @xmath109 the numerator of the fraction is the probability that a source at any radius lies within the mass bin @xmath33 , while the denominator is the probability of it lying within magnitude bin @xmath45 . note that @xmath35 and @xmath110 need not be identical . the first quantity is the estimated number of sources covering all possible masses . the second is the observed number in the magnitude range under consideration . in practice , this range is extensive enough that the two are nearly the same . we thus write @xmath111 using this last equation , along with equation ( 33 ) , equation ( 32 ) now becomes @xmath112 for those terms where , equation ( 34 ) tells us to replace the ratio @xmath113 by @xmath114 . summing @xmath115 over all @xmath33 and dividing by the area of the annulus gives the projected surface number density of sources as a function of radius . the total mass in each annulus is @xmath116 where @xmath117 is the sum of the masses of both binary components in the appropriate bin . division of @xmath118 by the annulus area gives the projected surface mass density . under the assumption of spherical symmetry , the corresponding volume densities are then found by the standard transformation of the abel integral ( * ? ? ? * section 4.2.3 ) . we begin with the observational data . figure 1 is a dereddened @xmath119 color - magnitude diagram for the pleiades , taken from the recent compilation by @xcite . shown are all sources which have high membership probability , as gauged by their colors , radial velocities , and proper motions ( see , e.g. , @xcite for one such proper motion study . ) the lower open circles correspond to probable brown dwarfs ; we exclude such objects from our study . most brown dwarfs are too faint to be observed , and the population , in any case , is more sparsely sampled . ( the magnitude cutoffs corresponding to a @xmath120 object are and . ) after also exluding the 11 bright , post - main - sequence stars , shown here as large , filled circles , we have a total sample size of @xmath121 . the solid curve near the lower boundary of the stellar distribution is a combination of the theoretical zero - age main sequence for @xcite and , for lower - mass stars , a pre - main - sequence isochrone @xcite .- band magnitudes to make them consistent with the 2mass @xmath122-band used in stauffer s catalog . see @xcite for this transformation . ] the isochrone is that for the measured cluster age of 125 myr . our basic assumption is that the observed scatter about this curve stems from two effects - binarity and intrinsic errors in the photometric measurements . we do not consider , for now , possible uncertainty in the cluster s age . ( see section 5 for the effect of this uncertainty . ) we also ignore the finite duration of star formation . this duration is roughly @xmath123 yr @xcite , or about 10% of the cluster age . of 0.12 . we therefore constructed figure 1 by applying uniformly the corresponding @xmath124- and @xmath125-values of 0.06 and 0.01 , respectively . ] after doing a polynomial fit to the mass - magnitude relations found by @xcite and @xcite , we have analytic expressions for @xmath126 and @xmath127 , the absolute @xmath43- and @xmath15-magnitudes for a binary consisting of a primary mass @xmath3 and secondary @xmath5 . here , the superscripts indicate that the magnitudes are theoretically derived . both @xmath128 and @xmath129 are calculated by appropriately combining the individual absolute magnitudes for @xmath3 and @xmath5 . we do not actually observe @xmath128 or @xmath129 for any source . what we have are dereddened , apparent magnitudes in these wavebands . using the pleiades distance of 133 pc , these apparent magnitudes are readily converted to absolute ones , @xmath130 and @xmath131 . the salient question is : given a source with intrinsic magnitudes @xmath128 and @xmath129 ( or , equivalently , with masses @xmath3 and @xmath5 ) , what is the probability that it is observed to have magnitudes @xmath130 and @xmath131 ? here we confront the issue of photometric errors . we assume the errors in the two wavebands to be normally distributed . then the relevant probability density is @xmath132\,\,.\ ] ] here , is the probability of observing a source in magnitude bin @xmath45 , centered on , and having widths @xmath133 and @xmath134 . the quantities @xmath135 and @xmath136 in equation ( 39 ) are the standard deviations of the photometric measurements . according to @xcite , the average standard deviation in the @xmath43-band is about 0.15 . figure 2 , constructed from table 2 of @xcite , shows that @xmath136 is generally lower , and rises steeply with @xmath131 for the dimmest sources . occurs because the observed @xmath15-magnitudes are approaching the sensitivity limit of the observations . many of the @xmath43-band measurements come from poss ii plates , for which the limit is 18.5 @xcite . another large source of data was the observations of @xcite , whose limiting magnitude was 19.7 . our lower cutoff for brown dwarfs corresponds to an apparent @xmath43-magnitude of 17.7 , so the rise in our @xmath135 should be modest . ] the two branches of the curve presumably represent the results from two different observations . we do a polynomial fit to the upper , majority , branch , and thus have an explicit expression for @xmath137 . suppose now that @xmath3 and @xmath5 are centered within a mass bin @xmath33 , which has widths @xmath138 and @xmath139 . then the response matrix @xmath49 is obtained by integrating @xmath140 over the magnitude bin , then averaging over the mass bin : @xmath141 the magnitude integrals can be expressed in terms of error functions if we reinterpret @xmath136 as being a function of @xmath129 rather than @xmath131 . the remaining numerical integrals over @xmath3 and @xmath5 are performed by finding , for each @xmath13 pair , the @xmath128- and @xmath129-values from our polynomial fits to the mass - magnitude relations . with the response matrix in hand , we are ready to input the catalog of source magnitudes . before turning to the pleiades itself , we first employed a number of synthetic datasets , in order to test various aspects of the code . we shall describe these tests shortly . first , however , we summarize the standard procedure we adopted for the analysis of any cluster , real or synthetic . the basic problem , we recall , is to guess the optimal values of @xmath21 , @xmath24 , and @xmath31 that maximize the function @xmath68 , as given in equation ( 18 ) . the entropy part of @xmath68 , labeled @xmath64 , is directly a function of @xmath31 ( eq . ( 17 ) ) , while the modified likelihood function @xmath142 depends on the observed magnitude distribution @xmath55 and the guessed one @xmath50 ( see eq . the guessed @xmath31 does not yield @xmath50 itself , but the guessed mass distribution @xmath51 , through equations ( 7)-(9 ) . it is in this transformation that the binary fraction @xmath21 and correlation coefficient @xmath24 appear . finally , @xmath50 is obtained from @xmath51 via the response matrix ( eq . ( 11 ) ) . we begin the maximization procedure by first setting the regularization parameter @xmath67 to zero . since in this case , the optimal set of @xmath31-values will be uniformly distributed , while subject to the normalization constraint of equation ( 6 ) . we guess @xmath21 , @xmath24 , and @xmath31 , and vary them to maximize @xmath68 . for the actual maximization , we employ a standard simplex algorithm ( * ? ? ? * chapter 10 ) . the resulting best - fit parameters are then perturbed and the maximization rerun . this check , which may be redone several times , is done both to confirm convergence and to avoid becoming trapped in small , local maxima of the function @xmath68 . we record the relevant covariances and biases , to be used in estimating errors in predicted quantities and to set the optimal @xmath67-value . the next step is to increase @xmath67 slightly . we maximize @xmath68 in the same way as before , again recording covariances and biases . we again increase @xmath67 , repeating the entire procedure until @xmath143 , the weighted sum of the biases , starts to become acceptably small . at this point , the best - fit @xmath21 , @xmath24 , and @xmath31 have been established . as a first test of the procedure , we introduced an artificial cluster whose single - star probability , @xmath27 , we selected beforehand . sources were chosen randomly to have masses according to this distribution . in a certain fraction of the sources , our preset binary fraction @xmath21 , a second star was added to the source . the mass of this object was also randomly chosen from @xmath27 . thus , the correlation coefficient @xmath24 was initially zero . given both masses in a source , its intrinsic @xmath128 and @xmath129 are readily obtained . these magnitudes are smeared out into neighboring bins according to gaussian errors with the appropriate standard deviations @xmath135 and @xmath136 . thus , the observed " magnitude distribution , @xmath55 , is established . figure 3 shows two representative examples . in the left panel , the chosen @xmath27 , is a power law : . on the right , we used a log - normal distribution : @xmath144 \,\,,\ ] ] where @xmath145 is the normalization constant . the central mass was chosen as and the logarithmic width @xmath146 was 0.4 . the binary fraction @xmath21 was chosen to be 0.30 in the power - law example , and 0.68 for the log - normal distribution . the total source number @xmath42 was 10,000 in both cases . the smooth curve in both panels is @xmath147 , while the data points are the best - fit values of @xmath148 , where @xmath28 is the bin width . shown also is the estimated error for each value . this was derived from the covariance matrix @xmath92 , introduced in section ( 2.1 ) . specifically , the plotted error is @xmath149 . we divide each @xmath29 and its associated error by @xmath28 because @xmath29 is integrated over the bin ( eq . ( 5 ) ) . it is evident that the code reproduces well the assumed @xmath27 in these two examples . note that most of the scatter seen in both plots , especially in the left panel , was already present in the input data , which were finite realizations of the analytic distributions . the derived ( i.e. , predicted ) binary fractions , and , respectively , are also in good agreement . we had similar success when we reduced @xmath42 to 1245 , the actual number in our pleiades source catalog . in this smaller sample , the errors in our predicted mass function and binary fraction increased , roughly as @xmath150 . figure 4 , taken from a dataset with , shows in more detail how the regularization parameter @xmath67 was chosen . the figure also illustrates some of the subtlety involved in this procedure . plotted here , as a function of @xmath67 , is @xmath143 , defined in equation ( 26 ) . as @xmath67 is gradually increased , @xmath143 takes a sudden , sharp dip . after climbing back , @xmath143 then more slowly declines , eventually falling below , the number of tunable parameters in this maximization ( @xmath21 , @xmath24 , and 19 @xmath31-values ) . it is the second threshold ( in this case ) that marks the true stopping point . the earlier dip in @xmath143 is due , not to a decrease in the biases , but to a sharp _ increase _ in the covariances @xmath88 . this increase commonly occurs when the likelihood term @xmath66 starts to become comparable to the entropy @xmath64 in the full function @xmath68 . at that point , @xmath31 makes an abrupt shift away from its earlier , nearly uniform , distribution . with further increase in @xmath67 , @xmath31 settles down gradually to its optimal form . continuing our synthetic data tests , we next introduced a correlation between the primary and secondary masses . first , we generated uncorrelated pairs , as above . generalizing the prescription of @xcite , we then altered the secondary mass in each source according to @xmath151 here , @xmath152 , a preset number between 0 and 1 , represents our imposed degree of correlation . thus , setting yields the previous , uncorrelated case , while , for , every binary has equal - mass components . we ran our routine with a variety of input single - star mass functions , binary fractions , and degrees of correlation . our general result was that the predicted @xmath31 still reproduced well the synthetic @xmath27 . the binary fraction @xmath21 was similarly accurate . most significantly , the predicted @xmath24-value tracked the input quantity @xmath152 . figure 5 shows this relation . we conclude that our statistical model , while crudely accounting for correlation by inserting a fraction of equal - mass pairs , nevertheless mimics a smoother correlation , such as would be found naturally . the shaded patch in the figure is the probable region occupied by the real pleiades ; section 3.4 below justifies this assessment . , equations ( 7 ) and ( 8) would have been numerical integrals , and the derivative matrix @xmath83 in equation ( 25 ) would also have required numerical evaluation . ] one price we paid for our simplified account of correlation was that our matching of @xmath27 was less accurate than for randomly paired input binaries . consider , for example , the log - normal function of equation ( 41 ) . while our best - fit @xmath31 still reproduced @xmath27 reasonably well , the output function peaked at too high a mass compared to @xmath153 . the filled circles in figure 6 shows that this shift , @xmath154 , increased with the input @xmath152-value . concurrently , our output function was too narrow compared to the input @xmath146 . the ( negative ) difference , @xmath155 , displayed as open circles in figure 6 , was also more pronounced at higher @xmath152 . these systematic errors need to be considered when analyzing a real cluster . the two patched areas in the figure again represent the likely regime of the pleiades , as we explain shortly . we now present the results of applying our maximum likelihood analysis to the pleiades itself , i.e. , to the @xmath43- and @xmath15-magnitudes of 1245 sources from the catalog of @xcite . our best - fit binary fraction was , while the correlation coefficient was . ( these and other uncertainties represent only random statistical error , and do not include systematic effects ; see section 5 . ) we will discuss the implications of these findings in the following section . first , we examine the global distribution of stellar mass . the data points in figure 7 are the best - fit values of each @xmath148 . as in figure 3 , these points are a discrete representation of the single - star mass function @xmath27 . the large error bars on the two points at highest mass are due to the small number of sources gauged to be in the respective bins . the smooth , solid curve in figure 7 is a log - normal mass function that best matches the empirical @xmath31 . referring again to equation ( 41 ) , we find that and . the presence of a finite binary correlation affects both estimates . judging from figure 6 , our @xmath153 is overestimated by about 0.06 , while @xmath146 should be raised by 0.08 . each of our mass bins has contributions from both the primary and secondary components of binary pairs , as quantified by equation ( 4 ) . integrating the full stellar mass probablity @xmath12 over all secondary masses , we obtain @xmath156 , the probability distribution of primary masses : @xmath157 note that this distribution includes the possibility that the star is single ( ) . the solid curve in figure 8 is a log - normal fit to the empirical @xmath156 . shown for comparison as a dashed curve is the fit for @xmath27 from figure 7 . relative to the latter function , the primary distribution falls off at lower masses . this falloff simply reflects the fact that less massive objects are more likely to be part of a binary containing a higher - mass star , and thus to be labeled as secondaries . " by integrating @xmath12 over @xmath3 , from @xmath5 to @xmath9 . the function @xmath158 has an excess of low - mass stars and drops very steeply at high masses , as most such objects are primaries . ] in any event , we now see why the peak of @xmath156 , , is elevated with respect to the peak of @xmath27 . similarly , the primary distribution is also slightly narrower than the single - star mass function , with . the parameters of our log - normal approximation to @xmath156 may be compared to those of @xcite . these authors fit the entire mass function . since , however , they did not account for binarity , their results are more closely analogous to our primary distribution . their best - fit @xmath153 of 0.25 is close to ours , while their @xmath146 of 0.52 is higher , mostly because of their inclusion of the highest - mass members . these parameters are also close to those given by @xcite in his log - normal fit to the field - star initial mass function ( , ) . in figure 9 , we compare our single - star distribution @xmath27 to the field - star initial mass function ( dashed curve ) . the latter , which has been raised in the figure for clarity , is taken from @xcite , who did correct for binarity . it is apparent that @xmath27 itself veers away from the imf for both low- and high - mass objects . when these are added in , the resemblance improves . the open circles in figure 9 are pleiades low - mass stars and brown dwarfs found by @xcite . we have normalized their data , taken from a limited area of the cluster , so that their total number of stars matches ours within the overlapping mass range . no such normalization was necessary for the 11 b - type stars ( filled circles ) , which are from the catalog of @xcite but not included in our maximum likelihood analysis . adding both these groups not only improves the match to the imf , but also reveals a gap in the stellar distribution between about 2 and @xmath159 . a similar gap is seen in the pleiades mass function of ( * ? ? ? * see their figure 1 ) . our estimate for the total cluster mass , based solely on the 1245 catalog sources , is @xmath160 , with a 4% uncertainty . adding in the brightest stars brings this total to @xmath161 , with the same relative error . tests with synthetic data indicate that the systematic bias due to binary correlation raises this figure by roughly @xmath162 , to @xmath163 . addition of the brown dwarfs would cause a further , relatively small , increase . for comparison , @xcite found @xmath164 in stars , and an upper limit of @xmath165 for the brown dwarf contribution . @xcite used the virial theorem to estimate a total mass of @xmath166 , with a 28% uncertainty . direct integration of their mass function gave @xmath167 , with an 18% fractional error . the global binary fraction , , obtained in our analysis represents most , but not all , of the full binary population . omitted here are spatially resolved systems . for these , the primary and secondary appear as separate sources in the catalog of @xcite . counting resolved pairs raises the total fraction to about 76% , as we now show . the smallest angular separation between stars in the catalog is @xmath168 . at the pleiades distance of 133 pc , the corresponding physical separation is 1400 au . an edge - on circular binary of exactly this orbital diameter will still be unresolved , since the components spend most of their time closer together . the true minimum separation in this case is 2200 au . here , we have divided 1400 au by @xmath169 , which is the average of @xmath170 , for @xmath171 randomly distributed between 0 and @xmath172 . of course , only a relatively small fraction of binaries have separations exceeding 2200 au . the average total mass of our unresolved systems is @xmath173 . a binary of that total mass and a 2200 au diameter has a period of @xmath174 . what fraction of binaries have even longer periods ? our average _ primary _ mass is @xmath175 , corresponding to a spectral type of m1 . @xcite studied the period distribution of binaries containing m - type primaries . they claimed that this distribution was indistinguishable from that found by @xcite for g - type primaries . in this latter sample , 11% of the systems had periods greater than our limiting value . if the pleiades periods are similarly distributed , then the total fraction of binaries - both resolved and unresolved - becomes . even without this augmentation , our total binary fraction appears to be inconsistent with the available direct observations of the pleiades . thus , @xcite found visual pairs with periods between 40 and @xmath176 yr . using the period distribution of @xcite to extrapolate their observed binary fraction of 28% yields a total fraction of 60% . @xcite observed spectroscopic pairs with periods under 3 yr . a similar exercise again yields 60% . we note , however , that this ostensible concurrence of results is based on very broad extrapolations from limited data . ( see fig . 4 of @xcite . ) our derived binary fraction also exceeds that found in the field - star population . @xcite found that 57% of g stars are the primaries of binary or higher - order systems . note that our @xmath21 represents the total probability that a star is in a binary , whether as the primary or secondary component . since g stars are rarely secondaries , the comparison with @xcite is appropriate . on the other hand , m stars _ are _ frequently secondaries , so we would expect the fraction of binaries with m - type primaries to be reduced . @xcite has found that only 25% of m - stars are the primary components of binaries . our own analysis yields a binary fraction of 45% for m - star primaries , still in excess of the field - star result . if our finding of a relatively high binary fraction proves robust , it may provide a clue to the progenitor state of the pleiades and other open clusters . a similar statement applies to the correlation between component masses within binaries . our adopted method of gauging this correlation - inserting a fraction of equal - mass pairs in the mass function - is admittedly crude . nevertheless , the strong result ( ) is significant . referring back to figure 5 , we find that the pleiades correlation is equivalent to setting @xmath152 equal to about 0.65 in the alternative description of equation ( 42 ) . whatever the origin of the pleiades binaries , the primaries and secondaries were not formed by completely independent processes . we now employ the procedure outlined in section 2.4 to investigate both the surface and volumetric density as a function of the projected distance from the cluster center . the filled circles in figure 10 , along with the associated error bars , represent the surface number density of sources , measured in pc@xmath177 . the solid curve is a density profile using the empirical prescription of @xcite . here , the core radius is 2.1 pc , while the tidal radius is 19 pc . for comparison , @xcite also fit the surface number density profile of their low - mass stars to a king model , with a core radius of 2.3 - 3.0 pc . our profile is also at least roughly consistent with the cumulative number distribution displayed by @xcite . our best - fit tidal radius is slightly larger than the 17 pc cited by these authors . the surface mass density is plotted in an analogous fashion , again as a function of the projected radius . we show both the data points ( small open circles ) and , as the dashed curve , the best - fit king model . here , the core radius is 1.3 pc , and the tidal radius is 18 pc . note that the mass density profile falls off more steeply than the number density . thus , stars near the center are abnormally massive , a trend we shall explore more extensively below . figure 11 displays the corresponding volumetric densities . as we indicated , the deconvolution from surface profiles assumes spherical symmetry . in fact , the pleiades is slightly asymmetric , with a projected axis ratio of 1.2:1 @xcite . this ellipticity is thought to stem from the tidal component of the galactic gravitational field @xcite . under the spherical assumption , the filled circles and solid , smooth curve show the number density . here , the king model is the same used for the surface number density in figure 10 , but deprojected into three - dimensional space . figure 11 also shows , as the small open circles and dashed curve , the mass density as a function of spherical radius . again , the king model here is the deprojected version of that from figure 10 . the relatively rapid falloff in the mass , as opposed to the number , density is another sign of the tendency for more massive stars to crowd toward the center . the information we used in obtaining these profiles also gives us the spatial variation of the binary fraction @xmath21 . that is , we first used equation ( 37 ) to obtain @xmath178 , the predicted mass distribution in each radial annulus . recall that the distribution refers to both primaries and secondaries , as well as single stars . the binary fraction can thus be computed locally . to within our uncertainty , about @xmath179 at each radial bin , we find no variation of @xmath21 across the cluster . we have mentioned , in a qualitative manner , that more massive cluster members tend to reside nearer the center . in figure 12 , we explicitly show this trend . here , we plot @xmath180 , the average system mass ( primary plus secondary ) as a function of the projected cluster radius . it is apparent that @xmath180 monotonically falls out to about 4 pc . beyond that point , the average mass is roughly constant . the pattern here is consistent with mass segregation , but is not a clear demonstration of that effect . the problem is that figure 12 gives no indication of the relative populations at different annuli . if the outer ones are occupied by only a small fraction of the cluster , is mass segregation present ? to gauge any variation in the mass distribution of stars , that distribution must be calculated over an adequate sample size . previous authors have also claimed evidence of mass segregation , using various criteria . @xcite looked at the distribution of surface and volumetric number densities for a number of different mass bins . @xcite divided the population by magnitude into relatively bright and faint stars . they calculated the cumulative number as a function of radius for both groups , and found the bright stars to be more centrally concentrated . finally , @xcite fit king profiles to the surface density of various mass bins . as the average mass increases , the core radius shrinks . figure 13 gives a simpler and more clear - cut demonstration of the effect . here , we consider @xmath181 , the number of sources enclused in a given projected radius , divided by the total number of sources in the cluster . we also consider @xmath182 , the analogous fractional mass inside any projected radius . the figure then plots @xmath182 versus @xmath181 . in the absence of mass segregation , @xmath182 would equal @xmath181 at each annulus . this hypothetical situation is illustrated by the dotted diagonal line . in reality , @xmath182 rises above @xmath181 , before they both reach unity at the cluster boundary ( see upper smooth curve , along with the data points ) . this rise , as already noted , indicates that the innermost portion of the cluster has an anomalously large average mass , i.e. , that mass segregation is present . moreover , the _ area _ between the solid and dotted curves is a direct measure of the effect . in the case of perfect " mass segregation , a few stars near the center would contain virtually all the cluster mass . the solid curve would trace the upper rectangular boundary of the plot , and the enclosed area would be 0.5 . we thus define the _ gini coefficient _ , @xmath183 , as _ twice _ the area between the actual @xmath184 curve and the central diagonal . is also half the mean mass difference of radial shells , where that mass difference is normalized to the average system mass in the cluster . ] for the pleiades , we find that . it is possible , at least in principle , that this effect is due entirely to a few exceptionally massive stars located near the center . in fact , this is _ not _ the case . we have artificially removed the 11 brightest sources ( all late - b stars ) and recalculated @xmath182 versus @xmath181 . the result is shown by the dashed curve in figure 13 . while the rise above the diagonal is diminished , it is still present . that is , the intermediate - mass population exhibits segregation , as well . an interesting contrast is presented by another populous group , the orion nebula cluster ( onc ) . the distribution of stellar masses in this far younger system was recently studied by @xcite . figure 2 in that paper compares the stellar populations in the inner and outer halves of the cluster . apart from a few high - mass objects , the two populations are essential identical . we may also construct an @xmath184 curve for the onc , as shown here in figure 14 . the solid curve again lies well above the fiducial diagonal , ostensibly indicating mass segregation . however , removal of just the four trapezium stars gives a dramatically different result ( dashed curve ) that is virtually indistinquishable from the diagonal . all stars except this tiny subset are similarly distributed . the cluster is too young to have undergone true mass segregation , a conclusion drawn previously from n - body simulations @xcite . the trapezium represents a special population , one that probably formed just prior to cloud dispersal @xcite . in this paper , we have applied a versatile statistical tool , the maximum likelihood technique , to assess the distribution of stellar mass and the incidence of binaries in the pleiades . we began with a near - infrared catalog of cluster members . our basic assumption was that all cluster members share the same evolutionary age , and that any dispersion in the color - magnitude plane stems from binarity and random photometric errors . we were then able to infer the most probable distribution of masses , both for the cluster as a whole , and as a function of distance from its center . finally , we introduced a simple method for gauging the degree of mass segregation in the cluster . one of our surprising results is the relatively high fraction of binaries . we estimate that 68% of all systems in the cluster are unresolved binaries ; this figure climbs to about 76% if resolved pairs are included . these fractions are significantly higher than the accepted field - star result , so we should scrutinize them carefully . could they stem from an underestimate of the photometric error at faint magnitudes ? since the error in @xmath43 is greater than @xmath15 , we artificially increased the dispersion @xmath135 . we kept @xmath135 at 0.15 until , below which we increased it linearly , reaching at . after redoing the maximum likelihood analysis , the global binary fraction @xmath21 for unresolved pairs is unchanged . another potential difficulty is our neglect of the physical thickness of the cluster . we have assigned all members a distance of 133 pc , although there will naturally be some variation . however , this effect is also relatively small . from section 4.1 , the volumetric number density falls off with radius approximately as a king model with core and tidal radii of 2.1 and 19 pc , respectively . consider the front half of a spherical cluster with such a density distribution . it may readily be shown that the mean distance from the plane of the sky of any cluster member is . for a cluster at mean distance @xmath185 , the induced magnitude spread is , which is 0.04 in our case . although the actual spread in magnitudes is not gaussian , we have added this figure in quadrature to both @xmath135 and @xmath136 , and rerun the analysis . again , the binarity is unaffected . the errors due to both photometry and finite cluster thickness induce a symmetric spread in stellar magnitudes . that is , they scatter as many sources below the fiducial isochrone as above it . thus , they can not reduce the estimated binarity , which stems from an excess of stars above the isochrone . one systematic error that _ would _ affect @xmath21 is an overestimation of the cluster distance . if @xmath185 were lowered , the absolute magnitudes of all sources would decrease equally , as would the inferred @xmath21-value . quantitatively , the distance would have to decrease by about 15 pc to bring the binary fraction down to the field - star result for g - dwarf primaries . an error of this size for the average distance is excluded by current observations , for which the estimated uncertainty is only 1 pc @xcite . since our method relies solely on photometry to assess binarity , we can not distinguish between physically linked pairs and chance alignments . as mentioned in section 3.4 , the resolution limit of our data is @xmath168 , or at the distance of the pleiades . consider a star at a radius @xmath99 from the cluster center . its average number of neighbors within @xmath186 is , where @xmath187 is the projected surface number density of the cluster . since each ring of width @xmath188 contains stars , integration over all members yields the total number of chance alignments : @xmath189 where @xmath0 is the cluster s outer radius , using @xmath187 from figure 10 , we find . thus , chance alignments have no quantitative impact . yet another source of systematic error is the cluster age . we have adopted the lithium - based figure of 125 myr from @xcite . earlier estimates , using the main - sequence turnoff , yielded a range of answers . for example , @xcite found 100 myr . even this minor reduction affects our results , since it lifts the low - mass end of the isochrone toward higher luminosity . for an age of 100 myr , our analysis gives and . the binary fraction is augmented to 0.64 when we include resolved pairs . from figure 3 of @xcite , a 100 myr age corresponds to a lithium edge at , or at the pleiades distance . such a result seems incompatible with the lithium data shown in figure 2 of @xcite , but the total number of observations is relatively small . we conclude that the enhanced binarity is a real effect . what this fact tells us about the origin of open clusters remains to be seen . our next step in addressing this basic issue is to try and account theoretically for the empirical properties just obtained through our statistical analysis . we will ascertain , using direct numerical simulations , the range of initial states that can relax dynamically to the present - day pleiades . such a study will bring us one step closer to understanding the molecular cloud environments that give rise to open clusters . we are grateful to eric huff , who first suggested the use of the maximum likelihood technique for this problem and provided continued insight . we also benefited from conversations with james graham and geoff marcy . the referee s comments helped to improve the final presentation . this research was supported by nsf grant ast-0639743 . in section 4.2 , we introduced the gini coefficient geometrically , as twice the area between the @xmath184 curve and the diagnonal line representing zero mass segregation . alternatively , @xmath183 may be defined in terms of the mean mass difference between radial shells in the cluster . here we describe more precisely , and prove the equivalence of , this second interpretation . altering our previous notation , we now let @xmath190 be the average system mass ( primary plus secondary ) in a shell with outer radius @xmath99 . if there are many shells , then we may define a system number density @xmath191 , such that the number of systems between @xmath99 and @xmath192 is @xmath193 . the total number of systems at all radii is @xmath194 this total was called @xmath35 in the text . the average system mass throughout the entire cluster is @xmath195 we will be concerned with the _ relative mean difference _ in the mass of shells . this is @xmath196 we will prove that the gini coefficient , as defined in section 4.2 , is also @xmath197 . in the geometric definition of @xmath183 , we utilized the cumulative fractional number @xmath181 and cumulative fractional mass @xmath182 . these may be written in terms of the system number density : @xmath198 we will later need the differentials of @xmath181 and @xmath182 , which are @xmath199 in terms of @xmath181 and @xmath182 , the geometric definition of @xmath183 is then @xmath200 if the mass is centrally concentrated , as expected in a real stellar cluster , then at all radii , and . hypothetical clusters in which larger masses are preferentially located farther from the center would have . we now return to equation ( a3 ) and manipulate it to obtain @xmath183 , as defined above . a central assumption we will make is that @xmath190 declines monotonically with @xmath99 . we may then expand the righthand side of equation ( a3 ) . we split the integral over @xmath201 into two parts , one for and the other for . under our assumption , in the first integral , and in the second . after pulling @xmath191 from the @xmath201-integration , we further split the integrands to find @xmath202 the first two terms of the integrand are @xmath203 the third term is @xmath204 while the fourth is @xmath205 putting equations ( a10 ) , ( a11 ) , ( a14 ) , and ( a17 ) back into equation ( a9 ) yields @xmath206\ , n ( r)\,dr \\ & = & 2\,\left ( \int_0 ^ 1\!f_m\,d f_n \,-\ , \int_0 ^ 1\!f_n\,d f_m \right ) \,\,,\end{aligned}\ ] ] where we have used both equations ( a1 ) and ( a2 ) to cancel the last two terms on the right side of equation ( a18 ) , and equations ( a6 ) and ( a7 ) to transform the remaining two terms . the second integral in the last equation is @xmath207 since @xmath182 and @xmath181 attain their upper and lower bounds simultaneously . using this result , equation ( a19 ) becomes @xmath208 adams , j. d. , stauffer , j. r. , monet , d. g. , skrutskie , m. f & beichman , c. a. 2001 , , 121 , 2053 baraffe , i. , chabrier , g. , allard , f. , & hauschildt , p. h. 1998 , , 337,403 bihain , g. et al . 2006 , , 458 , 805 binney , j. & merrifield , m. galactic astronomy , princeton : princeton u. press bonnell , i. a. & davies , m. b. 1998 , , 298 , 93 bouvier , j. , rigaut , f. & nadeau , d. 1997 , , 323 , 139 brown , a. g. a. 2001 , , 11 , 89 chabrier , g. 2003 , , 115 , 763 cohen , m. , wheaton , m. a. , & megeath , s. t. 2003 , , 126 , 1090 cowen , g. 1998 , statistical data analysis , oxford : oxford u. press deacon , n. r. & hambly , n. c. 2004 , , 416 , 125 de la fuente marcos , 1996 , , 314 , 453x dias , w.s . , alessi , b.s . , moitinho , a. , & lepine , j.r.d . 2002 , , 389 , 871 duquennoy , a. & mayor , m. 1991 , , 88 , 241 fischer , d. a. & marcy , g. w. 1992 , , 396 , 178 garcia , b. & mermilliod , j. c. , 368 , 122 haisch , k. e. , greene , t. p. , barsony , m. & stahler , s. w. 2004 , , 127 , 1747 hambly , n. c. , hawkins , m. r. s. & jameson , r. f. 1993 , , 100 , 607 hoffleit , d. et al . 1991 , bright star catalogue , new haven : yale university press huff , e. h. & stahler , s. w. 2006 , , 644 , 355 huff , e. h. & stahler , s. w. 2007 , , 666 , 281 king , i. 1962 , , 67 , 471 kroupa , p. 2001 , , 322 , 231 lada , c. j. 2006 , , 640 , 63 mason , b. d. , henry , t. j. , hartkopf , w. i. , ten brummelaar , t. & soderblom , d. r. 1998 , , 116 , 2975 mermilliod , j .- c . , rosvick , j. m. , duquennoy , a. & mayor , m. 1992 , , 265 , 513 meynet , g. , mermilliod , j .- c . , & maeder , a. 1993 , , 98 , 477 moraux , e. , bouvier , j. , stauffer , j. r. , & cuillandre , j .- c . 2003 , , 400 , 891 moraux , e. , kroupa , p. , & bouvier , j. 2004 , , 426 , 75 palla , f. & stahler , s. w. 2000 , , 540 , 255 pinfield , d. j. , jameson , r. f. , & hodgkin , s. t. 1998 , , 299 , 955 pinfield , d. j. , hodgkin , s. t. , jameson , r. f. , cossburn , m. r. , hambly , n. c. , & devereux , n. 2000 , , 313 , 347 pinfield , d. j. , dobbie , p. d. , jameson , r. f. , steele , i. a. , jones , h. r. a. , & katsiyannis , a. c. 2003 , , 342 , 1241 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 2002 , numerical recipes in c : second edition , cambridge : cambridge u. press raboud , d. & mermilliod , j .- c . 1998 , , 329 , 101 romani , r. w. & weinberg , m. d. 1991 , , 372 , 487 schwartz , m. j. & becklin , e. e. 2004 , , 130 , 2352 sen , a. 1997 , on economic inequality , oxford : clarendon press siess , l. , dufour , e. , & forestini , m. 2000 , , 358 , 593 soderblom , d. , nelan , e. , benedict , g. , mcarthur , b. , ramirez , i. , spiesman , w. , & jones , b. 2005 , , 129 , 1616 stauffer , j. r. , schultz , g , & kirkpatrick , j. d. 1998 , , 499 , 199 stauffer , j. r. et al . 2007 , , 172 , 663 steele , i. a. & jameson , r. f. 1995 , , 272 , 630 tassis , k. 2007 , , 379 , l50 wielen , r. 1974 , in stars and the milky way system , ed . l. n. mavridis ( berlin : springer - verlag ) , p. 326 zinnecker , h. & mathieu , r. d. ( eds ) 2001 , the formation of binary stars , san francisco : asp

we present deep polarimetric observations at 1420 mhz of the european large area _ iso _ survey north 1 region ( elais n1 ) as part of the dominion radio astrophysical observatory _ planck _ deep fields project . by combining closely spaced apertures synthesis fields , we image a region of 7.43 square degrees to a maximum sensitivity in stokes and ofjy beam , and detect 786 compact sources in stokes . of these , 83 exhibit polarized emission . we find that the differential source counts ( ) for polarized sources are nearly constant down tojy , and that these faint polarized radio sources are more highly polarized than the strong source population . the median fractional polarization is% for polarized sources with stokes flux density between 1 and 30 mjy ; approximately three times larger than sources with mjy . the majority of the polarized sources have been identified with galaxies in the _ spitzer _ wide area infrared extragalactic survey ( swire ) image of elais n1 . most of the galaxies occupy regions in the irac 5.8/3.6 m vs. m color - color diagram associated with dusty agns , or with ellipticals with an aging stellar population . a few host galaxies have colors that suggests significant pah emission in the near - infrared . a small fraction , 12% , of the polarized sources are not detected in the swire data . none of the polarized sources in our sample appears to be associated with an actively star - forming galaxy .

0705.2736

observation of polarized radiation at radio wavelengths is one of the prime means to study the roles of magnetic fields in astrophysics , through synchrotron emissivity which samples magnetic fields in relativistic plasmas , and through faraday rotation , created by radiation propagation through magnetized thermal plasmas . understanding the origin and evolution of magnetic fields is a key science goal of the the square kilometre array , a next - generation radio telescope under development by the international community @xcite . the primary observational tool to study the magnetic universe will be an ska all - sky rotation measure survey of background radio sources and diffuse galactic emission down to polarized flux density levels of @xmath11jy @xcite . however , while modern source counts approach flux density sensitivities of @xmath12jy in total intensity @xcite , very little is known about the polarization properties of the faint radio source population . the most extensive analyses of polarization of compact extragalactic sources were carried out by @xcite and @xcite , who used the nvss data @xcite to derive statistical polarization properties for @xmath1330,000 sources with @xmath14 mjy . similarly , @xcite used nvss sources with total flux density greater than 80 mjy to extrapolate polarized source counts to @xmath9jy levels . @xcite found that the mean fractional polarization of radio sources in the nvss brighter than 80 mjy was anti - correlated with flux density , especially for steep - spectrum radio sources ( @xmath15 , for @xmath16 ) . @xcite confirmed this result for the median of the fractional polarization for steep - spectrum sources only ( 87% of their sample ) , but found no significant trend for flat spectrum sources ( 13% of their sample ) . @xcite also noted that the flat shape of the polarized source counts indicates a dependence of the fractional polarization on flux density . radio sources with @xmath14 mjy are predominantly associated with flat or steep spectrum active galactic nuclei . star forming galaxies begin to be a significant fraction of the population at flux densities less than a few mjy ( see e.g. hopkins 2000 , windhorst 2003 ) . however , there is still ongoing debate about the fraction of the sub - mjy radio sources which is radio - quiet active galactic nuclei ( agn ) @xcite . it is thus uncertain that polarization properties derived from the strong radio source population are applicable to the sub - mjy radio sources . we have begun a project called the drao planck deep fields to explore the high latitude sky at high sensitivity in polarized radio continuum and in atomic hydrogen emission . the project uses the drao synthesis telescope at 1.4 ghz to create deep images of two fields , one with a very low column of foreground material , the elais n1 region @xmath17 , and a larger region of highly structured infrared cirrus emission at @xmath18 . this paper reports initial results from the first 30% of observations of elais n1 . the elais n1 ( european large area _ iso _ survey north 1 ) field @xcite , is an area of approximately 2 square degree chosen for a mid - infrared survey of distant galaxies with _ iso_. the field was selected to minimize confusion with galactic cirrus and the zodiacal background , being one of the lines of sight to the extragalactic sky with minimum _ iras _ 100 @xmath9 m emission . a larger area that includes elais n1 was later observed by the _ spitzer _ wide area infrared extragalactic survey ( swire ) @xcite . the large amount of archival data from these extragalactic surveys make this region ideal for studies of the faint polarized radio source population . the drao synthesis telescope ( drao st ) is described in detail in @xcite . the telescope is a seven - element east - west array of 9-m diameter antennas . three antennas are moved to provide complete sampling of the uv plane from the shortest baseline ( 12.86 m ) to the longest baseline ( 617.18 m ) after a full synthesis of 12 times 12 hours . the array has a primary beam size 1072 ( fwhm ) at @xmath19 , which makes it an effective instrument for wide - field surveys . the first sidelobe of the synthesized beam is at the 3% level , and side - lobes farther from the main lobe of the beam are less than 0.5% . the first grating ring of the synthesized beam appears at @xmath20 from the main lobe at 1420 mhz , which is outside the field of view defined by the 10% sensitivity level of the primary beam . the telescope observes simultaneously the hi 21 cm line and continuum at 408 mhz and full polarimetry in four 7.5 mhz wide frequency bands centered around @xmath19 . the antennas have prime focus feeds and at 1420 mhz receive both right - hand ( r ) and left - hand ( l ) circularly polarized radiation . the observations and data processing techniques used in this paper follow those that have been used to obtain high - fidelity wide - field polarimetric images of the galactic plane with the drao st @xcite . the system temperature of the telescope as described by @xcite was @xmath21 , leading to an rms noise in a synthesized image of a single field of @xmath22jy beam@xmath3 at field center . however , starting in 2003 , the sensitivity was enhanced by a series of improvements , completed by the time the current observations began in 2004 . these improvements comprised installation of new low - noise amplifiers , modifications to telescope structures to reduce ground noise , and installation of shielding fences to further block ground radiation from entering the aperture . the system temperature of the seven individual antennas now spans the range @xmath23 to @xmath21 , and the overall system temperature is @xmath24 , leading to a field - center rms noise of @xmath25jy beam@xmath3 ( @xmath26 ) at 1420 mhz . a higher sensitivity and larger field of view are obtained by creating a mosaic of overlapping fields . the most uniform sensitivity across a mosaic is obtained if the fields are centered on a hexagonal grid . the field centre separations for the survey presented here is @xmath27 , which is @xmath28 of the fwhm diameter of the primary beam . observations for the drao elais n1 survey began in august 2004 and will continue to create a final mosaic of 30 fields . this paper presents 21-cm continuum polarimetry of the first 10 fields . figure [ surveyarea - fig ] shows the location of the 10-field mosaic in relation to the _ iso _ elais n1 field , and the swire survey of elais n1 . the theoretical maximum 1-@xmath29 sensitivity in our mosaic of 10 fields is @xmath30jy . complex gains for the center of each field were determined by observing the unresolved and unpolarized sources 3c147 and 3c295 between 12-hour observing runs . the absolute polarization angle was calibrated by observing the highly linearly polarized source 3c286 once every 4 days . the polarization images do not contain much flux , making self - calibration ineffective . therefore r and l gain solutions derived from self - calibration of the stokes @xmath4 images were applied to the polarization data as well . the resulting visibility data sets for each field were then corrected for the effects of instrumental polarization across the field of view ( which leads to leakage of stokes @xmath4 power into @xmath0 and @xmath1 ) . instrumental polarization is a complex quantity , with phase and amplitude terms . these were measured for the seven individual antennas with a holographic technique , using the unpolarized source 3c 295 , on a @xmath31 grid across the primary beam . after interpolation , these measurements were used to predict conversion of @xmath4 into @xmath0 and @xmath1 at any point in the beam based on clean component models from processing the @xmath4 image for a field . residual errors in instrumental polarization for an individual field are estimated at 0.25% for the field center , growing to 1% at a distance of @xmath32 . with a field separation in the mosaic image of @xmath33 the instrumental polarization in the central 3.6 square degrees of the mosaic remains less than 0.5% . towards the edge of the mosaic instrumental terms may be as large as 1% . after these initial procedures there are usually still confusing arc - like structures left in the images , centered on bright sources inside and outside the primary beam . these are the result of residual complex gain errors at large distance from the field center . effects from these are removed from the visibilities using a procedure called modcal , which is in principle a direction - dependent self - calibration @xcite . the antenna sidelobes are highly polarized , and sources outside the primary beam can produce strong spurious polarized signals . the sun is seen in the sidelobes whenever it is above the horizon , but its effects are usually confined to the shorter baselines because of its large extent . the effects of the sun were removed by making images centered on the sun s position and removing the response from the visibilities . terrestrial interference , which is always polarized , is another source of spurious polarization . interference effects in images appear as a spurious source concentrated around the north celestial pole , and can be largely eliminated by making an image at the pole and correcting the visibilities . radiation from the ground appears to be polarized , and can appear as a correlated signal in visibilities corresponding to short baselines ; this effect is more difficult to remove . in some particularly bad cases the data for affected interferometer spacings were simply flagged and removed . figure [ en21-fig ] shows the deep 21 cm continuum images of the elais n1 area in stokes @xmath4 , @xmath0 , and @xmath1 . the images are centered on @xmath34 = @xmath35 , @xmath36 = @xmath37 and cover an area of 7.43 square degrees ( the area within the thick gray line in figure [ surveyarea - fig ] ) . the angular resolution varies with the declination over the images and is given by @xmath38 . at the mosaic center the resolution is @xmath39 . the noise near the center of the mosaic ( white square in figure [ en21-fig ] ) is measured to be @xmath2jy in @xmath0 , and @xmath1 ( see [ detect - sec ] ) . the stokes @xmath4 image is not limited by confusion , but the noise is slightly higher at @xmath40jy near the center of the mosaic , probably because of a contribution from faint sources . the final images are virtually free of artifacts , so the sensitivity is limited by the noise . the rms brightness sensitivity is @xmath41 . the dynamic range near the center of the mosaic is more than 3000:1 in stokes @xmath4 . with very few exceptions , sources in figure [ en21-fig ] appear as compact ( nearly ) unresolved sources , that can be characterized by their peak intensity and a single polarization angle . this is consistent with expectations from the angular size flux density relation for extragalactic radio sources @xcite ; the median angular size of a @xmath42 source is @xmath43 . flux densities and positions of all sources in the pilot deep field images were measured with a source extraction algorithm that fits a two - dimensional gaussian and a background level to each source . the mosaic images were multiplied by the primary beam response function of the mosaic to obtain an image with a uniform noise level , equal to the noise level at the center of the image . this operation retains the correct signal - to - noise ratio for each source , but the resulting uniform noise level greatly facilitates automated source extraction . the inverse primary beam correction is applied to the measured flux densities of sources from the uniform noise map to transform back to true flux density . the rms noise level in the uniform noise images was measured by fitting a gaussian function to the distribution of amplitudes in the image . the result is shown in figure [ fig : q_unoise ] . the noise distribution in both the @xmath0 and @xmath1 images is well fitted by a gaussian with dispersion @xmath44jy . the polarized flux density image ( @xmath45 ) was searched for polarized sources . for gaussian noise in the @xmath0 and @xmath1 images with rms amplitude @xmath29 , the statistical distribution of the noise in a measurement of @xmath46 for a source with an intrinsic polarized flux density of @xmath47 is a rice distribution @xcite @xmath48 here @xmath49 is the modified bessel function of the first kind . for @xmath50 , the distribution reduces to a rayleigh distribution , @xmath51 which gives the probability distribution of pixel amplitudes in the @xmath46 image in the absence of polarized emission . the noise in the @xmath46 image has a non - zero mean and has higher probability of positive peaks above a given detection threshold than gaussian noise . we searched the @xmath46 map down to a level of @xmath52 , which has an equivalent probability for false positive signals to the @xmath53 level for a gaussian distribution . the measured polarized flux density @xmath46 was corrected for noise bias to obtain an estimate of @xmath54 through the relation @xmath55 , which is a good approximation if the signal to noise ratio is larger than 4 @xcite . the 83 sources detected are listed in table [ table_polsources ] , which gives the position of each source , the integrated total flux density , noise bias - corrected peak polarized intensity , polarization position angle , fractional polarization defined as the ratio of the bias - corrected peak polarized intensity to the peak total intensity , and the spectral index of the total flux density between @xmath56 and @xmath19 if the source appears in the wenss catalogue @xcite . to ensure possible spurious polarized sources due to instrumental polarization are not included , we conservatively remove sources with observed fractional polarization , @xmath57 , less than 1% . only two sources were removed from the sample for this reason . the intrinsic fractional polarization @xmath58 of radio sources provides astrophysical information about the nature of the polarized sources . however , the observed fractional polarization @xmath57 is sensitive to the noise in @xmath46 and in @xmath4 . in addition to @xmath46 being a biased estimator of @xmath54 , the error distribution of the ratio @xmath59 has strong non - gaussian wings , so @xmath57 is not a very accurate estimate of @xmath58 even for relatively high signal - to - noise sources . this is illustrated in figure [ polstat_demo ] , which shows the relation between @xmath57 and stokes @xmath4 flux density for an artificial sample of sources , all with @xmath60% and gaussian noise added with equal amplitudes in stokes @xmath4 , @xmath0 , and @xmath1 . only those sources with a @xmath46 flux density more than @xmath61 are shown . error bars represent standard error propagation in @xmath57 from the errors in @xmath46 and @xmath4 . the high values of @xmath57 at signal - to - noise ratio less than @xmath13100 is a result of the detection threshold in @xmath46 and the non - gaussian error statistics of @xmath57 . the effect is much larger than the polarization noise bias alone . these problems highlight the need for careful analysis of the effects of noise and polarization detection thresholds in studies of the fractional polarization of faint sources . previous studies have focused on the fractional polarization of bright radio sources in the nvss . @xcite , @xcite , and @xcite considered polarized sources in the nvss with stokes @xmath4 flux density stronger than 80 mjy , 100 mjy , and 80 mjy respectively . the high flux density thresholds in these studies were adopted to achieve a good level of completeness in @xmath57 down to the limit set by residual instrumental polarization ( @xmath62 ) . these studies found a @xmath57 distribution that decreases monotonically with increasing @xmath57 , and with a median @xmath63 . for these bright sources , noise effects are small , and the @xmath57 distribution should be close to the intrinsic @xmath58 distribution down to limits set by residual instrumental polarization . in this paper we investigate the shape of the @xmath58 distribution for much fainter stokes @xmath4 flux densities than those considered in previous work . the present data have angular resolution similar to the nvss , so differences in the measured degree of polarization because of a difference in resolution are not expected . our results can be compared directly with results based on the nvss . noise effects such as those illustrated in figure [ polstat_demo ] were taken into account by a monte carlo analysis . a set of simulated catalogs was generated to accurately represent the effects of noise and the polarized flux density detection threshold in the data . stokes @xmath4 flux densities of simulated sources were drawn from the fit to observed source counts by @xcite . we assume in these simulations that the stokes @xmath4 source counts of polarized sources have the same shape as those for all radio sources . this is a reasonable assumption because @xmath64 of radio sources in the nvss display significant polarization @xcite . the intrinsic stokes i flux density is multiplied by the degree of polarization , @xmath58 , drawn from an assumed @xmath58 distribution to obtain the intrinsic polarized intensity @xmath54 . the intrinsic stokes @xmath49 and @xmath54 of a simulated source are transformed to observed flux densities @xmath4 and @xmath46 by adding noise with statistical properties identical to the properties of the noise in the data . first , @xmath54 is converted into intrinsic stokes @xmath65 and @xmath66 , assuming a random polarization angle . the error in the flux density is assumed to consist of a part that is proportional to the noise in the image at the location of the source , and a part that is proportional to the flux density of the source , added in quadrature . the error in the flux density @xmath67 is evaluated as @xmath68 where @xmath67 represents the intrinsic flux density @xmath49 , @xmath69 , or @xmath66 , @xmath29 is the rms noise in the image and @xmath70 and @xmath71 are constants . the value of @xmath70 was determined from the rms variation of the flux density of bright sources in the ten fields after field registration @xcite . flux densities of sources brighter than @xmath72 varied by @xmath73 ( rms ) over the ten fields . from this we adopt @xmath74 . the value of the constant @xmath75 was taken to be that found by @xcite from monte carlo simulations for flux density errors in the westerbork northern sky survey ( wenss ) . the value of @xmath29 is different for each source , to represent variation of the noise with location in the mosaic . the distribution of polarized intensities of the resulting simulated sources , have the same statistical effects as the observed polarized intensities , including the effects of noise bias , the detection threshold and the variation of the noise with position in the mosaic . to derive the @xmath58 distribution directly from the data , we compare the distribution of the data in a @xmath76 diagram with the probability distribution for simulated source catalogues . figure [ beck - fig ] shows the observed data points , and the model probability distribution assuming the @xmath58 distribution function of @xcite . as expected , sources brighter than @xmath77 are represented well by this model . however , fainter sources in our sample are more highly polarized than predicted by this distribution . this is clearly visible in figure [ beck - fig ] for sources with @xmath78 , where an offset exists between the distribution of observed sources and the predicted ridge of maximum source density . a 2-dimensional kolmogorov - smirnov test @xcite rejected the hypothesis that the data were drawn from the simulated distribution at the 99.9% confidence level . the best fit @xmath58 distribution was derived by fitting source probability density distributions to the data in the @xmath79 plane for trial @xmath58 distributions . the trial @xmath58 distributions were represented by a low - order gauss - hermite series , also called a gram - charlier series , following the description of @xcite , @xmath80 . \label{ghseries - eq}\ ] ] we consider only modest deviations from a gaussian , since previous work on bright nvss sources suggests that the shape of the distribution is nearly gaussian . assuming that the @xmath58 distribution peaks at zero and declines monotonically with increasing @xmath58 , we use only the fourth term ( @xmath81 ) in equation [ ghseries - eq ] , which results in symmetric deviations . a coefficient @xmath82 means that the wings of the distribution are stronger than the wings of a gaussian distribution , as shown graphically by @xcite . higher order terms were not considered because of the limited size of our data sample at this time . the @xmath58 distribution defined by equation [ ghseries - eq ] thus has two free parameters , the gaussian standard deviation , @xmath83 , and the amplitude of the lowest order symmetric deviation from a gaussian , @xmath84 . the fits maximize the likelihood @xmath85 of the data as a function of these parameters , @xmath86 with the probability of an observed ( @xmath87 ) for a given @xmath83 and @xmath84 @xmath88 the product over @xmath89 is over all sources in the data , @xmath90 . whereas the sum over @xmath91 is over all simulated sources in the catalog , @xmath92 , for a particular @xmath83 and @xmath84 . typically @xmath93 . the values @xmath94 and @xmath95 are the observed errors in stokes @xmath4 and polarized flux density for the @xmath96 source . the maximum likelihood @xmath58 distribution was found through a grid search over the parameter space . the best fitting model has parameters @xmath97 , and @xmath98 . the probability density function of this model is shown along with the data in figure [ bestfit_pi ] . the uncertainty in the best - fitting parameters was evaluated empirically . three hundred randomly selected samples , each containing on average the same number of sources as the observed sample , were drawn from the best fitting simulated catalog . each of these samples was fitted with the maximum likelihood fit to evaluate the spread of the best - fitting parameters . two thirds of these fits yielded a @xmath83 within 1% of the maximum likelihood value @xmath99 . we conclude that @xmath100 . the fitted value of @xmath84 is not independent of @xmath83 , because a larger @xmath84 can partially compensate for a smaller @xmath83 . from the same 300 experimental fits , two thirds yielded a value @xmath101 . although the data are consistent with @xmath102 , those fits with @xmath84 contrained to be zero yield an average @xmath103 . the data therefore suggest that the @xmath58 distribution may be somewhat broader than a simple gaussian with @xmath104 . the best - fitting model was also subjected to a 2-dimensional kolmogorov - smirnov test . the hypothesis that the data were drawn from the best - fit distribution in figure [ bestfit_pi ] was rejected at the 98% confidence level . this is much better than the result for the @xcite distribution , but it is still suggestive that all the data are not well represented by the maximum likelihood model distribution . this is entirely the result of the fact that the best fitting gauss - hermite distribution does not fit the bright sources in the sample very well . we were able to produce a better fit by creating a set of hybrid simulated catalogs that use the @xcite distribution for brighter sources and our best fitting distribution for faint sources . the transition between these regimes was made smooth , with equal contributions from the two distributions at a flux density of 30 mjy . these hybrid catalogs fitted the data significantly better , with the best fitting model , using @xmath105 and @xmath98 at low flux density , passing the 2-dimensional kolmogorov - smirnov test . the median fractional polarization of the best fit @xmath58 distribution for polarized sources with stokes @xmath4 below @xmath106 is @xmath107)% . to derive source counts of the polarized sources , the effect of the varying noise level over the mosaic on source detection probability ( completeness correction ) was measured as a function of flux density by repeating the source detection on 1000 simulated images having the same noise characteristics and source density as the data . separate simulations were performed for total intensity and for polarized intensity images . each simulated image contained random gaussian noise smoothed to the resolution of the drao images , and the same uniform rms amplitude as the data . for each simulated polarization image , two independent noise images were created to represent the @xmath0 and @xmath1 images . sources were placed at random positions , with flux densities drawn from the source - count curve derived by @xcite between 0.1 and @xmath108 . sources below the detection limit were included in the simulations to account for crowding in the field , and the possibility that faint sources are observed above the detection limit because of noise . the resulting synthetic stokes @xmath4 and polarized intensity images were then searched for sources in an identical manner as for the observed images . figure [ dndpmodel - fig ] shows the derived differential source counts for total intensity ( log @xmath109 log @xmath4 ) and polarized flux density ( log @xmath109 log @xmath46 ) at @xmath19 , normalized to the euclidean expectation in the conventional way . the polarized source counts are also listed in table [ polcounts - tab ] . counts were derived in bins starting at @xmath110jy to avoid sources with completeness correction greater than 10 . this resulted in the removal of the fifteen faintest sources from the counts . the stokes @xmath4 source counts show good agreement with the windhorst curve , although we find somewhat higher numbers around @xmath111 . the polarized source counts are nearly flat in the flux density range observed , consistent with the increased fractional polarization of the faint radio sources . figure [ dndpmodel - fig ] shows predicted polarized source count curves derived by convolving windhorst s stokes @xmath4 counts with the strong - source @xmath58 distribution from @xcite ( dashed curve ) , as well as our hybrid @xmath112 distribution fit to the @xmath113 distribution in [ maxlike - sec ] ( solid curve ) . the observed polarized source counts show an excess over the @xcite extrapolation for @xmath114 , and are consistent with the prediction based on the derived higher @xmath58 for these sources . the data point at the lowest flux density lies marginally above our predicted curve . this may suggest a continuing trend toward even higher fractional polarization in the sub - mjy population . the complete drao 30-field survey of the region , and other deep polarization surveys , will test this . the positions of the polarized radio sources were examined in the _ spitzer _ swire images of elais n1 . accurate positions for the sources were obtained from the vla first survey images @xcite , which provides 1.4 ghz continuum stokes @xmath4 images at @xmath115 resolution with a 1-@xmath29 sensitivity of @xmath116 . the sensitivity of the first images is sufficient to detect a stokes @xmath4 counterpart for every polarized source in the elais n1 deep field . approximately 35% of the polarized radio sources showed resolved structure on scales of 2 to 30 in the first images . a polarized source may be associated with a radio lobe instead of the core of a radio galaxy . visual inspection avoided misidentification in such cases . as a comparison to the polarized sources we also searched the swire images for identification of sources with stokes @xmath4 flux density larger than @xmath117 but no detectable polarization . in total , 54 polarized sources were unambiguously identified with _ spitzer _ objects . another 17 polarized sources had uncertain identifications , i.e. there was more than one possible counterpart within the errors of the radio position . two sources were not covered by swire . the remaining 12% of the sources ( 10 objects ) have no counterpart in the swire images . similar statistics resulted from the search for counterparts of the stokes @xmath4 sources with no detectable polarized emission . a wide range in flux density and angular size was found among the identified _ spitzer _ galaxies . although some faint galaxies appear unresolved in the _ spitzer _ images , the identified polarized sources seem to be mainly associated with elliptical galaxies . three of the polarized sources have counterparts in the _ spitzer _ images that are too faint to appear in the _ spitzer _ elais n1 source catalog . these sources were not included in the subsequent analysis . figure [ spitzercol - fig ] shows a near - infrared color - color diagram of _ spitzer _ galaxies identified with radio sources that had catalogued flux densities in all four irac bands at 3.6 , 4.5 , 5.8 , and @xmath118 m . this includes 41 of the polarized sources . since many of the sources are identified with extended galaxies , we used the isophotal fluxes in each band . our analysis of this color - color diagram is based on the modeling of @xcite , which divided the diagram into the four regions separated by dashed lines in figure [ spitzercol - fig ] . boundaries between these regions were defined so as to separate galaxies depending on the strength of near - infrared pah bands and the slope of the near - infrared continuum . region 1 is mainly populated by sources with a continuum that rises with wavelength . the near - infrared spectrum of these sources is usually dominated by non - equilibrium emission of stochastically heated very small dust grains , interpreted as pah destruction by the hard ultraviolet spectrum of an agn . region 2 is occupied mainly by dusty star - forming galaxies with strong pah bands at red - shift @xmath119 . this is because the strongest pah features at low redshift contribute to the flux in irac bands at 3.6 and @xmath118 m . region 3 is occupied by galaxies with fainter pah emission , or by dusty starforming galaxies at redshift @xmath120 . a conspicuous concentration of galaxies in region 3 is the blue clump near @xmath121 , @xmath122 . galaxies in the blue clump have a near - infrared continuum that declines with wavelength , since the spectrum is dominated by the starlight of an old stellar population . these are elliptical galaxies , at a wide range of redshift . region 4 is populated by pah - dominated sources at redshift @xmath123 . the number of sources by region in figure [ spitzercol - fig ] is listed in table [ table_irac_colcol ] . the host galaxies of the polarized sources occupy mainly regions 1 and 3 . the source in region 4 can not be considered a convincing high - redshift galaxy in view of uncertainties in the photometry and its proximity to the boundary with regions 1 and 3 . the majority ( 71% ) of the host galaxies of polarized sources in the _ spitzer _ color - color diagram are either in region 1 or in the blue clump . in both cases , the radio emission is interpreted as emission from an agn . eleven polarized sources ( 27% ) are found in the region of pah - dominated galaxies , with @xmath124 , a few dex above the blue clump . their location in the color - color diagram suggests either pah - dominated galaxies at redshift @xmath125 , or galaxies with faint pah bands at lower redshift @xcite . inspection of the first and swire images of these sources shows that all appear to be elliptical galaxies with a smooth morphology , and some have resolved symmetric radio sources , suggestive of radio lobes . this suggests that the polarized radio emission in these galaxies is associated with agn activity , while the pah emission originates from dust at a substantial distance from the agn , where it is shielded from the hard ultraviolet radiation that would destroy the pahs . dust in elliptical galaxies is a common phenomenon @xcite . it can be produced in the envelopes of cool red giant stars or it can be acquired through a merger with a gas - rich galaxy . the stokes @xmath4 sources with no detectable polarized emission generally occupy the same areas of the color - color diagram as the polarized sources , with two exceptions . first , a significant fraction ( 15% ) of the stokes @xmath4 sources is located in region 2 , the area where star forming , pah - dominated galaxies are expected . these objects are generally the faintest stokes i radio sources , and none would be detectable in polarization at our sensitivity level . the stokes @xmath4 sources in region 2 are likely members of a population of star - forming galaxies that is believed to make up a large fraction of the radio source population below @xmath131 mjy . the second exception is that table [ table_irac_colcol ] indicates an excess of polarized sources ( 27% ) relative to sources with no detected polarization ( 14% ) in region 3b , associated with galaxies having pah emission . this difference is significant if poisson errors are assumed . confirmation will require a more complete sample of polarized radio sources identified with _ spitzer _ galaxies . the polarized sources found in the elais n1 field have a median stokes @xmath4 flux density of 12 mjy . models of radio source populations fitted to the total radio source counts suggest that most radio sources with @xmath19 flux density @xmath126 are steep - spectrum radio galaxies , whose power is ultimately derived from accretion onto a compact object . however , it is not clear a priori that a faint polarization - selected sample of radio sources is representative of the entire population . in principle , a highly polarized population of faint radio sources may constitute a significant fraction of a sample of faint polarized sources . in section [ maxlike - sec ] we presented evidence that faint extragalactic polarized radio sources are on average more highly polarized than bright sources , with a median fractional polarization approximately three times higher . from an analysis of sources with stokes @xmath8 mjy in the nvss , @xcite also noted that the median fractional polarization of radio sources increases with decreasing flux density , from 1.05% for stokes @xmath127 mjy to 1.84% between @xmath128 . from a similar analysis of the nvss , @xcite showed that the anti - correlation between flux density and percentage polarization occurs only for steep - spectrum sources . they found that the median percentage polarization for steep - spectrum sources increased from 1.14% for flux densities greater than @xmath129 to 1.77% between @xmath130 . our result extends this to much lower flux densities and indicates a much stronger effect for faint sources , resulting in a median polarization of @xmath131 at stokes @xmath132 . analysis of table [ table_polsources ] shows that these faint polarized emitters are dominated by steep - spectrum sources ; all but one of the polarized sources for which a 325 @xmath19 spectral index exists ( 75% ) has @xmath133 . polarized sources that can be identified with galaxies in the _ spitzer _ elais n1 deep field , are associated with elliptical galaxies . most of the host galaxies have near - infrared colors typical for dust emission from the vicinity of an agn , or an old stellar population with no significant dust emission . the majority of the polarized sources is associated with agn activity for this reason . the remaining 11 polarized sources associated with galaxies having pah emission are also likely to contain an agn . some are clearly resolved double - lobed objects in the first images , and all appear to be elliptical galaxies with a smooth brightness distribution . we conclude that there is no evidence for galaxies in our sample of polarized sources in which the radio emission is powered by star formation . the higher fractional polarization of faint radio sources may be related to a population of radio - quiet agn in which fainter radio emission correlates with conditions that favor increased polarization , for example more ordered magnetic fields or less internal faraday depolarization . high - resolution polarimetry of these objects will provide more insight into their nature . we present sensitive observations of a complete sample of compact polarized radio sources , as part of a deep integration of the elais n1 region made with the synthesis telescope at the dominion radio astrophysical observatory . a total of 83 polarized sources was detected in the ten - field mosaic . the distribution of fractional polarization of faint polarized sources was investigated with a monte - carlo analysis that generates synthetic source lists with the same noise statistics and observational selection criteria as the data . maximum - likelihood fits of the synthetic source lists to the data in the @xmath134 - @xmath135 plane yielded a best fitting gauss - hermite function with @xmath136 , @xmath137 for the distribution of intrinsic fractional polarization . the data demonstrate a trend of increasing fractional polarization with decreasing flux density . polarized source counts from the elais n1 deep field are presented down to @xmath138 . we find that the euclidean - normalized polarized counts remain flat below @xmath117 . the distribution of fractional polarization derived from our monte carlo analysis is convolved with the total - intensity source counts to produce a prediction of the polarized source counts . the predicted euclidean - normalized polarized counts are nearly flat to @xmath139 , in good agreement with the data . however , the data at the faintest polarized flux densities suggests a continuing trend of increased polarization fraction with decreasing flux density . the near - infrared color - color diagram for host galaxies identified with the polarized sources in the elais n1 field shows that most of the host galaxies are ellipticals , or galaxies for which the near - infrared spectrum is dominated by stochastically heated very small grains , presumably from the vicinity of an agn . some of the host galaxies appear to have pah bands in their near - infrared spectrum , but the morphological resemblance with ellipticals , and the fact that some of these polarized sources are resolved radio galaxies in the first survey , indicates that these objects also harbour agn . we suggest that the higher degree of polarization indicates a difference between agn observed at a flux density of hundreds of mjy , and fainter agn . observations and research on the drao planck deep fields are supported by the natural sciences and engineering research council of canada and the national research council canada . ev sheehan of drao was instrumental in improving the sensitivity of the drao synthesis telescope , with the outcome recorded in this paper . we are indebted to him for his skill and dedication to this difficult task . the dominion radio astrophysical observatory is operated as a national facility by the national research council of canada . landecker , t. l. , dewdney , p. e. , burgess , t. a. , gray , a. d. , higgs , l. a. , hoffmann , a. p. , hovey , g. j. , karpa , d. r. , lacey , j. d. , prowse , n. , purton . c. r. , roger r. s. , willis , a. g. , wyslouzil , w. , routledge , d. , & vaneldik , j. f. 2000 , , 145 , 509 sadler , e. m. , ricci , r. , ekers , r. d. , ekers , j. a. , hancock , p. j. , jackson , c. a. , kesteven , m. j. , murphy , t. , phillips , c. , reinfrank , r. f. , staveley - smith , l. , subrahmanyan , r. , walker , m. a. , wilson , w. e. , de zotti , g. 2006 , , 371 , 898 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 cccrrrrrr 1 & 16 02 05.52 @xmath140 0.29 & 54 54 43.2 @xmath140 3.0 & 16.82 @xmath140 0.56 & 1.74 @xmath140 0.11 & 26 & 10.4 @xmath140 0.7 & @xmath1410.78 @xmath140 0.06 + 2 & 16 02 34.37 @xmath140 0.48 & 54 54 01.8 @xmath140 4.2 & 7.95 @xmath140 0.40 & 0.98 @xmath140 0.16 & 37 & 11.8 @xmath140 1.9 & @xmath1410.75 @xmath140 0.11 + 3 & 16 02 42.00 @xmath140 0.44 & 55 10 01.2 @xmath140 4.4 & 8.51 @xmath140 0.78 & 0.78 @xmath140 0.10 & 71 & 19.9 @xmath140 2.7 & @xmath1410.23 @xmath140 0.22 + 4 & 16 03 11.52 @xmath140 0.34 & 55 39 07.2 @xmath140 3.8 & 4.65 @xmath140 0.40 & 1.37 @xmath140 0.11 & 27 & 28.2 @xmath140 2.4 & @xmath1420.64 & + 5 & 16 03 34.08 @xmath140 0.44 & 54 29 20.4 @xmath140 3.9 & 4.53 @xmath140 0.37 & 0.79 @xmath140 0.10 & 36 & 17.2 @xmath140 2.3 & @xmath1411.46 @xmath140 0.07 + 6 & 16 04 42.24 @xmath140 0.14 & 54 38 45.6 @xmath140 1.4 & 19.42 @xmath140 0.55 & 1.88 @xmath140 0.11 & 33 & 10.5 @xmath140 0.7 & @xmath1410.99 @xmath140 0.04 + 7 & 16 05 05.52 @xmath140 0.15 & 55 00 46.8 @xmath140 1.4 & 27.86 @xmath140 0.74 & 1.25 @xmath140 0.11 & 7 & 5.3 @xmath140 0.5 & @xmath1411.33 @xmath140 0.03 + 8 & 16 05 23.04 @xmath140 0.20 & 54 29 31.2 @xmath140 1.8 & 6.39 @xmath140 0.29 & 1.16 @xmath140 0.11 & @xmath14112 & 17.0 @xmath140 1.6 & @xmath1420.43 & + 9 & 16 05 38.88 @xmath140 0.04 & 54 39 28.8 @xmath140 0.4 & 187.61 @xmath140 4.74 & 8.42 @xmath140 0.24 & 21 & 5.0 @xmath140 0.2 & @xmath1411.00 @xmath140 0.03 + 10 & 16 05 38.88 @xmath140 0.20 & 54 41 27.6 @xmath140 1.9 & 2.00 @xmath140 0.35 & 1.28 @xmath140 0.11 & @xmath14137 & 47.3 @xmath140 4.5 & @xmath1421.22 & + 11 & 16 06 01.44 @xmath140 0.04 & 54 54 10.8 @xmath140 0.5 & 209.87 @xmath140 5.29 & 5.24 @xmath140 0.17 & 15 & 2.9 @xmath140 0.1 & @xmath1410.91 @xmath140 0.03 + 12 & 16 06 08.40 @xmath140 0.15 & 55 16 04.8 @xmath140 1.6 & 5.06 @xmath140 0.28 & 0.86 @xmath140 0.10 & 16 & 15.8 @xmath140 2.0 & @xmath1410.69 @xmath140 0.19 + 13 & 16 06 13.44 @xmath140 0.23 & 55 01 58.8 @xmath140 2.3 & 1.14 @xmath140 0.18 & 0.71 @xmath140 0.11 & @xmath14114 & 52.6 @xmath140 9.3 & @xmath1421.59 & + 14 & 16 06 35.52 @xmath140 0.10 & 54 35 02.4 @xmath140 1.0 & 15.96 @xmath140 0.51 & 1.61 @xmath140 0.11 & 44 & 12.1 @xmath140 0.9 & @xmath1410.73 @xmath140 0.06 + 15 & 16 06 58.80 @xmath140 0.32 & 54 43 12.0 @xmath140 3.0 & 4.31 @xmath140 0.21 & 0.41 @xmath140 0.10 & @xmath1419 & 8.6 @xmath140 2.2 & @xmath1420.69 & + 16 & 16 07 22.56 @xmath140 0.22 & 55 31 04.8 @xmath140 2.8 & 12.16 @xmath140 0.45 & 0.73 @xmath140 0.10 & 1 & 8.0 @xmath140 1.2 & @xmath1411.05 @xmath140 0.05 + 17 & 16 08 21.36 @xmath140 0.11 & 56 13 51.6 @xmath140 0.8 & 221.05 @xmath140 5.59 & 3.65 @xmath140 0.15 & 54 & 2.0 @xmath140 0.1 & 0.30 @xmath140 0.03 + 18 & 16 08 28.56 @xmath140 0.28 & 54 10 37.2 @xmath140 3.7 & 19.94 @xmath140 0.56 & 0.59 @xmath140 0.10 & 26 & 3.0 @xmath140 0.5 & @xmath1410.68 @xmath140 0.06 + 19 & 16 08 38.64 @xmath140 0.29 & 54 14 34.8 @xmath140 3.3 & 2.41 @xmath140 0.21 & 0.51 @xmath140 0.10 & @xmath14119 & 17.0 @xmath140 3.5 & @xmath1421.09 & + 20 & 16 08 47.76 @xmath140 0.20 & 56 11 16.8 @xmath140 1.6 & 27.09 @xmath140 1.03 & 1.66 @xmath140 0.11 & 8 & 7.8 @xmath140 0.6 & @xmath1411.19 @xmath140 0.03 + 21 & 16 08 58.32 @xmath140 0.36 & 55 56 27.6 @xmath140 2.2 & 9.29 @xmath140 0.48 & 0.69 @xmath140 0.10 & 69 & 12.0 @xmath140 1.8 & @xmath1410.82 @xmath140 0.09 + 22 & 16 09 04.32 @xmath140 0.26 & 56 10 33.6 @xmath140 2.2 & 12.68 @xmath140 0.60 & 1.33 @xmath140 0.11 & 23 & 11.9 @xmath140 1.0 & @xmath1411.13 @xmath140 0.05 + 23 & 16 09 11.04 @xmath140 0.18 & 55 26 31.2 @xmath140 2.2 & 4.51 @xmath140 0.22 & 0.78 @xmath140 0.10 & 4 & 16.9 @xmath140 2.3 & @xmath1410.86 @xmath140 0.17 + 24 & 16 09 22.80 @xmath140 0.15 & 56 15 03.6 @xmath140 1.4 & 32.46 @xmath140 0.95 & 2.14 @xmath140 0.12 & 21 & 8.0 @xmath140 0.5 & @xmath1410.63 @xmath140 0.04 + 25 & 16 09 31.68 @xmath140 0.12 & 55 25 04.8 @xmath140 1.3 & 15.32 @xmath140 0.47 & 1.41 @xmath140 0.11 & 25 & 11.4 @xmath140 0.9 & @xmath1411.05 @xmath140 0.05 + 26 & 16 09 36.24 @xmath140 0.18 & 55 27 03.6 @xmath140 1.5 & 10.95 @xmath140 0.37 & 1.12 @xmath140 0.10 & 8 & 11.7 @xmath140 1.1 & @xmath1410.71 @xmath140 0.09 + 27 & 16 09 44.40 @xmath140 0.38 & 54 37 51.6 @xmath140 2.9 & 6.61 @xmath140 0.24 & 0.35 @xmath140 0.10 & 59 & 5.3 @xmath140 1.6 & @xmath1420.40 & + 28 & 16 09 52.56 @xmath140 0.24 & 55 07 08.4 @xmath140 2.5 & 0.68 @xmath140 0.14 & 0.42 @xmath140 0.10 & 1 & 44.1 @xmath140 12.0 & @xmath1421.95 & + 29 & 16 10 03.12 @xmath140 0.10 & 55 52 37.2 @xmath140 1.1 & 96.83 @xmath140 2.47 & 1.59 @xmath140 0.11 & 30 & 2.0 @xmath140 0.2 & @xmath1410.79 @xmath140 0.03 + 30 & 16 10 27.12 @xmath140 0.14 & 54 12 54.0 @xmath140 1.6 & 8.51 @xmath140 0.28 & 1.24 @xmath140 0.11 & 30 & 14.4 @xmath140 1.3 & @xmath1410.43 @xmath140 0.17 + 31 & 16 10 57.84 @xmath140 0.08 & 55 35 24.0 @xmath140 0.7 & 19.46 @xmath140 0.53 & 1.61 @xmath140 0.11 & 5 & 9.7 @xmath140 0.7 & @xmath1410.63 @xmath140 0.06 + 32 & 16 11 00.48 @xmath140 0.06 & 54 42 03.6 @xmath140 0.6 & 29.41 @xmath140 0.76 & 2.03 @xmath140 0.11 & 19 & 7.8 @xmath140 0.5 & @xmath1410.74 @xmath140 0.04 + 33 & 16 11 20.40 @xmath140 0.08 & 55 28 44.4 @xmath140 0.9 & 18.22 @xmath140 0.50 & 1.39 @xmath140 0.11 & 3 & 8.8 @xmath140 0.7 & @xmath1410.88 @xmath140 0.05 + 34 & 16 11 21.12 @xmath140 0.30 & 54 31 55.2 @xmath140 3.3 & 3.62 @xmath140 0.34 & 0.43 @xmath140 0.10 & @xmath1415 & 9.9 @xmath140 2.4 & @xmath1420.81 & + 35 & 16 11 29.04 @xmath140 0.29 & 55 51 36.0 @xmath140 2.6 & 13.22 @xmath140 0.41 & 0.44 @xmath140 0.10 & 25 & 4.3 @xmath140 1.0 & @xmath1410.87 @xmath140 0.06 + 36 & 16 11 37.92 @xmath140 0.29 & 53 59 34.8 @xmath140 4.2 & 13.82 @xmath140 0.44 & 0.82 @xmath140 0.10 & @xmath14118 & 5.9 @xmath140 0.8 & @xmath1411.03 @xmath140 0.05 + 37 & 16 11 38.16 @xmath140 0.26 & 55 59 52.8 @xmath140 2.7 & 4.50 @xmath140 0.23 & 0.48 @xmath140 0.10 & 14 & 11.3 @xmath140 2.4 & @xmath1411.19 @xmath140 0.11 + 38 & 16 11 50.88 @xmath140 0.22 & 55 00 54.0 @xmath140 1.0 & 9.14 @xmath140 0.41 & 0.82 @xmath140 0.10 & 14 & 16.3 @xmath140 2.1 & @xmath1410.63 @xmath140 0.12 + 39 & 16 12 12.24 @xmath140 0.02 & 55 22 48.0 @xmath140 0.2 & 312.36 @xmath140 7.86 & 20.37 @xmath140 0.52 & 29 & 8.0 @xmath140 0.3 & @xmath1411.11 @xmath140 0.03 + 40 & 16 12 24.00 @xmath140 0.13 & 55 26 02.4 @xmath140 1.5 & 9.57 @xmath140 0.34 & 0.74 @xmath140 0.10 & 32 & 8.7 @xmath140 1.2 & @xmath1410.56 @xmath140 0.12 + 41 & 16 12 28.56 @xmath140 0.34 & 55 06 46.8 @xmath140 2.0 & 10.06 @xmath140 0.30 & 0.37 @xmath140 0.10 & 59 & 4.1 @xmath140 1.1 & @xmath1411.13 @xmath140 0.06 + 42 & 16 12 31.68 @xmath140 0.31 & 54 18 10.8 @xmath140 2.5 & 9.81 @xmath140 0.36 & 0.43 @xmath140 0.10 & 21 & 4.5 @xmath140 1.1 & @xmath1410.73 @xmath140 0.10 + 43 & 16 12 35.28 @xmath140 0.04 & 56 28 19.2 @xmath140 0.4 & 176.88 @xmath140 4.51 & 11.47 @xmath140 0.31 & 39 & 7.8 @xmath140 0.3 & @xmath1410.96 @xmath140 0.03 + 44 & 16 12 47.52 @xmath140 0.33 & 55 02 31.2 @xmath140 2.9 & 7.68 @xmath140 0.25 & 0.35 @xmath140 0.10 & 51 & 4.9 @xmath140 1.4 & @xmath1410.74 @xmath140 0.12 + 45 & 16 12 51.36 @xmath140 0.20 & 56 03 50.4 @xmath140 2.4 & 2.59 @xmath140 0.19 & 0.61 @xmath140 0.10 & 25 & 20.9 @xmath140 3.7 & @xmath1421.04 & + 46 & 16 13 02.40 @xmath140 0.15 & 54 32 27.6 @xmath140 1.3 & 6.48 @xmath140 0.25 & 0.73 @xmath140 0.10 & 56 & 11.5 @xmath140 1.7 & @xmath1410.86 @xmath140 0.12 + 47 & 16 13 16.80 @xmath140 0.36 & 56 08 13.2 @xmath140 3.8 & 3.85 @xmath140 0.36 & 0.56 @xmath140 0.10 & @xmath14128 & 14.1 @xmath140 2.6 & @xmath1420.77 & + 48 & 16 13 19.20 @xmath140 0.09 & 54 16 40.8 @xmath140 1.0 & 5.78 @xmath140 0.43 & 1.34 @xmath140 0.11 & 5 & 31.5 @xmath140 2.8 & @xmath1420.50 & + 49 & 16 13 25.92 @xmath140 0.24 & 55 39 39.6 @xmath140 2.4 & 3.06 @xmath140 0.18 & 0.35 @xmath140 0.10 & 17 & 12.2 @xmath140 3.6 & @xmath1420.93 & + 50 & 16 13 26.64 @xmath140 0.10 & 55 15 46.8 @xmath140 1.1 & 14.60 @xmath140 0.14 & 0.80 @xmath140 0.10 & 14 & 6.5 @xmath140 0.9 & @xmath1411.19 @xmath140 0.04 + 51 & 16 13 28.80 @xmath140 0.27 & 56 17 49.2 @xmath140 2.1 & 17.29 @xmath140 0.57 & 0.87 @xmath140 0.10 & 26 & 6.8 @xmath140 0.8 & @xmath1410.53 @xmath140 0.07 + 52 & 16 13 30.72 @xmath140 0.06 & 54 27 21.6 @xmath140 0.6 & 80.29 @xmath140 2.03 & 2.67 @xmath140 0.12 & 15 & 3.7 @xmath140 0.2 & @xmath1410.85 @xmath140 0.03 + 53 & 16 13 36.72 @xmath140 0.36 & 54 11 16.8 @xmath140 2.7 & 1.70 @xmath140 0.36 & 0.59 @xmath140 0.10 & 1 & 22.9 @xmath140 4.2 & @xmath1421.33 & + 54 & 16 13 41.76 @xmath140 0.19 & 56 11 49.2 @xmath140 1.4 & 101.05 @xmath140 7.62 & 2.80 @xmath140 0.12 & 22 & 10.9 @xmath140 0.6 & 0.09 @xmath140 0.04 + 55 & 16 13 48.48 @xmath140 0.27 & 54 14 13.2 @xmath140 2.5 & 8.12 @xmath140 0.40 & 0.64 @xmath140 0.10 & 18 & 7.3 @xmath140 1.2 & @xmath1411.10 @xmath140 0.07 + 56 & 16 13 56.16 @xmath140 0.25 & 54 57 28.8 @xmath140 2.5 & 0.45 @xmath140 0.09 & 0.28 @xmath140 0.10 & @xmath1411 & 50.6 @xmath140 19.2 & @xmath1413.97 @xmath140 0.03 + 57 & 16 13 56.84 @xmath140 0.22 & 55 02 08.2 @xmath140 2.3 & 0.81 @xmath140 0.13 & 0.50 @xmath140 0.02 & @xmath14131 & 49.8 @xmath140 5.7 & @xmath1421.82 & + 58 & 16 14 00.96 @xmath140 0.20 & 53 57 21.6 @xmath140 1.8 & 14.75 @xmath140 0.49 & 1.08 @xmath140 0.10 & 32 & 7.5 @xmath140 0.8 & @xmath1410.41 @xmath140 0.10 + 59 & 16 14 16.80 @xmath140 0.29 & 55 42 57.6 @xmath140 3.5 & 1.14 @xmath140 0.36 & 0.46 @xmath140 0.10 & 56 & 64.0 @xmath140 17.2 & @xmath1421.60 & + 60 & 16 14 21.12 @xmath140 0.06 & 55 36 39.6 @xmath140 0.6 & 35.41 @xmath140 0.93 & 3.00 @xmath140 0.13 & 18 & 9.9 @xmath140 0.5 & @xmath1410.74 @xmath140 0.04 + 61 & 16 14 32.64 @xmath140 0.20 & 55 38 31.2 @xmath140 2.5 & 1.27 @xmath140 0.39 & 0.47 @xmath140 0.10 & @xmath14179 & 90.8 @xmath140 27.5 & @xmath1421.52 & + 62 & 16 15 27.36 @xmath140 0.13 & 54 27 10.8 @xmath140 1.2 & 4.51 @xmath140 0.20 & 0.97 @xmath140 0.10 & 5 & 20.1 @xmath140 2.3 & @xmath1420.66 & + 63 & 16 15 30.96 @xmath140 0.16 & 54 52 30.0 @xmath140 2.0 & 5.43 @xmath140 0.22 & 0.68 @xmath140 0.10 & 20 & 13.1 @xmath140 2.0 & @xmath1411.16 @xmath140 0.09 + 64 & 16 15 36.72 @xmath140 0.19 & 53 46 37.2 @xmath140 2.4 & 54.20 @xmath140 1.46 & 2.64 @xmath140 0.12 & 19 & 5.6 @xmath140 0.3 & @xmath1410.85 @xmath140 0.03 + 65 & 16 15 49.68 @xmath140 0.06 & 55 16 40.8 @xmath140 0.7 & 27.80 @xmath140 0.72 & 1.75 @xmath140 0.11 & 35 & 7.4 @xmath140 0.5 & @xmath1410.68 @xmath140 0.04 + 66 & 16 16 23.04 @xmath140 0.14 & 55 27 00.0 @xmath140 1.3 & 13.10 @xmath140 0.40 & 0.90 @xmath140 0.10 & 27 & 8.8 @xmath140 1.0 & @xmath1410.89 @xmath140 0.06 + 67 & 16 16 23.52 @xmath140 0.17 & 54 57 43.2 @xmath140 1.6 & 10.35 @xmath140 0.48 & 0.60 @xmath140 0.10 & 10 & 6.0 @xmath140 1.0 & @xmath1410.97 @xmath140 0.07 + 68 & 16 16 37.92 @xmath140 0.18 & 55 45 14.4 @xmath140 2.1 & 74.66 @xmath140 1.90 & 0.85 @xmath140 0.10 & 51 & 1.3 @xmath140 0.2 & @xmath1410.61 @xmath140 0.03 + 69 & 16 16 39.36 @xmath140 0.11 & 53 58 12.0 @xmath140 0.9 & 351.23 @xmath140 9.13 & 5.14 @xmath140 0.19 & 23 & 1.7 @xmath140 0.1 & @xmath1410.85 @xmath140 0.03 + 70 & 16 16 40.08 @xmath140 0.18 & 56 20 38.4 @xmath140 1.5 & 18.64 @xmath140 0.70 & 1.37 @xmath140 0.11 & 24 & 9.3 @xmath140 0.8 & @xmath1410.52 @xmath140 0.07 + 71 & 16 17 57.60 @xmath140 0.22 & 54 51 36.0 @xmath140 3.0 & 14.65 @xmath140 0.46 & 0.68 @xmath140 0.10 & @xmath14117 & 5.9 @xmath140 0.9 & @xmath1411.11 @xmath140 0.04 + 72 & 16 18 06.72 @xmath140 0.36 & 54 42 46.8 @xmath140 2.4 & 5.31 @xmath140 0.26 & 0.65 @xmath140 0.10 & 24 & 12.3 @xmath140 2.0 & @xmath1420.55 & + 73 & 16 18 32.64 @xmath140 0.05 & 54 31 44.4 @xmath140 0.5 & 48.68 @xmath140 1.27 & 4.56 @xmath140 0.15 & @xmath1411 & 10.3 @xmath140 0.4 & @xmath1410.78 @xmath140 0.03 + 74 & 16 18 57.57 @xmath140 0.12 & 54 29 26.2 @xmath140 1.3 & 132.06 @xmath140 5.88 & 2.55 @xmath140 0.14 & 21 & 2.3 @xmath140 0.2 & @xmath1411.39 @xmath140 0.03 + 75 & 16 18 59.28 @xmath140 0.18 & 54 52 40.8 @xmath140 1.7 & 40.08 @xmath140 1.07 & 1.86 @xmath140 0.11 & 21 & 6.0 @xmath140 0.4 & @xmath1410.50 @xmath140 0.04 + 76 & 16 19 15.36 @xmath140 0.20 & 55 05 13.2 @xmath140 1.4 & 16.65 @xmath140 0.50 & 0.69 @xmath140 0.10 & @xmath1412 & 5.7 @xmath140 0.9 & @xmath1410.30 @xmath140 0.11 + 77 & 16 19 19.20 @xmath140 0.08 & 55 35 56.4 @xmath140 0.8 & 53.48 @xmath140 1.36 & 1.98 @xmath140 0.11 & 15 & 4.1 @xmath140 0.3 & @xmath1410.44 @xmath140 0.04 + 78 & 16 19 19.44 @xmath140 0.06 & 54 48 25.2 @xmath140 0.6 & 57.59 @xmath140 1.46 & 3.32 @xmath140 0.13 & 11 & 6.1 @xmath140 0.3 & @xmath1410.67 @xmath140 0.03 + 79 & 16 19 24.24 @xmath140 0.18 & 55 50 52.8 @xmath140 1.6 & 2.18 @xmath140 0.21 & 0.79 @xmath140 0.10 & 19 & 26.8 @xmath140 3.7 & @xmath1421.16 & + 80 & 16 21 13.68 @xmath140 0.10 & 55 23 42.0 @xmath140 0.9 & 58.11 @xmath140 1.49 & 2.62 @xmath140 0.12 & 50 & 4.8 @xmath140 0.3 & @xmath1410.94 @xmath140 0.03 + 81 & 16 21 18.72 @xmath140 0.28 & 55 38 27.6 @xmath140 2.3 & 7.97 @xmath140 0.40 & 0.74 @xmath140 0.10 & 7 & 8.5 @xmath140 1.2 & @xmath1410.98 @xmath140 0.08 + 82 & 16 21 45.36 @xmath140 0.10 & 55 49 37.2 @xmath140 1.0 & 31.10 @xmath140 1.01 & 2.05 @xmath140 0.11 & 60 & 6.9 @xmath140 0.4 & @xmath1410.62 @xmath140 0.04 + 83 & 16 22 08.64 @xmath140 0.14 & 55 24 28.8 @xmath140 1.6 & 14.23 @xmath140 2.15 & 4.55 @xmath140 0.15 & 6 & 27.7 @xmath140 1.2 & @xmath1411.57 @xmath140 0.03 + lcccc 0.71 & & 29 & & 1.38 @xmath140 0.26 + 1.42 & & 20 & & 1.26 @xmath140 0.28 + 2.50 & & 11 & & 1.10 @xmath140 0.33 + 4.61 & & 5 & & 1.01 @xmath140 0.45 + 9.94 & & 2 & & 1.46 @xmath140 1.03 + 20.4 & & 1 & & 2.32 @xmath140 2.32 + ccccc 1 & 8 & 20@xmath1407 & 32 & 19@xmath1403 + 2 & 0 & 0 & 25 & 15@xmath1403 + 3a & 21 & 51@xmath14011 & 80 & 47@xmath1405 + 3b & 11 & 27@xmath1408 & 24 & 14@xmath1403 + 4 & 1 & 2@xmath1402 & 9 & 5@xmath1402 +

we analyze a large , complex equatorial coronal hole ( ech ) and its immediate surroundings with a focus on the roots of the fast solar wind . we start by demonstrating that our ech is indeed a source of the fast solar wind at 1 au by examining in situ plasma measurements in conjunction with recently developed measures of magnetic conditions of the photosphere , inner heliosphere and the mapping of the solar wind source region . we focus the bulk of our analysis on interpreting the thermal and spatial dependence of the non - thermal line widths in the ech as measured by _ soho_/sumer by placing the measurements in context with recent studies of ubiquitous alfvn waves in the solar atmosphere and line profile asymmetries ( indicative of episodic heating and mass loading of the coronal plasma ) that originate in the strong , unipolar magnetic flux concentrations that comprise the supergranular network . the results presented in this paper are consistent with a picture where a significant portion of the energy responsible for the transport of heated mass into the fast solar wind is provided by episodically occurring small - scale events ( likely driven by magnetic reconnection ) in the upper chromosphere and transition region of the strong magnetic flux regions that comprise the supergranular network .

1011.3066

equatorial coronal holes ( ech ) present spectroscopists with an unparalleled opportunity to study the source of the fast solar wind . these regions , while being shorter - lived , present us with a significantly less - complex line - of - sight than their well studied , polar cousins ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this paper we revisit an ech observed by the sumer instrument @xcite on board the _ solar and heliospheric observatory _ ( _ soho _ ; * ? ? ? * ) that was the primary subject of @xcite , while also providing support to the results of @xcite . @xcite used 770 emission line spectroheliograms formed in the upper transition / low - corona plasma to demonstrate the magnetic complexity of the ech and its surroundings . they noticed that the regions of strong outflow ( blue doppler shifts ) are overlying photospheric regions of largest magnetic unipolarity that have an open coronal - scale magnetic field above . @xcite expanded on the initial work , investigating the relationship between the magnetic flux distribution in the photosphere and the transition region doppler patterning on the supergranular scale . using simultaneously observed spectroheliograms in the 1548 emission line they built a picture of a truly dynamic energy release into the plasma driven by the relentless evolution of the photospheric magnetic field , or `` magnetic carpet '' ( e.g. , * ? ? ? * ) and placed the `` footprint '' of the outflow from the ech in the middle of the transition region . in this paper we study a spectral feature of this ech that has been ignored in our previous analyses of this region ; the dependence on temperature of the non - thermal line widths of the transition region spectral lines in and around the coronal hole . to facilitate this investigation we will exploit the recent detection of chromospheric and coronal alfvn waves @xcite , and the relationship of the former with dynamic spicules ( e.g. , * ? ? ? * ) . the energetic and mass - loading signature of these heating events at low heights can be extended to the transition region ( e.g. , * ? ? ? * ) , the active corona @xcite , the quiet corona @xcite , and to a staple of coronal hole morphology - polar plumes @xcite . the diagnostic information provided by the non - thermal line widths provides further information about the footprint of the region which , as we will see , maps to a very high flow speed region in the solar wind as observed by the _ advanced composition explorer _ ( ace ; * ? ? ? * ) . in section sanal2 we present some contextual observations of the region around the ech and establish the physical connectivity of the region to the solar wind observed at 1au following . following that , in sect . obs , we present the multi - thermal evolution of the non - thermal line widths in the ech through the transition region and assess the impact of the measured line profile asymmetries on the line widths measured . finally , in sect . results , we discuss the data analysis and results presented that offer us a look at the footprint of fast solar wind . the sumer spectroheliograms of the ech on 1999 november 6 and 7 , as demonstrated in @xcite and sect . 5 of @xcite , allowed us to begin a study of the coronal hole at disk center , minimizing interpretative error of both the line - of - sight spectral properties and magnetograms . supporting observations of the magnetic environment of the ech region are provided by abstractions of the soho / mdi @xcite los magnetic field : the supergranular flux balance and magnetic range of influence ( mroi ) as defined , developed , and discussed in @xcite . figure f1 provides context of the extreme - uv and magnetic environment of the ech from the extreme - ultraviolet imaging telescope ( eit ; * ? ? ? * ) and mdi respectively . in each panel the ech is outlined by a large closed contour determined simply from the 150 data number ( dn ) level of the 195 eit image and the rectangular regions enclose the regions sampled by the individual sumer spectroheliograms . we should reiterate that the supergranular flux balance map ( middle panel ) provides a measure of polarity `` dominance '' on a spatial scale of 20 mm while the mroi map ( bottom panel ) allows us to estimate the distance that separates partners in closed magnetic flux concentrations . however , in our previous papers we made no attempt to establish that this ech maps to a high speed flow region of the solar wind at 1 au . @xcite noticed in their recent analysis that the footpoint of the sub - terrestrial field line , determined from potential field source surface extrapolations ( e.g. , * ? ? ? * ) , jumps from patch to patch of locally high mroi , and walks smoothly within each patch ( e.g. , fig f1 in that paper ) . the upper panel of fig . [ f4 ] shows a map of mroi based on a merged synoptic magnetogram ( downsampled by a factor of four from @xmath0 pixels to @xmath1 pixels ) comprising the trailing half of carrington rotation 1955 and the leading half of carrington rotation 1956 . in yellow , we show the same proxy for the ech boundary : the smoothed 150 dn level of eit 195 . the blue dots track the sub - terrestrial field line footpoints for the first two weeks of november 1999 ( one dot per 96-min mdi magnetogram ) . repeating the analysis of @xcite , we trace 32 additional field lines ( arranged on the perimeter of an ellipse on the source surface with semimajor axes @xmath2 ) back to the photosphere . the additional field lines allow us to compute the standard deviation of the separation between their footpoints and that of the sub - terrestrial field line - providing error estimates for the location of each sub - terrestrial footpoint . to indicate the progression of time in the plot we label the magnetogram closest to noon each day . we see that the footpoint does not progress smoothly across the map rather , it jumps from one point to the next . it suddenly jumps ( by @xmath3 of longitude ) into the equatorial coronal hole after spending much of the first two days close to ar 8742 . a similarly large jump is seen on the november 13 after the footpoint has spent much of november 9 - 12 close to ar 8760 . the lower panels of fig . [ f4 ] shows in situ data observed at 1 au by ace , from top to bottom : the solar wind speed ; magnetic field strength and ( color - coded ) azimuth ( mag ; * ? ? ? * ) ; proton temperature & density , and oxygen charge state ratio ( swims ; * ? ? ? the time shown in these panels is that _ at ace _ , and runs from right to left , as in the synoptic magnetogram above them . notice the five vertical dashed lines that correspond to the time of the first magnetogram after five significant jumps : blue for the south - north crossing of the heliospheric current sheet ( 1999 november 2 , 12:51ut ) ; red for when the footpoint first enters the coronal hole ( 1999 november 4 , 14:23ut ) , leaves the eit - intensity defined boundary ( 1999 november 8 , 03:12ut ) , and the jump from the coronal hole to ar 8760 ( 1999 november 9 , 12:47ut ) ; and again in blue the second crossing of the heliospheric current sheet ( 1999 november 13 , 04:47ut ) . we note that while the first current - sheet crossing is muddied by the major change in wind parameters associated with fast coronal hole outflow , the second shows the classic signature of a current - sheet crossing : as the magnitude increases ( i.e. , field - line packing ) and the field direction rotates smoothly through over @xmath4 . the propagation delays from the sun to 1 au are calculated from the solar wind forecast model of @xcite . the predictions for the current sheet crossings are both about half a day too late based on field and plasma signatures , but correlate well with a small increase in ( unfortunately , the second current sheet crossing occurs during a gap in swics coverage ) . however , the jump into the coronal hole manifests itself with low conditions ( bottom panel ) at almost precisely the right time . simultaneously , we see an interplanetary shock , a step in wind speed , and associated spikes in the proton density and temperature , as the shock compresses and heats the plasma , and increases the field strength . the low conditions typical of a coronal hole outflow are bracketed by the first two red lines . there is a small shock and a discontinuity in field strength at the third red dashed line that is associated with the footpoint s jump to ar 8760 where it remains until the second current sheet crossing . in summary , the physical connectivity of the ech to the solar wind observed at 1 au is clear . the spectroheliograms studied in @xcite were taken as part of a sumer campaign to study the evolution of an ech as it transited the disk from 1999 november 3 - 8 - each has 3 stepping and an exposure time of 150s , with each `` pixel '' representing the cumulative spectra of nine individual 1 sumer spectra ( in a 3@xmath53 array ) , that is needed to boost the signal - to - noise of the dataset ( see also appendix sec : blends ) . for each observing sequence in the campaign , a spectroheliogram is formed in three emission lines that span the upper chromosphere , transition region , and low - corona ( 1534 - 1548 - 770 in second spectral order ) which have been calibrated ( absolutely ) in wavelength using the technique of @xcite . the first ( coolest ) of these lines is omitted from the present analysis due to the presence of a strong blend in the red wing of the line , and reduced signal - to - noise . figures f2 and f3 provide the detailed spectral context for the two sumer spectroheliograms ( 1999 november 6 14:00 - 00:00ut and 1999 november 7 ( 00:35 - 10:16ut ) ) that span the equatorial coronal hole studied ( the rectangular regions outlined in fig . f1 ) . in each case we show the eit 195 image and a composite mdi line - of - sight ( los ) magnetogram , the line intensities , non - thermal line widths ( @xmath6 ; determined from the recipe in * ? ? ? * and below ) and absolutely calibrated doppler velocities of the 1548 and 770.4 lines in the left and right columns respectively . in each case the spectral information is overlaid with the 150 count intensity contour of the intensity as a proxy for the supergranular network vertices . we note that the visibility of the network vertex pattern in ( panel d ) is not immediately obvious in either example ( a point discussed in other papers pertaining to this line , e.g. , * ? ? ? comparing the composite magnetograms and maps of @xmath6 we see a general correspondence between @xmath6 and the los field strength in both emission lines . in both figures , for , we see a mix of enhanced @xmath6 in the supergranular vertices where the strongest regions of magnetic flux are located and in regions straddling the intensity - defined network vertices ( see , e.g. , the region around 475 , -275 in panel e of fig . the periphery of the network vertices is exactly where we would reasonably expect to have regions of highly inclined magnetic field . further , we notice a strong visual correspondence between the regions of significant blue doppler shift and enhanced @xmath6 that are not only isolated to the supergranular network vertices . the values of @xmath6 in the region of enhanced emission in the center of panel f in fig . f3 , and particular those in the network vertices are reduced ( @xmath720km / s ) compared to the regions of strong outflow ( 30 - 35km / s ) . based on the analysis of @xcite this central portion of enhanced emission is , to all intents and purposes , quiet sun , not magnetically open and should be different from the open regions of the field - of - view . taking the two figures together , we see that the first and second moments ( line position / shift , and width ) of the line are strongly influenced by the open coronal field and outflow of the ech . investigating the relationship between the magnetic environment on the global ( and supergranular ) scales , the @xmath6 and profile symmetry in both spectral lines motivates the remainder of this section . non - thermal line widths are extracted from the emission line profiles of and using the method discussed in sect . 2 of @xcite . using a single gaussian genetic algorithm fit to the emission line profiles in each case ( e.g. , * ? ? ? * ) we compute the non - thermal line width @xmath6 @xmath8 ) of the line profile at 1/e of the peak intensity where @xmath9 is determined from the fwhm of the fitted gaussian ( @xmath10 ) , @xmath11 for an ion of mass @xmath12 and peak formation temperature of @xmath13 - assuming that the ion and electron temperatures are equal ) is the thermal width , and @xmath14 is the sumer instrumental width which is determined from the sumer solarsoft routine con_width_funct_3.pro . note that for this part of the sumer spectrum the velocity scale in each spectral pixel is of the order of 12 km / s / px and for the typical 25 pixel wide spectral window we can centroid the gaussian profile and determine its 1/e width to within 0.1 pixels @xcite . figure f5 shows the variation of the line peak intensity ( top ) , doppler velocity ( middle ) and @xmath6 ( bottom ) in the two sumer 1548 spectroheliograms studied . comparing the top and middle panels we can clearly see the persistent transition region red - shift and its enhancement in the supergranular network ( see , e.g. , * ? ? ? * and many others ) throughout the region . we also note the spatial variation of the doppler patterning as there is a considerable increase in the area of the images showing blue - shifted emission . most obvious in the upper left and lower right regions of the field of view , this change in doppler velocity `` partitioning '' was the subject of the discussion in sect . 5 of @xcite and occurs in the regions of largest magnetic flux imbalance and mroi . we interpreted the increased number of localized blue - shifts to indicate the presence of small concentrations of outflowing material on the supergranular boundary interiors and believe that these were most likely macrospicules , driven by magnetic carpet forced reconnection . comparing the top and bottom panels we see the strong relationship between the @xmath6 and line intensity - again , the cell interiors show values around 20km / s which increase to @xmath727km / s on the boundaries with some larger concentrations of enhanced @xmath6 ( @xmath735km / s ) in the upper left and lower right of the ech region with noticeably less network / interior contrast in the intermediate ( mixed polarity ) region . figure f6 shows the variation of the line peak intensity ( top ) , doppler velocity ( middle ) and @xmath6 ( bottom ) in the two sumer 770 spectroheliograms in a broader context , adding significantly to the versions published in @xcite and @xcite . immediately , we see the correspondence between the regions of strong blue doppler shift , enhanced @xmath6 and reduced line intensities . in the ech , the largest blue - shifts ( @xmath715km / s ) have @xmath6 of the order of 35km / s , but we see that the region ( inside the eit 195 determined ech boundary ) where the line intensity reaches quiet sun levels is exactly where we see a net reduction in the doppler velocity ( @xmath70km / s ) and @xmath6 ( @xmath722km / s ) . this correspondence between the intensity and doppler velocity was first noted in @xcite where the strength of the the blue - shift observed in and the strength of photospheric magnetic imbalance and openness " were linked . in the middle region of the euv ech , the emission is enhanced , the magnetic flux imbalance is low and the plasma appears to close locally ( small values of the mroi ) . this region also shows values of @xmath6 that appear to be anti - correlated with the bright emission present - vastly different from those seen in the open regions of large magnetic imbalance . we note that the other regions of enhanced emission inside the euv ech show reduced doppler velocity and @xmath6 , although the reduction is not as pronounced . the close correspondence in the spatial behavior of the spectral diagnostics ( the variation in intensity can be easily connected to doppler velocity and @xmath6 ) hints that the plasma at this temperature may be more simply structured than the plasma emitting the cooler emission . the discussion presented in @xcite and @xcite focuses on the apparent discrepancy between the definition of where the ech is rooted ( using spectral diagnostics ) and where the euv corona would have you believe that it is . as we have seen , the spectroscopic diagnostics observed in the and lines can differ significantly in the dark regions of the euv corona versus the brighter regions - suggesting , naturally , that the spectroscopic measurements provide a better indication of magnetic openness of the region than the apparent column depth of the euv corona does . in order to quantify the differences in @xmath6 imposed by changes in the global magnetic topology we construct scatter plots of line intensity , @xmath6 and doppler velocity . unlike the work of @xcite , we will not attempt to rigidly differentiate between the spectroscopic diagnostics of the supergranular cell interior and boundaries using ( subjective ) intensity thresholding of - we are more interested in the impact of the open and closed magnetic topologies on the @xmath6 values of the upper tr plasma . a further reason not to differentiate between cell boundaries and interiors here is dictated by the significantly reduced supergranular network boundary / interior contrast of ( or in most lines formed above @xmath7600,000k see , e.g. , * ? ? ? further , by studying the continuity of the emission across the whole supergranule we will be able to investigate the subtle changes in intensity , velocity , and @xmath6 contrast that are representative of the physical conditions of the boundary and interior together with the global scale variations in a single three - dimensional plot with reasonable confidence . first , we must consider the spectroscopic diagnostics of as they provide the most obvious delineation of the open ech plasma . under the assumption that we can safely combine the statistics of both absolutely calibrated spectroheliograms we show the scatterplots of the line intensity , doppler velocity , and @xmath6 in fig . notice the appearance of two distinct populations , or clusters of points , in the intensity-@xmath6 and intensity - velocity panels that can be separated at an intensity level of 250 counts . we see that the lower intensity set of points have typically larger values of @xmath6 and blue - shifted velocities while the mean doppler velocity of the brighter pixels is @xmath70km / s and the profiles are narrower than the higher intensity set . in the earlier papers @xcite this last point was designated as the key difference between the ech and quiet sun plasmas , adding the significantly enhanced @xmath6 to the picture provides important information and consistency checks towards understanding the driver of the wind streams likely to originate from the region . from this point on we will adopt the 250 count intensity level of as the ech boundary for the transition region plasma . in the two lower rows of fig . f7 we show the decomposition of the top row distributions into quiet sun ( i @xmath15 250 ; center ) and coronal hole ( i @xmath16 250 ; bottom ) components . if this intensity threshold marks the coronal hole boundary in then it appears to form a spatially sharp boundary region in that the spectral quantities change rapidly across the intensity threshold . the panels of fig . f8 show the scatter plots of the line spectral diagnostics in the re - designated quiet sun ( top ) and coronal hole ( bottom ) regions . in the quiet sun we see the strong dependence of intensity and red doppler shifts - characteristic of most transition region lines in the supergranular network ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? we also see the strong relationship between the line intensity and @xmath6 as was demonstrated for this line by @xcite although , unlike @xcite , we also see a relationship between plasma red - shifts and enhanced values of @xmath6 . the coronal hole scatter plot has a similar shape to that from the quiet sun region although , in general all of the distributions tend to shift to lower intensities and doppler velocities . we again see a strong @xmath6 - intensity relationship that shows a steeper gradient than its quiet sun counterpart . we also see a steeper relationship between doppler velocity and @xmath6 which is , in part , due to the reduction of the mean network redshift in the coronal hole region - also discussed in @xcite . the steeper relationship between the intensity and @xmath6 inside the coronal hole enforces our earlier ( visual ) impression that the network cells of the ech show a higher interior / boundary contrast in @xmath6 . it appears also that the brightest coronal hole network vertices in show enhancements in @xmath6 of the order of a few km / s over their quiet sun counterparts . it remains , however , that the coronal hole and quiet sun distributions are largely similar ; a detail that we will discuss in sect . sec : cons . extending our study of the non - thermal broadening we need to assess the degree of profile symmetry , or asymmetry , observed in the ech using the technique discussed in @xcite and @xcite . these papers tied the appearance of high - velocity , blue wing asymmetries of hot uv and euv emission lines to the mass loading of the active region , coronal holes , and the quiet corona that often seem to be associated with chromospheric `` type - ii '' spicules @xcite observed with the solar optical telescope ( sot ; * ? ? ? * ) of _ hinode _ @xcite . these tall spicules inhabit predominantly unipolar flux concentrations , have an apparent velocity between 40 and 150km / s , a dynamic timescale of 50 - 100s ( much faster than the 20 - 40km / s of classical spicules that have 3 - 10 minutes timescales , e.g. , * ? ? ? * ; * ? ? ? * and many others ) , and are likely caused by magnetic reconnection @xcite . type - ii spicules exhibit only upward motion , with the feature fading along its entire length in a matter of seconds , suggestive of rapid heating to tr temperatures , a feature exhibited in the analysis of @xcite . such heating is also suggested in so - called rapid blueshifted events that are thought to be the disk counterparts of type - ii spicules @xcite . as in @xcite we perform a single gaussian fit ( with a constant background ) to the line profile in each spectroheliogram pixel to determine its centroid position . then we interpolate the line profile to a wavelength scale that is ten times finer and calculate the total number of counts in the interpolated profile in two narrow range of wavelengths / velocities ( two original sumer spectral pixels wide , or 24 km / s at 1548 ) that are symmetrically positioned from the line centroid position , one in the blue and one in the red wing . the difference of the counts in the two wing positions , red - blue ( r - b ) , provides a measure of the asymmetry of the profile at the chosen wavelength or velocity offset . of course , a measure of zero indicates that the profile is symmetric in that velocity range . the r - b measure is related to the third ( skewness ) and fourth ( kurtosis ) moments of the the line profile and is a robust measure of asymmetry , provided that no spectral blends are present in the lines , an issue that we will return to below . applying the r - b asymmetry measure to the two emission lines we produce the series of images shown in figs . f9 and f10 ( and accompanying movies ) for a range of reference velocities in the two portions of the ech observation . in both figures , and in both spectral lines at low velocities ( below sound speed expected for the relevant formation temperatures ) , we see that the r - b asymmetry is dominated by the red wing . this dominance is clear in the supergranular boundaries , increasing further over the strongest magnetic flux concentrations that comprise the supergranular vertices . moving to higher velocities , we see that the r - b asymmetry in the supergranular network switches sign - indicating that the blue wing of the line profile is becoming dominant , which is compatible with a preponderance of high - velocity upflows in the strong magnetic field regions . the change in sign is rapid for , and occurs about 10 - 20km / s higher for . comparing the r - b asymmetry for at 50km / s and at 70km / s we see a striking similarity between the two panels . even though the supergranular network pattern is not obvious in the line intensities , the r - b map at 70km / s shows that the blue wing asymmetries are strongly correlated with the locations of network vertices . the behavior of the r - b maps is identical to those derived from quiet sun regions in these lines , plus several others @xcite . we also note that there is no clear delineation of the coronal hole / quiet sun boundary in the r - b asymmetry maps of figs . f9 and f10 . the lack of a clear quiet sun / coronal hole boundary has significant implications for the mass and energy transport in the upper solar atmosphere as we will discuss below . + in an effort to understand the relationships shown above , fig . f11 shows example ( normalized ; @xmath17 ) line profiles for magnetic network elements in the coronal hole ( blue solid line ; located at 420 , -325 ) and quiet sun ( red solid line ; located at 200 , -250 ) in ( left ) and ( right ) . in the case of we can immediately understand why the values of @xmath6 are so similar between the quiet sun and coronal hole network ; the relative reduction of the coronal hole network peak intensity ( which is the total intensity minus the continuum intensity in fig . f11 ) is of the order of 5% ( small , assuming that the continuum emission is not significantly affected between quiet sun and coronal hole ) and pushes the 1/e intensity level upward compared to the quiet sun profile . this compensates for the shallower gradient ( with wavelength ) in the wings of the coronal hole line which _ should _ increase the measured values of @xmath6 . combining these two effects results in only a small difference in the dropped black lines at the 1/e intensity level of each profile . we compare the blue and red vertical dashed lines which , again , show the relative shift to the blue of the coronal hole line profile . in the parlance of @xcite this shift of the profile blueward may be caused by a ( significant ) reduction amount of the lower tr plasma that is cooling out of the corona , in the magnetically open network compared to that of the closed quiet sun network . this is compatible with recent results from 3d mhd simulations that suggest that the pervasive redshifts in the transition region are related to plasma draining from the corona after the field lines are disconnected from the energy injection site @xcite . however , conflicting results have been obtained in recent numerical 3d mhd simulations which suggest that the pervasive redshifts in the transition region may be a natural consequence of episodic heating events in which upper chromospheric plasma is heated to coronal temperatures at low heights . in the latter scenario , the subtle difference in redshifts between coronal hole and quiet sun may be caused by a difference in the distribution of heights and amplitude of the heating events @xcite . further numerical simulations for varying magnetic field configurations will be required to help understand the observed differences . given the visual appearance of the data ( figs . f3 , f6 ) and the scatterplots ( fig . f7 ) we would expect that the differences between profiles of the quiet sun and coronal hole network are considerable , and they are . perhaps the most dramatic difference is the shoulder in the red wing of the line , but this does nt significantly alter the results presented here . we point the interested reader to appendix a where we explore the curious shape of the ech line profile in a little more detail discussing the impact of blends on the line shape , derived r - b measure , and @xmath6 diagnostics . the significant reduction in the peak intensity in the coronal hole network ( a factor of @xmath72.5 ) could be associated with several effects . for example , it may be indicative of a significantly reduced contribution to the line profile from material returning from the overlying open corona , i.e. , a large portion of the material that gets heated in the network to the formation temperature of ( or higher ) is injected into the fast solar wind . on the other hand , the open field conditions are expected to lead to very different contributions to the energy balance from the thermal conductive flux and enthalpy flux , which would lead to a different temperature stratification and thus tr emission measure distribution with temperature @xcite . a significant reduction in returning mass from the open corona would profoundly impact the line profile . this is compatible with what we observe . we see a large ( 10km / s ) shift of the line centroid to the blue , as the balance of mass - transport shifts to predominantly upflowing material . further , this reduction in intensity of the core profile component will compound with the net increase of the blue - wing asymmetry relative to it ( assuming that it does not change considerably in magnitude in a coronal hole , as we have seen ) . this combination will naturally lead to a noticeably ( 5 - 10km / s ) larger 1/e width of the single gaussian fit to the profile . these points again illustrate the perils of using only the first ( line center ) and second ( line width ) moments of the profile to infer the properties of a truly dynamic transition region plasma , as noted in the conclusion of @xcite . we also notice that the coronal hole profiles of and both show gradients in their wings that are systematically shallower than their quiet sun counterparts . this is indicative of an additional broadening source . we speculate that this symmetric excess component of the line - of - sight broadening is the result of the transverse ( alfvnic ) motion of the sub - resolution motion of the type - ii spicules that exist in these magnetic regions @xcite . based on the measurements of @xcite the spicules are mostly vertical , but have a mean inclination of 20 - 30 degrees from the normal , and for a transverse wave amplitude of 10 - 20km / s @xcite this would produce a contribution in the line of sight of @xmath73 - 7km / s which could increase to @xmath75 - 10km / s in the rarer low - coronal plasma . of course , in a coronal hole , the rapid density drop of the plasma with height would produce systematically larger amplitude wave motions @xcite in both lines , furthering the symmetric broadening of the profiles by a few km / s . the alfvnic motions may also help explain the enhancements of @xmath6 that straddle the network boundaries in ( figs . f2 , f3 ) . it is tempting to speculate that the sub - resolution alfvnic motion of the spicules on the larger inclination of the field spanning out into the internetwork will give rise to much larger values @xmath6 than elsewhere @xcite . we have studied the line broadening and profile asymmetry of the 1548 and 770.4 emission lines in a large equatorial coronal hole as an extension of the analysis presented in @xcite . we see that there are subtle changes in the network line profiles in an equatorial coronal hole over those in the quiet sun : the line - center red - shift of the emission is reduced by @xmath74km / s ; the peak emission is dimmer by 5% and ; the line wings have a shallower gradient . these effects contribute , as we have seen above , to an apparently invariant network @xmath6 between coronal hole and quiet network regions . for the low corona we see : strongly reduced intensities ; blue - shifted line centers ; and excessive line broadening in the coronal hole compared to the quiet sun . for the first time , we also analyze in detail the pervasive blue wing profile asymmetries in and around the network in a coronal hole . we have found these blueward asymmetries in a wide variety of spectral lines that span the entire transition region : 1548 and 770 ( in second order ) , as well as 1031 , and 770 in first spectral order ( see appendix b ) . our observations of the dependence of intensity , line centroid , width and asymmetries as a function of temperature and magnetic field configuration ( open vs. closed ) provide strict constraints to theoretical models aimed at explaining the structure and properties of the transition region . such constraints are crucial for these models : while they are rapidly advancing in complexity , with both 3d mhd models @xcite and sun to earth 1.5d models becoming available ( e.g. , * ? ? * ; * ? ? ? * ) , significant discrepancies and inconsistencies continue to exist , with some models providing contradictory explanations for , e.g. , pervasive redshifts in the transition region ( see 3.3 ) . none of the current models provide a compelling scenario that can explain the blueward asymmetries in and around the network . based on previous observational work , we have proposed a scenario in which heating events low in the atmosphere ( chromosphere or transition region ) above network regions accelerate and heat plasma from chromospheric or low tr temperatures up to coronal temperatures , for the quiet corona and fast solar wind alike . we speculate that this mass transfer may be driven from below through magnetic reconnection . depending on the exact height at which the non - thermal energy is deposited , we expect to see different diagnostic effects : if the energy is dumped at chromospheric heights , the upflows would be visible as type - ii spicules in the magnetic network of the chromosphere @xcite , whereas an event occurring at tr heights would lack a chromospheric counterpart , but nevertheless cause a faint , upflowing component in tr / coronal lines . generally , this scenario of heating at low heights is consistent with a picture presented in @xcite . one very interesting observational finding is that we see no difference in the magnitude of the asymmetry of the line profiles upon crossing from the open to closed magnetic regions . the apparent invariance of the asymmetry in open and closed magnetic region has implications for the relative importance of magnetic heating at coronal `` heights '' ( e.g. , * ? ? ? * ; * ? ? ? * ) compared to energy released in the lower atmosphere . based on the arguments of @xcite and @xcite , were the chromospheric mass being driven upward as a result of energy release at coronal heights , the signature of the mass transport seen in the lower atmosphere of a magnetically open region should look quite different to that in closed field regions . this is because the open coronal magnetic field is expected to inhibit the occurrence of coronal field - line braiding . in addition , the changing mix of thermal conduction and enthalpy flux in the energy balance in open vs. closed field regions would be expected to also significantly affect the magnitude of the high velocity upflows in coronal holes @xcite . that change would lead to a pronounced `` boundary '' in the r - b analysis when crossing from closed to open field . however , this is _ not _ observed . like that presented in @xcite and @xcite , the scenario we propose for the small - scale heating / upflow events is related to the model put forth by @xcite who suggested that a small fraction of the mass in chromospheric spicules is heated to coronal temperatures - providing the corona with hot plasma . following the example of @xcite , and assuming that the upward mass flux in the quiet sun and coronal hole regions are identical ( 1 - 4 type ii spicules of density 10@xmath18@xmath19 per arcsec@xmath20 per second in a magnetic network element ) , the spicular mass flux into the solar wind is sufficient to balance the observed mass - loss even if only 0.5 - 2.5% of the chromospheric spicule reaches temperatures greater than 600,000k @xcite . as we have stated above , the impact of these mass - loading events requires much more detailed study using new observing tools such as the atmospheric imaging array ( aia ) on the _ solar dynamics observatory _ and the upcoming interface region imaging spectrograph ( iris ) . such follow - on studies will be crucial to determine the potential impact of these events on the composition and speed of the fast solar wind . for example , we do not know whether the temperature of the plasma in these high velocity upflows is significantly in excess of the equilibrium formation temperatures @xcite . numerous challenges to our understanding arise as a result of the presented material . if the temperatures reached by the spicules is far in excess of 1 mk - readily visible in _ sdo_/aia observations of ech and qs @xcite and early , high cadence , _ stereo _ observations at the limb @xcite - then we would expect that the significant thermal pressure ensures that the super - heated portion of the spicule material overcomes the sun s gravitational pull to become a constituent of the solar wind rooted in the open magnetic network @xcite . are the alfvnic motions associated with these spicular upflow events @xcite triggered by the quasi - periodic mass - loading at the bottom of the magnetic field lines , and do those waves represent the significant source of low - frequency wave energy needed to accelerate the plasma to a typical fast wind speed at heights beyond the first few mm above the sun s surface @xcite ? finally , the most important question , what is the physical process that drives the type - ii spicules upward and heats a significant fraction of the mass that it contains to temperatures greater than one million kelvin ? the observed spatial and thermal dependence of intensity , line centroid , width and line asymmetries presented in this paper provide strict constraints to models for the roots of the fast solar wind . the transition region line profiles we observe are the result of an intricate balance of hot and cold material at a broad range of velocities . the newly discovered line asymmetries suggest that a significant portion of the energy release responsible for the transport of heated mass into the solar wind is provided by episodic heating and high speed upflow events that occur above the network at low heights ( sometimes observable as type - ii spicules ) . we suggest that detailed studies of the magneto - convective driving mechanism behind these heating events , and in particular the chromospheric spicules is essential . the magnitude , location , and timing of the energy release in the lower atmosphere will have considerable impact on the ionization balance and line formation . we anticipate that such an investigation will help us to understand the composition , initial acceleration , and nascent speed of the solar wind . swm thanks marco velli , egil leer and tom holzer for very illuminating discussions about this , and related , work . we are indebted to klaus wilhelm for comments on the manuscript and discussions regarding the work of the sumer team with respect to the si blend impact on the line profiles presented . the material presented was supported by the national aeronautics and space administration under grants to swm and rjl issued from the living with a star targeted research & technology program ( nnx08au30 g , nnh08cc02c respectively ) . in addition , part of the work presented here is supported by grants nnx08al22 g , nnx08ba99 g and nnx08ah45 g to swm and bdp . _ soho _ is a mission of international cooperation between esa and nasa . finally , we would like to acknowledge the ( anonymous ) referee who s comments strengthened the argument presented in this manuscript . the national center for atmospheric research is sponsored by the national science foundation . 72 athay , r. g. & holzer , t. e. , 1982 , , 255 , 743 beckers , j. m. , 1968 , , 3 , 367 brekke , p. , 1993 , , 87 , 443 chae , j. , schhle , u. , & lemaire , p. , 1998 , , 505 , 957 chae , j. , sun , h. s. , & poland , a. i. , 1998 , , 114 , 151 cranmer , s. r. & a. a. van ballegooijen , 2005 , , 156 , 265 cranmer , s. r. , a. a. van ballegooijen & edgar , r. j. , 2007 , , 171 , 520 curdt , w. , et al . , 2001 , , 375 , 591 dammasch , i.e. , wilhelm , k. , curdt , w. , & hassler , d.m . 1999 , , 346 , 285 davey , r. d. , mcintosh , s. w. , & hassler , d. m. 2006 , , 165 , 386 de pontieu , b. , et al . , 2007a , science , 318 , 1574 , b. , et al . , 2007b , pub . jap . , 59 , 655 de pontieu , b. , mcintosh , s. w. , hansteen , v. h. , & schrijver , c. j. 2009 , , 701 , l1 de pontieu , b. & mcintosh , s. w. , 2010 , `` quasi - periodic propagating signals in coronal loops : signatures of waves or flows ? '' , in press de pontieu , b. , mcintosh , s. w. , carlsson , m. , hansteen , v. h. , tarbell , t. d. , et al . 2010 , `` how is the solar corona filled with hot plasma ? '' , submitted science ( available to the referee if required ) delaboudinire , j .- , et al . 1995 , , 162 , 291 dere , k.p . , bartoe , j .- d . f. & brueckner , g.e . , 1984 , , 281 , 870 dere k.p . , bartoe j .- d.f . , brueckner g.e . , 1989 , 123 , 41 doschek , g.a , feldman , u. , 2004 , , 600 , 1061 feldman , u. , doschek , g. a. , & patterson , n. p. 1976 , , 209 , 270 , b. , domingo , v. & poland , a. i. 1995 , _ the soho mission _ , kluwer gebbie , k. b. et al . , 1981 , 251 , l115 geiss , j. , et al . , 1995 , science , 268 , 1005 gloeckler , g. et al . , 1998 , space sci . , 86 , 497 habbal , s. r. , scholl , i. f. & mcintosh , s. w. , 2008 , , 683 , l75 , b. n. , acton , l. w. , kankelborg , c. c. , et al . 1999 , , 187 , 229 hansteen , v. h. , & leer , e. 1995 , , 100 , 21577 hansteen , v. h. , hara , h. , de pontieu , b. , & carlsson , m. 2010 , to be submitted to . arxiv:1001.4769 hassler , d.m . , dammasch , i.e. , lemaire , p. , brekke , p. , curdt , w. , mason , h.e . , vial , j .- c . , & wilhelm , k. 1999 , science , 283 , 810 howard , r.a . , et al . , , 136 , 67 innes , d. e. , inhester , b. , axford , w. i. , wilhelm , k. , nature , 386 , 811 leamon , r. j. & mcintosh , s. w. 2007 , , 659 , 738 leamon , r. j. & mcintosh , s. w. 2008 , , 679 , l147 leamon , r. j. & mcintosh , s. w. 2009 , , 697 , l28 klimchuk , j. , 2006 , , 234 , 41 kosugi , t. , et al . , 2007 , , 243 , 3 mazzotta , p. , mazzitelli , g. , colafrancesco , s. , & vittorio , n. 1998 , , 133 , 403 , s. w. , et al . , 1998 , , 132 , 145 mcintosh , s. w. , davey , a. r. , & hassler d. m. 2006 , , 644 , l87 mcintosh , s. w. , et al . , 2007 , , 654 , 650 mcintosh , s. w. , de pontieu , b. & tarbell , t.d . , 2008 , , 673 , l219 mcintosh , s. w. , 2009 , , 693 , l1306 mcintosh , s. w. & de pontieu , b. , 2009a , , 706 , l80 mcintosh , s. w. & de pontieu , b. , 2009b , , 707 , 524 mcintosh , s. w. , & de pontieu , b. 2009 , in proc . second _ hinode _ science meeting , boulder , co , sep 30 - oct 3 , 2008 . arxiv:0901.2814 mcintosh , s. w. , innes , d. e. , de pontieu , b. , & leamon , r. j. 2010a , , 510 , l2 mcintosh , s. w. , de pontieu , b. , leamon , r.j . , 2010b , , parker , e. n. 1958 , , 128 , 664 parker , e. n. 1988 , , 330 , 474 parker , e. n. 1989 , , 121 , 271 parker , e. n. 1991 , , 372 , 719 patsourakos , s. , & klimchuk , j. , 2006 , , 647 , 1452 peter , h. & judge , p.g . , 1999 , , 522 , 1149 peter , h. , 2000 , , 360 , 761 pneuman , g. w. & kopp , r. a. , 1978 , , 57 , 49 popescu , m. d. , doyle , j. g. , & xia , l. d. 2004 , , 421 , 339 roberts , w. o. , 1945 , , 101 , 136 rouppe van der voort , l. , leenaarts , j. , de pontieu , b. , carlsson , m. , & vissers , g. 2009 , , 705 , 272 , p. h. , et al . 1995 , , 162 , 129 schrijver , c. j. et al . , 1997 , , 487 , 424 schatten , k. h. , wilcox , j. m. , & ness , n. f. 1969 , solar phys . , 6 , 442 shibata , k. & uchida , y. , 1986 , , 103 , 299 smith , c. w. , et al . 1998 , space sci . , 86 , 613 stone , e. c. , frandsen , a. m. , mewaldt , r. a. , et al . 1998 , space sci . rev . , 86 , 1 suzuki , t. & inutsuka , s. , 2005 , , 632 , l49 suzuki , t. , 2006 , , 640 , l75 tian , h. , curdt , w. , marsch , e. , & schhle , u. 2009 , , 504 , 239 tian , h. , teriaca , l. , curdt , w. , & vial , j .- c . 2009 , , 703 , l152 , s. , et al . , 2007 , science , 317 , 1192 tsuneta , s. , et al . , 2008 , , 249 , 167 tu , c .- y . , zhou , c. , marsch , e. , xia , l.d . , zhao , l. , wang , j .- x . & wilhelm , k. 2005a , science , 308 , 519 tu , c .- y . , zhou , c. , marsch , e. , wilhelm , k. , zhao , l. , xia , l.d . , & wang , j .- x . 2005b , , 624 , l133 verdini , a. & velli , m. , 2007 , , 662 , 701 verdini , a. , et al . 2010 , , 708 , l116 wilhelm , k. , et al . , 1995 , , 162 , 189 wilhelm , k. , et al . , 2000 , , 353 , 749 withbroe , g. l. , 1983 , , 267 , 825 xia , l. d. , marsch , e. & curdt , w. 2003 , , 399 , l5 xia , l. d. , marsch , e. & wilhelm , k. 2004 , , 424 , 1025 yun , g.s . , bellan , p.m. , 2010 , phys . plasma , 17 , 062108 zacharias , p. , bingert , s. , & peter , h. 2009 , memorie della societa astronomica italiana , 80 , 654 in the right hand panel of fig . f11 we see an obvious shoulder in the red wing of the mean 770.4 profile in the ech network . is this the signature of a blend in the line profile and what impact does it have on the presented analysis ? to estimate the impact of the blends on the @xmath6 and r - b asymmetry maps derived in this paper we use quiet sun data from the hrts spectral atlas @xcite since it only contains uv spectral lines observed in first order ( i.e. , it lacks the line ) . we find that there are two blends ( 1540.74 and 1540.96 ) that are of similar magnitude ( of order 100 erg @xmath21 s@xmath22 st@xmath22 @xmath22 above the continuum level ) near the line center of the line ( see fig . f12 , and fig . 2 of * ? ? ? in quiet sun , the line as observed with sumer in second order has an intensity of a magnitude between those of 1548 and 1550 @xcite . to determine the relative size of the blends to that of the in the sumer data we compare the hrts measurements of the blends with those of 1548 and 1550 . the latter are at levels of order 2000 to 4000 erg @xmath21 s@xmath22 st@xmath22 @xmath22 for quiet sun ( in hrts ) . this means that in quiet sun , the are at a level of of order 3% of the quiet sun intensity of the as observed with sumer in second order . in coronal hole network regions the line has about a third of the quiet sun network brightness , while the photospheric lines will have the same brightness irrespective of location . this means that , in the ech , the blends should have a higher relative magnitude . we use these estimates to investigate the impact of these blends on our analysis of the data . in the two panels of fig . f13 we use a forward model that is based on a three gaussian composite profile ( two gaussian lines of the same brightness separated by 36km / s and estimating the gaussian from the residual ) and compare it with our average observed profiles in the ech network ( left ) and quiet sun ( right ) . in both cases we find that the blends have the same intensity ( above the continuum ) as the continuum background itself . for the ech we see that the blends could be of order 10 - 20% of the line brightness , while in the quiet sun they are at most a few percent . for the ech , we see that the short wavelength blend and the lie almost on top of one another with the longer wavelength blend creating the shoulder in the red wing . as a result , the centroid of the composite profile is shifted by 8 km / s to the red , compared with the centroid of the actual line ( 21km / s ) . for the quiet sun case , the two blends are almost symmetrically distributed about the profile , which is blue - shifted by a couple of km / s . to illustrate the potential impact of these blends on the r - b measure used for in the main body of the paper we use the fitted component parameters for fig f13 . we have added the 5% amplitude component centered on 60km / s in the blue wing , with a 1/e width of 25 km / s , that is needed to describe the measured profile , to conduct a simple experiment . in this experiment , for which the results for the ech and quiet sun profiles are shown in the left and right panels of fig . f13b , we compare the r - b profile ( as a function of velocity ) of the measured network profiles ( @xmath23 ) with those for which the blends are 50% larger ( @xmath24 ) and 50% smaller ( @xmath25 ) in magnitude than those determined from the corresponding panels of fig f13 . we see that the r - b profiles for the measured ech and quiet sun network profiles are very similar . the largest impact on the r - b analysis with modified blend amplitudes is on the ech line profile , as the three quiet sun profiles are indistinguishable . in the case of the ech r - b profile we see that the main change is the magnitude of the asymmetry observed ( of the order of a few percent ) with the larger amplitude blends having a larger effect in the red wing ( enhancing the profile shoulder ) , but little on the blue wing . this is somewhat different for the lower amplitude blends where there is a noticeable difference ( 2% ) in both wings . the velocity position of the blue asymmetry remains relatively stable ( as we might expect since we have not moved the modified blends in velocity - space ) shifting by a few km / s further to the red as the blends get weaker . this is also expected and is due to the decreased weighting of the red wing shoulder such that we are able to `` see '' the blue wing asymmetry at lower velocities . this experiment lends support to our belief that the blend configuration does not significantly impact the faint ( 2 - 5% ) asymmetry in the blue wing of the profile in either case , although the blends and profile shape affect centroiding the raw profile accurately . in quiet sun , the strength of the line compared to that of the blends makes the r - b analysis more robust . the weak , and roughly symmetrically positioned , blends around the rest wavelength will mostly cancel each other in r - b asymmetry maps of the quiet sun . in ech , the incorrect centroiding of the line profiles in the ech by a few km / s to the red will skew the r - b asymmetry measure by a few km / s to the blue . for this will result in the r - b maps having a blue hue , see figs . f9 , f10 and f13b where the r - b profile of the example does not return to zero . however , we now show that this blue `` hue '' ( caused by the blends ) can not explain all of the blueward asymmetries we see in in the ech , especially the stronger blue asymmetries in the ech network . if the blueward asymmetries were caused only by the effect of the blends on the centroid of the profile , then we would expect the normalized r - b measure to be of the same magnitude in the network and in the internetwork . from fig . [ f13b ] we know that stronger blends could , in principle , create a spurious excess blueward asymmetry . however , the network / internetwork contrast for the is of order 2 @xcite , which is very similar to what can be expected for a chromospheric line such as . we note that the contrast for lines that are formed at similar temperatures like and is of order 1.5 - 2 @xcite . as a result , the relative impact of the blends on should be of the same size for network and internetwork , which would lead to `` flat '' normalized r - b map that shows no difference between network and internetwork . this is _ not _ what we observe : fig . [ f13_normalized ] shows significantly larger blueward normalized asymmetries of in and around the network , with many internetwork regions lacking asymmetric profiles . further , the individual sumer spectra have relatively low count rates - the key motivating point behind our decision to add the spectra over the 3@xmath5 3 region @xcite . low signal - to - noise in the line spectra adversely impacts the reliability of any estimate of the line profile asymmetry . to investigate this property of the line spectra we construct an approximate measure of the signal - to - noise at each pixel in the form @xmath26 . comparing this signal - to - noise estimate for each portion of the ech observation to the r - b asymmetry maps ( figs f9 and f10 ) in fig . fs2n we see that regions of strong blue asymmetry generally correspond to regions where the signal - to - noise measure is larger than 10 ( enclosed by contours ) . those are places where we might logically expect the magnitude of the r - b asymmetry to be of order 10% to be reliably detected . from the bottom panels of the figure we see that those regions of large profile r - b asymmetry are , at least , twice as large in magnitude as the square root of the peak line intensity at that pixel and so we assume that the determination of the profile asymmetry is real signal , and not the result of noise in the spectra . based on the visual correspondence between regions of strong blue asymmetry and higher signal - to - noise shown in the figure we deduce that the blue - asymmetries observed are significant , albeit marginally above the noise threshold in the coronal hole , and are the signature of high velocity plasma upflows at those locations , as we may be led to believe by recent _ sdo_/aia observations @xcite , and the _ trace _ observations of this ech that follow in appendix sec : trace_imaging . a final , independent confirmation in support of the ubiquity of the faint blue wing asymmetry in magnetic regions comes from more recent , lower signal - to - noise sumer observations of quiet sun which show the same faint blue - shifted component in 770 spectra when observed in first order ( i.e. , without the blends * ? ? ? sec : fuckyou we show that similar first order measurements of 770 _ in ech _ confirm the existence of faint blue wing asymmetries . these first order measurements are immune to the effects of the blends . further , the non - thermal broadening of the first order ( fig . f17 ) is of a magnitude commensurate with those shown in earlier figures of the second order observations ( figs . f2 - f6 ) adding further support to the fact that the blends do not sufficiently alter the diagnostic value of the measurements shown , consistent with the deduction of @xcite . we also consider potential blends in the 1548 line : we can clearly see in the 70km / s r - b asymmetry maps of figs . f9 and f10 that there is a signature of another spectral line on the red side of the line . based on the hrts and sumer @xcite spectral atlases , this is another blend . however , this blend is sufficiently far from the line center and weak that it does not impact the diagnostics presented other than to mask the true extent of the blue wing asymmetry . an unfortunate , but significant , casualty of these blends is the 1534 line . it would have allowed for study of much cooler plasma , but the blend in the red wing of that line is place at about the 1/e width of the line from line center and would have significantly impacted both r - b and @xmath6 diagnostics . it is not clear what impact these blends have on other high - profile sumer observations of the line in second order and results derived from them ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the careful analysis presented in @xcite concludes that the coronal hole plasma , and the network line profiles in particular , in second order are _ not _ at the mercy of blends and that the network profiles are not singularly impacted by the si blends . we note that caution is needed when interpreting these line profiles , there are at least two components contributing to the line profiles we observe and a single gaussian fit to those profiles is fraught with danger ( e.g. , * ? ? ? we believe the only way to properly assess the impact of the blends is to perform a first / second order observation and compare the line spectra taken of the same region at approximately the same time ( e.g. , * ? ? ? * ) . unfortunately , such a test is not possible with sumer and so we must rely on the first order diagnostics to illustrate the presence of the same spectroscopic characteristics as observed in the body of the present work . while sumer observed the ech the transition region and coronal explorer ( _ trace _ ; * ? ? ? * ) observed it s north - west portion in the 171 passband . this passband consists of emission lines of and , i.e. , it is sensitive to emission of plasma over 1mk . the exposure times for this sequence were variable due to the automatic exposure control of the telescope , but modulated between 3 and 14 seconds for the 12 hours of observation . the image shown in fig . trace has an exposure time of 13.77s and the position of the coronal hole boundary in the field of view is shown as a contour familiar from many of the figures above . inspection of the online edition movie supporting this figure suggests there are several regions where very weak propagating disturbances at coronal temperatures ( @xmath27 1mk ) are visible ( similar to those described in * ? ? ? these `` blobs '' , or propagating disturbances , are readily and persistently seen in the active regions ( top left of the field of view ) and have similar speeds in other imaging experiments ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? there are many other locations showing the same behavior ( encircled in red ) in the field of view , even some in the brighter emission regions of the coronal hole . indeed , the majority ( most clearly that centered on -210 , 140 ) of these encircled locations are locations of strong blue asymmetries in the and lines ( see , e.g. , fig . unfortunately , the erratic behavior of the _ trace _ exposure time does nt permit the type of space - time plot analysis that we have performed for other _ trace _ @xcite , or _ stereo _ observations @xcite , and so any further comparison between the _ trace _ image sequence and sumer r - b analysis is purely qualitative . the sample figure and movie is included in this appendix for the sake of completeness and to illustrate the presence of propagating disturbances in some locations of r - b asymmetries when the _ trace _ observation characteristics permit their observation . in this appendix we will present observations of line profile asymmetries in ech from the 1031.9 ( cf . sect . 2.3 of * ? ? ? * ) and in the unblended , first spectral order , 770 line profiles observed with sumer s detector b. as noted in @xcite , high ( @xmath2720 ) signal - to - noise is essential to robustly measure these small amplitude profile asymmetries . observations with the coated region of sumer detector b have considerably higher signal - to - noise than the long - exposure detector a observations shown in the body of this paper without the potential for impact on the spectra by cool , very weak blends . the observations presented in figs . f13_o6 through f16 provide supporting evidence that the profile asymmetries in coronal holes have the same characteristic nature in ech plasma in the 1031.9 line . the observations taken by sumer of a small patch of ech were taken on october 19 - 20 1999 ( 21:13 - 01:15ut ) and october 22 1999 ( 01:57 - 06:00ut ) , the first to the east and the second to the west of the equator . the observations in both cases used detector a , a @xmath28 slit , 1.13 horizontal steps , and 100s exposures f13_o6 and figs . f15 show context ( supplied by eit 195 images ) and the result of a single gaussian fit to the sumer spectra . the location of the small rasters are shown as an orange box on the eit image ( left ) image with the peak line intensity , ( relative - not absolutely calibrated ) doppler velocity , and line width for that region shown top - to - bottom on the right . on inspection of the r - b analysis of the line profiles in the region ( in figs . f14 and f16 ) we see little evidence that would allow us to draw a coronal hole boundary in the region , the reason why one is not drawn . the asymmetry of the profiles behaves as we have seen before : red at low velocities , but a more pronounced blue asymmetry going out to greater than 100 km / s . at such high velocities we expect that there may be some very high velocity explosive events @xcite around some of the ( presumed ) network elements . this was noted for quiet sun spectra of the same line ( sect . 2.3 of * ? ? ? figures f17 through f18 show the context and analysis of first order ( detector b ) observations in an ech ( including an euv bright point ) in the 770 line on october 19 - 20 1996 ( 20:22 - 03:24ut ) with an exposure time of 60s . unfortunately , the malfunction in the scanning mechanism meant that this observation requires the sun to rotate underneath the sumer slit to `` map '' the region in a ` drift - scan ' - in this case the sumer slit was held fixed at the central meridian and spectra were acquired allowing the sun to rotate underneath and an image to be formed over time . we note that this observation includes supporting information from the 765 line to provide some network context . figure f17 shows the context eit 195 from 23:00ut ( left ) with the slit position indicated . the panels on the right side of the figure show the peak line intensity , doppler velocity , non - thermal line width and the peak intensity to isolate the network using an iso - intensity contour of 200 . the passage of the euv bright point and a few network elements inside the coronal hole are clear . figure f18 shows the r - b analysis of the first order 770 observation . we notice that the non - thermal line widths of the first order observation are consistent with those observed in the ech in second order - indicating that the blends may not play a strong role in modifying the second order spectra . similarly , we see a progression from red asymmetries to blue at higher speeds than for in the coronal hole . more importantly , we find that many of the network elements ( as outlined by intensity contours ) indeed show blueward asymmetries . this supports the results of the main body of the paper . the timeseries also suggests that the blueward asymmetries are temporally intermittent , which we have noted before for quiet sun @xcite . the lack of blends in first order for this line permits observation of the extended blue wing of the line to higher velocities . indeed , it seems as though the core of the euv bright point ( y = -160 ) still shows a strong blue - wing asymmetry beyond 120km / s . this may indicate that the velocity range of the blue asymmetries depends on the magnetic field strength , which would suggest that the speed of the disturbances ( e.g. , spicules ) in the lower atmosphere is dependent on how strong the field is . alternatively , the larger velocities in the bright point may be because of a favorable viewing angle .

we study the quantum phase diagram and excitation spectrum of the frustrated- spin-1/2 heisenberg hamiltonian . a hierarchical mean - field approach , at the heart of which lies the idea of identifying _ relevant _ degrees of freedom , is developed . thus , by performing educated , manifestly symmetry preserving mean - field approximations , we unveil fundamental properties of the system . we then compare various coverings of the square lattice with plaquettes , dimers and other degrees of freedom , and show that only the _ symmetric plaquette _ covering , which reproduces the original bravais lattice , leads to the known phase diagram . the intermediate quantum paramagnetic phase is shown to be a ( singlet ) _ plaquette crystal _ , connected with the neighbouring nel phase by a continuous phase transition . we also introduce fluctuations around the hierarchical mean - field solutions , and demonstrate that in the paramagnetic phase the ground and first excited states are separated by a finite gap , which closes in the nel and columnar phases . our results suggest that the quantum phase transition between nel and paramagnetic phases can be properly described within the ginzburg - landau - wilson paradigm .

0809.4035

one of the primary goals of modern condensed matter physics is the characterization of strongly correlated quantum systems . a large class of such materials is represented by frustrated antiferromagnets , which are believed to exhibit a variety of novel states of matter at sufficiently strong coupling . growing experimental evidence indicates that layered materials such as @xmath2 @xcite , @xmath3 @xcite and @xmath4 @xcite can be adequately described by an antiferromagnetic heisenberg model with frustrating next- and next - next - nearest neighbor interactions . as a result , the study of low - dimensional magnets and their frustration - driven quantum phase transitions have attracted a lot of theoretical attention in the last decade @xcite . a paradigmatic system , illustrating the effects of frustrating couplings , is the spin-@xmath5 heisenberg model on a square lattice with competing nearest ( @xmath0 ) , and next - nearest ( @xmath1 ) neighbor antiferromagnetic ( af ) interactions ( @xmath0-@xmath1 model ) . despite numerous analytical and numerical efforts , its phase diagram , which exhibits a two sublattice nel af , quantum paramagnetic , and a four sublattice columnar af states , continues to stir certain controversy ( for a review of recent achievements , see ref . ) . while existence of the nel - ordered phase at small frustration ratio @xmath6 , and of the columnar af state at large @xmath6 is widely established , properties of the intermediate non - magnetic phase , which occurs around the maximum frustration value @xmath7 , are still under debate . particularly , the correlated nature of the intermediate state and the kind of quantum phase transition separating it from the nel state , attract most attention . various methods have been recently applied to characterize the quantum paramagnetic phase , such as green s function monte carlo @xcite , coupled cluster methods @xcite , series expansions @xcite and field - theoretical methods @xcite . as a result , several possible candidate ground states were proposed , namely : spin liquid @xcite , preserving translational and rotational symmetries of the lattice , as well as various lattice symmetry breaking phases , out of which the dimer @xcite , and the _ plaquette resonating valence bond _ phases @xcite are worth mentioning . not surprisingly , the nature of the quantum phase transition separating the nel - ordered and quantum paramagnetic phases is also under scrutiny . the most dramatic , and at the same time original , scenario @xcite is believed to violate the ginzburg - landau - wilson paradigm of phase transitions @xcite which revolves around the concept of an order parameter . such point of view is based on the observation that there are different spontaneously broken symmetries in the nel and quantum paramagnetic phases , which thus can not be connected by a group - subgroup relation . the former , of course , breaks the @xmath8 invariance of the hamiltonian and lattice translational symmetry @xmath9 @xcite , but preserves the four - fold rotational symmetry of the square @xmath10 . on the other hand , the paramagnetic phase is known to restore the spin - rotational symmetry and is believed to break @xmath9 and @xmath10 , due to spontaneous formation of dimers along the links of the lattice @xcite . it follows then , that these two phases can not be joined by the usual landau second - order critical point . this phase transition can either be of the first - order @xcite ( the latest coupled cluster calculations @xcite , however , seem to rule out this possibility ) , or represent an example of a second order critical point , which can not be described in terms of a bulk order parameter , but rather in terms of emergent fractional excitations ( spinons ) , which become deconfined right at the critical point @xcite . however , evidence regarding the structure of the non - magnetic phase is quite controversial . indeed , the results of spin - wave calculations @xcite , large-@xmath11 expansions @xcite , and calculations using the density matrix renormalization group combined with monte - carlo simulations @xcite are believed to indicate the emergence of a dimer order . on the other hand , monte - carlo @xcite and coupled cluster calculations @xcite , and analytical results @xcite seem to support the presence of @xmath10 symmetry ( plaquette - type ordering ) in the paramagnetic phase . in the absence of a reliable numerical or analytical proof of existence of any particular order in the non - magnetic region , there is no apparent reason to believe in the exotic deconfined quantum criticality scenario . although there apparently exists numerical evidence @xcite , at the moment of writing the authors are unaware of a local hamiltonian in space dimensions larger than one , rigorously proven to exhibit the type of quantum critical point discussed in ref . . interestingly , it was demonstrated in ref . that a two - dimensional ( 2d ) lattice model can possess a _ first order _ quantum critical point , which exhibits deconfined excitations . all in all , the complexity of methods used to infer properties of the paramagnetic phase and the variety of different conclusions have created a certain degree of confusion . our goal in the present paper is to try to clarify some of this controversy by proposing a controlled and manifestly symmetry preserving method , geared to computing ground state properties of the @xmath0-@xmath1 model . our approach is based on the recently proposed systematic methodology to investigate the behavior of strongly coupled systems @xcite , whose main idea consists of identifying _ relevant _ degrees of freedom and performing an educated approximation , called the hierarchical mean - field ( hmf ) , to uncover the phase diagram and other properties of the system of interest . in a future work these ideas will be coupled to a new , variational with respect to the energy , renormalization group approach , which thus adapts to the concept of relevant degrees of freedom . -@xmath1 model , summarizing our results . in each phase we show spontaneously broken ( framed symbols ) and unbroken symmetries ( usual symbols ) . the translational invariance is broken along both directions in the nel and paramagnetic phases , and only along the @xmath12direction in the columnar phase this fact is indicated by the subscripts @xmath13 and @xmath12 after @xmath9 . conclusions regarding the order of the phase transition , separating nel and plaquette crystal phases , as well as symmetries of various phases , are based upon extrapolation of our results towards the thermodynamic limit . ] in the present work we construct hmf approximations for the @xmath0-@xmath1 model . the crux of our method is the identification of a _ plaquette _ ( spin cluster @xmath14 or even larger @xmath15 ( _ superplaquette _ ) symmetry - preserving cluster ) as the relevant elementary degree of freedom , which captures necessary quantum correlations to represent essential features of the phase diagram . the importance of this degree of freedom was realized only recently in the present context @xcite , and somewhat earlier in connection with @xmath16 spin - orbital @xcite , and hubbard @xcite models . besides being variational , our formalism has the attractive feature of preserving fundamental lattice point symmetries and the @xmath8 symmetry of the hamiltonian , by utilizing the schwinger boson - type representation and racah algebra technology . remarkably , such simple mean - field calculation already yields all known results , concerning the phase diagram of the @xmath0-@xmath1 model , with a good accuracy , namely : existence of a nel - ordered phase with antiferromagnetic wavevector @xmath17 and spin - wave type excitations for @xmath18 , a non - magnetic intermediate gapped phase , separated by a second order quantum phase transition , and a first order transition point , which is characterized by the discontinuous disappearance of the energy gap and connects the paramagnetic state with the columnar antiferromagnetic phase at @xmath19 and @xmath20 for @xmath21 . we emphasize that our investigation primarily focuses on the symmetry analysis of the various phases . out of many possible coarse graining scenarios , such as covering of the 2d lattice with plaquettes , dimers and crosses , only the @xmath10-_symmetry preserving _ plaquette ( or superplaquette ) covering ( which reproduces the original bravais lattice ) displays the correct phase diagram . in particular , the intermediate paramagnetic phase is shown to be a _ plaquette crystal _ , which preserves spin and lattice rotational symmetries . for all other scenarios , including dimerized ( bond - ordered ) phases , we were unable to reproduce all known quantum phase transition points of the model . we notice that the hmf coarse graining procedure leads to an explicit breaking of a particular translational symmetry . as a result , one can not draw rigorous conclusions on the order of the phase transitions , based solely on a fixed coarse graining . nevertheless , it is still possible to make some predictions , using a finite - size scaling of the relevant degree of freedom towards the thermodynamic limit , where the effects associated with coarse graining should disappear . next two sections are devoted to the formulation of the hmf approach . then , we present results of our calculations and close the paper with a discussion . our main conclusions are summarized in fig . [ fig_phase_diagram ] , which emphasizes symmetry relations between different phases of the model . we consider the spin-@xmath5 antiferromagnetic heisenberg model with frustrated next - nearest neighbor interactions @xmath1 , defined on a 2d bipartite lattice with @xmath11 sites : @xmath22 as mentioned already in the introduction , we choose the plaquette , fig . [ fig_plaquette ] , as our _ elementary _ degree of freedom . then , assuming that @xmath11 is chosen appropriately , the entire lattice can be covered with such plaquettes in a sub - exponentially @xcite ( @xmath23 ) large number of ways . aiming at illustrating the main idea of the method , in this section we consider in detail only the symmetric covering of the lattice with @xmath14 plaquettes , which preserves the @xmath10 lattice symmetry , see fig . [ symm_plaquette_covering ] , although later the displaced covering ( fig . [ displ_plaquette_covering ] ) , which breaks @xmath10 down to @xmath24 ( two - fold symmetry axis ) , and the case of larger plaquettes ( superplaquettes , fig . [ fig_sp ] ) will be analyzed as well . it is convenient to take as a basis the states @xmath25 where @xmath26 and @xmath27 are total spins of the plaquette diagonals , while @xmath28 is the total spin of the entire plaquette and @xmath29 is its @xmath30-component . in this basis the hamiltonian of a single plaquette , @xmath31 is diagonal with eigenvalues @xmath32 + \nonumber \\ + & \frac{j_2}{2}\bigl[l_1\bigl(l_1 + 1\bigr)+l_2\bigl(l_2 + 1\bigr)-3\bigl ] . \label{plaquette_se}\end{aligned}\ ] ] we note that the basis of eq . ( [ basis ] ) is a natural one and allows us to explicitly label states with corresponding representations of @xmath8 . plaquette has each vertex occupied by a @xmath33 spin . the diagonal spins interact through a heisenberg term of strength @xmath1 , while nearest neighbor spins interact with strength @xmath0 . ] each plaquette is connected to 4 nearest and 4 next - nearest neighbors . ] the next step is to establish how a plaquette couples to the rest of the system . in fig . [ symm_plaquette_covering ] we show the symmetric plaquette covering of the 2d lattice . in the figure the vertices of every non - central plaquette are similarly labeled by the numbers @xmath34 , and total spins of diagonals are @xmath35 and @xmath36 . in the uncoupled basis matrix elements of the inter - plaquette interaction are : @xmath37 where @xmath38 ( @xmath39 ) corresponds to the nearest ( next - nearest ) neighbor interaction , @xmath40,@xmath41 ( @xmath42,@xmath43 ) represent initial ( final ) angular momenta of the two plaquettes and @xmath44 is their total angular momentum . in this equation we have introduced the notations @xmath45 , @xmath46 and @xmath47 , and similarly for the primed indices . because each plaquette has 4 nearest neighbors and 4 next - nearest neighbors ( see fig . [ symm_plaquette_covering ] ) , the symmetrized next - nearest neighbor interaction may be written as : @xmath48 while the symmetrized nearest neighbor plaquette interaction has the form : @xmath49 + \nonumber \\ & \qquad+2\bigl\langle\lambda_1^\prime\lambda_2^\prime , lm|h_{\rm int}^2| \lambda_1\lambda_2,lm\bigr\rangle , \nonumber\end{aligned}\ ] ] in eqs . ( [ c_4_nnn_int ] ) and ( [ c_4_nn_int ] ) the symbols @xmath50 denote reduced matrix elements of the @xmath51-th spin operator , and : @xmath52 where @xmath53 are wigner @xmath54 symbols ( or racah coefficients ) @xcite . let us now identify the plaquette degree of freedom with a schwinger boson which creates a specific state of the plaquette . then , the hamiltonian of eq . ( [ hamiltonian ] ) in the plaquette basis can be expressed as : @xmath55 where the operator @xmath56 creates a boson on site @xmath57 of the _ plaquette _ lattice ( which contains @xmath58 sites ) in the state , denoted by an index @xmath59 , running through the entire single - plaquette hilbert space ( of dimension @xmath60 ) and the summation is performed over doubly repeated dummy indices . the unphysical states are eliminated by enforcing the local constraint @xmath61 . in what follows , we impose periodic boundary conditions on the plaquette lattice . the bosonic operators @xmath62 define the hierarchical language @xcite for our problem . it will be used in the next section , where we develop an approximation scheme for diagonalizing the hamiltonian of eq . ( [ ham_second_quant ] ) . as it follows from eq . ( [ plaquette_se ] ) , the lowest single - plaquette state has the energy @xmath63 per spin , which , when @xmath64 , gives only the energy of a classical 2d antiferromagnet . thus , it is necessary to take into account the interaction term in eq . ( [ ham_second_quant ] ) . the hmf approximation is a mean - field approach , performed on the relevant degrees of freedom . in the present section we discuss only the simplest one a hatree - fock like ( hf ) approximation . a possible way to include fluctuation corrections is presented in appendix b. the hf approximation introduces the mixing of single - plaquette states which minimizes the total energy of the system and is based on a canonical transformation among the bosons , which we will restrict to be uniform ( plaquette independent ) : @xmath65 the real matrix @xmath66 satisfies canonical orthogonality and completeness relations : @xmath67 a _ translationally invariant _ variational ansatz for the ground state ( vacuum ) is a boson condensate in the lowest hf single - particle energy state ( @xmath68 ) : @xmath69 and since it has one boson per plaquette , there is no need to impose the schwinger boson constraint in the calculation . minimizing the total energy with respect to @xmath66 , we arrive at the self - consistent equation : @xmath70 where @xmath71 are the nearest- and next - nearest coordination numbers . the ground state energy ( gse ) per spin is then given by the expression : @xmath72 with @xmath73 being the lowest eigenvalue of eq . ( [ hmf_equation ] ) . another fundamental quantity to compute is the polarization of spins within a plaquette : @xmath74 where @xmath75 is the spin index , and the matrix elements ( determined from the wigner - eckart theorem ) are : @xmath76 this enables us to define the staggered and collinear ( along @xmath77 and @xmath12 axes ) magnetizations : @xmath78 notice the extreme simplicity of the hmf approximation . the reason why it is able to realize meaningful results is that the plaquette degree of freedom seems to contain the main correlations defining the physics behind the hamiltonian of eqs . ( [ hamiltonian ] ) , and ( [ ham_second_quant ] ) . to avoid confusion , we emphasize that the hf approximation and the fluctuation theory of appendix b are physically ( and obviously mathematically ) different from the spin - wave or canonical schwinger - wigner boson mean - field approach to spin systems @xcite . in particular , we make no assumption about the underlying ground state , thus allowing for an interplay of various quantum phases . moreover , it will be demonstrated , that the collective excitation spectra in each phase consistently reflect spontaneously broken symmetries , unlike the usual schwinger boson case @xcite , in which one obtains gapped excitations . symmetry is broken down to @xmath24 . ] symmetry is lowered to @xmath24 in both cases . ] symmetry , but the resulting lattice breaks it . ] our choice of the plaquette as an elementary degree of freedom remains unjustified at this point . in order to show its relevance we applied the analysis of two previous sections to several other coarse grainings ( besides the symmetric plaquette covering , case ( a ) , shown in fig . [ symm_plaquette_covering ] ) : ( b ) _ superplaquette _ ( spin cluster @xmath15 ) degree of freedom , covering the lattice in such a way that @xmath10 is preserved ( see appendix a for details ) ; ( c ) displaced plaquette covering of the lattice , fig . [ displ_plaquette_covering ] ; ( d ) symmetric and displaced _ dimer _ coverings , shown respectively in right and left panels of fig . [ dimer_covering ] ; ( e ) _ cross _ degree of freedom , fig . [ cross_covering ] . one should observe that symmetries of the original bravais lattice are preserved only in cases ( a ) and ( b ) . in cases ( c ) and ( d ) the lattice rotational symmetry @xmath10 is lowered to @xmath24 . case ( e ) is special in the sense that an isolated degree of freedom does not possess a singlet ground state . the information about a particular configuration is encoded in matrix elements of @xmath79 , whose calculation is elementary . other equations , presented in secs . ii and iii , retain their form . for each of the above cases we iteratively solve eq . ( [ hmf_equation ] ) and compute the gse ( [ hmf_energy ] ) , and staggered and collinear magnetizations ( [ magnetizations ] ) . the main message , which we would like to convey in this section is that only the plaquette degree of freedom ( of any size ) is relevant for constructing the phase diagram of the hamiltonian of eq . ( [ hamiltonian ] ) . let us focus first on cases ( a ) and ( b ) , i.e. symmetry - preserving coverings of the lattice with plaquette , fig . [ symm_plaquette_covering ] , and superplaquette degrees of freedom . the resulting gse as a function of @xmath6 is shown in fig . [ fig_gse_2x2_4x4 ] . one immediately observes a level - crossing at @xmath80 , indicating the first - order transition and a second - order quantum critical point at @xmath81 , which is supported by a jump of the second order derivative @xmath82 , fig . [ fig_gse_d2_2x2_4x4 ] . both nel and columnar phases are characterized by spontaneously broken @xmath8 symmetry . the former exhibits a nonvanishing staggered magnetization , @xmath83 , while the latter has nonzero collinear magnetization along the @xmath77-direction , @xmath84 . both order parameters become zero in the paramagnetic phase , suggesting that @xmath8 is restored . these results are summarized in fig . [ fig_mag_2x2_4x4 ] , from which it also follows that the phase transition at @xmath85 is continuous , while @xmath86 corresponds to a first - order transition point . we remind , in this connection , that our approach does not explicitly break the spin rotational symmetry , thus allowing for the treatment of competing ground states . as expected , considering a larger elementary degree of freedom superplaquette leads to a significant improvement of the gse and reduction of the magnetization , due to larger quantum fluctuations . the finite - size scaling ( insets in figs . [ fig_gse_2x2_4x4 ] and [ fig_mag_2x2_4x4 ] ) , using these two sizes ( @xmath14 and @xmath15 ) , indicates that @xmath87 and @xmath88 in the thermodynamic limit , a satisfying result for a hf approximation , which completely ignores fluctuations ( these numbers should be compared with well - known results of monte - carlo simulations @xcite : @xmath89 and @xmath90 ) . and @xmath15 plaquette elementary degrees of freedom . the inset shows finite - size scaling in the af phase at @xmath64 . ] for the @xmath14 and @xmath15 plaquette degrees of freedom . the discontinuity at @xmath91 is indicative of a second - order quantum phase transition . in the inset we present finite - size scaling for the jump @xmath92 . ] , for @xmath93 and collinear magnetization along the @xmath77-direction , @xmath94 , for @xmath95 ( for the @xmath14 and @xmath15 plaquette degrees of freedom ) , computed at the hf level . notice the continuous phase transition at @xmath91 and a first order transition at @xmath96 ( @xmath14 ) and @xmath97 ( @xmath15 ) . the inset shows finite - size scaling of @xmath98 at @xmath64 . ] next , we discuss in more detail symmetry properties of the various phases in figs . [ fig_gse_2x2_4x4 ] , [ fig_mag_2x2_4x4 ] . at all values of @xmath99 , the lattice translational symmetry @xmath9 is broken @xcite , but the rotational @xmath10 symmetry is preserved . for @xmath93 this corresponds to a nel - type long - range order with spontaneously broken @xmath8 . at large values @xmath95 , we observe the columnar ordering , which spontaneously breaks @xmath10 down to @xmath24 , and @xmath8 , but partially ( i.e. , along one direction ) restores the lattice translational symmetry . we present a more detailed discussion of the spatial symmetries later in this section . in the intermediate region @xmath100 the spin @xmath8 rotational symmetry is restored . in this paramagnetic phase the ground state wavefunction is a tensor product of individual plaquette ground states ( with quantum numbers @xmath101 , @xmath102 ) : @xmath103 . \nonumber\end{aligned}\ ] ] this ground state necessarily breaks the lattice translational symmetry , but preserves @xmath10 . in fact , the paramagnetic region on the phase diagram of figs . [ fig_gse_2x2_4x4 ] and [ fig_mag_2x2_4x4 ] is a trivial plaquette crystal : a set of _ non - interacting _ plaquettes , because the expectation value of the plaquette interaction ( see eq . ( [ ham_second_quant ] ) ) in the singlet state @xmath104 vanishes . an analogous situation is realized when the superplaquette is chosen as an elementary degree of freedom : the paramagnetic phase is a crystal of superplaquettes . it is interesting to note that in ref . a `` plaquette resonating - valence - bond state '' , exactly equal to ( [ plaquette_gs ] ) , has been proposed . however , later @xcite the intermediate phase was argued to be a spin liquid , i.e. , a state that preserves the lattice translational symmetry . in order to learn about spatial symmetries in various phases , we compare magnitudes of the several lattice symmetry - breaking observables proposed in the literature . we consider the following three , introduced in ref . ( in the notation of that paper ) : @xmath105 , \nonumber\end{aligned}\ ] ] where indices @xmath106 specify a spin in the 2d lattice . the operator @xmath107 probes the plaquette ordering , which preserves the lattice rotational symmetry , while @xmath108 and @xmath109 correspond to the columnar ordering . we note , however , that @xmath108 is already non - zero for an isolated plaquette ( or superplaquette ) . these functions can be combined in the complex `` order parameter '' , introduced in ref . . here we show details of the calculation of functions @xmath110 for the plaquette degree of freedom , case ( a ) , and only present the result for @xmath107 for the superplaquette case ( b ) . in the plaquette representation the above operators are written as : @xmath111 ; \nonumber \\ f_{2,4}=&\frac{1}{n}\sum_{i , j}\bigl[\bigl({\bm s}_{1;i , j}\mp{\bm s}_{4;i , j } \bigr)\bigl({\bm s}_{2;i , j}\mp{\bm s}_{3;i , j}\bigr)\pm \label{order_parameters } \\ & \,\quad\qquad\pm\bigl({\bm s}_{2;i , j}{\bm s}_{5;i+1,j}+{\bm s}_{4;i , j } { \bm s}_{7;i+1,j}\mp \nonumber \\ & \,\,\qquad\qquad\mp{\bm s}_{3;i , j}{\bm s}_{5;i , j+1}\mp{\bm s}_{4;i , j } { \bm s}_{6,i , j+1}\bigr)\bigr ] . \nonumber\end{aligned}\ ] ] in this equation the indices @xmath112 are coordinates of a plaquette in the lattice . ) , plotted as functions of @xmath6 for the plaquette and superplaquette degrees of freedom . due to the unbroken @xmath10 symmetry in the nel and paramagnetic phases , values of @xmath107 are twice larger than the corresponding values of @xmath108 , except in the columnar phase , where @xmath10 is spontaneously broken down to @xmath24 . the inset shows finite - size scaling for @xmath107 for three values of @xmath6 : @xmath113 ( circles ) , @xmath114 ( triangles ) and @xmath115 ( rhombs ) . ] expectation values of the functions eq . ( [ order_parameters ] ) in the hf ground state are shown in fig . [ fig_order_parameters ] . both phase transition points @xmath85 and @xmath86 are clearly seen from this plot . all functions change continuously across the second - order critical point @xmath85 and jump at the first - order transition point @xmath86 . except in the columnar phase the values of @xmath107 are everywhere exactly twice larger than those of @xmath108 , which is an indication of the unbroken four - fold rotational symmetry of the lattice in these regions . in the columnar phase , on the other hand , this symmetry is broken and the above relation does not hold . while in the nel and columnar phases nonlocal terms in eq . ( [ order_parameters ] ) are important , in the paramagnetic state the only contribution to either expectation value comes from isolated plaquettes ( local terms in eq . ( [ order_parameters ] ) ) , or superplaquettes . this observation is consistent with properties of the ground state in the non - magnetic phase , discussed earlier in this section . as mentioned in the introduction , any choice of degree of freedom breaks explicitly the lattice translational symmetry @xmath9 , with the result that links in the lattice become inequivalent . indeed , the functions of eq . ( [ order_parameters ] ) , defined on the links , have non - zero values even in the af phase at @xmath64 . however , this effect vanishes in the thermodynamic limit ( i.e. , as the size of the degree of freedom is increased ) . the finite - size scaling for @xmath116 is presented in the inset to fig . [ fig_order_parameters ] . the three extrapolated values : @xmath117 , @xmath118 and @xmath119 , suggest that the `` link - wise '' translational invariance is restored in the thermodynamic limit in the nel and columnar phases , but not in the plaquette crystal phase . moreover , the value of the jump @xmath120 , extrapolated to the thermodynamic limit ( see inset to fig . [ fig_gse_d2_2x2_4x4 ] ) remains finite : @xmath121 . in other words , these results imply that the critical point @xmath85 corresponds to the usual landau second - order phase transition . until now we have considered only ground - state properties of the model eq . ( [ hamiltonian ] ) . however , low - lying excited states are also of considerable interest . in particular , the paramagnetic phase is known to have gapped excitations , while nel and columnar phases exhibit goldstone modes . thus , the phase transition points @xmath85 and @xmath85 must be accompanied by the opening of a gap in the excitation spectrum : the former in a continuous and the latter in a discontinuous fashion . in appendix b we present a particular method to obtain the collective spectrum of the system . the main idea of this approximation is borrowed from the bogoliubov - fetter theory of superfluidity @xcite . namely , assume that on each plaquette the majority of schwinger bosons form a condensate in an appropriately chosen lowest energy state and neglect fluctuations in the number of condensed particles . we note , however , that due to the schwinger boson constraint , this quantity has the meaning of a probability to find a given plaquette in the lowest energy hf state , rather than the number of particles . nevertheless , we will call it the condensate fraction @xmath122 , which , in principle , should be determined self - consistently , and is a measure of the applicability of the entire approximation : it should satisfy the inequality @xmath123 . once the condensation part is separated from @xmath62 , what remains describes fluctuation corrections to the hf ground state . these fluctuations have rather strong effects near the phase transition points , leading to the modification of @xmath85 to a @xmath124-point and its considerable shift . the value of @xmath86 also changes , but much less significantly . these facts imply that our approximation breaks down near the phase transition points . indeed , in appendix b it is shown that close to the transition , the condensate is strongly suppressed . however , deep in each phase @xmath125 , thus allowing us to draw conclusions about general properties of the collective spectrum . the complete summary of the results is given in appendix b , here we only present the most interesting one : the gap in the excitation spectrum as a function of @xmath6 . although we focus only on case ( a ) ( @xmath14 plaquette ) , the superplaquette degree of freedom can be considered in a similar manner . the gap always occurs in momentum space at @xmath126 , which reflects translational invariance of the _ plaquette _ lattice . below we focus only on this point in the plaquette brillouin zone . in fact , there are @xmath127 collective branches and only some of them become gapless in the phases with spontaneously broken @xmath8 . however , in the paramagnetic phase all branches develop a gap . in fig . [ fig_gaps ] we show the energy gap @xmath128 for the two lowest excitation branches in a system of @xmath129 plaquettes , which approximates well the thermodynamic limit . the main panel shows results of the self - consistent solution of bogoliubov s equations . the inset compares it with the solution of the time - dependent gross - pitaeveskii equation ( [ gp_equation ] ) , which corresponds to the weak - coupling approximation . in the nel and columnar phases there are two spin wave - type goldstone modes , both of which acquire a gap in the paramagnetic phase , at @xmath85 and @xmath86 . however , as it follows from fig . [ fig_gaps ] , positions of these points change from their hf values to : @xmath130 and @xmath131 . the critical point @xmath85 was obtained by extrapolation of the staggered magnetization curve ( fig . [ fig_mag_sf ] ) to zero , while the first - order point @xmath86 by extrapolating the two gse curves in fig . [ fig_gse_sf ] until intersection . the single - plaquette physical picture , discussed previously in connection with the paramagnetic phase , remains valid , e.g. the condensation occurs again in the plaquette state @xmath132 . using this observation and symmetries of the matrix elements @xmath133 , one can rigorously show the existence of a gap in the non - magnetic region . in fact , we can say , that it is a property of our hmf approximation , rather than a numerical evidence . finally , we comment on the results for cases ( c)-(e ) , which , contrary to the configurations considered before , explicitly break lattice rotational symmetry , see figs . [ displ_plaquette_covering]-[cross_covering ] . the corresponding gses are shown in fig . [ fig_gse_dp_dd_cr ] . in contrast to the previously considered scenarios , these cases give _ qualitatively _ wrong phase diagrams . indeed , if we cover the lattice with displaced plaquettes or crosses , there exists no classical spin configuration , which gives the long - range nel order . on the other hand , such configuration exists for the columnar state . for the displaced plaquette covering the low-@xmath1 phase @xmath99 is an @xmath8 singlet , and spatially is a set of non - interacting plaquettes ( notice the coincidence of @xmath14plaquette energies in the paramagnetic phases of figs . [ fig_gse_2x2_4x4 ] and [ fig_gse_dp_dd_cr ] ) . thus , the phase transition to the columnar state is of the first - order . for the cross covering , on the other hand , @xmath8 is explicitly broken for all values of @xmath1 , but since the columnar phase partially restores the lattice translational invariance , it is again separated from the non - magnetic state by a first - order phase transition point . the two dimer configurations , case ( d ) , are complementary to each other in the sense that one of them has only the classical nel state and another only columnar phase . it follows that these configurations can have only one second - order critical point at which @xmath8 is restored and other symmetries remain broken . as a result , one obtains the phase diagram , shown in the lower panel of fig . [ fig_gse_dp_dd_cr ] , which is invariant under reflection in the plane @xmath134 . these observations imply that the coarse graining prescriptions ( c)-(e ) are probably a bad starting point for any approximation scheme . in this section we would like to put our main results in perspective by making several summarizing remarks . it should be emphasized , that the discussion below is based on our finite - size scaling results . first of all , from our analysis it follows that dimer ( bond ) order is always unfavorable in the non - magnetic phase . notice that even when the plaquette coverings were considered such an order did not occur , although spontaneous dimerization was not explicitly prohibited . instead , the quantum paramagnetic phase prefers to preserve the lattice rotational symmetry , which makes the phase transition separating it from the nel phase fit perfectly well within the ginzburg - landau - wilson paradigm . the data presented for the staggered magnetization , fig . [ fig_mag_2x2_4x4 ] , and symmetry - breaking observables , fig . [ fig_order_parameters ] , indicate that the symmetry group of the nel state is a subgroup of the symmetry group in the paramagnetic phase , as both phases break @xmath9 and preserve @xmath10 , but the latter also preserves @xmath8 . on the contrary , there is no such group - subgroup relation between the paramagnetic and columnar antiferromagnetic phases . consequently , the transition between these two states is first order . these observations are summarized in fig . [ fig_phase_diagram ] . indeed , starting from the known symmetry in the nel state and assuming validity of the landau theory , one can unambiguously rule out dimerized structures in the paramagnetic phase , since they break lattice rotational symmetry . therefore , our results do not favor the scenario of deconfined quantum criticality , advocated in refs . . as already discussed in the previous section , our method explicitly breaks a particular lattice translational symmetry : the ground state in eq . ( [ hf_wavefunction ] ) at @xmath64 ( af phase ) , is not invariant under a @xmath135 $ ] lattice translation . one way to cure this problem is to consider variational wavefunctions of the `` resonating plaquette '' type : @xmath136 which restore that symmetry , with @xmath137 the translation operator along the @xmath135 $ ] direction in the lattice . this state describes two resonating plaquette configurations , shifted with respect to each other along [ 11 ] . results of calculations using this wavefunction for systems up to @xmath138 spins indicate that for @xmath139 the ground state has long - range nel order and is paramagnetic for @xmath100 . the intermediate phase has a plaquette crystal order , but with partially restored translational invariance . the phase transition at @xmath85 is still of the second - order , which is not surprising , as it can be described solely in terms of the @xmath8 order parameter . however , based on these system sizes , we can not definitively conclude whether this phase transition remains of the second - order or becomes weakly first - order in the thermodynamic limit . next , we observe that despite profound differences between the 2d and 1d equivalent @xmath0-@xmath1 models , their non - magnetic phases present some similarities . the one - dimensional model is known to be quasi - exactly solvable @xcite at the point @xmath140 and exhibits a paramagnetic ground state with short - range correlations for @xmath1 above the critical value @xcite @xmath141 ( however , due to the peculiar physics in one dimension , the critical point @xmath142 is an essential singularity and , therefore , not obviously accessible for the hmf approximation of the type presented here ) . in this non - magnetic region the ground state is doubly degenerate , corresponding to two possible coverings of a 1d lattice with dimers , in accordance with the lieb - schultz - mattis theorem @xcite . unfortunately , a finite - size scaling calculation for the gap between the lowest and first excited energy levels , based on exact diagonalization of the @xmath14 and @xmath15 clusters with periodic boundary conditions , does not provide a definitive answer to the question on whether the ground state of the 2d @xmath0-@xmath1 model becomes degenerate in the region @xmath143 . this is indeed what one would expect on the basis of a generalization of the lieb - schultz - mattis theorem to higher space dimensions ( see , e.g. ref . ) . at the hf level , it is true that different _ plaquette _ coverings of the lattice have the same energy ( simply because each plaquette is in its singlet ground state ) . however , the total number of such configurations grows _ sub - exponentially _ @xmath23 , which should be contrasted with the dimer covering problem , where this number is known @xcite to be exponentially large . based on this distinction , one may speculate that if our plaquette picture is valid , there are not enough different plaquette configurations for the paramagnetic phase to become a spin liquid ( i.e. , a _ resonating plaquette state _ ) . this statement , certainly , requires a separate investigation . finally , we emphasize that the main goal of this work is to investigate the fundamental symmetries of the phases exhibited by the @xmath0-@xmath1 model . although the energies presented for the @xmath14 and @xmath15 plaquette cases are different from those , obtained by more sophisticated numerical methods , they can be systematically improved by considering correlated trial wavefunctions or by using more complex methods , which build upon the results reported here . however , we expect that symmetry - wise our conclusions will remain unchanged . one of such methods suited for computing the phase diagram of the model ( [ hamiltonian ] ) , which received significantly less attention , amounts to applying the wilson renormalization group procedure @xcite and the density matrix renormalization group ( dmrg ) method @xcite . application of the latter faces serious difficulties in 2d ( see ref . for a related discussion ) and requires a mapping of the 2d system onto a chain , which introduces a certain bias to the final results @xcite . while formulation of a practical and efficient dmrg approach to 2d systems is yet to be developed , we note that our results may provide a useful and rational initial input for such an algorithm . in summary , we analyzed the phase diagram of the 2d @xmath0-@xmath1 model on a square lattice , focusing on symmetries of the various phases . we showed that in this model the hierarchical language @xcite is defined by identifying the _ plaquette as a relevant degree of freedom_. using an unbiased and manifestly symmetry - conserving mean - field approach we compared it with several other possible candidates : dimer and cross degrees of freedom , as well as different ways to cover a 2d lattice with plaquettes . our results indicate that the plaquette ( and superplaquette ) covering , which preserves the lattice rotational symmetry has the best energy among considered coarse graining scenarios and it is the only one to reproduce known facts , such as the intermediate phase with gapped excitations , concerning the phase diagram of the model , while other configurations fail to exhibit all quantum phase transition points . superplaquette degree of freedom . each spin carries two indices : a @xmath14 plaquette number and a coordinate within this plaquette . ] consistent with the previous work on the subject , we found the quantum paramagnetic phase in the interval @xmath144 . main controversies revolve around the nature of this intermediate non - magnetic phase and of the quantum critical point separating it from the nel ordered state . we found that the paramagnetic phase is a _ plaquette crystal _ , preserving both lattice and spin rotational symmetries . extrapolation of our numerical results to the thermodynamic limit suggests that the ginzburg - landau - wilson paradigm of phase transitions is perfectly applicable in this case . indeed , within the hmf , there is a group - subgroup relation between symmetries of the non - magnetic and nel phases , which are thus separated by a second - order phase transition . on the contrary , such relation does not exist between the plaquette crystal and columnar antiferromagnetic phases , so the corresponding transition is of the first order . our plaquette crystal is quite different from the usually proposed dimerized ( bond - ordered ) phases in this non - magnetic region . we also proposed a way to include fluctuations around the hf ground state and showed that properties of the collective excitation spectrum are consistent with the overall picture of spontaneously broken symmetries . in particular , it was demonstrated that the quantum paramagnetic state is characterized by a finite gap in the excitation spectrum , which vanishes in the nel and columnar phases , producing a doubly degenerate goldstone mode . although currently there exists no known material whose ground state realizes the paramagnetic phase of the @xmath0-@xmath1 model , in the future momentum - resolved measurements , such as neutron diffraction methods , can be used to identify the plaquette crystal phase of the type proposed here . its experimental signature will be the unbroken four - fold lattice rotational symmetry on both sides of the second - order phase transition at the critical point @xmath85 . we acknowledge fruitful discussions with c. d. batista , c. esebbag and e. fradkin . this work was supported in part by the spanish mec under grant no . fis2006 - 12783-c03 - 01 . here we present details of the hf calculation which uses the @xmath15 superplaquette , shown in fig . [ fig_sp ] , as an elementary degree of freedom . it turns out that the full angular momentum basis is inconvenient , thus we use the @xmath14 plaquette product states in order to perform the mean - field calculations . each spin is characterized by two indices : the plaquette number @xmath145 and an index @xmath146 , which specifies a vertex in the plaquette . the singlet sector of the superplaquette hilbert space is spanned by the states : @xmath147 where prime indicates the constraint @xmath148 . using these states , we can write down matrix elements like @xmath149 in the compact form : @xmath150 with the symmetric matrices @xmath151 defined as : @xmath152 the hamiltonian of a single superplaquette consists of two parts : a diagonal one , involving only @xmath14 plaquette contributions and a non - diagonal part , which accounts for the plaquette interactions . the former is written down straightforwardly as a matrix : @xmath153 where @xmath154 is the plaquette self - energy , eq . ( [ plaquette_se ] ) . the non - diagonal part has the operator form : @xmath155 + \nonumber \\ & + j_2\bigl[{{\bm s}}_{12}{{\bm s}}_{23}+{{\bm s}}_{14}{{\bm s}}_{21}+{{\bm s}}_{23}{{\bm s}}_{42}+{{\bm s}}_{41}{{\bm s}}_{24 } + { { \bm s}}_{34}{{\bm s}}_{41}+{{\bm s}}_{32}{{\bm s}}_{43}+ \label{se_sp_nd } \\ & \quad\qquad+{{\bm s}}_{31}{{\bm s}}_{14}+{{\bm s}}_{13}{{\bm s}}_{32}+{{\bm s}}_{14}{{\bm s}}_{41}+ { { \bm s}}_{32}{{\bm s}}_{23}\bigr ] . \nonumber \end{aligned}\ ] ] let us now proceed with interaction terms in the @xmath0-@xmath1 hamiltonian ( [ hamiltonian ] ) . each superplaquette has 4 nearest and 4 next - nearest neighbors . within each neighboring superplaquette we enumerate @xmath14 plaquettes by the indices 5 , 6 , 7 , 8 , so that @xmath156 , @xmath157 , @xmath158 and @xmath159 . enumeration of vertices within a plaquette stays the same . in this manner we have the symmetrized nearest neighbor @xmath160 + \nonumber \\ & + \frac{j_2}{4}\bigl[{{\bm s}}_{11}({{\bm s}}_{74}+{{\bm s}}_{64})+{{\bm s}}_{12}({{\bm s}}_{73}+{{\bm s}}_{83})+ { { \bm s}}_{13}({{\bm s}}_{62}+{{\bm s}}_{82})+{{\bm s}}_{21}({{\bm s}}_{74}+{{\bm s}}_{84 } ) \label{sp_nn_int } \\ & \quad\qquad+{{\bm s}}_{22}({{\bm s}}_{53}+{{\bm s}}_{83})+{{\bm s}}_{24}({{\bm s}}_{51}+{{\bm s}}_{71})+ { { \bm s}}_{31}({{\bm s}}_{84}+{{\bm s}}_{64})+{{\bm s}}_{33}({{\bm s}}_{82}+{{\bm s}}_{52})+\nonumber \\ & \quad\qquad+{{\bm s}}_{34}({{\bm s}}_{51}+{{\bm s}}_{61})+ { { \bm s}}_{42}({{\bm s}}_{73}+{{\bm s}}_{53})+{{\bm s}}_{43}({{\bm s}}_{62}+{{\bm s}}_{52})+{{\bm s}}_{44}({{\bm s}}_{71}+ { { \bm s}}_{61})\bigr ] , \nonumber \end{aligned}\ ] ] and next - nearest neighbor @xmath161 \label{sp_nnn_int}\ ] ] superplaquette interactions . using eq . ( [ average ] ) , one can easily construct matrix elements of the operators ( [ se_sp_nd])-([sp_nnn_int ] ) , which are required to obtain the hf equation of the type ( [ hmf_equation ] ) . having computed the single - superplaquette ground state wavefunction @xmath162 , we can use it to determine the spin polarizations : @xmath163 where @xmath164 is given by eq . ( [ spin_matr_el ] ) . in this appendix we extend the analysis of sec . iii by considering fluctuations around the hf ground state . while not unique , a natural way to achieve this goal is to perform a superfluid - type mean - field approximation . as a result one can obtain the collective spectrum and corrections to the gse and magnetization . of primary interest is , of course , the energy gap in the excitation spectrum . the structure of the superfluid mean - field is similar to the fetter - bogoliubov approach to inhomogeneous bose liquids @xcite . although we shall present results only for the @xmath14 plaquette degree of freedom , it can equally be applied to the @xmath15 superplaquette case . let us return to the original hamiltonian ( [ ham_second_quant ] ) and explicitly separate out the condensate mode in the operators @xmath62 : @xmath165 the condensation will occur in a certain superposition of the single - plaquette states . the real - valued multiplet @xmath166 plays the role of a condensate wavefunction ( cwf ) @xcite . here it is chosen to be spatially homogeneous , but inhomogeneous phases can also be included . the cwf is normalized to the condensate fraction : @xmath167 the non - condensate bosonic operators @xmath168 describe fluctuation corrections to the hf solution . if they are neglected , we naturally return to the results of sec . it is important to observe , however , that the hf ground state corresponds to the bose condensation on _ each _ lattice site , not only in the @xmath126 mode . the superfluid mean - field approximation amounts to enforcing the schwinger boson constraint on average : @xmath169 neglecting fluctuations in the condensate channel , and retaining only terms quadratic in @xmath170 in the hamiltonian ( [ ham_second_quant ] ) : @xmath171+\nonumber \\ & + \sum_{i , a}(\epsilon_a-\mu)\beta_{ia}^\dag\beta_{ia}+4\sum_{i,\sigma}\bigl ( h_{\rm int}^\sigma\bigr)^{a_1^\prime a_2^\prime}_{a_1a_2}g_{a_2^\prime}g_{a_2 } \beta_{ia_1^\prime}^\dag\beta_{ia_1}+ \nonumber \\ & + \sum_{\sigma,\langle ij\rangle_\sigma}\bigl(h_{\rm int}^\sigma\bigr)^{a_1^ \prime a_2^\prime}_{a_1a_2}\bigl[g_{a_1^\prime}g_{a_2^\prime}(\beta_{ia_1}^ \dag\beta_{ja_2}^\dag+\beta_{ia_1}\beta_{ja_2})+ \nonumber \\ & \qquad\qquad\qquad\qquad+2g_{a_1}g_{a_2^\prime}\beta_{ia_1^\prime}^\dag \beta_{ja_2}\bigr ] , \label{bog_hamiltonian}\end{aligned}\ ] ] where we abbreviated @xmath172 and matrix elements of @xmath173 are given by eq . ( [ interaction ] ) . the cwf @xmath166 is determined by the gross - pitaevskii equation , similar to the hf equation ( [ hmf_equation ] ) : @xmath174 which defines the chemical potential @xmath175 and guarantees the disappearance of linear terms in @xmath170 from the hamiltonian of eq . ( [ bog_hamiltonian ] ) . it is clear that @xmath176 and @xmath177 . in other words , ( [ gp_equation ] ) reproduces the results of sec . iii , if @xmath122 is forced to be unity . naturally , the first line in eq . ( [ bog_hamiltonian ] ) coincides in this limit ( up to the chemical potential ) with the expression ( [ hmf_energy ] ) for @xmath178 . quadratic terms in eq . ( [ bog_hamiltonian ] ) represent fluctuation corrections to the hf results and constitute the focus of our analysis below . the next step is to transform the quadratic part ( @xmath179 ) of the hamiltonian in eq . ( [ bog_hamiltonian ] ) into momentum space : @xmath180 + \nonumber \\ & + 4g_{a_2^\prime}g_{a_2}\beta_{{{\bm k}}a_1^\prime}^\dag\beta_{{{\bm k}}a_1}\bigr\ } , \label{bog_hamiltonian_k}\end{aligned}\ ] ] where @xmath181 and @xmath182 is defined within the plaquette brillouin zone ( i.e. , there are @xmath183 @xmath182- states ) . this hamiltonian can be diagonalized by the bogoliubov s transformation : @xmath184 to a new set of bosonic operators @xmath185 , which represent quasiparticle excitations and annihilate the new ground state : @xmath186 . of course , only positive quasiparticle energies , labeled by @xmath187 , have physical meaning , however , in order to obtain closure relations for the wavefunction @xmath188 ( which is , obviously , even in @xmath182 ) , we need to include zero - energy eigenvectors as well @xcite . this completeness relation has the form , valid for _ all _ wavevectors : @xmath189 the amplitudes @xmath190 and @xmath191 are determined from bogoliubov s equations : @xmath192 where we have introduced symmetric matrices : @xmath193 it follows from eq . ( [ bog_eqs ] ) that at each @xmath182 the quasiparticle amplitudes obey the orthogonality conditions : @xmath194 for any value of @xmath182 bogoliubov s equations ( [ bog_eqs ] ) always have at least two zero eigenvalues , which correspond to the zero - norm eigenvector @xmath195 . this means that our case differs fundamentally from the canonical superfluid bose gas : instead of having a macroscopic number of particle in one particular energy state , we obtain a macroscopic number ( equal to @xmath183 ) of condensation modes , each containing less than one boson . the quasiparticle energy equals @xmath196 and the gse , condensate fraction and spin polarization are expressed in terms of @xmath197 and @xmath198 as : @xmath199 . \nonumber\end{aligned}\ ] ] in this expression @xmath182-summations are extended over the plaquette brillouin zone and @xmath187-summations over positive eigenvalues of eq . ( [ bog_eqs ] ) , as indicated by the primes . the condensate fraction @xmath122 should , in principle , be determined self - consistently . the approximation is reasonable if @xmath200 . however , close to the phase transition points this is not true , since fluctuations are very large in their vicinity . but deeply in each phase the approximation works reasonably well , because @xmath122 turns out to be of the order of @xmath201 . results of the numerical solution of eqs . ( [ gp_equation ] ) , ( [ bog_eqs ] ) and ( [ gse_cond_frac_mag ] ) for the symmetric covering of the lattice with @xmath14 plaquettes are shown in figs . [ fig_gse_sf]-[fig_omega ] . the system size is @xmath129 plaquettes and periodic boundary conditions are assumed . main panels correspond to the self - consistent solution and their insets give results after the first iteration , which is equivalent to solving the time - dependent gross - pitaevskii equation @xcite . due to bad convergence close to the transition points ( see , for instance , fig . [ fig_gse_sf ] ) the values of @xmath85 and @xmath86 were determined by extrapolation : @xmath130 and @xmath202 . the large shift of @xmath85 compared to the hf value is due to fluctuations in the @xmath170-channel , which renders this point to be a @xmath124-point , reduces the nominal value of the magnetization in the nel phase down to @xmath203 ( fig . [ fig_mag_sf ] ) , and causes a great suppression of the condensate , as shown in fig . [ fig_cf ] . technically , one may show that its very existence reflects the nature of the ground state in the paramagnetic phase . indeed , introducing linear combinations of the amplitudes @xmath204 and @xmath205 : @xmath206 and @xmath207 , bogoliubov s eq . ( [ bog_eqs ] ) can be rewritten in the form : @xmath208 in the non - magnetic phase the condensation occurs in the lowest plaquette state @xmath132 : @xmath209 and the chemical potential coincides with its energy : @xmath210 . moreover , the matrix @xmath211 vanishes . writing down the remaining matrices in ( [ u_matrices ] ) at @xmath126 , it is easy to see that there exists only one vector @xmath212 , which is annihilated by @xmath213 . outside the intermediate region this simple situation is not valid and there exist three eigenvectors @xmath212 , which correspond to @xmath214 . one of them is the condensate mode and should be discarded . the other two give doubly degenerate goldstone modes in the nel and columnar phases . here the self - consistent field , determined by @xmath166 , breakes the spin - rotational symmetry of the original hamiltonian . however , since the cwf @xmath166 belongs to the @xmath215 subspace , the generator @xmath216 remains an integral of motion . thus , there should be two goldstone modes associated with rotations around the @xmath77 and @xmath12 axes @xcite . our approximation correctly describes the excitation spectrum only at small @xmath182 . however , this is more than enough to observe that the collective modes are of the spin - wave type in the nel and columnar phases , while in the paramagnetic phase the excitation band is parabolic . these conclusions are summarized in fig . [ fig_omega ] , where we show the lowest branch @xmath217 along two directions @xmath218 $ ] and @xmath219 $ ] for three values of @xmath6 , chosen in different phases . 99 p. carretta , r. melzi , n. papinutto and p. millet , phys . lett . * 88 * , 047601 ( 2002 ) ; a. bombardi , j. rodriguez - carvajal , s. di matteo , f. de bergevin , l. paolasini , p. carretta , p. millet and r. caciuffo , _ ibid . _ * 93 * , 027202 ( 2004 ) . p. carretta , n. papinutto , c. b. azzoni , m. c. mozzati , e. pavarini , s. gonthier and p. millet , phys * b66 * , 094420 ( 2002 ) . r. nath , a. a. tsirlin , h. rosner and c. geibel , arxiv:0803.3535v1 [ cond-mat.str-el ] . s. sachdev , _ quantum phase transitions _ ( cambridge university press , cambridge , 1999 ) . g. misguich and c. lhuillier in _ frustrated spin systems _ , edited by h. t. diep ( world scientific , singapore , 2004 ) . l. capriotti and s. sorella , phys . rev . lett . * 84 * , 3173 ( 2000 ) . l. capriotti , f. becca , a. parola and s. sorella , phys . * 87 * , 097201 ( 2001 ) . m. s. l. du croo de jongh , j. m. j. van leeuwen and w. van saarloos , phys . * b62 * , 14844 ( 2000 ) . r. darradi , o. derzhko , r. zinke , j. schulenburg , s. e. krueger and j. richter , phys . * b78 * , 214415 ( 2008 ) . r. r. p. singh , z. weihong , c. j. hamer and j. oitmaa , phys . * b60 * , 7278 ( 1999 ) . k. takano , y. kito , y. ono and k. sano , phys . 91 * , 197202 ( 2003 ) . v. kotov , j. oitmaa , o. sushkov and z. weihong , phys . rev . * b60 * , 14613 ( 1999 ) . v. lante and a. parola , phys . rev . * b73 * , 094427 ( 2006 ) . n. read and s. sachdev , phys . lett . * 66 * , 1773 ( 1991 ) . t. senthil , l. balents , s. sachdev , a. vishwanath and m. p. a. fisher , phys . rev . * b70 * , 144407 ( 2004 ) . l. d. landau , e. m. lifshitz and l. p. pitaevskii , _ statistical physics _ , parts . 1 and 2 ( butterworth - heinemann , new - york , 1999 ) ; k. g. wilson and j. kogut , phys . * 12 * , 75 ( 1974 ) . e. manousakis , rev . phys . * 63 * , 1 ( 1991 ) . that the translational symmetry is broken in the nel phase is easy to see , if one _ assumes _ broken spin @xmath8 symmetry . then , due to the principle of weakening of correlations , the staggered correlation function satisfies : @xmath220 which can be constant only if @xmath221 depends on @xmath57 . j. sirker , z. weihong , o. sushkov and j. oitmaa , phys . rev . * b73 * , 184420 ( 2006 ) . a. w. sandvik , phys . lett . * 98 * , 227202 ( 2007 ) . c. d. batista and s. a. trugman , phys . rev . lett . * 93 * , 217202 ( 2004 ) . g. ortiz and c. d. batista , phys . rev . * b67 * , 134301 ( 2003 ) ; in _ condensed matter theories _ , vol . 18 , m. de llano et al . ( ed . ) , nova science publishers , 2003 . m. van den bossche , f .- c . zhang and f. mila , eur . j. * b17 * , 367 ( 2000 ) . e. altman and a. auerbach , phys . rev . * b65 * , 104508 ( 2002 ) . a. r. edmonds , _ angular momentum in quantum mechanics _ ( princeton university press , princeton , 1957 ) . a. auerbach , _ interacting electrons and quantum magnetism _ ( springer - verlag , new york , 1994 ) . n. trivedi and d. m. ceperley , phys . rev . * b40 * , 2737 ( 1989 ) . a. l. fetter , ann . phys . * 70 * , 67 ( 1972 ) . c. k. majumdar and d. k. ghosh , j. math * 10 * , 1388 ( 1969 ) . k. okamoto and k. nomura , phys . lett . * a169 * , 433 ( 1992 ) . e. h. lieb , t. d. schultz and d. c. mattis , ann . * 16 * , 407 ( 1961 ) . m. b. hastings , phys . b69 * , 104431 ( 2004 ) . m. e. fisher , phys . rev . * 124 * , 1664 ( 1961 ) . f. kruger and s. scheidl , europhys . lett . * 74 * , 896 ( 2006 ) . s. r. white , phys . 77 * , 3633 ( 1996 ) . j. p. blaizot and g. ripka , _ quantum theory of finite systems _ ( the mit press , cambridge , 1986 ) .

let denote the closed orientable surface of genus . we construct exponentially many mapping class group orbits of pairs of simple closed curves which fill and intersect minimally , by showing that such orbits are in correspondence with the solutions of a certain permutation equation in the symmetric group . next , we demonstrate that minimally intersecting filling pairs are combinatorially optimal , in the sense that there are many simple closed curves intersecting the pair exactly once . we conclude by initiating the study of a topological morse function over the moduli space of riemann surfaces of genus , which , given a hyperbolic metric , outputs the length of the shortest , minimally intersecting filling pair for the metric . we completely characterize the global minima of , and using the exponentially many mapping class group orbits of minimally intersecting filling pairs that we construct in the first portion of the paper , we show that the number of such minima grow at least exponentially in .

1312.0913

let @xmath4 denote the orientable surface of genus @xmath1 with @xmath5 boundary components , and let @xmath6 denote the _ mapping class group _ of the closed surface @xmath7 , the group of all orientation preserving homeomorphisms of @xmath7 , modulo isotopy ( see farb - margalit @xcite for background ) . a pair of curves on @xmath0 ( shorthand for @xmath7 ) are said to _ fill _ the surface if the complement of their union is a disjoint union of topological disks . let @xmath8 denote the set of all isotopy classes of simple closed curves on @xmath0 . then @xmath9 acts coordinate - wise on the product @xmath10 , and this action descends to an action on the quotient @xmath11.\ ] ] our first result establishes the exponential growth of the @xmath9-orbits of @xmath12 which fill and intersect minimally : [ thm:1 ] there exists a function @xmath13 satisfying the following : let @xmath14 denote the number of @xmath9-orbits of minimally intersecting filling pairs on @xmath0 . then @xmath15 in the statement of theorem [ thm:1 ] , @xmath16 denotes the following growth - rate relation : @xmath17 we compare theorem [ thm:1 ] to theorem @xmath18 of anderson - parlier - pettet @xcite , which says that if @xmath19 is a filling set of simple closed curves on @xmath0 pairwise intersecting no more than @xmath20 times , then @xmath21 thus , when @xmath22 , and indeed we show in section @xmath23 that a minimally intersecting filling pair intersects @xmath24 times if and only if @xmath25 ( when @xmath26 , @xmath27 intersections is minimal ) . given a pair of filling curves @xmath28 , there are potentially many ways of defining a sort of combinatorial `` complexity '' for the pair . one such candidate is to count the number of simple closed curves intersecting the filling pair a small number of times . heuristically , one expects a minimally intersecting filling pair to only `` barely '' fill @xmath0 . therefore , there should be many simple closed curves intersecting the union @xmath29 only once . indeed , we show : [ thm:2 ] let @xmath28 be a filling pair on @xmath30 and define @xmath31 to be the number of simple closed curves intersecting @xmath32 only once . then @xmath33 , with equality if @xmath28 is minimally intersecting . let @xmath34 denote the _ moduli space _ of riemann surfaces of genus @xmath1 , which is identified with the space of all complete , finite volume hyperbolic metrics on @xmath0 up to isometry . the _ systole _ function @xmath35 outputs the length of the shortest closed geodesic for a given hyperbolic metric @xmath36 . @xmath37 has been studied extensively , ( see buser - sarnak @xcite ; parlier @xcite , @xcite ; and schaller @xcite,@xcite , @xcite ) in part due to the fact that it is a _ topological morse function _ @xcite , a generalization of the notion of a classical morse function to functions which are not necessarily smooth , and for which versions of the morse inequalities also hold . thus a careful analysis of the critical points of @xmath37 can yield naturally arising cellular decompositions of @xmath34 . motivated by this program , we initiate the study of another topological morse function @xmath2 , the _ filling pair _ function , which outputs the length of the shortest minimally intersecting filling pair with respect to a metric @xmath38 . here , the length of a pair of curves is simply the sum of the individual lengths . we show : [ thm:3 ] @xmath2 is proper and a topological morse function . for any @xmath39 , @xmath40 where @xmath41 \right)\ ] ] denotes the perimeter of a regular , right - angled @xmath42-gon . furthermore , define @xmath43 then @xmath44 is finite and grows at least exponentially in @xmath1 . if @xmath45 , the injectivity radius of @xmath3 is at least @xmath46 * organization of paper and acknowledgements . * in section @xmath23 , we prove theorem [ thm:1 ] ; in section @xmath47 , we prove theorem [ thm:2 ] ; and in section @xmath27 , we prove theorem [ thm:3 ] . the authors would like to thank yair minsky for carefully reading through a draft of this project and for his many helpful suggestions . the first author would also like to thank dan margalit , chris arettines , subhojoy gupta , babak modami , and jean sun for many enlightening conversations regarding this work . in this section we prove theorem [ thm:1 ] . we begin with some preliminary notions and notation . all curves will be simple and closed . let @xmath16 denote the free homotopy relation . let @xmath48 denote the standard euler characteristic . if @xmath49 are two curves on @xmath0 , we define the _ geometric intersection number _ of the pair @xmath28 , denoted @xmath50 , by @xmath51 where the minimum is taken over all curves @xmath52 freely homotopic to @xmath53 . if @xmath54 , we say that @xmath53 and @xmath55 are in _ minimal position_. [ lm:1 ] suppose @xmath28 is a pair of curves which fill @xmath56 . then @xmath57 assume @xmath49 are in minimal position . that @xmath28 fills @xmath0 implies that the complement @xmath58 is a disjoint union @xmath59 of topological disks . thus we can view the @xmath27-valent graph @xmath32 as the @xmath18-skeleton of a cellular decomposition of @xmath0 . the number of @xmath60-cells is @xmath50 , and @xmath27-valency of @xmath29 implies that the number of edges is @xmath61 . therefore @xmath62 @xmath63 then since @xmath64 , we obtain the desired result . we remark that the proof of lemma [ lm:1 ] implies that @xmath65 exactly when @xmath66 consists of a single disk . given a minimally intersecting filling pair @xmath67 , choose an orientation for @xmath53 and @xmath55 . label the sub - arcs of @xmath53 from @xmath68 , separated by the @xmath24 intersection points in accordance with the chosen orientation for @xmath53 ( and after choosing an initial arc for @xmath53 ) , and similarly for @xmath69 . once labeled , we call such a pair an _ oriented filling pair_. a filling pair without the extra structure of an orientation is an _ ordinary filling pair_. the action of @xmath9 on ordinary filling pairs lifts naturally to an action on the set of oriented filling pairs . given an oriented filling pair , we obtain an ordinary one by forgetting the orientation . then associated to the oriented pair is a permutation @xmath70 , the permutation group on @xmath71 symbols , which we define as follows . we will identify @xmath72 as the permutations of the ordered set @xmath73 cutting along @xmath29 produces an @xmath42-gon @xmath74 , whose edges correspond to elements of @xmath75 . pick an initial edge of @xmath76 and orient it clockwise . then @xmath77 is defined by @xmath78 note that @xmath77 must be an @xmath42-cycle , and corresponds to the permutation of @xmath75 obtained by rotating @xmath74 counter - clockwise by @xmath79 . let @xmath80 denote the standard ordered @xmath42-cycle , which in cycle notation is @xmath81 note also that @xmath82 sends an edge to itself , oriented in the opposite direction . let @xmath83 denote the permutation @xmath84 which corresponds moving forward by one sub - arc in each of @xmath49 . in what follows , in order to simplify the relevant permutation equations , we will suppress the reference to @xmath1 and @xmath28 when referring to a permutation ( e.g. , @xmath85 will be written as @xmath86 and @xmath77 as @xmath3 ) . we say that a permutation @xmath87 _ respects parity _ , or is _ parity respecting _ , if @xmath88 [ lm:2 ] let @xmath28 be a minimally intersecting filling pair on @xmath0 . then the permutation @xmath89 satisfies the equation @xmath90 conversely , if @xmath91 is a parity respecting @xmath42-cycle which solves the above equation , then @xmath3 defines a minimally intersecting filling pair on @xmath0 . let @xmath92 . then @xmath93 is the labeling of the edge of @xmath94 immediately following the edge labeled with the @xmath95 element of @xmath75 . the ordering of @xmath75 has been chosen so that @xmath96 is precisely the element of @xmath75 identified with the inverse arc to @xmath97 . thus @xmath98 sends @xmath99 to the next arc along whichever curve @xmath99 is apart of , with respect to the chosen orientation of that curve , which is equal to @xmath100 . conversely , let @xmath3 satisfy the above permutation equation , and suppose further that @xmath3 is an @xmath42-cycle and is parity respecting . then associated to @xmath3 is a clockwise labeling of the edges of an @xmath42-gon ( up to cyclic permutation ) from the set @xmath101 , such that edges alternate between receiving even and odd labels . this is simply the labeling obtained by choosing an initial edge of the polygon , and labeling edges in order from the initial edge in accordance with the permutation @xmath3 written in cycle notation . this produces a labeling of the edges by the elements of @xmath75 , which we in turn interpret as a gluing pattern ; each edge glues to its inverse . thus we obtain a closed surface @xmath102 , which is orientable because each edge is glued to one which is oppositely oriented . it remains to check that the euler characteristic of the surface is @xmath103 , and that the boundary of the polygon projects to a minimally intersecting filling pair . @xmath104 is equivalent to there being @xmath24 equivalence classes of vertices under the gluing . indeed , the permutation equation ensures that each equivalence class contains precisely @xmath27 vertices ; as seen in figure @xmath18 , the vertex @xmath105 separating the edges labeled @xmath106 and @xmath107 glues to the vertex separating the edges labeled @xmath108 and @xmath109 . and @xmath107 share a vertex in common along the boundary of the polygon @xmath94 , then @xmath108 and @xmath109 must also share a vertex in common , and these two vertices are glued together to obtain @xmath102 . iterating this argument shows that each gluing class contains @xmath27 vertices.,width=336 ] similarly , the vertex separating @xmath108 and @xmath109 is glued to the vertex separating @xmath110 and @xmath111 , which in turn glues to the vertex separating @xmath112 and @xmath113 . finally , this fourth vertex glues to @xmath105 and the cycle is complete . by the same analysis , the terminal vertex of @xmath106 is identified to the initial vertex of @xmath109 , and therefore the @xmath53-arcs are concatenated in the quotient to produce a single simple closed curve @xmath53 , and similarly for the @xmath55-arcs . we also note that if a permutation in @xmath72 is an @xmath42-cycle and parity respecting , it must send each even number to an odd number . henceforth , any permutation satisfying the requirements of lemma [ lm:2 ] ( being an @xmath42-cycle , parity respecting , and satisfying the permutation equation ) will be called a _ filling permutation_. filling permutations parameterize oriented filling pairs . therefore , in order to count ordinary filling pairs , it suffices to count filling permutations , and divide by the number of distinct orientations which a fixed ordinary filling pair admits . lemma [ lm:3 ] characterizes which filling permutations correspond to the same ordinary filling pair : [ lm:3 ] if @xmath114 are two minimally intersecting filling pairs on @xmath0 in the same @xmath9-orbit , then @xmath115 , modulo conjugation by permutations of the form @xmath116 where @xmath117 @xmath118 @xmath119 if @xmath120 and @xmath121 are _ oriented _ filling pairs in the same @xmath9 orbit , there is a homeomorphism of @xmath0 taking @xmath120 to @xmath121 , which lifts to a homeomorphism of the @xmath42-gon that is simplicial in the sense that it fixes the boundary and sends edges to edges and vertices to vertices . the only option for such a map is a dihedral symmetry of the regular euclidean @xmath42-gon ; since the map must also be orientation preserving , this leaves rotation as the only option . thus the filling permutation for @xmath121 , written in cycle notation , is obtained from that of @xmath120 by a cyclic rotation , and thus determines the same underlying permutation of @xmath72 . however , in general there will exist oriented filling pairs which are not in the same @xmath9 orbit , but whose underlying ordinary filling pairs are . nevertheless , the fiber of the map from oriented pairs to ordinary pairs is in @xmath122 correspondence with the set of distinct orientations with which one can equip an ordinary pair . fixing an ordinary filling pair , any choice of orientation can be obtained from any other by simply choosing a different initial arc along @xmath53 ( corresponding to conjugating by a power of @xmath123 ) or @xmath55 ( conjugating by @xmath124 ) , reversing the direction of @xmath53 ( conjugating by @xmath125 ) , and interchanging @xmath53 and @xmath55 ( conjugating by @xmath126 ) . we call permutations of the form @xmath127 _ twisting _ permutations . in light of lemma [ lm:3 ] , to obtain a lower bound on @xmath14 , it suffices to find a lower bound on the set of filling permutations , and divide by an upper bound on the number of twisting permutations . since @xmath123 and @xmath124 both have order @xmath24 and @xmath128 are involutions , there are at most @xmath129 such permutations . let @xmath28 be an oriented minimally intersecting filling pair on @xmath0 , and label the intersections @xmath130 in accordance with the chosen direction along @xmath53 . we first describe a way of `` extending '' @xmath28 to a minimally intersecting pair on @xmath131 ( see remark [ rk:1 ] for an explanation of why we do not extend to @xmath132 ) . fix some @xmath133 . to create a minimally intersecting filling pair on @xmath131 , we first excise a small disk around @xmath134 , yielding @xmath135 . doing so turns @xmath53 and @xmath55 into arcs @xmath136 which intersect @xmath137 times and which fill @xmath135 . to complete the construction , we will glue a decorated copy @xmath138 of @xmath139 to @xmath135 along its boundary component ; by `` decorated '' , we mean that @xmath138 will come equipped with a special pair of oriented arcs @xmath140 , such that if we concatenate @xmath141 with @xmath142 , and @xmath143 with @xmath5 , we obtain a minimally intersecting filling pair on @xmath131 . we now describe the arcs @xmath140 . consider the @xmath47-component multicurve @xmath144 on @xmath145 pictured in red in figure @xmath23 below . the curve @xmath146 intersects each component of @xmath144 ; let @xmath147 denote the right dehn twist of @xmath146 around @xmath144 . [ lm:4 ] @xmath148 fills @xmath145 , and the complement of the pair consists of @xmath27 disks . furthermore , when @xmath149 are drawn in minimal position , there exists an intersection point which is on the boundary of all of these complementary disks . intersects each component of the red pants decomposition.,width=336 ] the proof is completely constructive . @xmath147 and @xmath146 are pictured in figure @xmath47 below ( @xmath147 in red , @xmath146 in black ) . the representatives drawn in figure @xmath47 intersect @xmath150 times ; thus both curves are subdivided into @xmath150 arcs : @xmath151 for @xmath146 and @xmath152 for @xmath147 , ordered cyclically with respect to some chosen orientation of @xmath147 ( resp . @xmath146 ) . to examine the complement @xmath153 , simply cut along the union of these two curves to obtain the @xmath27 labeled polygons in figure @xmath27 . @xmath145 is obtained by gluing these @xmath27 polygons back together along the oriented edge labelings . the four green vertices in figure @xmath27 are all identified together in this gluing . to complete the construction , we puncture @xmath145 at this green point by removing a small disk centered there to obtain @xmath138 , a copy of @xmath139 ; define the arc @xmath142 to be what remains of @xmath146 , and what remains of @xmath147 is defined to be the second arc @xmath5 . we then glue @xmath138 to @xmath135 along the boundary of the disk we removed centered at the intersection point @xmath134 , matching the endpoints of @xmath142 with those of @xmath141 and the endpoints of @xmath5 with @xmath154 , obtaining two simple closed curves @xmath155 on @xmath156 . here , the subscript @xmath99 denotes the fact that we obtained this pair of curves by excising a disk around the intersection point @xmath134 . in black and @xmath147 in red . the green point is the desired intersection point which is on the boundary of all @xmath27 complementary regions.,width=336 ] we claim that @xmath155 fill @xmath156 . assuming this , they must intersect minimally because by construction , @xmath157 . indeed , because the green point of intersection was on the boundary of all four complementary regions of @xmath158 , the number of complementary regions of @xmath159 is equal to the number of complementary regions of @xmath160 . therefore @xmath161 is connected , and by the same basic euler characteristic argument used in the proof of lemma [ lm:1 ] the complement of @xmath162 must be simply - connected . produces four complementary simply connected regions . when glued back up according to the oriented edge labelings , the four green vertices are identified.,width=288 ] [ rk:1 ] it is unclear how to attempt such a construction by extending a minimally intersecting filling pair on @xmath0 to one on @xmath132 . our construction works by finding a pair of simple closed curves on @xmath163 intersecting @xmath150 times , and such that any other simple closed curve intersects their union at least twice . the key point here is that the arcs on @xmath139 fill in the _ arc and curve sense _ , meaning that no essential arc or simple closed curve is disjoint from their union . to extend a minimally intersecting filling pair on @xmath0 to one on @xmath132 in this fashion , we would need a pair of simple closed curves on @xmath164 intersecting @xmath27 times such that any other simple closed curve intersects their union at least twice . one way of guaranteeing this is to choose a pair of curves that are distance @xmath47 in the farey graph . however , if a curve is distance @xmath47 or more from the @xmath165 curve , it must intersect @xmath165 at least @xmath166 times ; since @xmath167 acts transitively on the vertex set of the farey graph , the same is true for any vertex . it remains to show that by choosing different intersection points @xmath168 to excise and glue a copy of @xmath138 along the resulting boundary component , and by choosing different initial filling pairs on @xmath0 and carrying out this construction , we obtain distinct filling permutations . after gluing a copy of @xmath138 to @xmath0 along a minimally intersecting filling pair @xmath28 in the above fashion , we obtain a filling pair @xmath169 on @xmath131 which contains a copy of the arcs @xmath170 as sub - arcs of @xmath52 and @xmath171 , respectively . that is , there exists an open sub - arc @xmath172 of @xmath52 whose closure has boundary consisting of two intersection points of @xmath173 , and such that : 1 . it is comprised of @xmath150 consecutive arcs @xmath174 , ( since @xmath172 is open , we define @xmath175 to be open , and the other @xmath27 arcs to be closed ) ; 2 . there is a sub - arc @xmath176 of @xmath171 , comprised of @xmath150 consecutive @xmath171-arcs , such that @xmath177 , and the combinatorics of these intersections completely coincide with those displayed between the arcs @xmath142 and @xmath5 in figure @xmath47 . that is to say , the @xmath178 intersection point counted along @xmath172 coincides with the @xmath95 intersection point along @xmath176 if and only if this is the case for the arcs @xmath142 and @xmath5 on @xmath138 . we call @xmath176 the _ companion arc _ to @xmath172 . the union @xmath179 is called a _ @xmath138-piece_. when convenient , we will alternate between thinking of a @xmath138-piece as being a surface with boundary , or as being contained within a particular filling pair by identifying it with the arcs @xmath180 . [ rk:2 ] given a @xmath138-piece within a filling pair @xmath28 with @xmath53 and @xmath55 arcs labeled @xmath181 and @xmath182 , respectively as in figure @xmath47 , note that none of the following @xmath27 points can coincide on the surface : 1 . the initial point of @xmath183 ; 2 . the initial point of @xmath184 ; 3 . the terminal point of @xmath185 ; 4 . the terminal point of @xmath186 ; this observation is trivial , but it will play an important role in the proof of lemma [ lm:6 ] . note that there exists a filling pair @xmath187 on @xmath164 intersecting @xmath188 times , and at the end of this section we will demonstrate a filling pair @xmath189 on @xmath190 intersecting @xmath191 times . if @xmath28 on @xmath0 is a minimally intersecting filling pair obtained from one of these two``seed '' pairs by successively gluing on copies of @xmath138 , whether @xmath28 is obtained from @xmath187 or @xmath189 in this fashion is completely determined by the parity of @xmath1 . in the case that @xmath192 is even , we call @xmath28 a @xmath193-_child _ of the _ ancestor _ @xmath189 , and similarly if @xmath194 is odd . the purpose of the next lemma is to show that @xmath138-pieces are symmetric with respect to interchanging the roles of the arcs @xmath142 and @xmath5 in figure @xmath47 : [ lm:5 ] there exists a homeomorphism of a @xmath138-piece interchanging the arcs @xmath142 and @xmath5 and preserving the labeling , in the sense that sub - arc @xmath195 along @xmath142 maps to sub - arc @xmath196 along @xmath5 . consider the polygonal decomposition in figure @xmath27 ; if , for each @xmath99 , one relabels @xmath197 as @xmath198 , and @xmath196 as @xmath199 , then the concatenation @xmath200 displays the exact intersection combinatorics with the arc @xmath201 , as the oriented arc @xmath142 does with @xmath5 . given @xmath202 a @xmath138-piece , define @xmath203 to be the interior acs of @xmath53 contained in @xmath202 ; that is , @xmath203 is comprised of the sub - arcs @xmath204 . define @xmath205 similarly . if @xmath206 are two @xmath138-pieces in @xmath28 , let @xmath207 denote the set of arcs @xmath208 for @xmath209 , let @xmath210 denote the terminal point of the sub - arc @xmath197 in @xmath142 in the @xmath138-piece @xmath202 , and define @xmath211 similarly . the purpose of the next lemma is to show that distinct @xmath138-pieces can not share any @xmath53-arcs . [ lm:6 ] let @xmath212 be two @xmath138-pieces on a @xmath99-child @xmath28 for any @xmath213 . then @xmath214 without loss of generality , @xmath202 starts before @xmath215 along @xmath53 , and terminates before @xmath215 along @xmath53 . we claim that it suffices to show @xmath216 indeed , the arcs of @xmath53 contained in a @xmath138-piece determine uniquely the arcs of @xmath55 comprising the companion arc , and therefore @xmath217 @xmath218 identify @xmath215 with the @xmath138-piece pictured in figure @xmath47 ; that is , the @xmath53-arcs within @xmath215 are labeled @xmath181 , and the @xmath55-arcs are labeled @xmath182 in that order , respectively . * case 1 : @xmath219 . * note that , in figure @xmath47 , the concatenation of arcs @xmath220 forms a loop . therefore , if @xmath219 , the concatenation @xmath221 must also be a loop . by inspection , it is not . * case 2 : @xmath222 . * as seen in figure @xmath47 , the concatenation @xmath223 is a loop . then if @xmath222 , the concatenation of @xmath185 with the next two sub - arcs of @xmath53 also forms a loop , but this would imply that @xmath224 coincides with the initial point of @xmath183 , contradicting remark [ rk:2 ] . * case 3 : @xmath225 . * note that the concatenation of arcs @xmath226 forms a loop . then as in the previous @xmath23 cases , if @xmath225 , the concatenation @xmath227 must be a loop , but as seen in figure @xmath47 , it is not . * case 4 : @xmath228 . * the concatenation @xmath229 of @xmath230 with the forward direction of the next @xmath53 sub - arc can not be a loop , by remark [ rk:2 ] . however , @xmath226 is a loop , and therefore @xmath231 implies that @xmath229 is a loop . therefore , if @xmath232 , @xmath233 , and thus @xmath234 . since , by lemma [ lm:5 ] , @xmath138-pieces are symmetric with respect to the arcs @xmath142 and @xmath5 , lemma [ lm:6 ] immediately implies the following corollary : [ cor:1 ] let @xmath212 be two @xmath138-pieces on a @xmath99-child @xmath28 for any @xmath213 . then @xmath235 if , on the filling pair @xmath28 , @xmath202 and @xmath215 share @xmath55-arcs , then by lemma [ lm:5 ] , on the pair @xmath236 , @xmath202 and @xmath215 share @xmath52-arcs ; apply lemma [ lm:6 ] . putting lemma [ lm:6 ] and corollary [ cor:1 ] together , we obtain the following important fact , which states that distinct @xmath138-pieces can not overlap : [ lm:7 ] let @xmath206 be @xmath138-pieces on a @xmath99-child @xmath28 for any @xmath213 . then @xmath237 immediate from [ lm:6 ] , corollary [ cor:1 ] and the definition of @xmath207 . henceforth , we assume that @xmath194 is odd . the even case is completely analogous , and will therefore follow from the construction of a minimally intersecting filling pair on @xmath190 at the end of this section . given an odd integer @xmath194 , we define the set @xmath238 of integer sequences by @xmath239 let @xmath240 denote the set of all oriented minimally intersecting filling pairs on @xmath0 . [ lm:8 ] @xmath241 we show the existence of an injection @xmath242 . that is , given any integer sequence @xmath243 in @xmath238 , we will show how to construct a corresponding oriented minimally intersecting filling pair in a unique way . start with the filling pair @xmath187 on @xmath164 , and attach a @xmath138-piece along the only intersection point to obtain an oriented minimally intersecting filling pair @xmath244 on @xmath245 . as in the original construction at the beginning of this section , label the @xmath246 intersection points @xmath247 by choosing an initial intersection point and counting with respect to the orientation along @xmath248 . note that @xmath249 ; then attach a @xmath138-piece by excising a small disk centered around @xmath250 and gluing to obtain an oriented pair @xmath251 on @xmath252 . relabel the @xmath253 intersection points as follows : any of the original intersection points @xmath134 coming from @xmath254 with @xmath255 are labeled @xmath134 . intersection points @xmath168 coming from @xmath254 with @xmath256 are relabeled as @xmath257 . the @xmath166 intersection points coming from the @xmath138-piece are labeled , in order along the arc @xmath142 , @xmath258 . this produces a labeling of all @xmath259 intersection points which respects the orientation of @xmath260 . we then glue on an additional @xmath138-piece by excising the intersection point @xmath261 , obtaining the oriented pair @xmath262 on @xmath263 , and we relabel the now @xmath264 intersection points as above . we repeat this process for each @xmath265 , each time excising the intersection point labeled @xmath266 and gluing on another @xmath138-piece . this process terminates with an oriented minimally intersecting filling pair in @xmath240 , which we define to be @xmath267 . [ rk:3 ] it is unclear how many @xmath138-pieces the filling pair @xmath267 contains . however , it necessarily contains at least one @xmath138-piece : the one associated to the last index @xmath268 . it remains to show that @xmath269 is injective . . then we claim that @xmath271 suppose not ; then by remark [ rk:3 ] , the filling pair @xmath270 contains @xmath23 distinct @xmath138-pieces @xmath272 , associated to @xmath268 and @xmath273 , respectively . without loss of generality , @xmath274 . by lemma [ lm:7 ] , @xmath138 and @xmath275 can not overlap . therefore , if we remove @xmath138 by cutting along the separating curve in @xmath0 associated to its boundary and gluing in a disk , we obtain an oriented minimally intersecting filling pair on @xmath276 , which still contains @xmath275 . then since @xmath277 is a monotonically increasing sequence , by the same argument the filling pair obtained by successively cutting along the @xmath138-pieces associated to smaller and smaller indices of @xmath277 will still contain @xmath275 . eventually , we conclude that the filling pair @xmath187 on @xmath164 contains @xmath275 , a contradiction . hence , @xmath278 . applying the exact same argument inductively , we have @xmath279 for each @xmath280 as well . hence @xmath269 is injective . [ cor:2 ] for @xmath1 odd , @xmath281 by lemma [ lm:8 ] , it suffices to show that @xmath282 the lower bound on @xmath14 follows by dividing by an upper bound on the number of twisting permutations . moreover , we bound @xmath283 by counting the number of length @xmath284 integer sequences @xmath285 such that @xmath286 , and then dividing by the order of the permutation group @xmath287 to account for the fact that the sequences in @xmath238 are monotonically increasing . the following lemma then implies the exponential growth of @xmath14 : [ lm:9 ] @xmath288 immediate from the definition of @xmath16 after factoring a @xmath289 out from the numerator . as mentioned above , the proof of the lower bound for @xmath1 even , @xmath290 is exactly analogous . hence to complete the proof of the lower bound , it suffices to construct a minimally intersecting filling pair on @xmath190 ; equivalently , we demonstrate the existence of a filling permutation in @xmath291 : @xmath292 recall that the proof of lemma [ lm:8 ] relied on the assumption that there are no @xmath138-pieces in this filling pair . in section @xmath293 below , we show that there are no filling pairs on @xmath163 whose complement is connected . thus , if there was a @xmath138-piece within the filling pair on @xmath190 associated to the permutation above in @xmath291 , we could excise a copy of @xmath138 from @xmath190 to obtain a filling pair on @xmath163 whose complement is connected , a contradiction . the upper bound from theorem [ thm:1 ] is obtained by bounding the number of filling permutations from above . [ lm:10 ] let @xmath294 be a filling permutation . then @xmath295 , where @xmath229 is a square root of @xmath296 . conversely , if @xmath229 is a square root of @xmath296 such that @xmath295 is a parity respecting @xmath42-cycle , then @xmath3 is a filling permutation . suppose @xmath3 is a filling permutation . then @xmath297 conversely , if @xmath229 is a square root of @xmath296 , @xmath298 therefore , @xmath14 is bounded above by the number of square roots of @xmath296 ; the upper bound of theorem [ thm:1 ] is obtained by identifying a collection of such square roots @xmath229 such that @xmath299 is not an @xmath42-cycle , and subtracting this from the total number of square roots . note that @xmath296 is a composition of disjoint transpositions : @xmath300 the square roots of @xmath296 are obtained by partitioning the above transpositions into pairs and `` interleaving '' the pairs . for instance , one such root is obtained by pairing off @xmath301 with @xmath302 , and in general for each of @xmath303 , pairing off the @xmath304 transposition above with the @xmath305 , and converting each such pair into a single @xmath27-cycle by interleaving the @xmath27 numbers : @xmath306 note that for a particular pair of transpositions , there are two @xmath27-cycles we can obtain by interleaving . furthermore , because @xmath307 is parity respecting and sends even numbers to even numbers , the only way @xmath299 can be a parity respecting @xmath42-cycle is if it sends even numbers to odd ones . therefore there are exactly @xmath308 square roots @xmath229 of @xmath296 such that @xmath299 is parity respecting . we remark that the fact that @xmath308 is an upper bound for @xmath14 is immediate without the help of lemma [ lm:10 ] ( label the intersection points from @xmath18 to @xmath24 according to their order along @xmath53 and comparing this labeling to an analogous one obtained by ordering along @xmath55 produces a permutation in @xmath309 ; specifying the sign of each intersection accounts for the factor of @xmath310 ) . however , using lemma [ lm:10 ] and the structure of the square roots of @xmath296 , we can obtain strictly smaller lower bounds . in particular , consider the following @xmath311 permutations : pair the transposition @xmath301 with an arbitrary transposition @xmath312 of even numbers in the disjoint cycle notation for @xmath296 ; there are @xmath313 ways of doing this . if @xmath229 is any square root of @xmath296 with @xmath314 as one of its @xmath27-cycles , then @xmath299 maps @xmath18 to @xmath315 ( or @xmath71 if @xmath316 ) , the number representing the inverse direction of the arc corresponding to @xmath99 . note that by inspection of @xmath296 , @xmath317 and therefore we are free to pair the transposition containing @xmath318 with the transposition containing @xmath319 , and we choose the interleaving of these transpositions so that the resulting @xmath27 cycle sends @xmath315 to @xmath319 . then we can obtain a parity respecting square root @xmath229 by pairing off the remaining transpositions in any way , evens with odds . however , no square root @xmath229 obtained in this fashion has the property that @xmath299 is an @xmath42-cycle , because by construction , @xmath320 therefore , the number of admissible square roots is bounded above by @xmath321 note that the number of twisting permutations is bounded below by @xmath24 , since @xmath123 has order @xmath24 . this completes the proof of the upper bound of theorem [ thm:1 ] . @xmath322 in this section , we show that there are no filling pairs intersecting @xmath47 times on @xmath163 . for a filling pair with @xmath27 intersections on @xmath163 , see page @xmath323 of farb - margalit @xcite . [ thm:4 ] suppose @xmath28 is a filling pair on @xmath163 . then @xmath324 . suppose @xmath325 . we can schematically represent @xmath32 as a partially directed graph @xmath326 with @xmath150 vertices , @xmath47 undirected edges and @xmath150 directed edges as follows . to construct @xmath327 , begin with two disjoint @xmath47-cycles , one which is oriented clockwise and the other counter - clockwise . label the vertices on each cycle from @xmath18 to @xmath47 in cyclic order . the two directed @xmath47-cycles represent @xmath53 on each of its two sides , and the three intersection points between @xmath53 and @xmath55 correspond to the @xmath47 vertices on each cycle . the remaining @xmath47 undirected edges of @xmath327 correspond to the @xmath47 arcs of @xmath55 . if @xmath67 fill , their complement is a single disk , and this corresponds to the existence of a certain kind of eulerian cycle . concretely , there is a cycle @xmath328 on @xmath327 beginning with the oriented edge on the clockwise cycle whose initial point is vertex @xmath18 , satisfying the following : 1 . the edges along @xmath328 alternate between being directed and undirected ; 2 . every directed edge appears along @xmath328 exactly once ; 3 . every undirected edge appears along @xmath328 exactly once in each direction . as demonstrated below in figure @xmath166 , no undirected edge can connect vertices on the same @xmath47-cycle , or else @xmath53 and @xmath55 can not be in minimal position . therefore , without loss of generality we can assume that the first undirected edge connects vertex @xmath18 on the clockwise cycle to vertex @xmath47 on the counter - clockwise cycle . this will then determine @xmath327 , because vertex @xmath47 on the clockwise cycle can not then be connected to vertex @xmath18 on the counter - clockwise cycle , or else @xmath55 will close up without ever passing through vertex @xmath23 . therefore , @xmath327 must be as pictured in figure @xmath150 . note then that @xmath327 can not satisfy the eulerian condition described above ; starting at vertex @xmath18 and traveling along the clockwise cycle , one arrives at vertex @xmath23 . then the undirected edge incident to vertex @xmath23 on the clockwise cycle is connected to vertex @xmath18 on the counter - clockwise cycle at its other end . then we must travel counter - clockwise to vertex @xmath47 on the counter - clockwise cycle , and finally the undirected edge at this vertex connects back to vertex @xmath18 on the clockwise cycle . therefore , no single cycle can satisfy the three necessary requirements . above in the complement of @xmath29.,width=288 ] can not satisfy the eulerian condition described above.,width=288 ] in this section , we consider a potential way of measuring the combinatorial efficiency of a filling pair . namely , given a filling pair @xmath28 , let @xmath329 be the number of simple closed curves ( up to isotopy ) which intersect the union @xmath32 no more than @xmath99 times . we expect for an `` optimal '' filling pair to have large values of @xmath330 for small values of @xmath99 ; in other words , there should be many simple closed curves which only intersect the union a small number of times . specifically , we prove theorem [ thm:2 ] which shows that minimally intersecting filling pairs optimize the value of @xmath331 : * theorem 1.2 . * _ let @xmath332 be a filling pair on @xmath30 and define @xmath31 to be the number of simple closed curves intersecting @xmath32 only once . then @xmath33 , with equality if @xmath28 is minimally intersecting . _ suppose first that @xmath28 is a minimally intersecting filling pair . then @xmath333 , because for each of the @xmath24 sub - arcs @xmath106 of @xmath53 ( resp . @xmath55 ) , there is a single simple closed curve which intersects @xmath106 and no other sub - arc . this simple closed curve corresponds to the arc in the @xmath42-gon obtained by cutting along @xmath32 which connects the two edges which project down to the sub - arc @xmath106 . now suppose that @xmath28 do not intersect minimally , so that @xmath334 has multiple connected components . the number of components is equal to @xmath335 thus if we cut along @xmath32 we obtain a disjoint union of @xmath336 even - sided polygons @xmath337 , such that the total number of edges over all polygons is @xmath338 . @xmath339 is then equal to the number of pairs of edges which project to the same arc in @xmath0 , and which belong to the same polygon . for each @xmath97 , there must be at least one edge of @xmath340 which has the property that the inverse edge is not on @xmath340 ( or else @xmath340 will be disconnected from all of the other polygons after glued up ) . thus , for each @xmath97 there must be at least @xmath23 edges of @xmath340 whose inverse edges belong to other polygons , since @xmath340 is even - sided for each @xmath97 . suppose there exists @xmath341 such that there are exactly two edges of @xmath340 whose inverse edge is not on @xmath340 . then one must project to a sub - arc @xmath106 of @xmath53 , and the other to a sub - arc @xmath342 of @xmath55 . consider the arc @xmath343 which immediately precedes @xmath106 along @xmath340 , and assume @xmath344 ( otherwise , the arc immediately preceding @xmath342 can not be @xmath106 , so in this case we just switch the roles of @xmath342 and @xmath106 ) . by assumption , @xmath67 is not minimally intersecting , but there is still a map @xmath345 acting on the edges of @xmath346 playing the same role as the permutation @xmath229 discussed in section @xmath347 and in the proof of lemma [ lm:2 ] in section @xmath348 . namely , send an edge @xmath349 to the edge @xmath350 immediately following it in clockwise order along some @xmath351 , and then send @xmath350 to the edge which projects down to the same arc as @xmath350 does ( i.e. , the inverse edge for @xmath350 ) . this map @xmath345 has order @xmath27 , because it corresponds to a `` rotation '' by @xmath352 about an intersection point . note that @xmath353 , which by assumption is not on @xmath340 . then @xmath354 ( by @xmath355 , we simply mean the sub - arc immediately following @xmath343 in the direction of @xmath343 ; see figure @xmath191 ) . . arrows indicate the action of @xmath345 on each oriented edge.,width=336 ] we claim that @xmath356 , for assume not . then @xmath354 is not on @xmath340 . @xmath357 is immediately followed by @xmath358 . thus @xmath359 , which is also not on @xmath340 , because otherwise both @xmath106 and @xmath360 would have the property that their inverse edges are not on @xmath340 . however , @xmath345 must have order @xmath27 , which implies that @xmath361 immediately follows @xmath360 ( see figure @xmath191 ) , and that in particular , the inverse of @xmath343 is not on @xmath340 , which contradicts our choice of @xmath343 . therefore , @xmath356 . this in particular implies that @xmath113 immediately precedes @xmath362 on @xmath363 for some @xmath144 . we claim that @xmath342 must immediately follow @xmath106 on @xmath340 ( see figure @xmath364 ) . if not , then let @xmath365 denote the edge immediately preceding @xmath113 on @xmath363 . then @xmath366 on @xmath340 , and @xmath367 for some @xmath368 , also on @xmath340 . thus @xmath369 , also on @xmath340 , and since @xmath345 has order @xmath27 , @xmath370 immediately follows @xmath110 on @xmath340 , and therefore @xmath371 . but this is a contradiction , because @xmath365 and @xmath362 represent distinct edges on @xmath363 ; see figure @xmath364 . has order @xmath27 , we conclude that @xmath372 , but this is impossible because @xmath365 and @xmath362 are distinct edges of @xmath363 by assumption.,width=336 ] this implies that a sufficiently small neighborhood in @xmath0 around the vertex on @xmath340 separating @xmath106 and @xmath112 contains only two edges of @xmath32 . this is a contradiction , because @xmath32 is @xmath27-valent . since @xmath97 was arbitrary , we conclude that each polygon @xmath337 must have at least 4 edges whose inverses are not on the same polygon . this leaves at most @xmath373 edges whose inverse can potentially be on the same polygon , and therefore @xmath374 the example of a filling pair on @xmath163 shown in figure @xmath27 proves that one can not push this further , in the sense that there exists filling pairs with complementary polygons @xmath375 containing only @xmath27 edges whose inverse edge is not on @xmath375 . however , one can say more in the case that there exists a complementary region with a number of edges not dividing @xmath27 : [ cor:3 ] let @xmath28 be a filling pair having the property that one complementary region @xmath375 has a number of sides which is not divisible by @xmath27 . then @xmath376 . the proof of theorem [ thm:2 ] implies that each complementary region has at least @xmath27 sides whose inverse edge is not on the same polygon . furthermore , if there exists a polygon @xmath375 in the complement of @xmath32 having a number of edges not divisible by @xmath27 , it follows that the number of edges of @xmath375 projecting to @xmath53 is odd , and similarly for @xmath55 . thus , if @xmath377 are the @xmath23 @xmath53-edges of @xmath375 guaranteed by theorem [ thm:2 ] to have the property that their inverse edges are not on @xmath375 , the remaining number of @xmath53-edges on @xmath375 is still odd , and therefore they can not all glue together . hence , there are at least @xmath150 edges of @xmath375 whose inverse edges are on other polygons , and therefore @xmath376 . it would be interesting to analyze the statistics of @xmath378 as @xmath99 goes to @xmath379 . we pose the following question : what is the minimal @xmath99 such that @xmath330 is not maximized by a minimally intersecting filling pair ? in this section , we prove theorem [ thm:3 ] . in all that follows , let @xmath380 denote length in @xmath381 , the hyperbolic plane . as in the introduction , define @xmath382 to be the _ moduli space _ of @xmath0 , the space of all complete hyperbolic metrics on @xmath0 up to isometry . then @xmath383 is the function which outputs the length of the shortest minimally intersecting filling pair on a hyperbolic surface @xmath3 , where the length of a filling pair is the sum of the two individual lengths of the unique geodesic representatives of the simple closed curves in the pair . let @xmath384 be a continuous function . then @xmath385 is called an _ ordinary point _ of @xmath269 if there exists an open neighborhood @xmath386 of @xmath387 , a chart @xmath388 inducing real - valued coordinates @xmath389 , and an integer @xmath97 , @xmath390 such that for any @xmath391 , otherwise , @xmath387 is a _ critical point_. if @xmath387 is a critical point , @xmath387 is called a _ non - degenerate critical point _ of @xmath269 if there exists an open neighborhood @xmath386 of @xmath387 , a chart @xmath388 inducing real - valued coordinates @xmath389 , and an integer @xmath97 , @xmath393 such that for any @xmath391 , @xmath97 is called the _ index _ of the non - degenerate critical point @xmath387 . for our purposes , topological morse function _ is a real - valued continuous function @xmath269 on @xmath382 such that @xmath269 has only finitely many critical points , all of which are non - degenerate . * theorem 1.3 . * let @xmath395 . _ @xmath2 is proper and a topological morse function . for any @xmath39 , @xmath40 where @xmath41 \right)\ ] ] denotes the perimeter of a regular , right - angled @xmath42-gon . furthermore , define @xmath43 then @xmath44 is finite and grows exponentially in @xmath1 . if @xmath45 , the injectivity radius of @xmath3 is at least @xmath396 _ . let @xmath397 be a compact subset . then we claim that there exists @xmath398 such that @xmath399 is contained in @xmath400 , the @xmath401-_thick part _ of moduli space , defined to be the subset of @xmath382 consisting of hyperbolic surfaces with injectivity radius at least @xmath401 . to see this , note that this is implied by the statement @xmath402 as injectivity radius approaches @xmath60 . by the collar lemma ( see @xcite ) , there exists a function @xmath407 such that @xmath408 such that if @xmath3 is a hyperbolic surface with a closed geodesic @xmath87 of length @xmath409 , there exists an embedded tubular neighborhood around @xmath87 whose width is at least @xmath410 . since any filling pair must intersect @xmath87 at least once , the result follows . let @xmath412 be a collection of real - valued smooth functions defined over @xmath382 and indexed by some countable set @xmath413 . given @xmath385 and @xmath414 , define @xmath415 then the function @xmath416 is called a _ generalized systole function _ if for each @xmath385 and each @xmath414 , there exists a neighborhood @xmath417 of @xmath387 such that @xmath418 is a finite subset of @xmath413 . akrout shows that any generalized systole function is a topological morse function . define @xmath413 to be the collection of minimally intersecting filling pairs ( not taken up to homeomorphism ) , and for each such pair @xmath419 , let @xmath420 denote the length of the filling pair @xmath421 on the hyperbolic surface @xmath3 . then @xmath422 . fix @xmath423 and @xmath424 ; then by the collar lemma , only finitely many pairs in @xmath413 can have length less than @xmath425 on @xmath387 . let @xmath426 be ( one of ) the shortest simple closed geodesics on @xmath387 , and let @xmath427 denote the length of @xmath426 on @xmath387 . let @xmath386 be a sufficiently small neighborhood around @xmath387 so that for any @xmath428 , @xmath429 , for some small @xmath401 . thus any hyperbolic surface in @xmath386 has the property that @xmath426 has length at most @xmath430 , so by another application of the collar lemma , there can only be finitely many filling pairs having the property that there exists a point in @xmath386 on which the pair has length at most @xmath425 . therefore @xmath2 is a topological morse function . next , we identify the minima @xmath44 of @xmath2 . recall that given a minimally intersecting filling pair @xmath28 on a hyperbolic surface @xmath3 , cutting along @xmath32 produces a single hyperbolic @xmath42-gon @xmath431 . fix @xmath432 , @xmath433 , and let @xmath434 be such that there exists a hyperbolic @xmath435-gon with area @xmath425 ( that is , @xmath425 is less than the area of an ideal @xmath435-gon ) . then by a result of bezdek @xcite , the hyperbolic @xmath435-gon enclosing an area of @xmath425 with the smallest perimeter is the regular @xmath435-gon with area @xmath425 . therefore , the perimeter of @xmath375 is at least as large as the perimeter of a regular @xmath42-gon enclosing an area of @xmath436 ( the area of @xmath3 , by gauss - bonnet ) . the regular @xmath42-gon with area @xmath436 is right - angled . the perimeter of such a polygon is @xmath437 , and its perimeter is twice the length of the filling pair . now we demonstrate a uniform lower bound on the injectivity radius for any @xmath438 . we identify a simple closed geodesic @xmath87 on @xmath3 with its lift @xmath439 in the right - angled regular @xmath42-gon @xmath431 that one obtains by cutting along the minimally intersecting filling pair @xmath28 of minimal length @xmath440 on @xmath3 ; @xmath439 is a disjoint collection of geodesic arcs connecting edges of @xmath431 . assume first that every arc in @xmath439 connects adjacent edges of @xmath431 . since @xmath441 are all geodesics , they are in pairwise minimal position , and therefore no two of them form a bigon on the surface . if @xmath439 contains some arc @xmath387 with one endpoint on some edge @xmath142 of @xmath431 belonging to @xmath53 , and the other endpoint on an adjacent edge @xmath5 belonging to @xmath55 , there must be some arc @xmath442 of @xmath439 with an endpoint on the edge of @xmath431 corresponding to @xmath443 . by the previous paragraph , the other endpoint of @xmath442 can not be located on the edge @xmath328 belonging to @xmath53 immediately following @xmath142 ( see figure @xmath259 ) . but then the arc of @xmath87 obtained by concatenating @xmath387 and @xmath442 is longer than the edge @xmath5 , because @xmath431 is right - angled . all edges of @xmath431 have length @xmath444 \right)\ ] ] therefore , we can assume that there exists an arc @xmath368 in @xmath439 whose endpoints are not on adjacent edges of @xmath431 . let @xmath349 be an edge of @xmath431 containing an endpoint of @xmath368 . let @xmath446 denote the endpoints of the two edges adjacent to @xmath349 , which are not shared endpoints with @xmath349 , and let @xmath387 denote the geodesic segment in @xmath431 orthogonal to both @xmath349 and the geodesic connecting @xmath421 to @xmath105 ( see figure @xmath447 ) . we claim that the geodesic segment @xmath1 connecting @xmath421 to @xmath105 must seperate @xmath349 and its two neighboring edges from the rest of @xmath431 . indeed , if any part of @xmath448 crosses @xmath1 , then @xmath1 is not contained in @xmath431 , which contradicts the fact that @xmath431 is convex . thus , since by assumption the other endpoint of @xmath368 is not on an edge adjacent to @xmath349 , @xmath368 must cross @xmath1 , and therefore its length is at least the length of @xmath387 . finally , we adress the growth of @xmath455 . properness of @xmath2 implies that all minima are contained in a compact subset of @xmath382 , and the fact that @xmath2 is a topological morse function implies that the minima are isolated . therefore , @xmath456 . it remains to show that @xmath44 grows exponentially as a function of @xmath1 . each @xmath9 orbit of minimally intersecting filling pairs can be associated to an element of @xmath44 by simply gluing the edges of a regular right - angled @xmath42-gon together in accordance with the chosen orbit . however , this association need not be injective ; a priori , it may be the case that many minimal length , minimally intersecting filling pairs occur on the same hyperbolic surface . suppose @xmath3 is a hyperbolic surface admitting @xmath368 minimal length , minimally intersecting filling pairs . thus there are @xmath368 immersed right - angled @xmath42-gons on @xmath3 , and developing @xmath3 into @xmath381 , these polygons lift to @xmath368 distinct tilings @xmath457 of @xmath381 by regular , right - angled @xmath42-gons . let @xmath458 denote the holonomy of this developing map ; i.e. , @xmath458 is the discrete subgroup of @xmath459 identified with the fundamental group of @xmath3 . recall that the _ commensurator _ of @xmath458 , denoted @xmath460 , is the subgroup of @xmath461 consisting of elements @xmath1 such that @xmath462 is _ commensurable _ with @xmath458 , which is to say that @xmath463 is a finite index subgroup of @xmath458 and @xmath462 . note that @xmath464 . let @xmath465 denote the full isometry group of the tiling @xmath466 ; note that @xmath467 for each @xmath468 , and @xmath458 acts freely and transitively on the set of tiles . thus @xmath458 has finite index in @xmath466 , and therefore @xmath469 . for each @xmath471 , let @xmath472 be an element sending @xmath331 to @xmath330 . then @xmath473 is the group of isometries of @xmath330 , and by assumption this is also @xmath474 ; @xmath475 contains @xmath458 which is finite index in both , and therefore @xmath476 . suppose for @xmath477 that @xmath478 are in the same coset of @xmath479 in @xmath480 . then @xmath481 fixes @xmath331 . thus @xmath482 sends @xmath330 to @xmath331 , but by definition it also sends @xmath483 to @xmath331 , a contradiction unless @xmath484 or @xmath485 , either of which implies the other , and that @xmath486 . a fundamental polygon for the tiling @xmath331 can be subdivided into @xmath487 copies of a hyperbolic triangle with angles @xmath488 . as a consequence , @xmath479 is commensurable with the triangle group of signature @xmath489 , the subgroup of @xmath461 generated by reflections in the sides of one of these triangles , and for which a presentation is as follows : @xmath490 denote this group by @xmath491 . by a result of long - machlachlan - reid @xcite ( also takeuchi @xcite ) , there are only finitely many arithmetic triangle groups . by a theorem of margulis @xcite , a subgroup of @xmath461 is arithmetic if and only if its commensurator is not discrete . therefore , for all sufficiently large @xmath1 , @xmath492 is discrete . using lemma [ lm:13 ] , for sufficiently large @xmath1 , we can assume that the quotient space @xmath494 is a hyperbolic orbifold @xmath495 , and therefore the area of @xmath495 is at least @xmath496 ( @xcite ) . @xmath495 is covered by the orbifold @xmath497 , and the degree of the covering map is equal to the index @xmath498 $ ] , and is also equal to the ratio of the area of @xmath497 to the area of @xmath495 . the area of a fundamental domain for the action of @xmath479 is bounded above by the area of @xmath3 , and therefore @xmath499 thus , at most @xmath500 minimal length , minimally intersecting filling pairs can coincide on the same hyperbolic surface , and hence there are at least @xmath501 surfaces containing such a pair . since @xmath14 grows exponentially by theorem [ thm:1 ] , so must the number of such surfaces . [ rk:4 ] note that , more generally , we can also consider the functions @xmath502 , which , given a hyperbolic metric @xmath3 on @xmath0 , outputs the length of the shortest filling pair whose complement has @xmath99 or fewer complementary regions . in this notation , @xmath503 . from @xmath504 and @xmath505 . we lose a total of @xmath27 edges in the creation of @xmath506 : @xmath507 , and @xmath142 and @xmath5 concatenate smoothly because they represent adjacent arcs on the same geodesic , and therefore they are replaced by a single edge . by the same argument , so are @xmath328 and @xmath508 . , width=336 ] we claim that @xmath509 has the same minima as @xmath510 . to see this , let @xmath28 be a filling pair of geodesics on @xmath3 intersecting @xmath511 times , so that cutting along @xmath32 produces a pair @xmath512 of hyperbolic polygons . between both polygons , there is a total of @xmath513 edges . then choose an edge @xmath349 on @xmath504 which glues to an edge @xmath514 on @xmath505 as in figure @xmath259 . since @xmath349 and @xmath514 have the same hyperbolic length , we can glue @xmath504 and @xmath505 together along @xmath349 to produce a new polygon @xmath506 , which must have @xmath27 fewer sides than the total number of sides in @xmath515 , for we lose the two edges @xmath507 which lie in the interior of @xmath506 , and the edges @xmath516 concatenate together smoothly in @xmath506 to form a single long edge since they correspond to adjacent arcs along the same geodesic on @xmath3 . the same is true for the edges @xmath517 .

in this paper we study transport properties of electrons on the two - dimensional honeycomb lattice . we consider a half - filled system in the vicinity of a symmetry - breaking transition from a semimetallic phase towards an insulating phase with either charge density or spin density wave order . the effect of either order is to break the sublattice inversion symmetry which induces a finite gap for the electronic single - particle excitations . phenomenologically , such a scenario is described in the framework of a gross - neveu theory . we analyze two related formulations of the model by means of ( i ) a controlled renormalization group calculation and ( ii ) the large - n method , both of which in combination with a boltzmann transport equation . we determine the quantum - critical conductivity and also discuss crossover behavior from quantum critical behavior into the insulating and/or the semimetallic phases . we find that at asymptotically low temperatures the quantum - critical conductivity is given by a temperature independent universal number . over a large temperature window the temperature independent quantum critical conductivity is masked by a logarithmically temperature dependent contribution due to the marginally irrelevant long - range coulomb interaction . we discuss possible origins of this peculiarity in the two complementary formulations of the model . furthermore , we consider possible relations of our findings to recent experiments , with a special emphasis on the quantum - critical - to - insulator crossover . we find that our results are in remarkably good qualitative and quantitative agreement with a recent analysis of the data sets under the hypothesis of an underlying gap in the single - particle spectrum .

1012.0263

two dimensional electronic systems on a honeycomb lattice have seen tremendous activity over the last couple of years . this development was triggered by the experimental isolation of single layer graphene which comes with a variety of fascinating and unusual properties @xcite . commonly , graphene is considered an excellent conductor with extremely high mobilities @xcite . the effective low - energy theory describing the system is given in terms of massless dirac fermions moving at the fermi velocity @xmath0 instead of the speed of light @xmath1 ( @xmath2 ) . most of the experimental findings in graphene to date can be explained in terms of weakly interacting massless dirac quasiparticles @xcite . a most prominent exception is provided by the recently observed fractional quantum hall effect @xcite . despite this predominant lack of signatures of correlation effects , many authors have analyzed the role of local and long - range coulomb interaction and possible interaction driven insulating phases @xcite . in this paper we study transport in the vicinity of a semimetal - to - insulator quantum phase transition , in which the opening of the single - particle gap in the insulating phase is driven by spontaneous sublattice symmetry breaking . the sublattice inversion symmetry breaking is driven by collective instabilities of the charge density or spin density wave type . note that this scenario does not describe the mott transition in the traditional sense due to the additional symmetry breaking @xcite . in that sense the treatment presented here is closer to the excitonic insulator obtained due to long - range coulomb interaction and chiral symmetry breaking @xcite than it is to a pure hubbard model . within this paper we analyze the simplest possible transition from a semimetal to an insulator described by a hubbard like model and discard the possible existence of interesting intermediate phases such as a spin liquid phase @xcite . transport properties at quantum - criticality have traditionally been a very active branch of research in strongly correlated matter . in two dimensions microscopic calculations have been performed with the quantum - critical conductivity being given by a temperature independent universal number @xcite . this has to be expected for dimensional reasons if the critical point of the field theory has relativistic invariance and is located at finite interaction strength . in the context of clean intrinsic graphene , which shares a number of characteristics with a quantum - critical system @xcite , the critical conductivity was shown to diverge in logarithmic manner upon lowering temperature @xcite . this behavior is rooted in the renormalization group fixed point of the system lying at zero interaction , with the interaction parameter describing long - range coulomb interaction being marginally irrelevant , thus flowing to zero logarithmically upon lowering the energy scale . in our problem , the insulating phase is characterized by the presence of collective order of charge density ( cdw ) or spin density wave ( sdw ) type @xcite . the finite order parameter acts like a mass term for the dirac fermions and consequently opens a gap in the electronic excitation spectrum . both sorts of long - range order break sublattice inversion ( @xmath3 ) symmetry ( conventionally referred to as chiral symmetry ) , which is the driving force behind the finite electronic gap . sdw additionally breaks spin rotation symmetry ( su(@xmath4 ) ) . within this paper we choose a simple field theoretic formulation of the gross - neveu type @xcite , which captures the collective instability and the opening of the electronic quasiparticle gap . we consider two related formulations : ( i ) a landau - ginzburg type bosonic order parameter theory coupled to dirac fermions via a yukawa - type coupling ( henceforth referred to as model i ) and ( ii ) a locally interacting theory of dirac fermions ( henceforth referred to as model ii ) . in both cases there is a quantum phase transition of the semimetal - to - insulator type , which is second order and described by an interacting fixed point . in both cases an appropriate order parameter can easily be defined , which is zero in the semimetallic phase and finite in the insulating phase . thinking in terms of the landau - ginzburg description of phase transitions , the order parameter symmetry is @xmath5 with @xmath6 ( ising or @xmath3 ) for cdw and @xmath7 ( heisenberg ) for sdw . microscopically , local interactions can trigger these instabilities : on - site coulomb repulsion favors sdw , whereas repulsion between adjacent sites favors cdw . [ fig : phasediag ] sketches a phase diagram with the vertical axis being temperature and the horizontal axis the local interaction parameter . the strength of local repulsion , denoted @xmath8 , serves as tuning parameter for the quantum phase transition . in the case of a quantum phase transition towards cdw , @xmath8 denotes repulsive interaction between adjacent sites , whereas for the case of a transition towards sdw the repulsion is on - site . our analysis proceeds along the lines of a combination of a semiclassical boltzmann equation combined with renormalization group arguments ( model i ) and a large - n expansion ( model ii ) @xcite . throughout the paper all mathematical expressions are explicitly shown for the cdw case . the sdw expressions differ in combinatorial factors due to the difference in order parameter symmetry , but not in their structure . however , qualitative differences occur at finite temperature due to the absence of long - range order in the sdw case . in the cdw case a finite gap in the electronic excitation spectrum is stable at finite temperatures , whereas there is no hard gap for the sdw . this is discussed in great detail in the conclusions . we calculated the quantum - critical minimal conductivity of graphene at the semimetal - to - insulator transition using two different phenomenological models . furthermore , we estimated crossover functions for the conductivity which obtain upon lowering temperature and entering the semimetal or respectively the insulating phase . most interestingly , we estimate the crossover function for the quantum - critical - to - insulator crossover and compare it to experimental data in graphene . considering two slightly distinct models we find seemingly contrasting transport properties : \(1 ) within model i in contrast to other relativistically invariant critical points at finite interaction strength in two dimensional systems , the d.c . conductivity is not independent of temperature , but instead seemingly diverges upon lowering temperature @xcite . this surprising result obtains because of a conspiracy of matrix elements and kinematical constraints which prohibits electronic current relaxation from the yukawa coupling to _ all _ orders in perturbation theory . this statement , however , is only strictly true if the electrons and bosons are treated as sharp quasiparticles at all times . it turns out that the minimal conductivity ( at charge neutrality ) is eventually determined by a marginally irrelevant operator , namely long - range coulomb interaction . the critical transport to leading order in temperature turns out to be identical to that of a gas of hot dirac electrons interacting solely via long - range coulomb interaction @xcite . \(2 ) within model ii we find a universal temperature independent universal conductivity as has to be expected from dimensional reasoning . however , we find that the prefactor of the inverse relaxation time is extremely small compared to the one associated with the marginally irrelevant long - range coulomb interaction . in contrast to model i , the scattering from the cdw ( sdw ) order parameter fluctuations is not completely inactive , but still extremely small . a simple order of magnitude estimate suggests that for all experimental purposes the universal quantum - critical conductivity is masked by the current relaxation due to long - range coulomb interaction . we comment on the seeming discrepancy between the transport properties of model i and model ii and argue that the two different pictures are actually compatible and describe the same basic physics , with model i overestimating the kinematical blocking . on a more technical note another interesting result obtains within model i , in which bosons and fermions are directly coupled , namely the absence of boson - drag effects for the coupled boltzmann equations . boson - drag constitutes a serious complication in obtaining transport coefficients in field theories of electrons coupled to bosons , where the bosons are an effective degree of freedom , whose dynamics itself is governed by the underlying electrons . an often employed approximation for the calculation of electronic transport properties in such systems is that the bosonic system is assumed to be equilibrated on timescales relevant for electronic transport and drag effects are neglected . this approximation turns out to be exact in our case . finally , the crossover behavior of the conductivity for the the quantum - critical - to - insulator crossover is discussed for both forms of collective ordering , namely the cdw and sdw . as we argued before , in the case of cdw an ising degree of freedom condenses , which is allowed at finite temperature . thus the electronic degrees of freedom have a hard gap , which entails an exponentially suppressed conductivity . in the case of the sdw , on the other hand , a real quasiparticle gap only opens at temperature @xmath9 and finite temperature behavior is not governed by a hard gap , but rather by a `` pseudogap '' , leading to a power - law suppression of the conductivity . both scenarios allow to determine crossover curves which show remarkable similarities to experiments performed on suspended high - mobility graphene samples @xcite . it appears that they are qualitatively and quantitatively widely compatible with a recent analysis of the data sets under the hypothesis of an underlying electronic quasiparticle gap and might thus be a natural starting point for the identification of the fitting parameters discussed in ref . . in sec . [ sec : model ] we introduce the scenario for the semimetal - to - insulator transition and motivate two different phenomenological models , called model i and model ii . during the course of the paper we will switch between model i and model ii , since different aspects of the problem are more conveniently discussed in one or the other formulation . we review the renormalization group treatment of model i following ref . and also the large - n treatment of model ii following ref . we furthermore stress some of the main physical properties of the models . in sec . [ sec : boltzmann ] we introduce the general formalism of the boltzmann approach . we apply the formalism to both models . in sec . [ sec : boltzi ] we consider model i and solve a system of coupled boltzmann equations for the fermionic and bosonic degrees of freedom . the bosonic degrees of freedom capture the physics of the collective instability and are eventually responsible for the opening of the electronic quasiparticle gap . we explicitly show that in this system of coupled transport equations neglecting the effect of `` boson - drag '' is _ exact _ to linear order in the applied electric field and proceed to analyze the boltzmann equation for the massless dirac particles . we find that inelastic scattering due to the yukawa - coupling is kinematically forbidden ( see sec . [ sec : noleadingorder ] ) to _ all _ orders in perturbation theory . this results in the absence of a temperature independent minimal d.c . conductivity . in sec . [ sec : boltzii ] we solve the boltzmann equation of model ii . in contrast to model i we find that there is a finite universal conductivity which does not depend upon temperature . however , we find that the prefactor of the inverse scattering time is tiny . this leads us to sec . [ sec : boltzcomp ] where we compare the results of in model i with those in model ii and argue in which sense the two results are compatible . in sec . [ sec : qcrcon ] we discuss the minimal conductivity in the quantum - critical regime and make an estimate of crossover temperatures . in a final section ( sec.[sec : discussion ] ) we discuss crossover behaviors which obtain upon lowering of temperature . furthermore , we discuss the fundamental difference between the cdw and sdw cases with a special eye on pseudogap behavior in the sdw case . a discussion of experimental conditions and reference to puzzling results of recent experiments on suspended high mobility graphene samples concludes sec . [ sec : discussion ] . it is well known that a tight binding description of non - interacting electrons on the honeycomb lattice in its low - energy version can be cast in the form of the dirac theory for massless fermions . the hamiltonian for the electronic part reads @xcite @xmath10 with @xmath0 being the fermi velocity . we explicitly consider half - filling , _ i.e. _ , chemical potential @xmath11 throughout the paper . double indices are henceforth summed over unless stated otherwise . in the above notation we take @xmath12 to be a spinor whose two components are associated with the two sublattices a and b. the index @xmath13 denotes the spin degree of freedom whereas @xmath14 denotes the valley ( @xmath15 and @xmath16 point in the brillouin zone ) . the @xmath17-matrices are defined as @xmath18 , and @xmath19 , and @xmath20 , where @xmath21 are the pauli matrices acting in sublattice space . local interactions in this system can induce collective instabilities . spontaneous symmetry breaking leads to a gap in the electronic single - particle excitation spectrum . we concentrate on order parameters which break the sublattice inversion symmetry , _ i.e. _ , the two sites per unit cell become chemically distinguishable . two possible instabilities correspond to charge - density wave ( cdw ) and spin - density wave ( sdw ) states , whose order parameters in terms of electronic operators read @xmath22 where @xmath23 acts in spin space and is simply a vector composed of pauli matrices . in the above representation the singlet @xmath24 corresponds to staggered charge density ( cdw ) , and the triplet @xmath25 to staggered local magnetization ( sdw ) . microscopically , a finite @xmath26 ( @xmath25 ) may be induced by a large nearest - neighbor ( on - site ) repulsion . in the following we explicitly study the case of cdw and comment on differences with the sdw case as we go along . we start with introducing model i , which is a field theory of dirac fermions coupled to a landau - ginzburg order parameter theory via a yukawa - type coupling . for the following discussions we generalize the model to @xmath28 dimensions and adopt a lagrangian formulation . the generalization to @xmath28 dimensions eventually allows to treat the bosonic and fermionic sectors on an equal footing in a perturbative renormalization group treatment . we generalize the dirac hamiltonian to three dimensions such that matrix elements have a structure like in two dimensions : on a technical level this implies that we evaluate angular integrals as if they were in two dimensions , but keep the integrals over absolute values in @xmath28 dimensions . the following discussion closely mimics the main steps of ref . . we decompose the local interaction into separate dynamical fields via hubbard - stratonovich decoupling and formulate the effective action near the semimetal - to - insulator transition as @xmath29 , with @xmath30 , where the bosonic part is described by @xmath31 and the yukawa term that couples bosonic and fermionic fields reads @xmath32 we note that @xmath33 describes the cdw fluctuation mode , where @xmath34 is the tuning parameter , @xmath35 the self - interaction , and @xmath36 the velocity . it is thus a landau - ginzburg theory for an ising order parameter . the physical meaning of the yukawa term is that a cdw can decay into a pair of fermions . we also introduce the inverse bosonic propagator at this point @xmath37 since it is needed for the generic transport equations . it is important to note here that @xmath38 lacks the lorentz symmetry due to the bosonic velocity @xmath39 . interestingly , the above gross - neveu theory is renormalizable in @xmath40 dimensions , where _ both _ the yukawa coupling @xmath41 and the self - interaction coupling @xmath35 become dimensionless . it is then possible to study the effective yukawa theory in @xmath42 dimensions using renormalization group ( rg ) methods with @xmath27 being the control parameter for the perturbative series ( @xmath43 corresponds to the two - dimensional model ) . we note that other authors have considered similar field - theories using an rg framework in the context of d - wave superconductors @xcite . in the following discussion we sketch the rg treatment of ref . . we first restore lorentz invariance of the model , implying we set @xmath44 throughout this paragraph . this is a crucial point and we extensively comment on this step below . we take @xmath38 as the starting point and integrate out fermionic and bosonic modes within the momentum - shell defined by @xmath45 . we furthermore introduce dimensionless couplings @xmath46 and @xmath47 with @xmath48 yielding @xmath49 to one - loop order . the above flow equations obtain for the bosonic mass @xmath34 tuned to criticality . the intricate flow diagram is schematically depicted in fig . [ fig : flow ] . most interestingly , the usual wilson - fisher fixed point ( wf ) of the @xmath50-theory is unstable towards finite yukawa coupling @xmath51 leading to an additional non - trivial fixed point called smi located at @xmath52 this fixed point describes the semimetal - to - insulator transition and the associated tuning parameter is the bosonic mass parameter @xmath34 ( eq . ) . the correct theory thus is described by electrons and bosons . a simple order parameter theory would not suffice to describe the physics correctly , since there the critical point would essentially be given by wf . in the above picture we have assumed the equivalence of bosonic and fermionic velocities , _ i.e. _ , lorentz invariance . a closer look at the theory reveals that this is only asymptotically correct and breaking the relativistic invariance actually corresponds to an irrelevant perturbation with respect to the fixed point smi . the corresponding rg equation was derived in ref . and reads @xmath53 at @xmath54 , where @xmath55 . this parameter will play a vital role in the analysis of the leading scattering mechanism in sec . [ sec : noleadingorder ] . we note that in the case of sdw the rg equations remain structurally intact , with modified combinatorial factors coming from the order parameter being @xmath56 instead of @xmath3 . and @xmath57 are the yukawa and the bosonic coupling , respectively ; g denotes the unstable gaussian fixed point of the landau theory and wf the wilson - fisher fixed point , which is unstable towards smi ( semimetal - to - insulator fixed point ) upon switching on the yukawa coupling @xmath51 . the transition is tuned by the bosonic mass parameter @xmath34 which is perpendicular to the @xmath58 plane and not shown explicitly.,scaledwidth=45.0% ] it is know that in a half - filled electronic system on the honeycomb lattice long - range coulomb interaction is marginally irrelevant @xcite . it is thus also important to understand the role of long - range coulomb interaction at smi . in order to do this one has to generalize the coulomb interaction to @xmath28 dimension following ref . @xmath59 this form ensures that integration over @xmath60 induces a @xmath61 density - density interaction between fermions in any dimension . it is found @xcite that coulomb interaction obeys the following rg equation @xmath62 one observes that our choice of @xmath63 ( eq . ) indeed produces a marginally irrelevant long - range interaction in any dimension . this establishes long - range coulomb interaction as the least irrelevant coupling at smi . the physics of the quantum - critical point smi has been analyzed in great detail by herbut _ et al . _ @xcite and we only repeat one key feature . a prediction of the field theory is that right at the quantum - critical point the fermion propagator behaves as @xmath64 where @xmath65 is the anomalous dimension . in this sense there are no well - defined fermionic quasiparticles at the critical point . however , it was shown by damle and sachdev @xcite that a boltzmann transport analysis in terms of quasiparticles still captures the relevant physics in the hydrodynamic regime ( @xmath66 ) . in that sense the semiclassical boltzmann equation only constitutes a convenient starting point to solve the more complicated quantum - boltzmann equation , which we refrain from showing in full bloom in this paper . in the following we consider an alternative formulation of the physics described above . the advantage of the below model is that we can analyze the model directly in two dimensions . in order to do so we employ a generalization in the spirit of the large - n method . the large - n limit is obtained by generalizing the fermionic spinors with a new index @xmath67 . the model of the free electrons was introduced before in eq . . in the formulation of the model we follow ref . . the interaction is now modeled by contact interaction of the type @xmath68 the physical equivalence of the above model with model i is established upon noting that @xmath41 drives an instability towards cdw , whereas @xmath69 drives the instability towards sdw . in the following we again concentrate on the case of the cdw , implying that we set @xmath70 . for large @xmath71 , @xmath41 can give rise to a dynamically generated mass of the type @xmath72 which is determined by the gap equation @xmath73 which can be integrated to yield @xmath74 where @xmath75 is the ultraviolet cutoff . demanding invariance of @xmath76 under rescaling @xmath77 we obtain @xmath78 where @xmath79 was used ( note that eq . differs from ref . due to a slightly different choice in the cutoff scheme ) . @xmath80 represents a quantum - critical point , where for @xmath81 runaway flow into a phase with broken chiral symmetry , _ i.e. _ , @xmath82 obtains . this model represents an alternative starting point for studying the critical transport properties in the framework of the boltzmann equation . without going into the details of the derivation @xcite we simply state that the effect of long - range coulomb interaction has also been studied in the context of this model in the limit of large - n . in order to do so the coulomb interaction for @xmath83 is equally generalized to large @xmath71 and the flow equation of the charge reads @xmath84 which confirms that the long - range coulomb interaction is marginally irrelevant , in qualitative agreement with eq . from model i and ref . we close this section with a generic comment about recent numerical simulations performed for dirac fermions with only long - range interactions in terms of a lattice gauge theory simulation . it was shown @xcite that even without short - range interactions a quantum phase transitions as a function of the fine - structure constant can be driven from a semimetal to an insulator , where in the insulating phase chiral symmetry is observed to be broken . this transition happens at a critical value of the fine structure constant and numerical values for this critical coupling come close to the value expected in vacuum . within numerical accuracy the transition proves to be second order and more remarkably the critical exponents seem compatible with the above two models with short - range interactions exclusively . we note that for a purely hubbard type model there has been a recent work discussing the mott transition which is not driven by collective symmetry breaking , but a transition into an intermediate paramagnetic spin liquid phase is observed @xcite . only upon increasing the local interaction the model has a transition into an antiferromagnetic phase . the nature of this spin liquid phase is currently not fully understood and not subject of our discussion . the boltzmann transport equation in combination with rg methods and large - n has proven a valuable tool in analyzing transport properties of quantum - critical systems @xcite . in this section we introduce the basic notation and concepts for the calculation of the critical conductivity following closely ref . . it is well known that in momentum space one can diagonalize the hamiltonian using a unitary matrix of the form @xmath85 with @xmath86 and @xmath87 . expressing the hamiltonian @xmath88 in terms of particle and hole operators @xmath89 we obtain the diagonal form @xmath90where the sum @xmath91 extends over electron- and hole - band , denoted @xmath92 . we can also express the electrical current in terms of the @xmath93 . for the case of a spatially independent current the result can be written as @xmath94with @xmath95@xmath96 measures the current carried by the motion of the quasiparticles and quasiholes ; the prefactor @xmath91 accounts for these excitations having opposite charges . the operator @xmath97 is not shown explicitly ; it describes the creation of a quasiparticle - quasihole pair ( it corresponds to the so - called zitterbewegung , see ref . ) as in the problems studied in refs . , in a particle - hole symmetric situation a current carrying state with holes and electrons moving in opposite directions has a vanishing total momentum and the current can decay by creation or annihilation of particle - hole pairs without violating momentum conservation . this is the physical reason why at the particle - hole symmetric point , _ i.e. _ , at vanishing deviation of the chemical potential from the dirac point , the d.c . conductivity is finite even in the absence of momentum relaxing impurities . however , at finite deviation from particle - hole symmetry , driven by chemical potential or the presence of second neighbor hopping @xmath98 , a driving electric field always excites the system into a state with finite momentum which can not decay . this entails an infinite d.c . conductivity in absence of umklapp scattering or disorder . as a first step one defines the distribution function @xmath99 note , there is no sum over @xmath100 on the r.h.s .. we furthermore assume that the distribution functions are the same for all valleys and spins , which has to be expected for symmetry reasons . in equilibrium , _ i.e. _ , in the absence of external perturbations , the distribution functions are fermi functions @xmath101where we temporarily allow for a finite chemical potential @xmath102 . in principle , off - diagonal elements such as @xmath103 are also created by an electric field . however , they are not needed to evaluate @xmath104 , which is the part of the current that we focus on in the hydrodynamic regime , @xmath105 . furthermore , the off - diagonal elements feed back to the kinetic equation of the diagonal elements only to higher order in perturbation theory . in the presence of an external electric field @xmath106 acting as a driving force on quasielectrons and quasiholes , we find the semiclassical boltzmann equation @xmath107 where @xmath108 is the scattering integral . we parameterize the change in @xmath109 from its equilibrium value by @xcite @xmath110where we have performed a fourier transform in time to frequencies , @xmath111 , and introduced the unknown function @xmath112 . at the particle - hole symmetric point ( @xmath11 ) , an applied electric field generates deviations in the distribution functions having opposite sign for quasiparticles and quasiholes . formally , this is a consequence of the driving term in eq . being asymmetric under @xmath113 , and thus the solution has to be asymmetric as well . this reflects the fact that there is an increased number of quasiholes and quasiparticles moving parallel and antiparallel to the exciting field , respectively . as quasiparticles and -holes have opposite charges , their electrical currents are equal and add up , while their net momenta have opposing signs and subtract to yield zero . in this paper we consider instabilities , which do not break particle - hole symmetry , and thus the above parametrization is strictly justified . we can formulate the boltzmann equation for the electrons and holes ( we only show the equation for the electrons ; for holes it follows by symmetry ) to lowest order in the electron - boson coupling as @xmath114 where @xmath115 with @xmath116 @xmath117 is the bosonic propagator and @xmath118 the bose distribution function . there exists an analogous equation for the bosonic degrees of freedom . since the bosons do not couple to the electromagnetic field , the driving term ( l.h.s . ) is exactly zero . however , the scattering integral couples the non - equilibrium fermions to the bosons constraining the bosonic distribution function to solve @xmath119 this bosonic boltzmann equation has to be solved simultaneously with the fermionic one . in principle , there is also a boson only scattering term coming from the self - interaction ( @xmath120-term ) @xcite . this term is not shown explicitly for reasons of conciseness . it turns out that its neglect proves to be exact in the case at hand . this is rooted in the fact that the bosons remain in equilibrium , which is shown explicitly below . severe complications in calculating electromagnetic response functions in systems in which fermions and collective ( slave-)bosons are coupled to each other result from the fact that one has to solve coupled boltzmann equations ( eq . and eq . ) . the physical reason for that is simple : electrons transfer momentum to the bosons , which themselves are driven out of equilibrium . since the bosons also scatter from the fermions , there is a feedback effect which in principle has to be taken seriously . often this is circumvented by assuming the bosonic sector to be in equilibrium , which is equivalent to saying that the bosonic sector equilibrates due to lattice effects or impurities on a time scale faster than the electronic scattering time . this can be an oversimplifying assumption with no _ a priori _ justification . one can appreciate this most easily by considering a galilean invariant electronic system interacting with effective bosonic modes . in this situation neglecting the generic non - equilibrium situation for the bosonic sector yields fundamentally wrong results , namely a _ finite _ d.c . one can convince oneself that taking into account the boltzmann equation for the bosons reestablishes the physically correct result of infinite d.c . conductivity ( see chapter on phonon - drag in ref . ) . we now turn our attention to eq . and expand the fermionic and bosonic distribution functions to linear order in the applied external field @xmath121 using eq . . for the bosonic sector we stick to the symbolic @xmath122 with @xmath123 being linear in the applied field and @xmath124 as the equilibrium bose distribution function . a simple analysis shows that @xmath125 drops out of eq . and the remaining parts annihilate exactly , as they should for consistency . the above scattering integral is thus solved exactly by letting @xmath126 and using eq . for the fermions . we have thus shown that at the dirac point the boson - drag vanishes exactly and considering the bosons in thermodynamic equilibrium is exact . one can convince oneself that in a situation with finite chemical potential @xmath102 this ceases to be true and the bosons have to be treated as being out of equilibrium . in such a situation a finite amount of disorder is needed to relax the net momentum and the associated current . we have thus established that in clean graphene at @xmath11 the bosonic sector remains in equilibrium . we can understand this more intuitively as follows : we first consider an electronic band with quadratic dispersion interacting with phonons . in this situation the bosonic sector absorbs some momentum from the electrons which scatter and is itself driven out of equilibrium . in graphene we have no quadratic disperion and by virtue of the perfect particle - hole symmetry ( @xmath11 ) the charge carrying modes at the dirac point are modes with net momentum zero . thus the bosons do not absorb any momentum which they have to dissipate . this implies they are not driven out of equilibrium and treating the bosons in equilibrium is no approximation but exact to linear order in the applied external field @xmath127 . note that the above reasoning is exclusive to electrical transport and does not hold for thermal transport . this line of argument is not exclusive to graphene or the matrix elements of the above interaction , but also applies to other systems with perfect particle - hole symmetry which is unbroken by the interaction mediated by the bosons . as a timely example one could consider bilayer graphene @xcite where the quasiparticles interact with effective bosonic degrees of freedom and the reasoning still holds . for dimensional reasons one expects the solution to eq . to yield a quantum - critical conductivity which is independent of temperature @xcite . it turns out that this is not true . to quadratic order in the interaction @xmath51 the scattering integral is exactly zero due to a conspiracy of the matrix element and energy conservation . this in itself does not rule out a temperature independent conductivity , since contributions to all orders in @xmath51 provide scattering times independent of temperature . however , one can convince oneself that the kinematical constraint shown in fig . [ fig : vertex ] annihilates contributions to current relaxation due to the yukawa coupling @xmath51 to _ all _ orders in perturbation theory , provided @xmath128 . this can easily be seen by looking at the matrix elements eq . and imposing energy and momentum conservation as shown in fig . [ fig : vertex ] . the exact cancellation is only operational for @xmath129 . breaking of lorentz invariance , _ i.e. _ , allowing for @xmath130 allows to relax the current . this implies that current relaxation due to the collective cdw degree of freedom is determined not only by @xmath41 but also by an irrelevant parameter , namely @xmath131 , which was introduced as the dimensionless parameter measuring the breaking of lorentz invariance . since @xmath41 assumes its fixed point value , which is a pure number , the inverse scattering time vanishes upon lowering temperature at least like @xmath132 , see eq . , leading to a diverging d.c . one now has to determine the least irrelevant scattering mechanism among all possible ones . as we argued before , all scattering events coming from the interaction with the collective mode scale with additional powers of @xmath131 and thus are irrelevant . we have seen in sec . [ sec : lrc ] that long - range coulomb interaction constitutes another perturbation to smi , which is only marginally irrelevant and consequently vanishes more slowly as temperature is lowered , namely in a logarithmic manner . from the above analysis we conclude that the dominant current relaxation mechanism stems from long - range coulomb interaction . this implies that to leading order the behavior in the semimetal is the same as at smi . this statement in itself is very surprising . we are now going to reexamine and reinterpret it in the framework of model ii . for scattering of quasiholes ( @xmath133 , full line ) and quasielectrons ( @xmath133 ) from the collective mode ( wiggly line ) . one can work out that for @xmath129 the bosonic spectral function is composed of sharp quasiparticles and only collinear scattering is allowed ; collinear scattering , however , annihilates the associated matrix elements.,scaledwidth=35.0% ] the following analysis is carried out directly in 2 dimensions starting from the interaction term eq . . the collision term on the r.h.s of eq . can be determined using fermi s golden rule or equivalently from a quantum - boltzmann equation . we refrain from an explicit derivation and refer the reader to ref . @xcite , which gives an explicit derivation for the case of long - range interactions which is easily generalized to the present case . the generalization leads to where @xmath135 note that the collision kernels @xmath136 and @xmath137 both come with a prefactor @xmath71 . there are additional contributions coming from crossed diagrams , which are however down by a factor @xmath138 , see ref . . the terms proportional to @xmath139 represent collisions between oppositely charged particles , while those proportional to @xmath140 are collisions between like charges . other processes , where a particle - hole pair is created , turn out to have vanishing phase space : it is not possible to fulfill momentum and energy conservation at the same time . this is a peculiarity of the linear dispersion , _ i.e. _ , @xmath141 ( see also ref . ) . the scattering matrix elements are given by @xmath142 we now proceed to the linearization of ( [ trans1 ] ) by inserting the parametrization ( [ paramet ] ) ( we define @xmath143 ) and find @xmath144 \notag \\ & & + \frac{\delta(k + k_1 - |\mathbf{k } + \mathbf{q}| - |\mathbf{k}_1 - \mathbf{q}| ) r_2 ( \mathbf{k } , \mathbf{k}_1 , \mathbf{q})}{(e^{-v_f k / t } + 1)(e^{-v_f k_1/t}+1 ) ( e^{v_f |\mathbf{k } + \mathbf{q}|/t } + 1 ) ( e^{v_f |\mathbf{k}_1 - \mathbf{q}|/t } + 1 ) } \label{trans2 } \\ & & \times \left [ \frac{\mathbf{k}}{k } g_\lambda(k , \omega ) + \frac{\mathbf{k}_1}{k_1 } g_\lambda(k_1 , \omega ) - \frac{(\mathbf{k } + \mathbf{q})}{|\mathbf{k } + \mathbf{q}| } g_\lambda(|\mathbf{k } + \mathbf{q}| , \omega ) - \frac { ( \mathbf{k}_1 - \mathbf{q})}{|\mathbf{k}_1 - \mathbf{q}| } g_\lambda(|\mathbf{k}_1 - \mathbf{q}| , \omega ) \right ] \biggr\}\ , . \nonumber\end{aligned}\ ] ] for the subsequent discussion we take the limit @xmath145 and focus on the solution of the linearized transport equation in eq . ( [ trans2 ] ) for the function @xmath51 . the following discussion closely follows the presentation given in ref . . we can view the right hand side of eq . ( trans2 ) as a linear operator , the so - called collision operator @xmath146 , acting on the function @xmath147 . a key property of @xmath148 is that it is hermitian with respect to the natural inner product @xmath149this hermiticity follows @xcite from symmetry properties of @xmath139 and @xmath140 under exchanges between incoming and outgoing momenta , which are very similar to those used in establishing boltzmann s h - theorem . related to the above properties of the collision operator , we can introduce a functional @xmath150 $ ] , such that eq . ( [ trans2 ] ) is equivalent to finding its stationary point @xmath151}{\delta g}=0 . \label{stat}\]]with the explicit form of the functional given by in the previous analysis of a quantum boltzmann equation for massless dirac fermions in two dimensions @xcite , it was noted that the phase space for scattering of particles was logarithmically divergent in the collinear limit . for the interaction considered in that paper , the collinear scattering cross - section vanished , and so this singular phase space density had no important consequences . the collinear scattering does not vanish for the present interaction , and so we need to take this logarithmic divergence seriously @xcite . the physical origin of the divergent collinear scattering is related to the linear dispersion which implies that quasiparticles or -holes moving in the same direction share the same group velocity , independent of their energies . this leads to a diverging duration of collisions of nearly collinear particles , which is enhanced due to the low space dimensionality . to the extent that collinear scattering is very strong , and considering frequencies much smaller than the inelastic scattering rate , we may expect that quasiparticles and holes that move in the same direction in the plane will establish a pseudo - equilibrium characterized by an effective chemical potential and an effective temperature which , however , depends on the direction of motion . in linear response the deviations of these effective parameters from the equilibrium values @xmath102 and @xmath153 have to vary with @xmath154 for symmetry reasons . the remaining dominant mode of the function @xmath51 will correspond to an effective shift in chemical potential which translates into [ psimu ] g(k)= , where the prefactor has been chosen so as to make @xmath155 dimensionless . with this ansatz , which will be confirmed below , it simply remains to determine the prefactor @xmath156 , yielding the leading term in the non - equilibrium distribution . the whole reasoning presented here is given in much more detail in refs . . note that the effective chemical potential shift ranges between @xmath157 depending on the direction of motion . comparing this to the temperature allows us to estimate the threshold electric field strength , @xmath158 , below which non - linear effects should remain small . the ansatz can be justified along the lines of ref . . the phase space for scattering diverges upon the perpendicular momentum going to zero . however , it turns out that the scattering matrix is not zero for collinear scattering . this implies that the logarithmic divergence for the perpendicular momentum going to zero has to be cut off by higher - order self - energy corrections to the free fermions . this implies that modes which are not zero for the collinear scattering channel have an additional factor @xmath159 , which in the limit @xmath160 is a large number . it thus turns out that the diverging phase space for collinear scattering helps to essentially solve the problem exactly in the limit @xmath161 . we conclude that up to corrections of order @xmath162^{-1}$ ] , we can choose @xmath51 to be of the form @xmath163we insert this parameterization into the functional @xmath150 $ ] in eq . ( [ func ] ) ; the solution of the stationarity condition in eq . ( [ stat ] ) is then equivalent to requiring the vanishing of the derivative with respect to @xmath164 . we numerically evaluated the integrals in eq . ( [ func ] ) using an elliptic co - ordinate system to solve the energy conservation constraint @xcite , and obtained @xmath165=\frac{1}{t}\frac{\ln { 2}}{4\pi } \biggl[\kappa \frac{g_s^2 t^2}{v_f^2 n } c^{2}-2 c\biggr]\,,\]]with @xmath166 . from the stationarity condition we then obtain @xmath167the conductivity can be obtained from @xmath164 by combining eqs . ( defj1 ) , ( [ paramet ] ) and ( [ gc ] ) : @xmath168where we have re - inserted factors of @xmath169 and @xmath170 . this seems to diverge upon lowering temperature , however , we have to keep in mind that at criticality we have @xmath171 . identifying the running energy scale @xmath172 with the temperature @xmath173 we end up with @xmath174 this result is independent of temperature , as we expect for a quantum - critical point in two dimensions . it is worthwhile noting that the prefactor is @xmath175 , which is different from the large - n transport analysis of the @xmath5 rotor model shown in ref . , which only has a prefactor of @xmath71 . the difference stems from the fact that in our case all @xmath71 flavors couple to the electromagnetic field , whereas in the latter case only two components were coupled to the electromagnetic field . so in contrast to the analysis of the transport equations in model i we find that the universal conductivity is finite , however with a numerically huge prefactor . we will now comment on the relationship between the two results . we have seen that model i and model ii seemingly give contradictory answers to the question whether there is a finite universal conductivity . whereas in model i we find that a kinematical constraint knocks out scattering from the order parameter fluctuations right at the critical point , in model ii we find a finite scattering due to the electron - electron interaction . we will argue in the following that this can actually be interpreted consistently . in order to do so we have to analyze the reason for the kinematical constraint in model i. this discussion can most easily be carried out directly in two dimensions , where for the moment we put all subtleties about controlling the perturbation series aside . if we consider the r.h.s of eq . we realize that the reason for knocking out scattering processes stems from the fact that the order parameter fluctuations are described by sharp quasiparticles , _ i.e. _ , the spectral function assumes the form of delta peaks . one can now ask in which sense this describes the right physics . in order to do so we can pretend to derive model i from model ii via hubbard - stratonovich transformation and integrating out electrons . it turns out that the propagator of the order parameter field is actually _ not _ given by @xmath176 but instead by @xmath177 this has severe consequences : while eq . implies that the bosonic spectral function is sharply peaked , eq . describes a continuum , which still has its maximum weight at the former resonance . while a lot of the weight still is kinematically blocked , using eq . in the boltzmann approach would immediately render the r.h.s . of eq . finite , albeit small . in that sense the bosonic sector in model i does not faithfully describe the character of the bosonic spectral function in the real two dimensional system . this implies that in reality we expect the real bosonic spectral function to still be peaked around the resonances , but also to have a continuum background . the boltzmann analysis in model ii backs up this statement . directly in two dimensions we find a finite temperature independent inverse scattering time from interactions with the collective field , albeit with a tiny numerical prefactor . the prefactor being so small can consistently be interpreted as a remnant of the complete kinematical blocking in model i. our conclusion thus is that the boltzmann equation as it is presented for model i simply overestimates the kinematical blocking by annihilating the scattering completely . we thus conclude this section by stating that the asymptotic quantum - critical universal conductivity is numerically huge , but independent of temperature , as we expected to find . we will now discuss the implications of the results of sec . [ sec : boltzmann ] . we choose to carry out the discussion in the framework of model ii for reasons of enhanced clarity . we found that the universal quantum - critical conductivity within model ii is a pure number times @xmath178 . this central result is shown in eq . . however , as we argued before , the prefactor of the inverse scattering time is numerically tiny , consistent with the complete blocking found in model i. this naturally leads us to consider the influence of the remaining marginally irrelevant interaction , which is given by the long - range coulomb interaction . we established that in the clean system at zero chemical potential there are two main sources of current relaxation namely ( i ) short - range interaction which drives the phase transition and ( ii ) long - range coulomb interaction . in the following we compare the inverse scattering times due to the two independent processes . the discussion is simplified by the fact that we can restrict our discussion of the inverse scattering times to one hydrodynamic mode , see sec . [ sec : boltzmann ] . we know that in the framework of the boltzmann equation collision integrals add . thus , by virtue of the most relevant relaxation processes for both types of interaction living on the same hydrodynamic mode inverse scattering times also simply add up . this is what is usually called matthiessen s rule @xcite within the relaxation time approximation , but here it is essentially exact . in the following we again lend from ref . and simply take the numerical value of the inverse scattering time on the zero mode and compare it to the one obtained in eq . . since for practical purposes the large - n parameter @xmath71 is a number of order on we can safely drop it from our discussion in the physically relevant situation . schematically , we can write the collision integral composed of the sum of the two processes as @xmath179 where @xmath180 is the running long - range coupling . the perturbation theory of both terms is formally controlled in the small parameter @xmath138 . however , there is a big difference : while the first term in brackets is temperature - independent , the second term scales to zero logarithmically upon lowering temperature . this implies that asymptotically in the very low temperature limit the second term vanishes and the first term determines the universal temperature independent quantum - critical conductivity , in agreement with the previous discussion . for higher temperatures , however , the contribution due to long - range coulomb interaction will win . we can make a very crude order of magnitude estimate for the temperature @xmath181 , where the crossover from universal quantum - critical conductivity to the regime , in which conductivity is determined by long - range coulomb interaction , takes place . we know that in graphene the ultraviolet cutoff is of the order ev , which translates to @xmath182 . furthermore , the bare long - range coulomb interaction is a dimensionless number of order one . we showed in sec . [ sec : boltzmann ] that @xmath183 and ref . found @xmath184 . we can estimate the crossover temperature @xmath181 as the temperature , where both contributions are comparable in size . this crossover temperature can conservatively be estimated to be of the order @xmath185 . this implies that for all practical purposes the universal quantum - critical conductivity will be masked by the long - range coulomb interaction . interestingly , this also implies that the quantum - critical conductivity at the semimetal - to - insulator critical point has the same characteristic transport properties as the one in the semimetal , _ i.e. _ , for a gas of hot dirac fermions interacting via long - range coulomb interaction . furthermore , this implies that the conclusions of the hydrodynamic transport relations obtained in a series of papers @xcite also hold at the quantum - critical point . we simply cite the result for the coulomb interaction limited minimal conductivity which has been analyzed elsewhere @xcite . there , a boltzmann equation was solved directly in two dimensions . this procedure has been presented at length in ref . and we only highlight the final result for the d.c . conductivity reading @xmath186 using eq . in eq . we observe that upon lowering the temperature the conductivity diverges as @xmath187 scales to zero logarithmically . in eq . @xmath188 is a number of order one and @xmath172 constitutes the high energy cutoff set by the lattice . we end this discussion by saying that we expect this to be the correct minimal conductivity in region ii in fig . [ fig : phasediag ] for all physically relevant situations . this implies that within our approach for realistic temperatures we can not distinguish the minimal conductivity in region i from region ii in fig . [ fig : phasediag ] . in this paper we have investigated quantum - critical transport of electrons on the honeycomb lattice in the vicinity of the semimetal - to - insulator transition . on a technical level , we have used a combination of the semiclassical boltzmann transport equation with rg and large - n to analyze two related gross - neveu type field theories . one of the central results of this paper is given in eq . which is the conductivity in the quantum - critical region ( region ii in fig . [ fig : phasediag ] ) for physically realistic temperatures . the most important characteristic of this minimal d.c . conductivity is that it is set by scattering off the long - range coulomb interaction , which is a marginally irrelevant coupling at the semimetal - to - insulator transition . the result is surprising since the critical conductivity weakly but explicitly depends upon temperature through the flow of the dimensionless coulomb interaction parameter . in general , for a lorentz invariant quantum - critical point in two dimensions described by an interacting fixed point one would expect a conductivity independent of temperature . a further remarkable consequence of our calculation is that to leading order in temperature one can not distinguish the conductivity in fig . [ fig : phasediag ] in region i from region ii . we furthermore showed that below a very low temperature scale @xmath181 a temperature independent universal conductivity takes over , in agreement with expecations . we finally go back to fig . [ fig : phasediag ] and analyze possible crossover scenarios . we know that the quantum phase transition microscopically is driven by the interaction parameter @xmath8 . within the two models we consider this translates into different quantities . whereas in model i the quantum phase transition is driven by the bosonic mass parameter @xmath34 in the case of model ii the phase transition is driven by the interaction parameter @xmath51 . the crossovers of the system are shown in fig . [ fig : phasediag ] by doted lines with the associated crossover temperature @xmath189 , which is set by @xmath190 . another more formal interesting result of our analysis of model i is that it is possible to show that the so - called boson - drag in our setup is exactly zero meaning that an equilibrium treatment of the bosonic degrees of freedom within the fermionic collision integral is justified to leading order in the applied external field @xmath121 . this statement is more general than that and can be extended to systems with perfect particle - hole symmetry , in which the collective bosonic modes do not break the particle - hole symmetry . for the case of the cdw we have an ising type transition . this implies that due to the absence of goldstone modes symmetry breaking is possible at finite temperatures and a finite electronic gap is stabilized . for the sake of simplicity we concentrate on a crossover function , which captures the essential physics deep in the quantum - critical regime and in the insulating regime . one such function is of the type @xmath192 with @xmath193 being a number . the activated form of the conductivity in the insulating regime iii accounts for the fact that we have a finite gap separating the conduction band from a valence band , link in an ordinary semiconductor . a typical crossover curve is shown in fig . [ fig : resistivity ] , where a comparison of the quantum - critical resistivity ( inverse conductivity right at the critical coupling ) is compared with the crossover function for the crossover from region ii to region iii . the full line in fig . [ fig : resistivity ] shows the quantum - critical conductivity , whereas the dashed line shows the crossover function . . the curve for the crossover ii - iii stays mostly flat over a wide temperature range and eventually takes off exponentially . in the quantum - critical region ii the resistivity approaches zero upon lowering temperature in logarithmic manner . the plotted curves are shown for numerical values @xmath194 ( eq . ) , @xmath195 , and @xmath196.,scaledwidth=45.0% ] in the case of a cdw instability a discrete ising degree of freedom is spontaneously broken and long - range order at finite temperature is possible . for the sdw an additional breaking of spin rotation symmetry is required , thus there can be no true long - range order at finite temperature by virtue of the hohenberg - mermin - wagner theorem . this implies that the fermionic quasiparticles are only truly gapped at zero temperature . the question is whether the conductivity can still show signatures of a pseudogap in the conductivity which diverges as temperature is lowered . the role of a finite temperature pseudogap in scaling functions has been analyzed before in the context of the hertz - millis theory in two dimensional systems @xcite . in our qualitative consideration , however , we go along the lines of lee _ et al . _ @xcite , who considered fluctuation effects at the peierls transition in one dimensional systems . it is most convenient to think of the problem at very low temperatures in terms of a spin - fermion model like model i. we concentrate in the following on the low - temperature region which is usually called the renormalized classical regime @xcite . there , the self energy of the electrons is given by the zero - frequency matsubara mode of the bosonic propagator , _ i.e. _ , @xmath197 where @xmath198 is the finite correlation length which diverges upon lowering temperature . in the framework of a boltzmann theory this implies that the former inelastic scattering of the electrons off the bosonic mode has become elastic and corresponds to the scattering off long - range impurities of the @xmath199-type where @xmath200 is the distance . in order for this to be true we assume @xmath201 here , which is a non - singular limit in the boltzmann - equation and corresponds to the `` renormalized classical '' regime . the temperature behavior of such a type of impurities has been analyzed before @xcite and it was found that the associated transport scattering time behaved as @xmath202 again , we assume a crossover function which correctly captures the transport characteristics in the quantum - critical and the renormalized classical regime of the form @xmath203 the respective crossover function is plotted and compared to the quantum - critical resistivity with notation as above in fig . [ fig : resistivitysdw ] . the curve for the crossover ii - iii stays mostly flat over a wide temperature range and eventually takes off as a power - law . in the quantum - critical region ii the resistivity approaches zero upon lowering temperature in logarithmic manner . the plotted curves are shown for numerical values @xmath194 ( eq . ) and @xmath204.,scaledwidth=45.0% ] one element we have not yet discussed is the role of particle - hole symmetry breaking and related to this disorder . if the system does not have perfect particle - hole symmetry some rigorous statements , such as the one about vanishing boson - drag and finite conductivity without disorder , cease to be true . in practice , there are several ways to break particle - hole symmetry . one is to allow for a finite chemical potential , which however brings us away from half - filling . another way is to take into account second neighbor hopping @xmath98 . if particle - hole symmetry is absent , there is no current relaxation in absence of impurities , since the momentum mode is excited in that case . also , in that case the boson - drag can not be neglected any more . however , we still expect our theory to be valid in some limit . the regime of validity of our theory is identical to the one which has been worked out in great detail ref . . for a finite chemical potential we need @xmath205 for our theory to hold . for particle - hole symmetry breaking due to next - nearest - neighbor hoppings , such as provided by @xmath98 @xcite we have to require that for the temperature window we are considering that we are still within energies , where the spectrum behaves linearly . within this paper we have analyzed the resistivity in the vicinity of the semimetal - to - insulator transition on the honeycomb lattice . more specifically , we have studied the crossover from region ii to region iii in fig . [ fig : resistivity ] and fig . [ fig : resistivitysdw ] . we now argue that our findings bear some relevance for the understanding of experiments carried out on suspended graphene samples . remarkably , the curves shown in fig . [ fig : resistivity ] and fig . [ fig : resistivitysdw ] mimic the almost constant behavior of the resistivity upon decreasing temperature over a large temperature window followed by an upturn reminiscent of an insulator as observed by bolotin _ et . al _ @xcite on suspended samples ( see ref . , fig . 4 inset ) . the samples considered in those experiments exhibit ultrahigh mobilities and enormous mean free paths with respect to impurity scattering ( concerning impurity scattering the samples are ballistic with an associated mean free path longer than the sample length ) . these high mobilities and long elastic mean free paths are obtained after repeated annealing removing impurities from the samples . this makes the inelastic scattering mechanism discussed in this note a viable candidate to set the limiting scattering process for the conductivity . most strikingly , bolotin _ et al . _ find that upon lowering temperature the resistivity increases . it is tempting to attribute this behavior to the existence of a quasiparticle gap . it is well known that increasing the interaction on the honeycomb lattice , both short- or long - range , can lead to a finite quasiparticle gap in the system . samples on substrates do not seem to show this tendency , but one has to keep in mind that suspending the sample modifies the dielectric environment and renders the system more strongly interacting . one can thus imagine that suspending graphene samples modifies the dielectric environment strongly enough to drive the system towards strong interactions rendering the charge gap finite . at finite densities ( @xmath206 ) a qualitative explanation of the experimental findings has been given in ref . however , here we concentrate on the zero doping case and to the best of my knowledge the data is not fully understood . recently , the data at charge neutrality have been thoroughly reanalyzed by drut _ et al . _ @xcite who speculated about an underlying , possibly interaction driven gap in the system . the authors determined a couple of phenomenological parameters to fit the data . more specifically , a phenomenological form of the conductivity @xmath207 where the subscript q stands for quasiparticle and bg for background was introduced . the background conductivity was set by an unspecified scattering process . thus we do not speculate about the nature of this part but concentrate on the quasiparticle part . for the interpretation and the fitting of @xmath208 the authors considered a gap as well as another unspecific scattering process : concerning the unspecified scattering process , the ratio of the inverse scattering time to temperature , _ < @xmath209 was found to be independent of temperature . up to logarithmic accuracy this is indeed the case in our scenario , due to the inverse scattering time due to inelastic scattering being given by @xmath210 . furthermore , the authors find that within the relevant temperature range the mean free path associated with this scattering process is on the order of a tenth of micrometers . this is also in agreement with our results . the mean free path corresponding to inelastic scattering due to long - range coulomb interaction has been estimated before in ref . and is given by @xmath211 } $ ] . in this sense long - range coulomb interaction is a natural candidate for one of the current relaxation mechanisms needed to explain the measurements faithfully . concerning the gap of the quasiparticle , the authors argued that the data is not only compatible with a hard gap but also with a power - law type contribution . the hard gap naturally emerges from the cdw scenario , whereas the pseudogap behavior stems from the sdw scenario . both mechanisms result in a diverging resistivity at low temperatures . obviously , there are additional other scattering sources , which could play a role . it turns out that phonons due to the high fermi velocity play a role at rather elevated temperatures ( @xmath212 ) and thus are not a candidate to modify the above picture of the transport characteristics @xcite . another candidate would be the corrugation or ripples which are observed in suspended graphene . it turns out that on a formal level ripples share some characteristics with long - range coulomb interaction . to our knowledge no thorough study of the effect of ripples on the conductivity in suspended graphene exists to date and a detailed analysis is called for @xcite . we thus conclude this discussion by stating that a proximity of free standing graphene to a quantum - critical point for a transition from the semimetal to an insulator described in this note could provide a natural explanation for the ( temperature ) behavior of the resistivity observed in the experiment of bolotin _ et al . _ @xcite on ultraclean high - mobility samples . this , however , would also imply that signatures of collective symmetry breaking of the sdw or cdw type should be observable in those samples , which to my knowledge has not been attempted . this provides a necessary independent check of the scenario of graphene in vacuum being an excitonic insulator . k. i. bolotin , k. j. sikes , z. jiang , g. fudenberg , j. hone , p. kim , and h. l. stormer , sol . . comm . * 146 * , 351 ( 2008 ) . s. y. zhou , g .- h . gweon , j. graf , a. v. fedorov , c. d. spataru , r. d. diehl , y. kopelevich , d .- h . lee , s. g. louie , and a. lanzara nature phys . * 2 * , 595 ( 2006 ) . i. f. herbut , v. jurii ' c , and b. roy , phys . b * 79 * , 085116 ( 2009 ) . i. f. herbut , v. jurii ' c , and o. vafek , phys . b * 80 * , 075432 ( 2009 ) . s. sorella and e. tosatti , europhys . lett . * 19 * , 699 ( 1992 ) .

the supernova remnant ( snr ) n49 in the large magellanic cloud has been observed with the advanced ccd imaging spectrometer ( acis ) on board the _ chandra x - ray observatory_. the superb angular resolution of the _ chandra_/acis images resolves a point source , the likely x - ray counterpart of soft gamma - ray repeater sgr 0526 , and the diffuse filaments and knots across the snr . these filamentary features represent the blast wave sweeping through the ambient interstellar medium and nearby dense molecular clouds . we detect metal - rich ejecta beyond the main blast wave shock boundary in the southwest of the snr , which appear to be explosion fragments or `` bullets '' ejected from the progenitor star . the detection of strong h - like si line emission in the eastern side of the snr requires multi - phase shocks in order to describe the observed x - ray spectrum , whereas such a multi - phase plasma is not evident in the western side . this complex spectral structure of n49 suggests that the postshock regions toward the east of the snr might have been reheated by the reverse shock off the dense molecular clouds while the blast wave shock front has decelerated as it propagates into the dense clouds . the x - ray spectrum of the detected point - like source is continuum dominated and can be described with a power law of 3 . this provides a confirmation that this point - like x - ray source is the counterpart of sgr 0526 in the quiescent state .

astro-ph0211552

n49 is a bright supernova remnant ( snr ) in the large magellanic cloud ( lmc ) @xcite . the low metal abundances of n49 @xcite indicated a substantial mixture of the supernova ( sn ) ejecta material with the lmc interstellar medium ( ism ) , which is consistent with the relatively old dynamical age as derived to be @xmath25000 @xmath0 10000 yr @xcite . the most remarkable feature of n49 is strong enhancements of the optical / x - ray surface brightness in the south - east . this asymmetric intensity distribution has been attributed to the interaction of the blast wave with the nearby molecular clouds located in the east of the snr @xcite . the radio images show similar intensity distribution to the x - ray morphology , but with a complete shell in the west as well as in the east @xcite . the radio spectral index of @xmath3 = @xmath00.58 @xcite is typical for a shell - type snr . although the mean fractional polarization over the whole remnant is low ( @xmath45% ) , the western side of the snr is highly polarized . the highly organized magnetic field in the west of the snr is supportive of the less - disturbed blast wave expanding into the low density ism in the west @xcite . vancura et al . ( 1992 ) have performed extensive optical / uv spectroscopy and have found that n49 is interacting with dense clouds ( @xmath5 @xmath2 150 @xmath6 on average ) while the preshock intercloud density is @xmath20.9 @xmath6 . this dense ism is most likely associated with the co clouds detected in the east of the snr @xcite . the bright optical / uv emission arises from `` sheets '' of emission generated by slow radiative shocks of @xmath7 @xmath8 140 km s@xmath9 as the blast wave interacts with the dense ambient material , while the x - ray emission appears from the fast shock ( @xmath7 @xmath2 730 km s@xmath9 ) propagating in the intercloud medium . the x - ray spectrum of n49 as obtained with _ asca _ was fitted with a non - equilibrium ionization ( nei ) model with @xmath10 @xmath2 0.6 kev and @xmath11 @xmath2 10@xmath12 @xmath6 s @xcite . the fitted electron temperature and the ionization age are consistent with the sedov evolution without significant alteration of the preshock ism by the stellar winds from the progenitor ( the model suggested by shull et al . [ 1985 ] ) , which supports the overall picture proposed by vancura et al . the overall x - ray spectrum was fitted with low metal abundances , which is also consistent with the optical / uv spectral analysis . the position of the outburst of the soft gamma - ray repeater ( sgr ) sgr 0526@xmath066 in 1979 march 5 has been known to be within the boundary of n49 @xcite and is consistent with the unresolved x - ray source rx j05260.3@xmath0660433 @xcite , suggesting the identification of this source as the x - ray counterpart of sgr 0526@xmath066 . this unresolved x - ray source appears spatially associated with snr n49 based on its positional coincidence within the boundary of the snr although an incidental alignment along the line of sight between the two can not be ruled out @xcite . although an 8 s pulsation from sgr 0526@xmath066 has been detected during the @xmath13-ray burst in 1979 @xcite , the x - ray timing / spectral analysis of this unresolved x - ray source has been limited with previous detectors ( e.g. , marsden et al . no counterparts in the infrared and radio bands have been found for sgr 0526@xmath066 @xcite . these results from sgr 0526@xmath066 share typical characteristics of the anomalous x - ray pulsars ( axp ) , provided that the detection of the x - ray pulsar is confirmed . high resolution x - ray imaging with the moderate spectral capability of _ chandra _ observations would be critical to confirm the detection of the x - ray counterpart of sgr 0526@xmath066 . we here report the results from the imaging and spatially resolved spectral analysis of snr n49 from the _ chandra _ observations . while this work is primarily devoted to the analysis of the diffuse x - ray emission from snr n49 , we also report some preliminary results from the detected point source , the x - ray counterpart of sgr 0526@xmath066 . we describe the observation and the data reduction in [ sec : obs ] . the image and spectral analyses are presented in [ sec : analysis ] , and a discussion in [ sec : disc ] . a summary will be presented in [ sec : summary ] . snr n49 was observed with the advanced ccd imaging spectrometer ( acis ) on board the _ chandra x - ray observatory _ on 2001 september 15 as part of the _ chandra _ guaranteed time observation ( gto ) program ( obsid 1041 ) . the acis - s3 chip was chosen to utilize the best sensitivity and energy resolution of the detector in the soft x - ray bands . the pointing ( @xmath3@xmath14 = 05@xmath15 25@xmath16 [email protected] , @xmath18@xmath14 = @xmath065@xmath19 59@xmath20 06@xmath219 ) was roughly toward the x - ray centroid of the nearby lmc snr n49b with the aim point shifted by 245 detector pixels ( @xmath22@xmath20 ) from the nominal s3 chip aim point in order to simultaneously observe both of n49 and n49b on the same ccd chip . this pointing direction results in an angular offset of @xmath26@xmath225 from the aim point at the position of n49 . the results from the data analysis of the pointed target snr n49b are presented elsewhere @xcite . we have utilized the data reduction techniques developed at the pennsylvania state university for correcting the spatial and spectral degradation of the acis data caused by radiation damage , known as charge transfer inefficiency ( cti ) @xcite . the expected effects of the cti correction include an increase in the number of detected events and improved event energies and energy resolution @xcite . we screened the data with the flight timeline filter and then applied the cti correction before further data screenings by status and grade . we removed `` flaring '' pixels and selected _ asca _ grades ( 02346 ) . the overall lightcurve was examined for possible contamination from time - variable background . the overall lightcurve from the whole s3 chip shows relatively high background ( by a factor of @xmath232 ) for the last @xmath210% of the observation time . this contamination from the variable background is however negligible for the n49 source region ( @xmath21@xmath224 diameter ) and we ignore it in the data analysis . after these data reduction process , the effective exposure was 34 ks that contains @xmath2241 k photons in the 0.3 @xmath0 8 kev band for the whole remnant within a 1@xmath224 diameter circular region . during _ chandra _ cycle 1 ( 2000 january 4 ) , n49 was observed with the acis - s3 as a guest observer ( go ) program , which is now publicly available in the archive ( obsid 747 ) . the pointing was toward the position of sgr 0526@xmath066 ( @xmath3@xmath14 = 05@xmath15 26@xmath16 [email protected] , @xmath18@xmath14 = @xmath066@xmath19 04@xmath20 35@xmath210 ) . in this observation , instead of the whole ccd and the standard 3.2 s frame time , a 128-row subarray with a 0.4 s frame time was used for the temporal study of the target object sgr 0526@xmath066 . due to the selection of the subarray of the ccd chip , the coverage of the snr was incomplete and only the northern parts of the snr were observed . the total exposure was @xmath240 ks . although the angular coverage is incomplete , this on - axis observation provides better angular resolutions than our gto data , thus we utilize these archival data as supplements in the analysis . we have reprocessed the raw archival data in exactly the same way as the gto data were processed , including the cti correction . the `` true - color '' x - ray images of snr n49 from the _ chandra_/acis observations are presented in figure [ fig : fig1 ] . figure [ fig : fig1]a is the image from the gto observation and figure [ fig : fig1]b is the one from the archival go data . the differences in the point spread function ( psf ) between two images due to the angular offset of @xmath26@xmath225 from the aim point in the gto data are apparent . the high resolution on - axis archival data reveal the detailed diffuse filamentary structure , particularly in the east of the snr , far better than the off - axis gto data . the detection of the point source as the counterpart of sgr 0526@xmath066 is evident in the middle of the northern part of the snr as shown by a bright white spot ( figure [ fig : fig1]b ) . this point source is also detected in the off - axis image ( figure [ fig : fig1]a ) but appears extended because of the psf degradation . despite the superior angular resolution of the on - axis data , the selection of the subarray of the ccd allows only a partial coverage of the snr : i.e. , about a third of the snr in the south - west was not covered ( figure [ fig : fig1]b ) . the imaging and spectral analysis over the whole remnant is important to understand the global nature of the snr and we primarily utilize our gto data for the data analysis . in order to take advantage of the high angular resolution , we use the archival data for supplementary purposes . in figure [ fig : fig1 ] , the soft x - ray emission ( red - yellow ) reveals various filamentary structures across the snr . the circumferential shell - like soft x - ray features around the snr are spatially coincident with the radio shells ( figure [ fig : fig2]a ) , representing the blast wave shock front . by comparison , the hard x - ray emission features show little correlation with the radio intensities ( figure [ fig : fig2]b ) . based on astrophysical considerations , we believe that it is reasonable to suppose that shocks are reflecting off the denser molecular gas in the east and may be propagating back into the interior of n49 . this could be a partial explanation for the higher temperatures we find in the spectral analysis of the eastern region ( [ subsubsec : east ] ) , although a definitive conclusion would require a detailed investigation of the hydrodynamics of n49 that is beyond the scope of this work . other evidence , albeit more speculative , in support of reflected shocks in the eastern region of n49 include ( 1 ) the morphology of the hard x - rays which peak closer to the interior on the eastern side than the soft x - rays do ( figure [ fig : fig2 ] ) , and ( 2 ) the clear arc - like feature in the si ew map ( [ subsec : ew ] and figure [ fig : fig5]f ) which traces hot gas , that perhaps delineates the edge of the inward moving reflected shock . the western half is , for the most parts , dominated by hard x - ray emission ( green - blue ) , which may indicate that the blast wave is propagating through a low density ism , as suggested by radio data @xcite . the brightest diffuse emission is in the southeast , as recognized by previous x - ray observations , where the snr is interacting with dense molecular clouds . the overall bright x - ray emission in the eastern half of the snr is significant in all colors and likely has complex spectral structure . while filamentary soft , red - yellow emission prevails in the eastern half of the snr , there are apparent blue regions in the east ( primarily to the immediate north of the southeast brightest region ; figure [ fig : fig1 ] ) . this diffuse x - ray emission with spectral variations appears localized to produce spatially distinctive layers of x - ray colors in the eastern half of the snr . the interactions of the blast wave shock with the clumpy dense molecular clouds in these regions appear responsible for the observed spatio - spectral substructures of the diffuse x - ray emission ( see [ subsec : spec ] ) . for example , the bright filametary emission features from the optical oiii images are largely anti - correlated with the blue , hard x - ray intensity and more correlated with the soft , red x - ray emission ( figure [ fig : fig3]a and figure [ fig : fig3]b ) . the optical emission from n49 is believed to originate primarily from slow radiative shocks entering dense molecular clouds , while the x - ray emission is in general from fast shocks propagating through the intercloud medium @xcite . the high resolution x - ray images from the acis data reveal that the hard x - rays are indeed from intercloud regions where little optical emission originates , and that the soft x - ray emission is spatially more coincident with the bright optical emitting regions . these features support the interactions of the blast wave with the clouds as the origin of the x - ray spectral variations observed in the color images . we detect an interesting , previously unknown feature of n49 with our _ chandra _ gto data : i.e. , the blue , hard x - ray knot in the southwest , beyond the main boundary of the snr ( figure [ fig : fig1]a ) , from which faint radio emission is also associated ( figure [ fig : fig2]b ) . the extended shape of this feature appears caused by the off - axis psf and the true morphological nature of this emission is uncertain with the current off - axis data . we can not utilize the high resolution archival data , because this feature lies outside the field of view ( figure [ fig : fig1]b ) . we find no proper counterparts for this feature in other wavelengths , which suggests that it is unlikely a projection of a background / foreground point source ( [ subsubsec : bullet ] ) . the x - ray spectrum of this feature indicates metal - rich thermal emission with strong atomic emission lines rather than a power law dominated spectrum ( [ subsubsec : bullet ] ) . these spectral features support the proposition that this hard x - ray knot is a part of the snr , probably fragment(s ) of sn ejecta . the overall spectrum of n49 shows some broad atomic emission lines ( figure [ fig : fig4 ] ) and we generate _ equivalent width _ ( ew ) images of k - shell line emission from the atomic elements o , ne , mg , si , s , and the l - shell line emission of fe by selecting photons around the broad line complexes ( table [ tbl : tab1 ] ) , following the methods described by park et al . ( 2002a ) . images in the bandpasses of interest for both the lines and continuum were extracted with 2@xmath24 pixels and smoothed by a gaussian with @xmath25 = 7@xmath24 . the underlying continuum was calculated by logarithmically interpolating between images made from the higher and lower energy `` shoulders '' of each broad line . the estimated continuum flux was integrated over the selected line width and subtracted from the line emission . the continuum - subtracted line intensity was then divided by the estimated continuum on a pixel - by - pixel basis to generate the ew images for each element . in order to avoid noise in the ew maps caused by poor photon statistics near the edge of the snr , we have set the ew values to zero where the estimated continuum flux is low . we also set the ew to zero where the integrated continuum flux is greater than the line flux . we present the ew images for o , ne , mg , si , and fe ( figure [ fig : fig5 ] ) . the s ew image is unreliable due to the low photon statistics . the broad line complexes at @xmath20.8 @xmath0 1.1 kev are from various ionization states of ne k and fe l - shell line emission , and we present tentative identifications in figure [ fig : fig5 ] and table [ tbl : tab1 ] . although mg and si lines show he - like and h - like substructures in the overall spectrum ( figure [ fig : fig4 ] ) , we present the ew images from the combined broad lines in figure [ fig : fig5 ] since the h - like lines of these elements are too weak to separately produce useful ew images over the whole snr . the o ew distribution is generally consistent with the optical oi , iii intensity distributions and shows higher ew in the east and south of the snr ( figure [ fig : fig5]a ; vancura et al . 1992 ) . the fe ew ( figure [ fig : fig5]d ) shows similar distribution to the o map . ne / fe and mg ew ( figure [ fig : fig5]b , figure [ fig : fig5]c , and figure [ fig : fig5]e ) distributions appear to be opposite to o , showing higher intensities in the west of the snr . the si ew is relatively strong in both east and west ( figure [ fig : fig5]f ) . the o and fe ew enhancements are spatially correlated with the location of the dense molecular clouds where the x - ray and optical emission is generally bright ( figure [ fig : fig5]a and figure [ fig : fig5]d ) . since the x - ray emission in these regions is presumably dominated by the shocked dense ism , these high o and fe ews are unlikely due to abundance enhancements , but appear to trace shock temperature variations . optical spectroscopic studies have indicated that metal abundances of n49 are close to the lmc values @xcite and support temperature variations , rather than abundances , as the origin of these ew distributions . based on the plane - parallel shock model with a moderate ionization timescale ( @xmath11 @xmath2 10@xmath26 @xmath6 s ) and the lmc abundances , the fe l ew ( figure [ fig : fig5]d ) is in fact representative of the broad line complex at @xmath27 @xmath2 0.8 kev , which is characteristic of a `` cool '' plasma ( @xmath10 @xmath28 0.5 kev ) as opposed to the `` hot '' component ( @xmath10 @xmath2 1 kev ) that would produce a prominent fe l line complex at @xmath27 @xmath21 kev . in direct contrast , the h - like ne ( and/or hot fe l ) ew ( figure [ fig : fig5]c ) , which is sensitive to high temperatures of @xmath10 @xmath2 1 kev , is enhanced in the west where the blast wave is expanding into a low density ism . these ew distributions may thus indicate the presence of an _ undisturbed _ ( by dense ism ) blast wave shock in the west , which may have a higher temperature than that in the eastern side of the snr . if the dominant contributor to the enhanced ew at @xmath27 @xmath2 1.1 kev is ne , comparisons between he - like and h - like ne ew images ( figure [ fig : fig5]b and figure [ fig : fig5]c ) may suggest progressive ionizations of ne by the reverse shock as detected in other snrs @xcite . the si ew enhancements in both east and west are intriguing and may suggest the presence of hot phases in addition to cool phase shocks in the south - east parts of the snr ( see more discussion in [ subsec : spec ] ) . the si ew enhancements in the southeastern quadrant are largely shell - like , as indicated by a dotted line in figure [ fig : fig5]f . this may also be interpreted as hot , highly ionized plasma in the southeastern regions of the snr . the hard x - ray knot in the southwest beyond the snr boundary , as revealed with the true - color image , shows strong enhancements in the si , mg , and hot ne / fe ew images . the actual spectral analysis indicates that these high ews are caused by metal abundances as well as advanced ionizations and high temperature ( see [ subsubsec : bullet ] ) . the enhanced ews for the elemental species caused by high metal abundances support that this hard x - ray knot may not be a background extragalactic source , and that it is likely emission from metal - rich sn ejecta associated with n49 . the ew around the detected point source is , on the other hand , very low for all elemental species . these low ews are strong indications of the continuum - dominated x - ray spectrum for the point source . this is in good agreements with the presumed identification of the point source as the neutron star , or the x - ray counterpart of sgr 0526@xmath066 in the quiescent state . the high resolution acis data , for the first time , allow us to perform spatially - resolved spectral analysis in x - rays for the study of snr n49 . despite the psf degradations due to the large angular offset from the aim point , the gto data provide moderate angular resolution of a few arcsec , which are still sufficient to resolve major diffuse structure of n49 . moreover , the complete spatial coverage of the snr with the gto data is advantageous over the archival data for the efficient and consistent study of the global structures of the snr . combining two data sets may not be desirable because of the significant difference in the psf . we thus make a simple and secure approach for the spatially - resolved spectral analysis by utilizing the gto data as the primary data and the archival data as the supplementary data : i.e. , we perform all spectral analysis with the gto data and then verify the results with the archival data as necessary . we extract the spectrum from several small regions over the snr . the regional selections were based on the characteristics revealed by the broad band color images and the ew images , and are presented in figure [ fig : fig6 ] . in all spectral fittings , we made a low energy cut at the photon energy of @xmath27 = 0.5 kev because of unreliable calibrations of the acis spectrum in the soft bands . in most cases , we bin the data to contain a minimum of 20 counts for the spectral fitting . for the spectral analysis of our cti corrected data , we have utilized the response matrices appropriate for the spectral redistribution of the ccd , as generated by townsley et al . ( 2002b ) . the low energy ( @xmath27 @xmath4 1 kev ) quantum efficiency ( qe ) of the acis has been degraded because of the molecular contamination on the optical blocking filter . we have corrected this time - dependent qe degradation by modifying the ancillary response function ( arf ) for each extracted spectrum by utilizing the idl version of acisabs software developed by chartas et al . ( 2002 ) . we select three regions from the western half of the snr ( regions 1 , 2 , and 3 ; figure [ fig : fig6 ] ) . these regions are where the broad band surface brightness is relatively low , while the ew images reveal some noticeable features ( figure [ fig : fig5 ] and figure [ fig : fig6 ] ) . region 1 is where o and ne ews are enhanced and contains @xmath23500 photons . region 2 is where ne / fe , mg and si are enhanced containing @xmath28100 counts . region 3 shows low ew for most of the elements and contains @xmath23200 photons . we fit these spectra with the plane - parallel shock model @xcite as presented in figure [ fig : fig7]a , figure [ fig : fig7]b , and figure [ fig : fig7]c . the elemental abundances were fixed for he (= 0.89 ) , c (= 0.30 ) , n (= 0.12 ) , ca (= 0.34 ) and ni (= 0.62 ) at values for the lmc @xcite ( hereafter , all abundances are with respect to the solar ; anders & grevesse 1989 ) since the contribution from these species in the spectral fitting is expected to be insignificant in the selected energy range . other elements were allowed to vary . the results of the spectral fittings for the west regions of the snr are summarized in table [ tbl : tab2 ] and table [ tbl : tab3 ] . these spectra in the western half of the snr can be described by a thermal plasma with the best - fit @xmath10 of 0.52 @xmath0 0.65 kev . the implied absorptions are @xmath20.6 @xmath02.7 @xmath29 10@xmath30 @xmath31 . these results are consistent with the results from the _ asca _ data @xcite . the fitted abundances are generally consistent with the ew distributions . the results from the region 2 spectrum are perhaps the most interesting . this is where ne / fe , mg , and si ews are enhanced and the best - fit abundances for those elemental species ( except for fe ) are also relatively high , although the magnitudes of the enhancements are moderate only at around the solar level . the fitted abundances for this region suggest that the high ew responsible for the emission line at @xmath27 @xmath2 1.1 kev ( figure [ fig : fig5]c ) is due to highly ionized h - like ne rather than hot fe . the best - fit shock temperature and the ionization timescale for this region are higher than region 1 in which he - like ne ew is enhanced ( figure [ fig : fig5]b ) . this supports the advanced ionization in region 2 as suggested by the ew images . the region 3 spectrum is on the other hand fitted with low lmc - like abundances for all elements . the broad band x - ray surface brightness is also low for this region , indicating x - ray emission from the shocked low density lmc ism . we select two regions from the eastern half of the snr ( regions 4 and 5 ; figure [ fig : fig6 ] ) . region 4 is within the bright eastern rim which is evidently dominated by the hard x - ray emission as displayed in figure [ fig : fig1 ] . a total of @xmath210000 photons are extracted from region 4 . region 5 is the brightest knot in the southeast , where we extract @xmath217000 counts . we fit these spectra in the same way as we did for the western regions : i.e. , we use the plane - parallel shock model with o , ne , mg , si , s , and fe abundances variable in the fittings . both of the spectra in the eastern half of the snr show significant si ly@xmath3 line ( @xmath27 @xmath2 2.0 kev ) as well as strong soft x - ray emission ( @xmath32 1 kev ) ( figure [ fig : fig7]d and figure [ fig : fig7]e ) . in order to adequately describe these spectral features , a hot component shock with @xmath10 @xmath23 1 kev is required in addition to the soft component with @xmath10 @xmath2 0.4 @xmath0 0.5 kev . the hot components have larger ionization timescales ( @xmath11 @xmath23 10@xmath26 @xmath6 s ) than the cool components ( @xmath11 @xmath2 10@xmath12 @xmath6 s ) . the fitted ionization timescales for the hot component are not constrained in the upper bounds and a thermal plasma in the complete collisional ionization equilibrium ( cie ) may also fit the data with the equivalent statistical significance . the implied absorbing columns are @xmath33 @xmath2 3 @xmath0 4 @xmath29 10@xmath30 @xmath31 for these regions . the soft component model for the region 5 spectrum indicates low metal abundances of @xmath28 0.5 solar for fitted elemental species . this soft component dominates the x - ray emission at @xmath27 @xmath28 1 kev , which consistently supports that the enhanced o and fe ews ( figure [ fig : fig5]a and figure [ fig : fig5]d ) are not caused by high abundances . the best - fit model for the hard component implies overabundances ( typically @xmath2 1 @xmath0 2 solar ) compared with the lmc abundances for some elemental species . the statistical uncertainties for the fitted abundances are however large enough that the reality of these high abundances are inconclusive with the current data . for example , this region 5 spectrum may also be fitted with subsolar abundances , for both of the soft and hard components of the shock , with equivalent statistical significance based on the f - test . the ews distributions in the east of the snr are therefore more likely dominated by the shock temperature and the ionization states rather than by metal abundances . the fitted model parameters for the region 4 spectrum are also generally consistent with the results from region 5 , which also supports the temperature and ionization states for the origins of the observed ew distributions . we note that region 4 ews appear relatively weak for the atomic lines at low photon energies ( o and fe ; figure [ fig : fig5]a and figure [ fig : fig5]d ) and stronger in higher energies ( si ; figure [ fig : fig5]f ) compared with region 5 , even though there is no significant difference in the fitted shock parameters between these two regions . these features may imply a difference in the local interstellar environments due to the clumpy structure of the interacting dense molecular clouds : i.e. , a density gradient of the clouds between these two regions might have caused deficient soft x - ray emission in region 4 as displayed by the lack of red , soft emission there ( figure [ fig : fig1 ] ) . the anticorrelations between optical and hard x - ray emission are supportive of this case . the fitted volume emission measure ( @xmath34 ) of the soft component plasma for the region 5 spectrum is indeed significantly higher than that for region 4 , which also supports this interpretation . the results of the spectral fittings for the east regions of the snr are summarized in table [ tbl : tab4 ] . the hard x - ray knot in the southwest beyond the main boundary of the snr ( region 6 in figure [ fig : fig6 ] ) contains @xmath2800 counts and we binned the spectrum to contain a minimum of 12 counts per bin . the spectrum can be fitted with a single - temperature plane - parallel shock ( figure [ fig : fig7]f ) . the spectrum features strong atomic emission lines and is most likely emission from metal - rich ejecta . the strong si ly@xmath3 line ( @xmath27 @xmath2 2.0 kev ) constrains the plasma temperature to @xmath10 @xmath23 1 kev . although statistical uncertainties are large due to the poor photon statistics of this faint feature , the best - fit model indicates strong enhancements in metal abundances at several times larger than solar abundances . the broad band line features between @xmath20.8 kev and @xmath21.2 kev represent fe l line complex as well as he - like and/or h - like ne k lines . with the photon statistics of the current data , it is difficult to identify these individual atomic lines . we have thus attempted fitting the spectrum by assuming dominant contributions from ne or fe as well as from both elements . the overall fits are statistically indistinguishable among these cases . we however note that the model fits fe as dominant contributor ( fe abundance @xmath2 0.8 , ne @xmath2 0 ) when both ne and fe are simultaneously allowed to vary . even with fe abundance fixed at lmc value , the fitted ne abundance is still low ( @xmath20.4 ) . we therefore obtain the best - fit model by allowing fe abundance to vary with ne fixed at the lmc abundance . the fitted parameters are summarized in table [ tbl : tab2 ] and table [ tbl : tab3 ] . although the metal - rich spectrum indicates that this feature is likely emission from shocked sn ejecta , the isolated morphology of this hard x - ray knot outside of the main boundary of the snr may suggest the possibility of a projected background object . the source position for the knot is @xmath3@xmath14 = 05@xmath15 25@xmath16 [email protected] , @xmath18@xmath14 = @xmath066@xmath19 05@xmath20 13@xmath215 , and the acis astrometric uncertainties at @xmath26@xmath20 off - axis is generally @xmath282@xmath24 ( e.g. , feigelson et al . the _ chandra_/acis position of sgr 0526@xmath066 based on the on - axis data has a conservative astrometric uncertainty of @xmath282@xmath213 ( kaplan et al . 2001 and references therein ) . the position of the sgr of our off - axis data as determined by _ wavdetect _ algorithm is @xmath21@xmath218 offset from the on - axis sgr position . the source positions with our off - axis data therefore appear to have @xmath44@xmath24 accuracy as a conservative limit . we have searched for counterparts around the source position in simbad and ned catalogs as well as @xmath2100 multi - wavelengh on - line catalogs available through heasarc . we find no candidate for an extragalactic counterpart within 1@xmath20 radius of the source position . the derived x - ray flux is @xmath35 @xmath2 6.0 @xmath29 10@xmath36 ergs s@xmath9 @xmath31 and @xmath37 @xmath2 1.5 @xmath29 10@xmath36 ergs s@xmath9 @xmath31 . based on the logn @xmath0 logs relations of the _ chandra _ deep field @xcite , these x - ray fluxes imply the probability of only @xmath2810@xmath38 within a generous area of 5@xmath24 radius around the source position for a coincidental detection of an extragalactic object . we find an lmc star , macs j0525@xmath0660#037 @xcite which has an angular offset from the position of the hard x - ray knot of @xmath26@xmath212 . considering the positional accuracy of the macs catalog ( @xmath20@xmath215 ; tucholke et al . 1996 ) , the apparent angular offset between the hard x - ray knot and optical position of the macs star is significant . we find another nearby lmc star within the ob association lh 53 as cataloged ( i d # 1630 ) by hill et al . the relatively small angular offset ( @xmath22@xmath24 ) of this star from the x - ray hard knot position suggests that this star could be the optical counterpart of the x - ray knot . the derived x - ray luminosity of @xmath39 @xmath2 3.4 @xmath29 10@xmath40 ergs s@xmath9 ( at the distance of 50 kpc ) for the x - ray knot is however higher than that of typical late - type young stellar objects ( @xmath39 @xmath2 10@xmath41 @xmath0 10@xmath42 ergs s@xmath9 ; e.g. , imanish et al . 2001 ) by a few orders of magnitude . a massive early - type star should be brighter ( @xmath39 @xmath2 10@xmath43 ergs s@xmath9 ; e.g. , seward & chlebowski 1982 ) , but may also not match the derived high @xmath39 of the hard x - ray knot . moreover , although there are @xmath280 lh 53 stars within 1@xmath20 radius of this hard x - ray knot position @xcite , none of them is detected with our acis observation at the flux level of this hard x - ray knot . it is unlikely for only one star to be detected with the acis at this particular position at this flux level . if this object is a foreground young late - type galactic star with a typical x - ray luminosity of @xmath39 @xmath2 10@xmath41@xmath010@xmath42 ergs s@xmath9 , this object _ must _ be nearby with a distance of @xmath230 @xmath0 300 pc , and a v magnitude of @xmath211 @xmath0 14 is expected from the measured x - ray flux . we have , however , found no such bright stars in galactic star catalogs . we find a faint star in the guide star catalog with an @xmath22@xmath24 offset from the source position , but the cataloged b magnitude is only 19.44 . the best - fit absorbing column from the x - ray spectral modeling may also be higher ( @xmath33 = 1.1 @xmath29 10@xmath30 @xmath31 ) than the galactic absorption toward the position of this hard x - ray knot ( @xmath33 @xmath2 6 @xmath29 10@xmath44 @xmath31 ; dickey & lockman 1990 ) . this hard x - ray knot is therefore most likely a feature spatially associated with snr n49 . this feature then appears to be fragments from the sn explosion ejected outside of the snr shock boundary or `` shrapnel '' as discovered in other galactic snrs such as vela and tycho @xcite . the true angular size and shape of this hard x - ray knot is however uncertain because of the poor off - axis psf . we note that an on - axis _ chandra_/acis observation with a full angular coverage of n49 , targeted on sgr 0526@xmath066 has recently been performed ( obsid 2515 ) . the detailed results from the diffuse x - ray emission of the snr have yet to be reported . the spectrum from the point - like source is extracted from region 7 ( figure [ fig : fig6 ] ) . due to the angular offset from the aim point , the psf at the position of the point source is significantly degraded ( @xmath232@xmath24 ) , and we select a relatively large area ( a circular region with a radius of 6@xmath24 ) , which contains @xmath214000 photons . although the total count rate ( @xmath20.4 counts s@xmath9 ) is relatively high , pileup is insignificant due to the large off - axis angle . since the source location is within fairly bright diffuse emission from the snr , the degraded psf also makes the background estimation difficult . we thus include a thermal component for the background spectrum in the spectral fitting instead of subtracting the background from the source spectrum . in a direct contrast to the diffuse emission spectra , the point source spectrum is dominated by continuum , as perhaps expected from a neutron star ( figure [ fig : fig8]a ) . the source spectrum can be described with a single power law model of photon index @xmath1 = 2.98@xmath45 . the embedded background thermal component is a @xmath10 @xmath2 0.33 kev plasma with lmc - like metal abundances , which is primarily responsible for the soft emission at @xmath27 @xmath4 1 kev . a black body model may also fit the data ( @xmath10 @xmath2 0.52 kev ) with acceptable statistics , but appears , qualitatively , inappropriate to describe the hard tail of the specrum . since the archival data are pointed in the direction of the position of the point source or sgr 0526@xmath066 , the on - axis high angular resolution allows us to manipulate the background subtraction with less confusion . we extract the source spectrum from a circular region with a radius of 2@xmath24 , which contains @xmath210000 photons . because of the utilized short time - frame ( 0.4 s ) , pileup is negligible with this on - axis observation . sgr 0526@xmath066 is on top of complex filamentary structure of the diffuse emission from the snr and the background emission for the sgr is non - uniform ( figure [ fig : fig1]b ) . we thus extract the background emission from the immediate annular region centered on the point source position ( inner radius of 2@xmath215 and outer radius of 4@xmath24 ) , in order to obtain an `` average '' spectrum over the variable emission features around the sgr . the sgr spectrum from the on - axis data can be fitted with a single power law model ( figure [ fig : fig8]b ) . the best - fit photon index is @xmath1 = 2.72@xmath46 which is in good agreement with the result from the off - axis data . the results from the fit with a black body model ( @xmath10 @xmath2 0.50 kev ) are also identical to the off - axis data and appear inadequate by the reason of the same hard - tail discrepancy . the implied x - ray luminosity is @xmath39 = 2.1 @xmath29 10@xmath47 ergs s@xmath9 in the 2 @xmath0 10 kev band with the off - axis data ( @xmath39 = 2.5 @xmath29 10@xmath47 ergs s@xmath9 based on the on - axis data ) , which is consistent with the asca upper limit @xcite and that of the galactic sgr 1806@xmath020 @xcite . the results from the spectral model fits of the point source are summarized in table [ tbl : tab5 ] . the x - ray color images and the ew images indicate complex spectral structure of the diffuse x - ray emission across the snr n49 . the filamentary and clumpy emission features in the eastern half of the snr show spatially distinctive bright structures in different colors . the spatial and spectral structure of the x - ray emission in the western half is relatively uniform , primarily represented by intermediate - to - hard emission ( green - blue ) with low surface brightness . the soft ( red - yellow ) x - ray shell - like emission features represent the blast wave shock front as observed in the radio band . the overall features of the ew images indicate enhancements of soft line emission ( @xmath32 1 kev ) in the south - east and hard emission lines ( @xmath48 1 kev ) in the west of the snr . these features are consistent with the overall density distributions of the ambient ism ; i.e. , the blast wave is propagating into a low density ism in the west and interacting with dense clouds in the east . the comparisons between the x - ray and optical images also suggest that the hard x - ray emission ( @xmath48 1.6 kev ) appears to be produced in less - dense intercloud regions while the soft x - rays ( @xmath32 0.75 kev ) originate near the dense clouds . the results from the spatially - resolved spectral analysis confirm the speculations inferred from the image analysis by revealing that shock temperatures and ionization states in the east of the snr are significantly different from those in the western half : i.e. , the western portion of the snr is described by a simple one - temperature plane - parallel shock with relatively large ionization timescales while two - temperature model with multiple ionization timescales is required in the eastern half . the global spectral structure of the snr may then be represented by a multi - phase shock model with three characteristic temperatures , i.e. , the `` cool '' ( @xmath10 @xmath2 0.44@xmath49 kev ) and `` hot '' ( @xmath10 @xmath2 1.25@xmath50 kev ) components in the east , and the `` intermediate '' ( @xmath10 @xmath2 0.59@xmath51 kev ) component in the west . the presence of multiple phases of the shock due to the interactions of the blast wave with the dense molecular clouds has been put forth by hester & cox ( 1986 ) . recently , based on the high resolution _ chandra_/acis observation of the galactic snr cygnus loop , levenson et al . ( 2002 ) reported evidence for the three distinct regions characterised as : ( i ) the forward shock decelerated by the clouds , ( ii ) the postshock snr interior , and ( iii ) the reflected shock that re - heated the postshock regions . in this context , we propose that the cool component of x - ray emission from n49 corresponds to the forward shock entering the dense clouds , the intermediate component to emission from the postshock region , and the hot component to emission from the re - heated postshock region , respectively . comparisons of the x - ray emission features with optical images indicate that the soft x - rays are spatially coincident with the bright optical knots while the hard x - ray emission is anti - correlated with the optical intensity distribution . since the bright filamentary optical emission features are believed to originate from the slow , cool radiative shock entering dense molecular clouds , the observed correlation / anti - correlation between x - ray and optical images provide good support for the suggested multi - phase shock interpretation . hester et al . ( 1994 ) have developed a simple 1-dimensional model in order to describe such a multi - phase shock structure of the nearby galactic snr cygnus loop , probably a `` magnified '' version of the distant snr n49 . our best - fit electron temperatures indicate that the ratios of temperatures between the re - heated postshock and the postshock without re - heating are @xmath52 = @xmath21.8 @xmath0 2.3 , which implies the preshock cloud density @xmath53 @xmath2 30 @xmath0 2500 @xmath54 based on the model by hester et al . ( 1994 ) , where @xmath54 is the preshock intercloud density . assuming the preshock intercloud density @xmath54 @xmath2 0.9 @xmath6 as estimated from the optical spectroscopy @xcite , we derive @xmath53 @xmath2 27 @xmath0 2300 @xmath6 for the interacting dense molecular clouds . this is in plausible agreements with the previous estimations based on optical / uv spectroscopy ( @xmath53 = 20 @xmath0 940 @xmath6 ; vancura et al . the best - fit @xmath34s in the eastern regions imply ( @xmath53/@xmath54)@xmath55(@xmath56/@xmath57 ) = 2 @xmath0 13 , where @xmath56 is the x - ray emission volume of the cool forward shock component and @xmath57 is the x - ray emission volume of the hot re - heated component ( we assume an x - ray emitting volume filling factor of @xmath21 , hereafter ) . provided that @xmath53/@xmath54 is in the order of @xmath210@xmath58 @xmath0 10@xmath59 , this implies @xmath56/@xmath57 @xmath28 0.01 . the soft x - ray emission from the cool forward shock is then from the blast wave entering small - volume , clumpy clouds while the hot component originates from the large - volume , re - heated intercloud postshock regions . for an adiabatic shock , the relations between the shock temperature and the shock velocity is @xmath60 = @xmath61 and can be written as @xmath62 where @xmath60 is the shock temperature , @xmath63 is the shock velocity , and mean atomic mass @xmath64 @xmath65 0.7@xmath66 ( @xmath66 = proton mass ) . the cool component shock velocity is then @xmath67 = 582@xmath68 km s@xmath9 , the hot component velocity is @xmath69 = 981@xmath70 km s@xmath9 , and the intermediate velocity is @xmath71 = 674@xmath72 km s@xmath9 . for comparisons , vancura et al . ( 1992 ) have derived a shock velocity of @xmath7 = 730 km s@xmath9 for the x - ray emission of n49 ( note : for simplicity , we assumed the electron and ion temperatures in equalibration in these calculations ) . the ionization timescale ( @xmath11 ) ratios between the cool and the hot components are ( @xmath11)@xmath73/(@xmath11)@xmath74 @xmath28 0.1 , which implies @xmath75 @xmath28 0.01@xmath76 , where @xmath75 is the ionization age of the shock entering the clouds and @xmath76 is the ionization age of the re - heated regions . assuming an sn explosion energy of @xmath77 = 10@xmath78 ergs , the snr radius of @xmath79 = 10 pc based on the x - ray image of n49 , and the preshock density @xmath54 = 0.9 @xmath6 , the implied sedov dynamical age @xcite , @xmath80 is @xmath2 6600 yr . assuming @xmath76 is on the order of @xmath80 or @xmath210@xmath81 yr , the measured @xmath11 ratios imply @xmath75 @xmath28 100 yr . this short ionization age suggests that the soft x - ray emission ( @xmath10 @xmath2 0.44 kev ) has been produced by a rapidly decelerated shock entering dense molecular clouds as was observed from the cygnus loop snr @xcite . the hard x - ray knot in the southwest beyond the snr blast wave shock front ( region 6 ; figure [ fig : fig6 ] ) is most likely associated with the snr rather than a background / foreground object . the x - ray spectrum is dominated by emission lines and can be described with a @xmath10 @xmath2 1 kev plasma with metal - rich abundances , which supports the shocked ejecta origin . the fitted plane - parallel shock model indicates a highly advanced ionization state ( @xmath11 @xmath25 @xmath29 10@xmath82 @xmath6 s@xmath9 ) and the spectrum may also be equally described with a thermal plasma in the cie . the high temperature and the large ionization timescale of this feature are also supportive of its origin as the high - speed ejecta fragments created at the time of the sn explosion . because of the psf degradation , the apparent angular extension of this feature ( @xmath25@xmath24 in radius ) is likely an artifact and the true morphology is uncertain with the current data . in order to estimate the electron density @xmath83 , we thus assume two extreme cases of a circular shape with a radius of 5@xmath24 ( for the apparent angular size and so 1.2 pc radius in physical size at the distance of @xmath84 = 50 kpc ) and 0@xmath215 ( for the case of an unresolved point - like source , and so the physical size of @xmath28 0.125 pc radius at @xmath84 = 50 kpc ) . the assumed range of the physical size of the knot ( @xmath20.1 pc @xmath0 1.2 pc in radius ) is reasonably consistent with the sn ejecta fragments detected in other galactc snrs of vela and tycho @xcite . assuming the distance of 50 kpc to the snr and a simple spherical geometry for the x - ray emitting volume , the best - fit @xmath34 implies @xmath83 of 2.3 @xmath6 to @xmath23 68 @xmath6 depending on the assumed emission volume . based on the best - fit abundances , we may derive the ejecta mass of @xmath20.001 @xmath85 @xmath0 0.03 @xmath85 . although this range of the ejecta mass is in reasonable agreements with other galactic sn ejecta fragments @xcite , we caution that these estimations are crude due to the large uncertainties in the fitted abundances and assumed emission volumes . the angular offset of the hard x - ray knot from the geometrical center of n49 is @xmath248@xmath24 . the spatial separation of this knot from the center of the snr is then @xmath86 = 11.6 @xmath87 ( @xmath88)@xmath9 pc , where @xmath87 is the distance to n49 in the units of 50 kpc , and @xmath89 is the angle between the line of sight and the moving direction of the knot . assuming that the sn explosion site , so the birth place of the observed ejecta fragments , is close to the geometrical center of the current x - ray remnant of n49 , the mean transverse velocity is @xmath90 = 1717 @xmath91 @xmath92 km s@xmath9 , where @xmath91 is the spatial separation of the knot from the geometrical center of the snr in the units of 11.6 pc , and @xmath93 is the age of the snr in the units of 6600 yr . the analysis on this hard x - ray knot is however substantially limited by the poor psf . follow - up on - axis _ chandra _ observations would be essential to unveil the detailed nature of this feature . chandra_/acis observations conclusively detect an x - ray point source at the position of sgr 0526@xmath066 within the boundary of snr n49 . the x - ray spectrum of this point source is dominated by continuum , which can be described with a power law model ( see [ subsubsec : sgr ] ) . this point source is thus confidently identified with the x - ray counterpart of quiescent phase sgr 0526@xmath066 . the single power law model indicates a soft spectrum with @xmath1 @xmath2 3 for sgr 0526@xmath066 , which resembles the typical spectrum of the axps ( mereghetti et al . 2002 and references therein ) , while the x - ray spectra from galactic sgrs are described with harder single power laws of @xmath1 @xmath2 2 @xcite . as x - ray spectra from the axps and sgrs can be described with the power law + black body models , we have made an attempt to fit the sgr 0526@xmath066 spectrum with such a multi - component model . with the off - axis gto data , the fitted black body temperature ( @xmath10 @xmath2 0.55 kev ) is consistent with typical axp / sgr spectra , and the improvement in the overall fit is also statistically significant based on the f - test . the power law component , with the addition of the black body component , however implies substantially harder spectrum ( @xmath1 @xmath2 1.7 ) than those of the axps ( @xmath1 @xmath2 3@xmath04 ; mereghetti et al . 2002 and references therein ) . nonetheless , the two component model is still in good agreements with galactic sgrs as they are fitted with @xmath1 @xmath2 1@xmath02 and black body temperature @xmath10 @xmath2 0.5 kev @xcite . we however note that this two component model is not required to describe the on - axis data . since the embedded confusion level caused by the background in the spectral model fitting is more severe for the off - axis data than the on - axis data , we consider that the presence of the black body component is inconclusive , although the improvement in the fit for the off - axis data is statistically significant with the additional black body component . recently , kulkarni et al . ( 2002 ) have reported the results from the spectral analysis of sgr 0526@xmath066 utilizing two _ chandra_/acis go observations : one ( obsid 747 ) of those two data sets was identical to that we uitilze in the current work . they found that the observed spectrum of the sgr was best described by a black body + power law model , which appears to be inconsistent with our results . although the origin of the inconsistency is unclear , we may attribute the descrepancy to the background estimations utilized in the spectral analysis . as kulkarni et al . ( 2002 ) pointed out , their sgr source spectrum apparently include residual emission from the diffuse snr , whereas such a contamination in our on - axis source spectrum appears to be negligible . the soft x - rays from the diffuse emission of the snr might have caused the slight differences in the results of the spectral analysis as presented by kulkarni et al . ( 2002 ) . we have also noticed that kulkarni et al . ( 2002 ) have not incorperated the cti and qe corrections in their data reduction . although the substantial contributions by the cti and the qe degradation effects on a point - like source in the observation taken in a relatively early phase of the _ chandra _ mission are unlikely , we speculate that potential contaminations from these artifacts in the spectral analysis of any acis data can not be completely ruled out . in addition , we would like to comment on the spectral analysis of some parts of the diffuse emission from n49 by kulkarni et al . ( 2002 ) . they have used a mekal model for such a spectral analysis , which should be inappropriate for the complicated diffuse snr emission as we presented in the current work . more complex , physically realistic models should be utilized in order for an appropriate spectral analysis of such data . the results from our simple spectral analysis may support the speculation that sgr 0526@xmath066 is a transition object bridging the axp and the sgr phases of neutron stars @xcite . extensive analyses with the on - axis observations , which are beyond the scope of the current work , would however be necessary to understand the detailed astrophysical characteristics of this object . for example , the measurements of the period ( @xmath94 ) and the period derivative ( _ p _ ) would be critical to determine the magnetic field strengh @xmath95 and the spin - down age @xmath96 of sgr 0526@xmath066 . this information can provide important clues on the nature of sgr 0526@xmath066 such as the identification with a high - field ( @xmath95 @xmath23 10@xmath97 g ) neutron star or the `` magnetar '' @xcite . the measurement of the spin - down age would also be useful to test the physical association of sgr 0526@xmath066 with snr n49 , which should provide important implications on the possible astrophysical connections between the sgrs and the axps ( e.g. , gaensler et al . 2001 ) . kulkarni et al . ( 2002 ) have recently reported a detection of the 8.04 s pulsation from sgr 0526@xmath066 by utilizing @xmath285 ks _ chandra_/acis observations . the derived magnetic dipole field strength is high ( @xmath95 @xmath2 7 @xmath29 10@xmath98 g ) and supports the magnetar interpretation . the estimated spin - down age ( @xmath94/2_p _ @xmath2 2000 yr ) is consistent with the dynamical age of snr n49 , which support the physical association between n49 and sgr 0526@xmath066 . the location of sgr 0526@xmath066 in the lmc and the likely association with n49 is also supported by the fitted columns : i.e. , the absorbing columns for n49 and sgr 0526@xmath066 are consistent . this sgr location may confirm the enormous energy release in the 1979 march @xmath13-ray burst . if sgr 0526@xmath066 is indeed the compact remnant of the sn explosion which has also created snr n49 and it was born near the center of n49 , we may estimate its transverse velocity . the angular offset of sgr 0526@xmath066 from the geometrical center of n49 is @xmath222@xmath24 , which implies that the spatial separation of the sgr from the center of the snr is @xmath86 = 5.3 ( @xmath88)@xmath9 pc at the distance of 50 kpc . assuming that sgr 0526@xmath066 is the neutron star born at the time of the sn explosion that produced n49 , and that the age of the sgr is @xmath96 @xmath2 6600 yr , the mean transverse velocity is @xmath90 @xmath23 787 km s@xmath9 . this would suggest that sgr 0526@xmath066 is a young , high velocity neutron star that is in a different class than typical axps instead of being the same class object as other axps , simply in a more evolved phase . snr n49 in the lmc has been observed with _ chandra_/acis . the high resolution `` true - color '' acis images reveal complex x - ray filamentary structures which represent the blast wave shock sweeping through the ambient ism and the shock interacting with the nearby dense molecular clouds . the overall features from the ew images of the detected elemental species indicate that soft line emission is enhanced in the south - east while hard line emission is enhanced in the west of the snr . these overall ew features are most likely dominated by the variations in the shock temperature and the ionization states of the plasma across the snr . the spatially - resolved spectral analysis of n49 suggests that the x - ray emission from the snr can be described with multiple - phase shocks representing the singly - postshock region , the re - heated postshock region by the reflected shocks , and the cool forward shock entering the dense molecular clouds . we detect a hard x - ray knot in the southwest of n49 beyond the snr shock boundary . the x - ray spectrum of this knot is emission line dominated and can be described with a metal - rich hot thermal plasma . this result implies that this hard knot appears to be ejecta fragments created at the time of the sn explosion that produced snr n49 . the psf degradation of the off - axis observation however restricts the data analysis of this feature and preclude resolving the angular structure of this knot . follow - up on - axis observations with a decent exposure might be necessary to address the detailed nature of this feature . the acis observations of n49 also firmly detect the x - ray counterpart of the quiescent phase of sgr 0526@xmath066 . the x - ray spectrum of sgr 0526@xmath066 is continuum dominated and can be fitted with a power law model . the implied photon index is @xmath1 @xmath2 3 . this power law spectrum is generally consistent with those from the axps although the observed @xmath13-ray burst of sgr 0526@xmath066 is a rare characteristic of typical axps . the authors thank j. dickel for providing his atca data . sp also thank a. bykov for a valuable discussion on the sn ejecta fragments . this work has been in parts supported by nasa under contract nas8 - 01128 for the _ chandra x - ray observatory_. jph acknowledges support from _ chandra _ grants go1 - 2052x , go2 - 3068x , and go2 - 3080b to rutgers . brandt , w. n. , alexander , d. m. , hornschemeier , a. e. , garmire , g. p. , schneider , d. p. , barger , a. j. , bauer , f. e. , broos , p. s. , cowie , l. l. , townsley , l. k. , burrows , d. n. , chartas , g. , feigelson , e. d. , griffiths , r. e. , nousek , j. a. , & , sargent , w. l. w. 2001 , aj , 122 , 2810 cline , t. l. , desai , u. d. , teegarden , b. j. , evans , w. d. , klebesadel , r. w. , laros , j. g. , barat , c. , hurley , k. , niel , m. , vedrenne , g. , estulin , i. v. , kurt , v. g. , mersov , g. a. , zenchenko , v. m. , weisskopf , m. c. , & grindlay , j. 1982 , apj , 255 , l45 mereghetti , s. , chiarlone , l. , israel , g. l. , & stella , l. 2002 , `` neutron stars , pulsars and supernova remnants '' , proceedings of the 270 . we - heraeus - seminar , bad honnef , january 2002 , eds . w. becker , h. lesch & j. trmper , mpe - report in press ( astro - ph/0205122 ) cccc o & 590 @xmath0 740 & 300 @xmath0 550 & 1200 @xmath0 1280 + fe l & 750 @xmath0 870 & 300 @xmath0 550 & 1200 @xmath0 1280 + ne ( he@xmath3 ) & 880 @xmath0 950 & 300 @xmath0 550 & 1200 @xmath0 1280 + ne ( ly@xmath3)/fe l & 980 @xmath0 1100 & 300 @xmath0 550 & 1200 @xmath0 1280 + mg & 1290 @xmath0 1550 & 1200 @xmath0 1280 & 1630 @xmath0 1710 + si & 1750 @xmath0 2100 & 1630 @xmath0 1710 & 2140 @xmath0 2280 + ccccccc 1 & 2.7@xmath99 & 0.52@xmath100 & 8.6@xmath101 & 2.01@xmath102 & 43.3 & 60 + 2 & 1.2@xmath103 & 0.65@xmath104 & 70.0@xmath105 & 2.79@xmath106 & 77.0 & 93 + 3 & @xmath4 2.5 ( 0.6 ) & 0.59@xmath107 & 31.5@xmath108 & 1.71@xmath109 & 59.9 & 56 + 6 & @xmath4 3.6 ( 1.1 ) & 1.04@xmath110 & @xmath111 41.0 ( 500 ) & @xmath4 0.27 ( 0.12 ) & 19.5 & 36 + ccccccc 1 & 0.90@xmath112 & 1.01@xmath113 & 0.53@xmath114 & 0.39@xmath115 & @xmath4 5.77 ( 1.91 ) & 0.17@xmath116 + 2 & 1.23@xmath117 & 1.80@xmath118 & 0.78@xmath119 & 0.86@xmath120 & 1.75@xmath121 & 0.25@xmath122 + 3 & 0.46@xmath123 & 0.30@xmath124 & 0.16@xmath125 & 0.19@xmath126 & @xmath4 4.02 ( 1.50 ) & 0.18@xmath127 + 6 & 5.71@xmath128 & 0.33 ( fixed ) & @xmath4 502.7 ( 1.11 ) & 5.92@xmath129 & @xmath4 1334.2 ( 3.98 ) & 0.94@xmath130 + cccccccc 4 & 3.9@xmath131 & 67.7 & 94 & soft & 0.47@xmath132 & 0.8@xmath133 & 3.81@xmath134 + & & & & hard & 1.08@xmath135 & @xmath111 14.6 ( 40.9 ) & 2.19@xmath136 + 5 & 3.6@xmath137 & 86.7 & 96 & soft & 0.41@xmath138 & 1.1@xmath139 & 19.83@xmath140 + & & & & hard & 1.41@xmath141 & @xmath111 9.3 ( 22.1 ) & 1.53@xmath142 +

the corot satellite supplied the scientific community with a huge data base of variable stars . among them the rr lyrae stars have intensively been discussed in numerous papers in the last few years , but the latest runs have not been checked to find rr lyrae stars up to now . our main goal was to fill this gap and complete the corot rr lyrae sample . we found nine unstudied rr lyrae stars . seven of them are new discoveries . we identified three new blazhko stars . the blazhko effect shows non - strictly repetitive nature for all stars . the frequency spectrum of the blazhko star corot 104948132 contains second overtone frequency with the highest known period ratio . the harmonic amplitude and phase declines with the harmonic order were studied for non - blazhko stars . we found a period dependent but similar shape amplitude decline for all stars . we discovered significant random period fluctuation for one of the two oversampled target , cm ori . after a successful transformation of the corot band parameters to the johnson _ v _ values we estimated the basic physical properties such as mass , luminosity , metallicity . the sample can be divided into two subgroups with respect to the metallicity but otherwise the physical parameters are in the canonical range of rr lyrae stars . [ firstpage ] stars : variables : rr lyrae stars : oscillations stars : interiors techniques : photometric space vehicles

1608.06418

the recent photometric space missions such as corot @xcite and _ kepler _ @xcite yielded long and quasi - uninterrupted time series with high precision for a huge amount of variable stars . although few people expected previously , many amazing discoveries have been made for classic high amplitude variables stars , as rr lyraes . among others these mission revealed the excitation of low amplitude modes for many blazhko rrab stars , one of which is responsible for the so - called period doubling phenomenon @xcite , while others show the famous @xmath00.61 period ratio of the rrc stars @xcite or just appear as an additional mode(s ) resulting in non - traditional double and triple mode stars @xcite . the corot satellite was one of the most successful in these discoveries . up to now about ten peer - reviewed scientific articles published corot rr lyrae results . the paper of @xcite gives a good overview of these . previous works discussed 13 stars observed in long runs only and till the end of the fourth long run towards the galactic center direction ( lrc04 ) . neither the short nor the late runs were investigated for searching rr lyrae stars . we make up for this incompleteness with the present work . @cccccccccc@ corot i d & name & @xmath1 & ra & dec & @xmath2 & @xmath3 & @xmath4 & @xmath5 & run i d + & & [ mag ] & hh : mm : ss.ss & d : mm : ss.s & [ d ] & [ mag ] & [ d ] & [ d ] & + 102326 * 020 * & & 14.654 & 06:10:59.87 & @xmath64:40:32.5 & 0.77029 & 0.1296 & 79.5 & 148.31 & lra03 + 104948 * 132 * & & 15.217 & 18:37:43.63 & @xmath64:01:44.2 & 0.58642 & 0.2420 & 28.1 & 87.26 & lrc05 + 205924 * 190 * & & 13.600 & 18:57:17.45 & @xmath74:16:39.4 & 0.72049 & 0.1053 & & 20.86 & src02 + 605307 * 902 * & & 15.170 & 06:08:42.38 & @xmath66:22:15.1 & 0.62418 & 0.1559 & & 77.6 & lra04 + 617282 * 043 * & cm ori & 12.889 & 06:03:54.87 & @xmath68:14:32.4 & 0.65593 & 0.2859 & & 90.48 & lra05 + 651349 * 561 * & & 14.708 & 19:14:03.69 & @xmath72:27:54.8 & 0.61179 & 0.1013 & 21.9 & 83.51 & lrc09 + 655183 * 353 * & & 14.911 & 19:14:41.21 & @xmath72:06:09.1 & 0.69426 & 0.1921 & & 83.51 & lrc09 + 657944 * 259 * & & 14.954 & 19:18:08.54 & @xmath72:34:43.9 & 0.57787 & 0.1876 & & 83.51 & lrc09 + 659723 * 739 * & v2042 oph & 15.186 & 18:32:46.97 & @xmath67:58:05.7 & 0.53849 & 0.2761 & & 81.22 + 83.47 & lrc07+lrc10 + since hundreds of papers appeared on the basis of the corot satellite data the basic facts about the satellite and its equipment are well - known , we only refer here the most important instrumental paper @xcite and on - line documentations ] where the reader can find the details of the mission . this work used the corot public archive n2 data of exo - fields . for simplicity , we used the entire data base in our rr lyrae search including the initial run ( ira01 ) , long runs ( lra01-lra07 , lrc01-lrc10 ) , and short runs ( sra01-sra05 , src01-src03 ) . here the letters ` c ' and ` a ' denote that the corot observed either toward the galactic centre or anti - centre directions , respectively the advantage of the use of the total observing material is that we can check our search efficiency when we compare our candidates in the early runs with the rr lyraes found by former studies . we used the automatic corot variable classifier ( cvc ; @xcite ) when we searched for rr lyrae stars separately for each subtype in the complete data base . the cvc has improved a lot since the beginning . our present search results illustrate this well : if we decrease the probability level assigned by cvc of rrab stars from @xmath850% to @xmath810% , then to @xmath80% , we found 15 , 16 and 18 candidates , respectively . investigating the candidates by visual inspection only one of them proved to be non - rr lyrae stars in the lowest probability group . all others ( 17 ) show clear rr lyr light curve properties . eight rrab stars were discussed by former studies ( @xcite , and references therein ) . two additional rrab stars were known in the corot sample ( corot 101315488 and corot 100881648 ) which have not been found by cvc . these stars are , however , heavily blended by nearby stars and therefore their variability have much smaller amplitude and distorted light curves compared to a normal rr lyrae star . the corot data of nine rrab stars have not been discussed anywhere yet . these stars constitute the present sample ( see table [ tab1 ] ) . [ zoo ] shows small light curve parts of each star . the rr lyrae nature of two stars : cmori ( corot 617282043 ) and v2042oph ( corot 659723739 ) have already been known but the other seven ones are completely new discoveries . the same process ( cvc with decreasing probability levels and visual light curve examination ) has been carried out for rrc subtype as well . for rrc stars we get 25 , 34 and 137 candidates for @xmath850% , @xmath810% , and @xmath80% probability levels , respectively . the two known rrc stars ( corot105036241 and corot105735652 ) were already contained in the highest probability group . the increasing candidate numbers of the lower probability level groups did not result in new rrc stars . the groups consist mostly of binary stars and some spotted and hads / sx phe candidates . in the case of double mode rr lyrae stars ( rrd ) the cvc yields different kind of results . for @xmath850% and @xmath810% probability levels we get only one candidate ( corot101368812 ) which is known as a bona fide rrd star @xcite , while for @xmath80% probability level the number of candidates abruptly increases to 7650 . we applied a semi - automatic method to handle this large number of light curves . we calculated the fourier amplitude spectra in the 110d@xmath9 range of all the 7650 light curves . we selected all peaks from these spectra that exceeded the 4@xmath10 amplitude , where @xmath10 was calculated as the 3d@xmath9 moving average of the amplitude spectrum . the classical double mode rr lyrae stars have a well - known period ratio between their radial fundamental and first overtone modes @xmath110.72 - 0.755 d@xmath9 which defines the petersen diagram ( see e.g. in @xcite ) . therefore , we selected those objects that have their highest peak @xmath12 in the 15d@xmath9 range and that have at least one further significant peak in either the @xmath13@xmath14 or the @xmath15@xmath16 range . these are those stars that have frequency ratio in the 0.720.755 range with the largest - amplitude peak . this selection yielded 113 objects . we investigated the light curves and fourier spectra of these 113 objects visually . apart from the known corot101368812 , we did not find any further rrd pulsator . note that 61 targets were identified as @xmath17dor candidates and 42 as eclipsing / heartbeat binary candidates . from now on , this paper refers corot stars with the last three digits ( boldfaced in the first column in table [ tab1 ] ) of their i d number for short notation . after we extracted the downloaded data files we applied trend and jump filtering , outlier removal , and a transformation into the magnitude scale as we described in detail in @xcite . two of our targets ( cm ori and # 190 ) were bright enough to be observed by corot in colour mode , but because of the uniform handling , we also used their integrated ( white light ) fluxes only . the nominal sampling is 512 sec for all the stars , except cm ori , where the the light curve was observed with the much denser oversampling mode ( 32 sec ) . the data set of cm ori is twice as dense as the _ kepler_/k2 short cadence observations . the main tool was a standard discrete fourier analyzer implemented by the program package mufran @xcite . it yields sinusoidal fourier decomposition , therefore , we used this one through this paper . the nyquist frequency of the normal and oversampled data are 84.375 d@xmath9 , and 1350 d@xmath9 , respectively . the stability of the periods were checked by o@xmath7c diagrams @xcite . the o@xmath7c diagrams were constructed from the maximum brightness times . the proper maximum times were determined by light curve maxima with 7 - 10 order polynomial fits . the fourier spectra of the analyzed nine stars are dominated by the main pulsation frequencies and their harmonics . simultaneous non - linear fits to the light curves with these frequencies resulted in the pulsation periods @xmath2 and the amplitudes @xmath3 given in table [ tab1 ] . , after we pre - whitened the data with @xmath18 and all significant harmonic frequencies . the inserts show narrower ranges @xmath19 of the spectra after a further pre - whitening step in which we removed all the significant modulation side frequencies as well . ] * # 020 * : the light curve shows a slight amplitude change ( see the top left panel in fig . [ bl_fig ] ) . after we pre - whitened the data with the main pulsation frequency and its significant harmonics we found significant peaks in the residual spectrum around the pre - whitened frequency positions . we can identify them as blazhko modulation frequencies as @xmath20 , ( @xmath21 is positive integer ) . the blazhko period @xmath22 given in table [ tab1 ] was calculated from the average of the side peak differences around the main peak . the side peak amplitudes are highly asymmetric , the right hand side peaks are much higher than the left hand side ones ( see the top panel in fig . [ fourier ] ) . this asymmetry , however , decreases with the increasing harmonic order : while the @xmath23 and @xmath24 are around the detection limit , @xmath25 and even @xmath26 are much less asymmetric . this behaviour has a natural explanation in the simultaneous amplitude and frequency modulation framework @xcite . in the o@xmath7c diagram ( top right in fig . [ bl_fig ] ) we indeed find a periodic signal with 80.7 d@xmath9 period and @xmath05 min amplitude which can be identified as the frequency modulation part of the blazhko effect . the pulsation period of the star # 020 is one of the longest ( @xmath27 d ) ever found for a blazhko rr lyrae star . its blazhko period is long enough to test the relation between the pulsation period and blazhko period discovered by @xcite . they found that the blazhko frequency , which was represented by the frequency separation between the main pulsation frequency and the side frequencies , has a maximal allowed value at a given pulsation frequency . these maximal values form a linear upper envelope curve in the pulsation frequency vs. blazhko frequency diagram ( see fig . 6 in @xcite ) . the position of # 020 is under this upper envelope line supporting the finding of the paper . both the amplitude and frequency modulation parts of the two observed blazhko cycles differ from each other ( top panels in fig . [ bl_fig ] ) . this raises the possibility of multiperiodic ( or chaotic ) nature of the modulation , but such a short observing span does not allow us to confirm or reject these hypotheses . * # 132 * : the light curve in the middle left panel in fig . [ bl_fig ] shows three clear blazhko cycles with decreasing amplitudes . the fourier spectrum contains evident triplet structures around the main pulsation frequency and all its harmonics . the low - frequency range of the spectrum contains three significant frequencies : @xmath28 d@xmath9 , @xmath29 d@xmath9 and @xmath30 d@xmath31 . the frequency @xmath32 can be interpreted as the blazhko frequency @xmath33 itself , the @xmath34 most probable belongs to the total time span , and @xmath35 is a linear combination of these two frequencies . after the first two pre - whitening steps in which we removed the main pulsation frequency , its harmonics and all significant blazhko side peaks the residual spectrum shows two groups of significant peaks between the harmonics ( middle panel in fig . [ fourier ] ) . these peaks are most robust between the main pulsation frequency and the first harmonic . the highest peak is at @xmath36 d@xmath9 while the second one is at @xmath37 d@xmath9 . the latter frequency can be identified as @xmath38 which belongs to the period doubling phenomenon . what about the frequency @xmath39 ? numerous studies @xcite found frequencies at this position : between the period doubling ( pd ) frequencies and the first overtone . these frequencies are explained by the @xmath40 second radial overtone mode or a non - radial mode which is excited close to the position of the second overtone mode . # 132 has the highest period ratio @xmath41 ever detected . the identification of the @xmath42 is verified with theoretical model calculations . [ dm_model ] shows a theoretical petersen diagram where the periods @xmath2 , @xmath43 , @xmath44 , @xmath45 , @xmath46 belong to the radial fundamental , first , second , third and fourth overtone modes , respectively . the different metallicity models are noted by different symbols . the parameters and details of the models are the same as it was described in @xcite . as we see the ratio of @xmath39 is in the allowed range of second overtone frequencies . that is the fourier spectrum of # 132 shows the pd and the radial second overtone mode frequencies simultaneously similarly to some other stars in the corot and _ kepler _ samples ( e.g. corot 101128793 @xcite ; v355 lyr , kic 7257008 , and kic 9973633 @xcite ) . the o@xmath7c diagram in the middle right panel in fig . [ bl_fig ] shows evident frequency modulation . the fourier spectrum of the o@xmath7c diagram contains @xmath33 and @xmath47 reflecting the non - sinusoidal nature of the variation . the consecutive blazhko cycles are different but do not show declining amplitudes as opposed to the light curve . the strictly mono - periodic nature of the blazhko effect , similarly to the case of the star # 020 , can be ruled out both on the basis of the light and o@xmath7c curves . * # 561 * : both the light and o@xmath7c curves ( bottom panels in fig . [ bl_fig ] ) show the blazhko effect . the amplitude modulation seems to be more regular than the frequency modulation part . the modulation side peaks in the fourier spectrum of the light curve are asymmetric : left hand side peaks have lower amplitudes than the right hand side ones . after we removed all the significant side frequencies , no additional frequencies have been found in the residual spectrum . interestingly , while all stars that show additional frequencies show the blazhko effect as well @xcite , the reverse statement is not true . there are known some corot @xcite and _ kepler_/k2 @xcite blazhko stars without any additional frequencies . as we show here the stars # 561 and # 020 seem to belong to this group . we can speculate whether this finding is a selection effect or a real feature . the selection effect explanation is supported by the time dependence of these additional frequencies . they have low and highly time dependent amplitude . so these frequencies could disappear from the frequency content for a while and then emerge again . c diagrams of cm ori ( top ) and v2042 oph ( bottom ) . the ( red ) plus symbols denote the historical data the ( blue ) x signs are the corot measurements . the green continuous curve in the top panel shows the quadratic fit to the data . the ( purple ) triangles in the bottom figure mean the residual o@xmath7c data after we removed the linear trend showed by green continuous line . the right hand side vertical scale in the bottom panel belongs to the residual values , while the dashed curve shows the quadratic fit to theses data [ o - c_long],title="fig : " ] c diagrams of cm ori ( top ) and v2042 oph ( bottom ) . the ( red ) plus symbols denote the historical data the ( blue ) x signs are the corot measurements . the green continuous curve in the top panel shows the quadratic fit to the data . the ( purple ) triangles in the bottom figure mean the residual o@xmath7c data after we removed the linear trend showed by green continuous line . the right hand side vertical scale in the bottom panel belongs to the residual values , while the dashed curve shows the quadratic fit to theses data [ o - c_long],title="fig : " ] as we noticed two of our stars ( cm ori and v2042 oph ) were known rr lyrae variables . cm ori was discovered early by @xcite while v2042 oph was found in the sonneberg plates by @xcite . the stars were targets of maxima observation from time to time . the geos rr lyrae data base @xcite contains 13 and 17 maxima values for cm ori and v2042 oph , respectively . we completed the data set of cm ori with three new maxima observed by tarot survey @xcite and the current corot observations , while we did not use maxima of @xcite because of their large uncertainties ( @xmath487 min ) . the constructed long - term o@xmath7c diagrams can be seen in the top panel in fig . [ o - c_long ] . the o@xmath7c points distribute along a parabolic function suggesting a continuous period increase . we fitted the simple quadratic formula @xmath49 where @xmath50 are constants , @xmath51 is the linear period change rate , @xmath52 is the time . the rate of the period increase is @xmath53 dd@xmath9 , or , @xmath54 dmy@xmath9 . the o@xmath7c diagram of v2042 oph ( bottom panel in fig . [ o - c_long ] ) consists of the geos data and points from the two corot runs . we shifted the 512 sec corot date with 224 sec as it was described by @xcite . the o@xmath7c diagram suggests that the geos period ( @xmath2=0.5385 d ) is too long . from the slope of a fitted linear we get @xmath55 d correction , which is good agreement with our new corot period ( @xmath56 ) . applying this correction ( viz . subtracting the fitted linear shown by the continuos line ) we detect a period increase again . from the quadratic fit ( dashed curve ) the period change rate is @xmath57 dd@xmath9 , or , @xmath58 dmy@xmath9 . these period change rates are in agreement with the canonical stellar evolution models @xcite which predict the red - ward evolution rate between @xmath59 and 10@xmath60 dd@xmath9 , where @xmath61 . in our cases @xmath62 , and 7.2@xmath60 dd@xmath9 for cmori and v2042oph , respectively . these actual values are also fit well to the empirical rate distributions determined for the globular cluster m3 by @xcite . after the investigation of the long term period changes we turn to the cycle - to - cycle stability of the non - blazhko stars . @xcite performed a stability analysis for the _ kepler _ sample . they calculated the time dependence of the main fourier parameters , namely @xmath63 , @xmath64 , @xmath65 , @xmath66 . all these functions proved to be constant with a small random scatter . @xcite followed a different approach when they found random period fluctuation in the _ kepler _ cepheid v1154cyg . they studied the o@xmath7c diagram , which suggested the cycle lengths scatter of 0.015 - 0.02 d ( @xmath6720 - 30 min ) , meaning @xmath68 of the pulsation period . similar magnitude of cycle - to - cycle light curve fluctuation would be 1 - 2 minutes for a typical rr lyrae stars . we have a chance to detect such a small time - scale effect only for the oversampled stars . two corot rrab stars cm ori and corot103800 * 818 * @xcite were observed in oversampled mode . here we investigate the data sets of these two stars in details . c diagrams of cm ori ( top ) and # 818 ( bottom ) constructed from the phase shifts . black dots represent the observed data , red triangles denote the synthetic data ( see the text for the details ) . for better visibility we show the error bars as separated symbols and we connected the consequtive points . we call the attention of the reader to the different vertical scales . [ o - c_short],title="fig : " ] c diagrams of cm ori ( top ) and # 818 ( bottom ) constructed from the phase shifts . black dots represent the observed data , red triangles denote the synthetic data ( see the text for the details ) . for better visibility we show the error bars as separated symbols and we connected the consequtive points . we call the attention of the reader to the different vertical scales . [ o - c_short],title="fig : " ] first we constructed the traditional o@xmath7c diagrams from the light curve maxima . both diagrams are constant lines with a standard deviation of @xmath69 d , and 0.00132 d for cm ori and # 818 , respectively . in the case of cm ori this scatter means 0.88 min , or , 0.09% of the period , while for # 818 these values are 1.9 min and 0.3% . to test the accuracy of this o@xmath7c method we prepared synthetic light curves using the fourier solutions ( amplitudes and phase ) of the real stars and adding to them gaussian noise . the noise was set that the rms of the synthetic light curve fits became equal to the measured light curves : 0.0037 mag for cm ori , and 0.0068 mag for # 818 , respectively . the standard deviation of the o@xmath7c diagrams of these stationary periodic synthetic data are @xmath700.00058 d ( cm ori ) and @xmath700.00121 d ( # 818 ) . since we recovered the observed scatters our conclusion is that the random fluctuation of the pulsation period is lower than the detection limit of this method . secondly a more sensitive method was used to measure phase shifts of each pulsational cycle and transform this to an o@xmath7c diagram . this approach uses the complete light curve instead of small parts around the maxima and could result in more precise o@xmath7c values than the traditional method . the method is the same as it was successfully used by @xcite for detecting the random period changes of the _ kepler _ cepheid v1154 cyg . a template curve was defined by fitting 36th order fourier polynomial to the phase diagram of one pulsational cycle . then we used it allowing only vertical and horizontal shifts to fit each phase in the light curve . combining the phase shifts and the period with the epoch , we calculated the o@xmath7c values which are plotted in fig . [ o - c_short ] with black dots . the same process has been done for the above mentioned synthetic data as well . the results are shown in fig . [ o - c_short ] with red triangles . the red and black symbols show different curve shapes for both stars . this difference is significant for cm ori ( top panel in fig [ o - c_short ] ) but it is also less than 0.0008 d ( 1.2 min ) . what is the reason of this deviation ? a frequency modulation ( blazhko effect ) is unlikely because no side peaks have been detected around the harmonics in the fourier spectrum . the rather irregular shape of these o@xmath7c curves rule out the binarity explanation . the most natural explanation is the intrinsic random period fluctuation . in other words : the rr lyrae stars are also not precise clocks . due to the cumulative nature of the o@xmath7c diagram we see here a random walk . a detailed discussion of the phenomenon and its appearance in o@xmath7c diagrams is given in @xcite . a possible irregular period jitter in cepheids and rr lyrae stars was proposed by several authors @xcite on different theoretical grounds . but up to now it has not been detected yet for rr lyrae stars . it is possible that our detected small period changes of cm ori are typical . the 1.2 min cumulative period change means no more than 1 - 2 seconds differences between the length of the subsequent pulsation cycles . such a small difference could be detect only in precise uninterrapted high cadence data . . ] the fourier amplitude of the harmonics decreases with the harmonic order . previously this decline was believed to be exponential but ( for at least non - blazhko stars ) it was not studied . satellite data which provide us high number of significant harmonics showed that this nearly exponentially decay can not necessarily be described by a single exponential function . both of the studied non - blazhko rrab stars ( corot101370131 , @xcite , corot 103800818 @xcite ) show similar behaviour : their amplitudes decline exponentially for a while then this decline shows a bump then the amplitude decreases again exponentially but with a smaller exponent than with the lower order harmonics . in this section we investigate how general this feature is . we mention here that the amplitude decline of the blazhko stars were investigated in several papers ( e.g. @xcite ) . these works concentrated mostly on the different declines of the harmonics and side peak amplitudes . recently @xcite found a hyperbolic fit for the harmonic decline to be better than the exponential ones . in the case of blazhko stars , however , the amplitudes of the harmonics are heavily affected by the frequency / phase modulation part of the blazhko effect @xcite . in other words , in the amplitude decline behaviour the basic physics of the pulsation and unknown aspects of the blazhko effect are mixed . to avoid this complication we restricted our study to the non - blazhko stars . our sample contains six non - blazhko rr lyrae stars . we completed this sample with the three formerly discussed non - blended corot stars ( corot 101370 * 131 * , corot 103800 * 818 * , corot104315 * 804 * , @xcite ) . the used fourier parameters @xmath71 and @xmath72 of these stars are given in an electonic - only table , the amplitude in corot magnitude , and the phase in radian . ] . if we plot the fourier amplitude ratios @xmath73 as a function of the harmonics @xmath74 for these nine corot non - blazhko stars we get fig . [ amp ] . in this logarithmic plot we see that the monotonic decay holds to the 5 - 15th harmonic order . the exact value varies from star to star . the distribution curves seem to build up a sequence from the stars # 804 , where only a slight break appears at the 5th harmonic , to the star # 818 , where a deep minimum exists at around the 15th harmonic . it can also be realized that the depth of the dip tends to increase with the increasing harmonic order . this trend seems to break at # 739 , but its dip is defined by only one small ( marginally significant ) amplitude of the 12th harmonic frequency , the extraordinarily deep minimum might cause this uncertainty . the order of stars settled by the diagram in fig . [ amp ] is almost the same as the order of decreasing pulsation period . the longest period non - blazhko star of the sample is # 804 ( @xmath75 d ) , while the shortest one is the star # 818 ( @xmath76 d ) . seven stars from the nine one follow this period length order . the two exceptions are # 902 and # 259 which are located in longer period positions of the sequence than their real period . the fourier amplitudes and phases collectively contain the complete information of a signal . although this is well - known , the phase spectra are studied much less frequently than the amplitude ones . in the right - hand - side panel of fig . [ amp ] we show the epoch independent phase differences @xmath77 of the nine studied rrab stars . the panel shows the phase differences according to the definition : @xmath78 ( @xmath74 integer ) , where we did not transform the differences to a restricted interval as it is common . the reason is illustrated with the fig . [ phase_traditional ] where we transformed the first seven @xmath77 quantities into the @xmath79 interval . here the @xmath80 distributions of different stars are mixed . the distributions are hardly traceable . by looking at the right panel in fig . [ amp ] we notice that these @xmath80 distribution functions are deviate from each other . the distance of the nearby functions increases with the increasing harmonic order forming a fan shape structure on the @xmath81 surface . this spread suggests that the fine structure of the individual stars light curve differ more than their main character described with the low order harmonics . the shape of these @xmath80 decline functions vary form the middle of the fan - shape distribution to the wings : from the pure linear ( # 043 ) to intensely crisscross shapes ( e.g. # 259 , # 190 ) , respectively . the fine structures of the light curves such as the amplitude distribution of the higher harmonics have never been studied theoretically . @xcite estimated fourier parameters @xmath82 and @xmath77 for @xmath83 using a simple non - adiabatic one - zone pulsation model . they did not investigate the amplitude ratio and phase difference distributions with respect to the harmonic orders , but their figures fig . suggest monotonic declining amplitudes with the increasing @xmath74 , except some model light curves , where their ` acuteness ' parameter is small . in these cases the higher order amplitudes ( @xmath84 ) could be higher than the lower ones . the phase plots in their fig . predict monotonic increasing phase differences ( within the interval @xmath85 ) at least until @xmath86 , which is evidently not the case for the observed light curves ( see fig . [ phase_traditional ] ) . the amplitude ratios and phase differences of theoretical light curves were investigated by @xcite using a more enhanced non - linear convective 1d pulsation approach but their calculations extended up to the 5th harmonic order only . the more realistic physical treatment gives more realistic predictions for phase differences ( see fig . 8 in the paper ) , @xmath87 ) for a direct comparison to our work . ] . reading the related values from the different panels we can recognize a similar crisscross behaviour of the phase difference distribution as we see in our fig [ phase_traditional ] . ( top panel ) and epoch independent phase differences @xmath77 vs. harmonic orders @xmath74 ( bottom panel ) of some synthetic light curves.,title="fig : " ] ( top panel ) and epoch independent phase differences @xmath77 vs. harmonic orders @xmath74 ( bottom panel ) of some synthetic light curves.,title="fig : " ] we compared the fourier amplitude and phase distribution of the observed and model light curves calculated from the photospheric temperature @xmath88 and gravitational acceleration @xmath89 curves by the florida - budapest hydrocode @xcite . since the models use the exact black - body condition we can use these photospheric quantities instead of the effective ones . we investigated three different model light curves denoted by a , b and c. the input parameters of the model a are : @xmath90 m@xmath91 ( stellar mass ) , @xmath9240 l@xmath91 ( luminosity ) , @xmath93 k ( the effective temperature of the input static model ) , z=0.0001 ( metal content ) . for model b these parameters are : @xmath94 m@xmath91 , @xmath9250 l@xmath91 , @xmath95 k , z=0.004 ; and model c : @xmath96 m@xmath91 , @xmath9240 l@xmath91 , @xmath95 k , z=0.004 . for the direct comparison we have to prepare synthetic corot light curves from the model outputs : @xmath97 , @xmath89 . using the basic formula of synthetic photometry @xcite we determine the detected stellar fluxes in photon numbers @xmath98 as @xmath99 where @xmath100 is the spectrum of a star in energy units , @xmath101 is the response function of the system , @xmath102 is the filter function , @xmath103 is the wavelength , @xmath104 is the planck constant and @xmath105 is the speed of light . the flux was computed by using the corot equipment s spectral response function @xmath101 given by @xcite and without filter function ( @xmath106 ) . let us denote the result of this computation by @xmath107 , and the artificial corot magnitude is @xmath108 , where @xmath105 is an arbitrary constant . we used the enhanced kurucz models @xcite provided by the spanish virtual observatory as theoretical flux distribution functions @xmath100 . a set of model spectra ( 360 synthetic spectra ) were chosen which cover the physical parameters ( @xmath109 , @xmath110 ) of different pulsation phases and the metallicity range of the rr lyrae stars , namely , @xmath109 is between 5750 and 8000 k , @xmath110 2.5 and 5.0 , while [ fe / h ] is between 0 and @xmath72.5 dex . although these static model atmospheres are not optimal for all rr lyrae pulsation phases ( see eg . @xcite ) without available , adequate dynamical model atmospheres we can use them as a first approximation . by using @xmath111 and @xmath89 functions from the pulsation model we assigned an interpolated @xmath112 value to each pulsation phase . generating the artificial corot light curves by this way corresponds to a phase dependent bolometric correction which was done by @xcite for _ v _ band . the fourier solutions of the synthetic light curves are presented in fig . [ amp_phase_modell ] in the same form as the observed ones in fig , [ amp ] . the amplitude distribution of model a in top panel show similar decline to the observed curves , however , the higher metallicity models b and c show double bumps . we have not seen such behaviour for any of the real stars . it is not clear , which one of the two dips correspond to the observed single dip . the phase difference distributions in bottom panel of fig [ amp_phase_modell ] are phenomenological similar to the observed data ( left panel in fig . [ amp ] ) , but the numerical @xmath77 values of model a are unusually high . we conclude that recent 1d hydrodynamic models can not reproduce the fine structure of the observed light curves of the rr lyrae stars even in the simplest non - modulated fundamental mode pulsating case . maybe the enhanced versions of the first successful multidimensional computations ( 2d/3d @xcite ) will provide us better synthetic light curves . for a perfect agreement between observation and theory we probably have to wait for sophisticated dynamical atmosphere models as well . we call the readers attention to the diagrams in fig . [ amp ] , which presents a potentially sensitive tool for quantitative study of the fine structures of rr lyrae light curves . the standard photometric _ v _ light curves of rr lyrae stars allow us to estimate their physical parameters such as metallicity , effective temperature , @xmath110 , mass , etc . by using empirical formulae ( e.g. @xcite ) . however , neither chromatic nor monochromatic corot light curves are calibrated to standard photometric systems . in principle , this calibration is possible , but no one has published such attempts yet . the basic idea of such a transformation along with some of its inherent problems is discussed in @xcite . similar transformation problems were successfully solved by @xcite who had to deal with light curves in _ kepler _ @xmath113 band . they compared the _ light curves with the existing johnson _ v _ light curves of three non - modulated rrab stars , then they determined numerical shifts for all useful fourier parameters . using these transformed parameters and formulae for the _ v _ filter they received noticeably good estimates for the physical parameters which were then verified by high - resolution spectroscopy @xcite . a different approach was developed by the ogle team . since the ogle - iii rr lyrae data @xcite are predominantly observed in cousins @xmath114 colour , empirical formulae for _ i _ filter data were needed for physical parameter estimation . the ogle team independently calibrated the empirical relations @xcite instead of transforming the existing formulae relevant for the _ v _ passband . a complete calibration of corot coloured and monochromatic data is out of the scope of this paper . instead of this we use two alternate transformations achieving the necessary data . our primary goal is to check the physical parameters of our sample and to select special stars if there are any for further investigation . for this purpose a rough physical parameter estimation is enough which can be available through these approximate transformations . \(1 ) due to the direct analogy of corot monochromatic and _ kepler _ unfiltered @xmath113 data we applied the method used by @xcite . unfortunately , _ v _ observations of corot rr lyrae stars are even more sparse than kepler stars measurements . we found one ( almost ) complete phase curve for corot 101370131 in @xcite . fourier parameters were obtained from superwasp data by @xcite for cm ori . from these two data sets we derived the two most important fourier parameters : @xmath115 where @xmath116 denotes the fourier amplitude of the main pulsation frequency , @xmath117 epoch independent phase difference is defined in the usual way : @xmath118 @xcite , where @xmath119 and @xmath120 are the fourier phases of the main frequency and the second harmonic . the upper indices denote the used filter system . ) and synthetic johnson _ v _ ( @xmath121 ) magnitudes . the insert shows a zoom from the plot demonstrating the weak dependence of the relation on secondary parameters ( metallicity , @xmath110 ) . ] \(2 ) because of the evident weaknesses of the previous transformation we alternatively applied the synthetic photometry approach ( eg . this method allow us to transform the corot monochromatic rr lyrae observations to the standard johnson _ v _ light curves to which we can apply the empirical relations directly . following the basic idea of synthetic photometry @xcite we determine the detected stellar fluxes in photon numbers @xmath98 as it was shown in eq . [ np ] . in our case , the flux was computed twice , once by using the corot equipment s spectral response function @xmath101 given by @xcite and second without the filter function . let us denote the result of this computation by @xmath107 . second , by using the same @xmath101 function combined with the bessell _ v _ filter function @xmath102 @xcite we get @xmath122 . we utilized the same set of model spectra ( 360 synthetic spectra ) was used for the computation of synthetic corot light curves in sec . [ sec : amp_phi ] . we correlated the synthetic corot magnitudes @xmath123 to the synthetic _ v _ magnitudes : @xmath124 . the average of all our light curves transformed to the magnitude scale are set to zero , that defines a kind of ` relative ' magnitude . by using these light curves we avoid the need to determine the absolute zero point of the synthetic photometry which is generally a difficult problem . here we can set the constants @xmath125 and @xmath126 arbitrarily . we have chosen these so that the resulted magnitude ranges are similar to the observed amplitudes of rrab stars and the zero point of the fitted line is zero . the correlation between the synthetic corot and the _ v _ magnitudes is shown in fig . [ corot - v ] . the best - fitted linear relation is @xmath127 the insert in fig . [ corot - v ] demonstrates the dependence of the relation on physical parameters such as metallicity , or @xmath110 . this virtual ` scatter ' is an intrinsic physical property of the stars and not a scatter due to the uncertainty of our method . using the ratio in eq . [ fit ] we derived the transformed _ light curves , from which we deduced the fourier parameters ( amplitudes and phase differences ) necessary to use the empirical formalae . @ccccccccc@ corot i d & [ fe / h]@xmath128 & @xmath129 & @xmath130 & @xmath110 & @xmath131 & @xmath132 & @xmath133 & rem . + & & [ mag ] & [ mag ] & & [ k ] & & [ m@xmath91 ] & + 100689 * 962 * & @xmath70.75,@xmath70.65 & 0.867,0.900 & 0.314,0.324 & 3.023 & 6730,6694 & 1.519,1.524 & 0.71,0.75 & bl , @xmath135 + 101128 * 793 * & @xmath70.75,@xmath70.64 & 0.737,0.767 & 0.326,0.335 & 2.872 & 6632,6600 & 1.546,1.534 & 0.52,0.52 & bl , @xmath135 + 101370 * 131 * & @xmath71.32,@xmath71.22 & 0.539,0.568 & 0.347,0.355 & 2.728 & 6436,6407 & 1.674,1.663 & 0.57,0.57 & @xmath135 + 102326 * 020 * & @xmath70.66,@xmath70.56 & 0.501,0.538 & 0.395,0.407 & 2.612 & 6248,6207 & 1.687,1.667 & 0.52,0.51 & bl , @xmath135 + 103800 * 818 * & @xmath70.97,@xmath70.86 & 0.675,0.699 & 0.305,0.312 & 2.879 & 6714,6694 & 1.578,1.564 & 0.54,0.53 & + 103922 * 434 * & @xmath71.22,@xmath71.11 & 0.612,0.641 & 0.330,0.338 & 2.799 & 6547,6520 & 1.627,1.612 & 0.56,0.55 & bl + 104315 * 804 * & @xmath70.55,@xmath70.45 & 0.557,0.594 & 0.387,0.399 & 2.646 & 6304,6263 & 1.661,1.641 & 0.51,0.50 & @xmath135 + 104948 * 132 * & @xmath70.97,@xmath70.87 & 0.605,0.635 & 0.344,0.353 & 2.757 & 6492,6461 & 1.635,1.619 & 0.54,0.53 & bl + 105288 * 363 * & @xmath71.13,@xmath71.02 & 0.643,0.677 & 0.354,0.364 & 2.775 & 6435,6398 & 1.613,1.595 & 0.55,0.54 & bl + 205924 * 190 * & @xmath71.28,@xmath71.18 & 0.514,0.552 & 0.392,0.403 & 2.648 & 6210,6167 & 1.682,1.667 & 0.56,0.56 & + 605307 * 902 * & @xmath71.03,@xmath70.92 & 0.605,0.640 & 0.367,0.377 & 2.724 & 6372,6333 & 1.636,1.623 & 0.54,0.54 & + 617282 * 043 * & @xmath71.39,@xmath71.28 & 0.484,0.512 & 0.348,0.355 & 2.697 & 6418,6393 & 1.700,1.685 & 0.57,0.56 & + 651349 * 561 * & @xmath70.26,@xmath70.15 & 0.702,0.740 & 0.375,0.387 & 2.735 & 6415,6371 & 1.588,1.567 & 0.49,0.48 & bl , @xmath135 + 655183 * 353 * & @xmath71.26,@xmath71.16 & 0.500,0.534 & 0.371,0.381 & 2.667 & 6312,6278 & 1.691,1.663 & 0.56,0.53 & + 657944 * 259 * & @xmath70.54,@xmath70.43 & 0.673,0.706 & 0.353,0.363 & 2.765 & 6497,6461 & 1.601,1.582 & 0.51,0.50 & + 659723 * 739 * & @xmath70.89,@xmath70.78 & 0.642,0.670 & 0.330,0.338 & 2.803 & 6581,6553 & 1.608,1.598 & 0.53,0.53 & 2011 , @xmath135 + 659723 * 739 * & @xmath70.94,@xmath70.84 & 0.656,0.687 & 0.337,0.346 & 2.803 & 6540,6508 & 1.604,1.588 & 0.54,0.53 & 2012 , @xmath135 + since no physical parameters have been derived for any of the corot rr lyrae stars , yet , we estimated the basic physical parameters of the complete corot rrab sample . only three blended stars ( corot 101315488 , 100881648 , and 101503544 ) were omitted , because of their small amplitude distorted light curves ( see @xcite ) resulted in unusable fourier parameters . although the empirical formulae were originally calibrated on the non - blazhko stars , since then some works showed that the formulae can be applied for well - sampled blazhko stars as well @xcite . the calculations were done for the above ( 1 ) and ( 2 ) transformed data sets , independently . the metallicity was determined by using the @xcite formula , which is defined on the @xcite scale . these metallicity values can be transformed to the @xcite scale as it was showed by @xcite . there are signs that the formula of @xcite gives systematically higher metallicity in the case of very low metallic content ( see . e.g. @xcite ) , although for example @xcite found no strong evidence suggesting this . since our sample does not contain lower metallicity stars than [ fe / h ] @xmath136 dex , we prefer the carretta - gratton metallicity scale as the best - fit one to the high resolution spectroscopy . the extinction free colour index @xmath137 and @xmath110 were calculated from the formulae of @xcite , while for the absolute magnitude determination we followed @xcite , @xcite , and @xcite who applied a 0.2 zero point shift to the formula of @xcite . the effective temperature was estimated by the formula of @xcite . the empirical formulae for the two fundamental parameters : the stellar mass and luminosity are yielded systematically different results depending on whether they come from the pulsation @xcite or the stellar evolution @xcite models . these discrepancies were discussed in the case of _ kepler _ rrab sample by @xcite in details . more recently , however , @xcite constructed consistent pulsation models where the stellar masses and luminosities were fixed according to the horizontal - branch evolutionary models . we took into account these new results when we computed the masses and luminosities from the @xcite formulae than we fixed either of these parameters and determine the other one by the van albada - baker type relation given by @xcite ( eq . 1 in the paper ) . the typical result of this cross check is that the pulsation mass defined by @xcite formula needs higher luminosity for consistency than that we get from the @xcite luminosity formula . or vice verse : @xcite relation yields lower mass if we use @xcite luminosities . but in the latter case the calculated small masses are out of the allowed parameter range where the stable pulsation persists . therefore , we kept the pulsation masses and determined the luminosities by the @xcite formulae . we mention here that for two stars ( # 793 and # 363 ) the alternate calculations provided consistent values for both masses and luminosities , and for # 962 the pulsation mass ( 0.52 m@xmath91 ) resulted in such low luminosity ( @xmath138=1.416 dex ) which is below the horizontal branch for the given metallicity . we accept the evolutionary mass and pulsation luminosity for # 962 . the results are summarized in table [ tab_par ] . in spite of the rather ad hoc character of the two colour transformations the estimated parameters are surprisingly consistent . the differences between the comma separated columns ( obtained from the two transformations ) indicate the accuracy of the estimated parameters . we checked the consistence of the light curve shapes with the calibration sample of the empirical formulae using the @xmath139 parameter @xcite . several blazhko stars yield @xmath139<3 values which supports that their estimated physical parameters should be as good as the non - blazhko stars ones . there is a rather surprising result of this test : for three non - blazhko stars ( # 131 , # 804 , and # 739 ) @xmath139>3 which suggests some light curve distortion . we compared the observed individual fourier parameters of these stars with the calculated ones by the interrelation of @xcite . we found that the observed amplitudes ( e.g. @xmath116 ) are always smaller that the predicted ones . this implies some flux loss for these stars . this would not be unprecedented . similar problems have been discussed for several _ kepler _ rr lyrae stars @xcite . the explanation of the possible flux lost is simple : the star fluxes are collected within pre - defined pixel masks for both corot and _ kepler_. in the preparatory phase of the missions photometric sky surveys had been conducted and the brightness values of the input catalogues ( exo - dat , kic ) define the size of the masks . since these missions are optimized to detect small brightness variations ( transiting exoplanets , solar type oscillations ) this strategy is generally appropriate . however , the rr lyrae stars have large amplitudes with highly non - sinusoidal light curve shapes , viz . they spend most of their time in faint phases . if a mask is fitted to this faint state , it is too small around maximal brightness resulting in flux loss . this hypothesis is strengthened by the fact that these problematic stars ( # 131 , # 804 , and # 739 ) are the faintest ( 15.6 , 15.94 , and 15.186 mag ) non - blazhko rr lyrae stars in the exo - dat catalogue . additionally , the star # 739 was observed using different masks for the two runs ( lrc07 and lrc10 ) . the light curves of the two runs are different , e.g. they show different amplitudes and other fourier parameters are also slightly different . if both masks contained the total flux no such differences would be expected . and the metallicity [ fe / h]@xmath140 . the blue dots define a moderate metallicity stellar group with the average metalliciy of < [ fe / h]>=-0.96 dex , while red squares show a higher metallic content group with < [ fe / h]>=-0.39 dex . the red square between the two groups shows the star # 259 with < [ fe / h]>=-0.43 dex . ] the original goal of this parameter estimation was to find any extreme stars in our sample . the physical parameters of all stars are , however , in the canonical ranges of the equilibrium values of rrab stars . the metallicity is between @xmath141 } > -1.39 $ ] , the absolute visual brightness is @xmath142 mag , the reddening free color index is @xmath143 , the effective temperature is @xmath144 k , the surface gravitational acceleration is @xmath145 , the luminosity is @xmath146 , and the mass is @xmath147 m@xmath91 . by reviewing some classical diagnostic diagrams of rr lyrae stars such as period vs. amplitude , @xmath137 vs. @xmath148 , etc . we found that the corot sample consists of two distinct groups . the separation of the two groups is the best seen in period vs. metallicity diagram ( see fig . [ fehp ] ) . the blue symbols of the figure define a moderately metal poor group with an average metallicity of < [ fe / h]>[email protected] dex , while the red filled squares denote the most metal rich stars of the sample ( # 020 , # 804 , # 561 , < [ fe / h]>=@xmath149 dex ) and the star # 259 which is located at an intermediate position between of these two groups . the metal rich group has significantly higher average period < @xmath2>=0.7013 d than the more populous metal poorer group < @xmath2>=0.5677 d. these stars are at the red edge when we plot the sample in the theoretical hr diagram ( @xmath150 , @xmath138 ) . this is not a surprise , since the instability strip moves red - ward with the increasing metallicity . we presented here the results of a rigorous rr lyrae search in the corot archive . we found nine rrab stars in the data base that have not been studied and seven of them are new discoveries . three stars show the blazhko effect . the cycle - to - cycle variation of these blazhko effects are evident either for amplitude or frequency modulations or both which rises the possibility of their multiperiodic ( or irregular / chaotic ) nature . the fourier spectrum of star # 132 contains small amplitude ` additional ' frequencies which can be identified as the consequence of the period doubling effect and the excitation of the second radial overtone mode . the later mode identification is strengthened by pulsation modeling , since the observed period ratio is unusually high ( @xmath41 ) , but linear pulsation models at least do not exclude this period ratio . the other two blazhko stars show no significant additional frequencies . the light curve stability and period changes were studied for non - blazhko stars . we detected a significant cycle - to - cycle fluctuation of the pulsation period ( cycle length ) of cm ori . this is the first case that random jitter has been detected for an rr lyrae star . the fluctuation rate is tiny : at longest 1 - 2 seconds per cycles . long term period changes were also investigated for the two known rr lyrae stars cm ori and v2042 oph . their o@xmath7c diagrams cover more than 80 and almost 70 years , respectively . both stars show a slight period increase with rates which agree well with the prediction of the stellar evolution theory for normal red - ward evolution . the fourier amplitude and phase difference distribution with respect to the harmonic order have been studied for the first time on a larger set of non - blazhko stars . we found a common amplitude decline feature . the phase differences show a split distribution . the observed distributions were compared theoretical works . our conclusion is that the recent model calculations reproduce the main features but the fine structures can not be described properly . these amplitude and phase distribution diagrams could be potential diagnostic tools for constraining pulsation models , because of their high sensitivity of the fine structure of the light curves . we successfully transformed the corot unfiltered light curves , and their fourier parameters to the johnson _ v _ curves , and parameters . as a by - product of the use of the interrelations we found possible flux loss for several stars . such an effect was shown for high amplitude variables of the _ kepler _ telescope , but has not documented yet in the case of corot satellite . by using empirical formulae we estimated the basic physical parameters of the complete corot rrab sample . the estimated physical parameters define two subgroups of the sample . ( 1 ) a shorter period and lower metallicity and ( 2 ) a longer period higher metallicity groups . the physical parameters of both groups , however , are within the canonical ranges of rr lyrae stars . this work was supported by the esa pecs grant no 4000103541/11/nl / kml and the hungarian national research development and innovation office nkfih k-115709 . s was supported by the jnos bolyai research scholarship of the hungarian academy of sciences . the research leading to these results has received funding from the european community s seventh framework programme ( fp7/2007 - 2013 ) under grant agreement no . 312844 . auvergne , m. , bodin , p. , boisnard , l. et al . 2009 , , 506 , 411 baglin , a. auvergne , m. , boisnard , l. et al . 2006 , in 36th cospar scientific assembly , ed . a. wilson , esa sp 1296 , ( esa , noordwijk ) , 3749 barcza , s. 2010 , , 406 , 486 barcza , s. & benk , j.m . 2014 , , 442 , 1863 benk , j. m. & szab , r. 2015 , , 809 , l19 benk , j. m. , kolenberg , k. , szab , r. et al . 2010 , , 409 , 1585 benk , j. m. , szab , r. , & papar , m. 2011 , , 417 , 974 benk , j. m. , palchy , e. , szab , r. , molnr , l. , & kollth , z. 2014 , , 213 , 31 bessell , m. s. 1990 , , 102 , 1181 bessell , m. s. 2005 , , 43 , 293 bono , g. , caputo , f. , & di criscienzo , m. 2007 , , 476 , 779 borucki , w. j. et al . 2010 , , 327 , 977 carretta , e. & gratton , r. g. 1997 , , 121 , 95 castelli , f , gratton , r. g. , & kurucz , r. l. 1997 , , 318 , 841 chadid , m. 2012 , , 540 , a68 chadid , m. , benk , j. m. , szab , r. et al . 2010 , , 510 , a39 cox , a. n. 1998 , , 496 , 246 deasy , h. p. & wayman , p. s. 1985 , , 212 , 395 debosscher , j. sarro , l. m. , lpez , m. et al . 2009 , , 506 , 519 deleuil , m. , meunier , j. c. , moutou , c. et al . 2009 , , 138 , 649 demarque , p. , zinn , r. , lee , y .- w . & yi , s. 2000 , , 119 , 1398 derekas , a. , szab , gy . m. , berdnikov , l. , et al , 2012 , , 425 , 1312 dorfi , e. a. & feuchtinger , m. u. 1999 , , 348 , 815 dorman , b. 1992 , , 81 , 221 geroux , c. m. , & deupree , r. g. 2015 , , 800 , id.35 girardi , l. , bressan , a. , bertelli , g. & chiosi , c. 2000 , , 141 , 371 hoffmeister , c. 1930 , astron . nachr . , 238 , 17 hoffmeister , c. 1949 , astron . ergnzungshefte z.d . , 12 , no 1 , a3 jurcsik , j. 1998 , , 333 , 571 jurcsik , j. & kovcs , g. 1996 , , 312 , 111 jurcsik , j. , szeidl , b. , nagy , a. , & sdor , . 2005a , , 55 , 303 jurcsik , j. , sdor , . , vradi , m. , et al . 2005b , , 430 , 1049 jurcsik , j. , szeidl , b. , sdor , . , 2006 , , 132 , 61 jurcsik , j. , hurta , zs . , sdor , . , et al . 2009a , , 397 , 350 jurcsik , j. , sdor , . , szeidl , b. et al . 2009b , , 400 , 1006 jurcsik , j. , hajdu , g. , szeidl , b. et al . 2012 , , 419 , 2173 koen , c. 2005 , in the light - time effect in astrophysics , ed . c. sterken , asp conf . 335 , ( asp , san francisco ) , 25 kolenberg , k. , szab , r. , kurtz , d. w. et al . 2010 , , 713 , l198 kollth , z. 1990 , konkoly observatory occasional technical notes , no . 1 , 1 kollth , z. , & buchler , j. r. 2001 , in stellar pulsation non - linear studies , eds . takeuti , m. & sasselov , d. d. , ass library series vol . 257 , ( kluwer academic publishers , dordrecht ) , 29 kollth , z. , buchler , j. r. , szab , r. , & csubry , z. 2002 , , 385 , 932 kollth , z. , molnr , l. , & szab , r. 2011 , , 414 , 1111 kovcs , g. 2005 , , 438 , 227 kovcs , g. & kanbur , s. 1998 , , 295 , 834 kovcs , g. & walker , a. r. 2001 , , 371 , 579 le borgne , j .- f . , paschke , a. , vandenbroere , j. et al . 2007 , , 476 , 307 le borgne , j .- f . , klotz , a. , poretti , e. et al . 2012 , , 144 , 39 lee , j .- w . lpez - morales , m. , hong , k. al . 2014 , , 210 , 6 marconi , m. , coppola , g. , bono , g. , et al . 2015 , , 808 , 50 molnr , l. , szab , r. , moskalik , p. et al . 2015 , , 452 , 4283 moskalik , p. , smolec , r. , kolenberg , k. et al . 2015 , , 447 , 2348 mundprecht , e. , muthsam , h. j. , & kupka , f. 2013 , , 435 , 3191 nemec , j. m. 2004 , , 127 , 2185 nemec , j. m. , smolec , r. , benk , j. m. et al . 2011 , , 417 , 1022 nemec , j. m. , cohen , j. g. , ripepi , v. et al . 2013 , , 773 , 181 papar , m. , szab , r. , benk , j. m. et al . 2009 , in stellar pulsation : challanges for theory and observation , aip conf . ser . vol . 1170 . , ed . j. a. guzik & p. a. bradley , 240 pietrukowicz , p. , udalski , a. , soszyski , i. et al . 2012 , , 750 , 169 poretti , e. papar , m. , deleuil , m. et al . 2010 , , 520 , a108 ross , f. e. 1925 , , 36 , 99 sandage , a. 2004 , , 128 , 858 sandage , a. 2006 , , 131 , 1750 simon , n. r. & lee , a. s. 1981 , , 248 , 291 skarka , m. 2015 , , 445 , 1584 smith , h. a. , barnett , m. , silbermann , n. a. , & gay , p. 1999 , , 118 , 572 smolec , r. 2005 , , 55 , 59 soszyski , i. udalski , a. , szymaski , m. k. et al . 2009 , , 59 , 1 soszyski , i. udalski , a. , szymaski , m. k. et al . 2014 , , 64 , 177 stellingwerf , r. f. & donohoe , m. 1987 , , 314 , 252 sterken , c. 2005 , in the light - time effect in astrophysics , ed . c. sterken , asp conf . 335 , ( asp , san francisco ) , 3 straiys , v. 1996 , baltic astron . , 5 , 459 sweigart , a.v . & renzini , a. 1979 , , 71 , 66 szab , r. , kollth , z. , molnr , l. et al . 2010 , , 409 , 1244 szab , r. , benk , j. m. , papar , m. et al . 2014 , , 570 , a100 weingrill , j. 2015 , astron . 336 , 125 zalian , c. , chadid , m. , vernin , j. , et al . 2016 , , 456 , 192 zinn , r. & west , m. j. 1984 , , 55 , 45

quantum pumping in closed systems is considered . we explain that the kubo formula contains all the physically relevant ingredients for the calculation of the pumped charge ( ) within the framework of linear response theory . the relation to the common formulations of adiabatic transport and geometric magnetism " is clarified . we distinguish between adiabatic and dissipative contributions to . on the one hand we observe that adiabatic pumping does not have to be quantized . on the other hand we define circumstances in which quantized adiabatic pumping holds as an approximation . the deviation from exact quantization is related to the thouless conductance . as an application we discuss the following examples : classical dissipative pumping by conductance control , classical adiabatic ( non dissipative ) pumping by translation , and quantum pumping in the double barrier model . in the latter context we analyze a 3 site lattice hamiltonian , which represents the simplest pumping device . we remark on the connection with the popular matrix formalism which has been used to calculate pumping in open systems .

cond-mat0307619

linear response theory ( lrt ) @xcite is the leading formalism to deal with driven systems . such systems are described by a hamiltonian @xmath2 where @xmath3 is a set of time dependent classical parameters ( `` fields '' ) . the kubo formula is the corner stone of lrt . it allows the calculation of the response coefficients , and in particular the _ conductance matrix _ ( @xmath4 ) of the system . if we know @xmath4 , we can calculate the charge ( @xmath0 ) which is transported through the system during one cycle of a periodic driving . this is called _ pumping_. pumping of charge in mesoscopic @xcite and molecular size devices is regarded as a major issue in the realization of future quantum circuits or quantum gates , possibly for the purpose of quantum computing . in order to explain the motivation for the present work , and its relation to the published literature , we have to give a better definition of the problem . for presentation purpose we focus on a model system with a ring geometry ( fig.1 ) . the shape of the ring is controlled by some parameters @xmath5 and @xmath6 . these parameters can be gate voltages that determine the location of some boundaries , or the height of some barriers . the third parameter is the flux through the ring : @xmath7 we shall use units such that the elementary charge is . note that the hamiltonian @xmath8 has gauge invariance for @xmath9 . another system with a ring topology is presents in fig . 1b , and its abstraction is represented in fig . 2c . the dot " can be represented by an @xmath1 matrix that depends on @xmath5 and @xmath6 . in fig . 1d also the flux @xmath10 is regarded as a parameter of the dot . if we cut the wire in fig.1d we get the open two lead geometry of fig.1e . finally we can put many such units in series ( no flux ) , hence getting the periodic system of fig.1f . in the latter case the hamiltonian is invariant for unit translations , and therefore the quasi momentum @xmath11 is a constant of motion . it follows that the mathematical treatment of a driven periodic structure reduces to an analysis of a driven ring system with flux . _ pumping " _ means that net charge ( or maybe better to say net integrated probability current " ) is transported through the ring per cycle of a periodic driving . using the common jargon of electrical engineering this can be described as ac - dc conversion . we distinguish between + * pumping in open systems ( such as in fig . * pumping in closed systems ( such as in fig . 1d ) * pumping in periodic systems ( such as in fig . 1f ) for a reason that was explained at the end of the previous subsection we regard the last category @xcite as mathematically equivalent to the second category . we also regard the first category @xcite as a special ( subtle ) limit of the second category : in a follow up paper @xcite we demonstrate that in the limit of open geometry the kubo formula reduces to the @xmath1-matrix formula of bttiker prtre and thomas @xcite . there are works in the literature regarding rectification " and ratchets " @xcite . these can be regarded as studies of pumping in periodic systems with the connotation of having _ damped _ non - hamiltonian dynamics . these type of systems are beyond the scope of the present paper . there is also a recent interest in hamiltonian ratchets@xcite , which is again a synonym for pumping in periodic systems , but with the connotation of having a _ non - linear _ pumping mechanism . we are going to clarify what are the conditions for having a _ linear _ pumping mechanism . only in case of linear pumping mechanism the kubo formula can be used , which should be distinguished from the non - linear mechanism of ref.@xcite . the purpose of this paper is to explain and demonstrate that the kubo formula contains all the physically relevant ingredients for the calculation of the charge ( @xmath0 ) which is pumped during one cycle of a periodic driving . in the limit of a very slow time variation ( small @xmath12 ) , the emerging picture coincides with the adiabatic picture of refs.@xcite . in this limit the response of the system is commonly described as a non - dissipative geometric magnetism " @xcite effect , or as adiabatic transport . a major objective of this paper is to bridge between the adiabatic picture and the more general lrt / kubo picture , and to explain how _ dissipation _ emerges in the quantum mechanical treatment . for one - parameter driving a unifying picture that bridges between the quantum mechanical adiabatic picture and lrt has been presented in @xcite . a previous attempt @xcite had ended in some confusion regarding the identification of the linear response regime , while @xcite had avoided the analysis of the mechanism that leads to dissipation in the quantum mechanical case . the presented ( kubo based ) formulation of the pumping problem has few advantages : it is not restricted to the adiabatic regime ; it allows a clear distinction between _ dissipative _ and _ adiabatic _ contributions to the pumping ; the classical limit is manifest in the formulation ; it gives a level by level understanding of the pumping process ; it allows the consideration of any type of occupation ( not necessarily fermi occupation ) ; it allows future incorporation of external environmental influences such as that of noise ; it regards the voltage over the pump as electro motive force , rather than adopting the conceptually more complicated view @xcite of having a chemical potential difference . of particular interest is the possibility to realize a pumping cycle that transfers _ exactly _ one unit of charge per cycle . in open systems @xcite this quantization " holds only approximately , and it has been argued @xcite that the deviation from exact quantization is due to the dissipative effect . furthermore it has been claimed @xcite that exact quantization would hold in the strict adiabatic limit , if the system were _ closed_. in this paper we would like to show that the correct picture is quite different . we shall demonstrate that the deviation from exact quantization is in fact of adiabatic nature . this deviation is related to the so - called thouless conductance " of the device . we give several examples for the application of the kubo formula to the calculation of the pumped charge @xmath0 : + * classical dissipative pumping * classical adiabatic pumping ( by translation ) * quantum pumping in the double barrier model the last example is the main one . in the context of open geometry it is known as pumping around a resonance " @xcite . we explain that this is in fact an diabatic transfer scheme , and we analyze a particular version of this model which is represented by a 3 site lattice hamiltonian . this is definitely the simplest pump circuit possible , and we believe that it can be realized as a molecular size device . it also can be regarded as an approximation for the closed geometry version of the two delta potential pump @xcite . in section 2 we define the main object of the study , which is the conductance matrix @xmath4 of eq.([e_8 ] ) . the conductance matrix can be written as the sum of a symmetric ( @xmath13 ) and an anti - symmetric ( @xmath14 ) matrices , which are later identified as the dissipative and the adiabatic contributions respectively . in the first part of the paper ( sections 2 - 8 ) we analyze the adiabatic equation ( section 3 ) , and illuminate the distinction between its zero order solution ( section 4 ) , its stationary first order solution ( section 5 ) , and its non - stationary solution ( section 6 ) . the outcome of the analysis in section 5 is eq.([e_30 ] ) for the conductance matrix @xmath4 . this expression is purely adiabatic , and does not give any dissipation . in order to get dissipation we have to look for a non - stationary solution . the standard textbook derivation of the kubo formula ( eq.([e_33 ] ) ) for the conductance matrix @xmath4 _ implicitly _ assumes a non - stationary solution . we show how to get from it eq.([e_34 ] ) for @xmath13 and eq.([e_35 ] ) for @xmath14 . the latter is shown to be identical with the adiabatic result ( eq.([e_30 ] ) ) . in section 7 we further simplify the expression for @xmath13 leading to the fluctuation - dissipation relation ( eq.([e43 ] ) ) . the disadvantages of the standard textbook derivation of kubo formula make it is essential to introduce a different route toward eq.([e43 ] ) for @xmath13 . this route , which is discusssed in section 8 , _ explicitly _ distinguishes the dissipative effect from the adiabatic effect , and allows to determine the conditions for the validity of either the adiabatic picture or lrt . in particular it is explained that lrt is based , as strange as it sounds , on perturbation theory to infinite order . in section 9 we clarify the general scheme of the pumping calculation ( eq.([e6 ] ) ) . section 10 and section 11 give two simple classical examples . in section 12 we turn to discuss quantum pumping , where the cycle is around a chain of degeneracies . the general discussion is followed by presentation of the double barrier model ( section 13 ) . in order to get a quantitative estimate for the pumped charge we consider a 3 site lattice hamiltonian ( section 14 ) . the summary ( section 15 ) gives some larger perspective on the subject , pointing out the relation to the @xmath1-matrix formalism , and to the born - oppenheimer picture . in the appendices we give some more details regarding the derivations , so as to have a self - contained presentation . consider the hamiltonian @xmath15 , where @xmath3 is a set of time dependent parameters ( `` fields '' ) . for presentation , as well as for practical reasons , we assume later a set of three time dependent parameters . we define generalized forces in the conventional way as @xmath16 note that if @xmath5 is the location of a wall element , then @xmath17 is the force in the newtonian sense . if @xmath6 is an electric field , then @xmath18 is the polarization . if @xmath10 is the magnetic field or the flux through a ring , then @xmath19 is the magnetization or the current through the ring . in linear response theory ( lrt ) the response of the system is described by a causal response kernel , namely @xmath20 where @xmath21 for @xmath22 . the fourier transform of @xmath23 is the generalized susceptibility @xmath24 . the conductance matrix is defined as : @xmath25}{\omega } \ = \ \int_0^{\infty } \alpha^{kj}(\tau ) \tau d\tau\end{aligned}\ ] ] consequently , as explained further in appendix b , the response in the dc limit " ( @xmath26 ) can be written as @xmath27 as an example for the applicability of this formula note the following standard examples for one parameter driving : let @xmath28 wall or piston displacement , then @xmath29 wall or piston velocity , @xmath30 friction coefficient , and @xmath31 is the friction force . another standard example is @xmath28 magnetic flux , @xmath32 electro motive force , @xmath30 electrical conductance , and hence @xmath31 is ohm law . it is convenient to write the conductance matrix as @xmath33 , where @xmath34 is the symmetric part of the conductance matrix , while @xmath35 is the antisymmetric part . in case of having three parameters we can arrange the elements of the antisymmetric part as a vector . consequently eq.([e_8 ] ) can be written in abstract notation as @xmath36 where the dot product should be interpreted as matrix - vector multiplication , which involves summation over the index @xmath37 . the wedge - product also can be regarded as a matrix - vector multiplication . it reduces to the more familiar cross - product in case that we consider @xmath38 parameters . the dissipation , which is defined as the rate in which energy is absorbed into the system , is given by @xmath39 only the symmetric part contributes to the the dissipation . the contribution of the antisymmetric part is identically zero . the adiabatic equation is conventionally obtained from the schrodinger equation by expanding the wavefunction in the @xmath40-dependent adiabatic basis : @xmath41 where following @xcite we define @xmath42 differentiation by parts of @xmath43 leads to the conclusion that @xmath44 is a hermitian matrix . note that the effect of gauge transformation is @xmath45 note that the diagonal elements @xmath46 are real , and transform as @xmath47 . associated with @xmath48 is the gauge invariant 2-form , which is defined as : @xmath49 this can be written in an abstract notation as @xmath50 . using standard manipulations , namely via differentiation by parts of @xmath51 , we get for @xmath52 the expressions : @xmath53 and hence @xmath54 } { ( e_m - e_n)^2}\end{aligned}\ ] ] we define the perturbation matrix as @xmath55 and @xmath56 for @xmath57 . then the adiabatic equation can be re - written as follows : @xmath58 if we neglect the perturbation @xmath59 , then we get the strict adiabatic solution : @xmath60 due to @xmath48 , we have the so called geometric phase . this can be gauged away unless we consider a closed cycle . for a closed cycle , the gauge invariant phase @xmath61 is called berry phase . with the above zero - order solution we can obtain the following result : @xmath62 in case of the standard examples that were mentioned previously this corresponds to conservative force or to persistent current . from now on we ignore this trivial contribution to @xmath63 , and look for the a first order contribution . for linear driving ( unlike the case of a cycle ) the @xmath48 field can be gauged away . assuming further that the adiabatic equation can be treated as parameter independent ( that means disregarding the parametric dependence of @xmath64 and @xmath65 on @xmath66 ) one realizes that eq.([e_24 ] ) possesses stationary solutions . to first order these are : @xmath67 note that in a _ fixed - basis representation _ the above stationary solution is in fact time - dependent . hence the notations @xmath68 , @xmath69 and @xmath70 are possibly more appropriate . with the above solution we can write @xmath63 as a sum of zero order and first order contributions . from now on we ignore the zero order contribution , and go on with the first order contribution : @xmath71 for a general _ stationary _ preparation , either pure or mixed , one obtains eq.([e_8 ] ) with @xmath72 where @xmath73 are weighting factors , with the normalization @xmath74 . for a pure state preparation @xmath73 distinguishes only one state @xmath75 , while for canonical preparation @xmath76 , where @xmath77 is the temperature . for a many - body system of non - interacting fermions @xmath73 can be re - interpreted as the fermi occupation function , so that @xmath78 is the total number of particles . thus we see that the assumption of a stationary first - order solution leads to a non - dissipative ( antisymmetric ) conductance matrix this is know as either adiabatic transport @xcite or geometric magnetism " @xcite . in the later sections we shall discuss the limitations of the above result . the kubo formula is an expression for the linear response of a driven system that goes beyond the stationary adiabatic solution of the previous section . the kubo formula has many type of derivations . one possibility is to use the same procedure as in section 5 starting with @xmath79 { \mbox{e}^}{-ie_mt } |m\rangle\end{aligned}\ ] ] for completeness we also give in appendix a a simple version of the standard derivation , which is based on a conventional fixed - basis first - order - treatment of the perturbation . the disadvantages are : + * the standard derivation does not illuminate the underlaying physical mechanisms of the response . * the stationary adiabatic limit is not manifest . * the fluctuation - dissipation relation is vague . * the validity conditions of the derivation are not clear : no identification of the _ regimes_. for now we go on with the conventional approach , but in a later section we refer to the more illuminating approach of @xcite and @xcite . then we clarify what is the regime ( range of @xmath12 ) were the kubo formula can be trusted @xcite , and what is the sub - regime where the response can be described as non - dissipative adiabatic transport . in order to express the kubo formula one introduces the following definition : @xmath80\rangle\end{aligned}\ ] ] we use the common interaction picture notation , where @xmath81 with @xmath82 . the expectation value assumes that the system is prepared in a _ stationary _ state ( see previous section ) . it is also implicitly assumed that the result is not sensitive to the exact value of @xmath40 . note that @xmath83 has a well defined classical limit . its fourier transform will be denoted @xmath84 . the expectation value @xmath85 is related to the driving @xmath3 by the causal response kernel . the kubo expression for this response kernel , as derived in appendix a , is @xmath86 where the step function @xmath87 cares for the upper cutoff of the integration in eq.([e_3 ] ) . the fourier transform of @xmath88 is the generalized susceptibility @xmath89 . the conductance matrix is defined as : @xmath90}{\omega } \ = \ \int_0^{\infty } k^{ij}(\tau)\tau d\tau\end{aligned}\ ] ] this can be split into symmetric and anti - symmetric components ( see derivation in appendix c ) as follows : @xmath91}{\omega } \\ \label{e_35 } { b}^{ij } = { \mbox{\small $ \frac{1}{2}$}}({g}^{ij}{-}{g}^{ji } ) & = & { -}\int_{-\infty}^{\infty}\frac{d\omega}{2\pi } \frac{{\mbox{re}}[\tilde{k}^{ij}(\omega)]}{\omega^2}\end{aligned}\ ] ] the antisymmetric part is identified @xcite ( appendix c ) as corresponding to the stationary solution eq.([e_30 ] ) of the adiabatic equation . the kubo formula for the symmetric part of the conductance matrix ( @xmath92 ) can be further simplified . if we take eq.([e_34 ] ) literally , then @xmath93 due to the simple fact that we have finite spacing between energy levels ( see @xcite for a statistical point of view ) . but if we assume that the energy levels have some finite width @xmath94 , then the smoothed version of @xmath84 should be considered . in common textbooks the introduction of @xmath94 is justified " by assuming some weak coupling to an environment , or by taking the limit of infinite volume . but we are dealing with a strictly isolated finite system , and therefore the meaning of @xmath94 requires serious consideration . we postpone the discussion of this issue to the next section . if the smoothed version of @xmath84 should be used in eq.([e_34 ] ) , then it is possible to obtain @xmath92 from power spectrum @xmath95 of the fluctuations . this is called the fluctuation - dissipation relation . the spectral function @xmath95 is defined as the fourier transform of the symmetrized correlation function @xmath96 we use again the interaction picture , as in the definition of @xmath83 . also this function has a well defined classical limit . there are several versions for the fluctuation - dissipation relation . the microcanonical version @xcite has been derived using classical considerations , leading to @xmath97\end{aligned}\ ] ] in appendix c we introduce its quantum mechanical derivation . the subscript emphasizes that we assume a microcanonical state with energy @xmath98 , and @xmath99 is the density of states . the traditional version of the fluctuation - dissipation relation assumes a canonical state . it can be obtained by canonical averaging over the microcanonical version leading to : @xmath100 the standard derivation of the dissipative part of the kubo formula , leading to the fluctuation - dissipation relation eq.([e43 ] ) , is not very illuminating physically . more troubling is the realization that one can not tell from the standard derivation what are the conditions for its validity . an alternate , physically appealing derivation @xcite , is based on the observation that energy absorption is related to having diffusion in energy space @xcite . in appendix e we outline the main ingredients of this approach . it should be clear that the diffusion picture of appendix e holds only in case of _ chaotic systems_. if this diffusion picture does not hold , then also the kubo formula for @xmath92 does not hold ! driven one - dimensional systems are the obvious example for the failure of linear response theory ( lrt ) . as in the case of the kicked rotator ( standard map ) @xcite there is a complicated route to chaos and stochasticity : by increasing the driving amplitude the phase space structure is changed . if the amplitude is smaller than a threshold value , then the diffusion is blocked by kolmogorov - arnold - moser curves , and consequently there is not dissipation . therefore the kubo formula is not applicable in such cases . the following discussion of _ dissipative response _ assumes that we deal with a quantized chaotic system . we would like to discuss the reason and the consequences of having an energy scale @xmath94 . in the standard derivation the assumption of having level broadening as if comes out of the blue . as we already noted it is customary in textbooks to argue that either a continuum limit , or some small coupling to an environment is essential in order to provide @xmath94 . but this is of course just a way to avoid confrontation with the physical problem of having a driven _ isolated finite mesoscopic system_. in fact the energy scale @xmath94 is related to the rate ( @xmath12 ) of the driving : @xmath101 where for simplicity of presentation we assume one parameter driving . we use @xmath102 to denote the mean level spacing , and @xmath103 is the root mean square value of the matrix element @xmath104 between neighboring levels . in order to derive the above expression for @xmath94 we have used the result of @xcite ( sec.17 ) for the core width `` at the breaktime @xmath105 of perturbation theory . the purpose of the present section is to give an optional ' ' pedestrian derivation " for @xmath94 , and to discuss the physical consequences . looking at the first order solution eq.([e_28 ] ) of section 5 one realizes that it makes sense provided @xmath106 . this leads to the adiabaticity condition @xmath107 . the adiabaticity condition eq.([e91 ] ) can be explained in a more illuminating way as follows : let us assume that we prepare the system at time @xmath108 at the level @xmath109 . using _ time dependent _ fopt we find out that a stationary - like solution is reached after the heisenberg time @xmath110 . this is of course a valid description provided we do not have by then a breakdown of fopt . the condition for this is easily found to be @xmath111 , where @xmath112 . this leads again to the adiabaticity condition eq.([e91 ] ) . another assumption in the derivation of section 5 was that we can ignore the parametric dependence of @xmath64 and @xmath65 on @xmath66 . the adiabaticity condition @xmath113 manifestly justify such an assumption : we should think of @xmath114 as the transient time for getting a stationary - like state , and we should regard @xmath115 as the parametric correlation scale . as strange as it sounds , in order to have dissipation , it is essential to have a breakdown of fopt . in the language of perturbation theory this implies a required summation of diagrams to _ infinite order _ , leading to an effective broadening of the energy levels . by iterating fopt , neglecting interference terms , we get a _ markovian approximation _ for the energy spreading process . this leads to the diffusion equation of appendix e. this diffusion can be regarded as arising from fermi - golden - rule transitions between energy levels . a simple ad - hoc way to determine the energy level broadening is to introduce @xmath94 as a lower cutoff in the energy distribution which is implied by eq.([e_28 ] ) : @xmath116 this constitutes a generalization of the well known procedure used by wigner in order to obtain the local density of states @xcite . however , in the present context we do not get a lorentzian . the width parameter @xmath94 is determined self consistently from normalization , leading to eq.([e_42 ] ) [ disregarding numerical prefactor ] . we can summarize the above reasoning by saying that there is a _ perturbative regime _ that includes an adiabatic ( fopt ) sub - regime . outside of the adiabatic sub - regime we need _ all orders _ of perturbation theory leading to fermi - golden - rule transitions , diffusion in energy space , and hence dissipation . thus the dissipative part of kubo formula emerges only in the regime @xmath117 , which is just the opposite of the adiabaticity condition . the next obvious step is to determine the boundary of the perturbative regime . following @xcite we argue that the required condition is @xmath118 . the bandwidth @xmath119 is defined as the energy width @xmath120 were the matrix elements @xmath104 are not vanishingly small . if the condition @xmath118 is violated we find ourselves in the non - perturbative regime where the kubo formula can not be trusted @xcite . we still have to illuminate why we can get in the perturbative regime a dissipative _ linear _ response in spite of the breakdown of fopt . the reason is having a separation of scales ( @xmath121 ) . the non - perturbative mixing on the small energy scale @xmath94 does not affect the rate of first - order transitions between distant levels . therefore fermi golden rule picture applies to the description of the coarse grained energy spreading , and we get linear response . the existence of the adiabatic regime is obviously a quantum mechanical effect . if we take the formal limit @xmath122 the adiabaticity condition @xmath123 breaks down . in fact the _ proper _ classical limit is _ non - perturbative _ , because also the weaker condition @xmath124 does not survive the @xmath122 limit . for further details see @xcite . in the non - perturbative regime the quantum mechanical derivation of kubo formula is not valid . indeed we have demonstrated @xcite the failure of kubo formula in case of random - matrix models . but if the system has a classical limit , then kubo formula still holds in the non - perturbative regime due to _ semiclassical _ ( rather than quantum - mechanical ) reasons . the discussion of _ dissipation _ assumes a generic situation such that the schrodinger equation does not have a stationary solution . this means that driven one - dimensional systems are automatically excluded . another non - generic possibility is to consider a special driving scheme , such as translation , rotation or dilation @xcite . in such case the time dependent hamiltonian @xmath15 possesses a stationary solution ( provided the velocity " @xmath12 is kept constant ) . consequently we do not have a dissipation effect . in section 11 we discuss the simplest example of pumping by translation , where the stationary adiabatic solution of section 5 is in fact exact , and no dissipation arises . so far we have discussed the response for driving in a very general way . from now on we focus on a system with a ring geometry as described in the introduction , and illustrated in fig . 1 . the shape of the ring is controlled by some parameters @xmath5 and @xmath6 , and @xmath10 is the magnetic flux . the generalized force @xmath19 which is conjugate to the flux is the current . the time integral over the current is the transported charge : @xmath125 in fact a less misleading terminology is to talk about probability current `` and ' ' integrated probability current " . from a purely mathematical point of view it is not important whether the transported particle has an electrical charge . disregarding a possible persistent current contribution , the expression for the pumped charge is : @xmath126_{k=3}\end{aligned}\ ] ] if we neglect the first term , which is associated with the dissipation effect , and average the second ( `` adiabatic '' ) term over the flux , then we get @xmath127 the integration should be taken over a cylinder of vertical height @xmath128 , and whose basis is determined by the projection of the pumping cycle onto the @xmath129 plane . we already pointed out that the berry phase @xmath130 is gauge invariant . therefore from stokes law it follows that @xmath131 is independent of the surface , and therefore @xmath132 with closed surface should be @xmath133integer . integrating over a cylinder , as in eq.([e7 ] ) , is effectively like integrating over a closed surface ( because of the @xmath134 periodicity in the vertical direction ) . this means that the flux averaged @xmath0 of eq.([e7 ] ) has to be an integer . the common interest is in pumping cycles in the @xmath135 plane . this means that the zero order conservative contribution to @xmath0 , due to a persistent current , does not exist . furthermore , from the reciprocity relations ( see appendix b ) it follows that @xmath136 , and @xmath137 , which should be contrasted with @xmath138 . this means that a pumping cycle in the @xmath135 plane is purely adiabatic : there is no dissipative contribution to @xmath0 . only the @xmath139 field ( second term in eq.([e6 ] ) ) is relevant to the calculation of the pumped charge , and its vertical component @xmath140 vanishes due to the time reversal symmetry . the absence of dissipative contribution for a cycle in the @xmath135 plane , does not imply that dissipation is not an issue . the symmetric part of the conductance matrix @xmath92 is in general non - zero , leading to an energy absorption rate which is proportional to @xmath141 . this implies that the energy absorption per cycle is proportional to @xmath142 . therefore we are able to minimize the dissipation effect by making the pumping cycle very slow . furthermore , if we get into the quantum - mechanical adiabatic regime , then @xmath92 becomes extremely small , and then we can neglect the dissipation effect as long as quantum - mechanical adiabaticity can be trusted . whenever the dissipation effect can not be neglected , one should specify whether or how a _ stationary operation _ is achieved . in case of pumping in open system the stationary operation is implicitly guaranteed by having equilibrated reservoirs , where the extra energy is dissipated to infinity . in case of pumping in closed system the issue of stationary operation is more subtle : in the adiabatic regime , to the extend that adiabaticity can be trusted , we have a stationary solution to the transport problem , as defined in section 5 . but outside of the adiabatic regime we have diffusion in energy space ( appendix e ) leading to a slow energy absorption ( dissipation ) . thus a driven system is heated up gradually ( though possibly very slowly ) . strictly speaking a stationary operation is not achieved , unless the system is in ( weak ) thermal contact with some large bath . another way to reach a stationary operation , that does not involve an external bath , is by having an effectively bounded phase space . this is the case with the mixed phase space example which is discussed in ref.@xcite . there the stochastic - like motion takes place in a bounded chaotic region in phase space . before we discuss the quantum mechanical pumping , it is instructive to bring simple examples for _ classical _ pumping . in the following we consider one particle ( @xmath143 ) in a two dimensional ring as in fig.1a . the first example is for classical _ dissipative _ pumping . the conductance @xmath144 can be calculated for this system @xcite leading to a mesoscopic variation of the drude formula . the current is given by ohm law @xmath145 , where @xmath146 is the electro - motive - force . consider now the following pumping cycle : change the flux from @xmath147 to @xmath148 , hence pumping charge . change the conductance from @xmath149 to @xmath150 by modifying the shape of the ring . change the flux from @xmath148 back to @xmath147 , hence pumping charge . consequently the net pumping is @xmath151 thus we have used the dissipative part of the conductance matrix ( first term in eq.([e6 ] ) ) in order to pump charge . in the quantum mechanical version of this example extra care should be taken with respect to the zero order contribution of the persistent current . the second example is for classical _ adiabatic _ pumping . the idea is to trap the particle inside the ring by a potential well , and then to make a translation of the trap along a circle . the result of such a cycle is evidently @xmath152 . we would like to see how this trivial result emerges form the kubo formula . let @xmath153 be the canonical coordinate of the particle in the ring , while @xmath129 are the center coordinate of a trapping potential . the hamiltonian is : @xmath154 \nonumber \\ & + & u_{{\mbox{\tiny trap}}}({r}_1{-}x_1(t),{r}_2{-}x_2(t))\end{aligned}\ ] ] where @xmath155 and @xmath156 are the components of the momentum along the ring and in the perpendicular ( transverse ) directions . the pumping is done simply by cycling the position of the trap . the translation of the trap is assumed to be along an inside circle of radius @xmath157 , @xmath158 in this problem the stationary solution of section 5 is an exact solution . namely @xmath159 are the eigenfunctions of a particle in the trap . eq.([e_psit ] ) is just galilei transformation from the moving ( trap ) frame to the laboratory frame . it is a - priori clear that in this problem the pumped charge per cycle is @xmath152 , irrespective of @xmath160 . therefore the @xmath139 field must be @xmath161 this can be verified by calculation via eq.([e29 ] ) . the singularity along the @xmath10 axis is not of quantum mechanical origin : it is not due to degeneracies , but rather due to the diverging current operator ( @xmath162 ) . we turn now to the quantum mechanical case . consider an adiabatic cycle that involves a particular energy level @xmath75 . this level is assumed to have a degeneracy point at @xmath163 . it follows that in fact there is a vertical chain of degeneracy points : @xmath164 these degeneracy points are important for the geometrical understanding of the @xmath139 field , as implied by eq.([e29 ] ) . every degeneracy point is like a monopole charge . the total flux that emerges from each monopole must be @xmath165integer for a reason that was explained after eq.([e7 ] ) . thus the monopoles are quantized in units of @xmath166 . the @xmath139 field which is created ( so to say ) by a vertical chain of monopoles may have a different _ near field _ and _ far field _ behavior , which we discuss below . the far field region exists if the chains are well isolated . later we explain that far " means @xmath167 , where @xmath168 is the thouless conductance . the far field is obtained by regarding the chain as a smooth line . this leads _ qualitatively _ to the same field as in eq.([e_52 ] ) . consequently , for a large radius " pumping cycle in the @xmath135 plane , we get @xmath169 . in the following we are interested in the deviation from exact quantization : if @xmath170 we expect to have @xmath171 , while if @xmath172 we expect @xmath173 . only for the @xmath11 averaged @xmath0 of eq.([e7 ] ) we get _ exact quantization_. the deviation from @xmath174 is extremely large if we consider a tight pumping cycle around a @xmath170 degeneracy . after linear transformation of the shape parameters , the energy splitting @xmath175 of the energy level @xmath75 from its neighboring ( nearly degenerated ) level @xmath176 can be written as @xmath177 where @xmath178 is a constant . the monopole field is accordingly @xmath179 where the prefactor is determined by the requirement of having a single ( @xmath166 ) monopole charge . assuming a pumping cycle of radius @xmath157 in the @xmath135 plane we get from the second term of eq.([e6 ] ) @xmath180_3 \ = \ \mp \pi \sqrt{g_t}\end{aligned}\ ] ] where @xmath181 is a practical definition for the thouless conductance in this context . it is used here simply as a measure for the sensitivity of an energy level to the magnetic flux @xmath160 . what we want to do in the next sections is to interpolate between the near field result , which is @xmath182 , and the far field result , which is @xmath183 . for this purpose it is convenient to consider a particular model that can be solved exactly . a simple example for quantum pumping is the double barrier model . an open geometry version of this model has been analyzed in @xcite using the @xmath1 matrix approach . the analogous closed geometry version is obtained by considering a one - dimensional ring with two delta barriers . as we are going to explain below , the pumping process in this model can be regarded as a particular example of an adiabatic transfer scheme : the electrons are adiabatically transfered from state to state one by one as in musical chair game " . the two delta barriers version of the double barrier model is illustrated in fig.2 . the length of the ring is @xmath184 , with periodic boundary conditions on @xmath185 . a dot region @xmath186 is defined by the potential @xmath187 it is assumed that @xmath188 and @xmath189 are small enough so one can classify the ring eigenstates into two categories : wire states , and dot states . the latter are those states that are localized in the dot region @xmath186 in the limit of infinitely high barriers . we define the fermi energy as the energy of the last occupied wire level in the limit of infinitely high barriers . the three parameters that we can control are the flux @xmath190 , the bias @xmath191 , and the dot potential @xmath192 which is related to @xmath193 . the energy @xmath194 correspond to the dot state which is closest to the fermi energy @xmath195 from above . we assume that the other dot levels are much further away from the fermi energy , and can be ignored . note that another possible way to control the dot potential , is simply by changing a gate voltage : that means to assume that there is a control over the potential floor in the region @xmath186 . _ the pumping cycle is assumed to be in the @xmath135 plane , so there is no issue of conservative persistent current contribution_. the pumping cycle is defined as follows : we start with a positive bias ( @xmath196 ) and lower the dot potential from a large @xmath197 value to a small @xmath198 value . as a result , one electron is transfered via the _ left _ barrier into the dot region . then we invert the bias ( @xmath199 ) and raise back @xmath6 . as a result the electron is transfered back into the wire via the _ right _ barrier . a closer look at the above scenario ( fig.2b ) reveals the following : as we lower the dot potential across a wire level , an electron is adiabatically transfered once from left to right and then from right to left . as long as the bias is positive ( @xmath196 ) the net charge being pumped is very small ( @xmath200 ) . only the lowest wire level that participate in the pumping cycle carries @xmath183 net charge : it takes an electron from the left side , and after the bias reversal it emits it into the right side . thus the pumping process in this model can be regarded as a particular example @xcite of an _ adiabatic transfer scheme _ : the electrons are adiabatically transfered from state to state , one by one , as in musical chair game " . for a single occupied level the net @xmath0 is the sum of charge transfer events that take place in four avoided crossings ( two avoided crossings in case of the lowest level ) . for many particle occupation the total @xmath0 is the sum over the net @xmath0s which are carried by individual levels . for a dense zero temperature fermi occupation the summation over all the net @xmath0s is a telescopic sum , leaving non - canceling contributions only from the first and the last adiabatic crossings . the latter involve the last occupied level at the fermi energy . rather than analyzing the two - delta - barriers version of the double barrier model , we consider below a simplified version that still contains the _ same _ essential ingredients . this is obtained by considering a three site lattice hamiltonian . the advantage is obviously the possibility to make an exact analytical treatment that does not involve approximations . the middle site in the three site lattice hamiltonian supports a single dot state , while the two other sites support two wire states . the hamiltonian is @xmath201 the three parameters are the bias @xmath191 , the dot energy @xmath202 , and the flux @xmath190 . for presentation purpose we assume that , and characterize the wire - dot coupling by the parameter @xmath203 . the eigenstates are @xmath64 . disregarding the interaction with the dot ( @xmath204 ) we have two wire states with . this implies degeneracies for @xmath205 . once we switch on the coupling ( @xmath206 ) , the only possible degeneracies are between the even dot state and the odd wire state of the mirror symmetric hamiltonian ( @xmath207 ) . the flux should be either integer ( for degeneracy of the dot level with the lower wire level ) , or half integer ( for degeneracy of the dot level with the upper wire level ) . thus we have two vertical chains of degeneracies : @xmath208 in order to calculate the @xmath139 field and pumped charge @xmath0 , we have to find the eigenvalues and the eigenvectors of the hamiltonian matrix . the secular equation for the eigenvalues is @xmath209 using the notations @xmath210 the roots of the above cubic equation are : @xmath211 where @xmath212 . the corresponding eigenstates are : @xmath213 where @xmath1 is the normalization , namely @xmath214 for the calculation of the pumped charge in the next paragraph it is useful to notice that for @xmath215 the normalization is @xmath216 , while for @xmath217 the normalization is @xmath218 . after some algebra we find that the first component of the @xmath139 field in the @xmath135 plane is @xmath219 which is illustrated in fig.3 . from here it follow that if we keep constant bias , and change only @xmath202 , then the pumped charge is : @xmath220 for a planar ( @xmath135 ) pumping cycle around the negative vertical chain the main contribution to @xmath0 comes from the two crossings of the @xmath221 line . hence we get @xmath222 where the thouless conductance in this context refers to the avoided crossing , and is defined as @xmath223 a similar calculation of the pumped charge for a planar cycle around the positive chain leads to @xmath224 with @xmath225 . in both cases we have approximate quantization @xmath226 for @xmath167 , while for a tight cycle either @xmath227 or @xmath228 depending on which line of degeneracies is being encircled . if the pumping cycle encircles both chains then we get @xmath229 . in the latter case @xmath230 for @xmath167 , with no indication for quantization we have shown how the kubo formalism can be used in order to derive both classical and quantum mechanical results for the pumped charge @xmath0 in a closed system . in this formulation the distinction between dissipative and non - dissipative contributions is manifest . within the framework of the kubo formalism ( disregarding non - linear corrections ) we have made a distinction between the following levels of treatment : + * strict adiabaticity + ( outcome of zero order treatment ) * adiabatic transport + ( outcome of stationary first order treatment ) * dissipation + ( the result of first order transitions ) in the adiabatic regime one can assume a _ stationary _ solution to the adiabatic equation , which implies no dissipation effect . this leads to the picture of adiabatic transport , where the berry phase is the outcome of a zero order treatment , while the geometric magnetism " of eq.([e_30 ] ) is the outcome of a first order treatment of the inter - level couplings . in some very special cases ( translations , rotations and dilations ) this assumption ( of having a stationary solution ) is in fact exact , but in generic circumstance this assumption is an approximation . outside of the adiabatic regime the stationary solution can not be trusted . assuming quantized _ chaotic dynamics _ one argues that fermi - golden - rule transitions between levels lead to ( slow ) diffusion in energy ( eq.([ed1 ] ) ) . this leads to the emergence of the dissipative part in the kubo formula . we have obtained an expression ( eq.([e_42 ] ) ) for the energy scale @xmath231 that controls the dissipative effect . we have explained that the dissipative contribution to the kubo formula is valid only in the regime @xmath232 . otherwise the dynamics is either of adiabatic nature ( @xmath123 ) or non - perturbative ( @xmath117 ) . in order to calculate the pumped charge @xmath0 we have to perform a closed line integral over the conductance ( eq.([e6 ] ) ) . this may have in general both adiabatic and dissipative contributions . for the common pumping cycle in the @xmath135 plane , only the adiabatic contribution exists . this follows from the reciprocity relations ( section 9 ) . still we have emphasized ( without any contradiction ) that in the same circumstances a dissipation effect typically accompanies the pumping process . the quantum adiabatic contribution to the pumping is determined by a line integral over a @xmath139 field which is created by _ monopoles_. the monopoles , which are related to the degeneracies of the hamiltonian , are located along vertical chains in @xmath40 space ( eq.([e_chain ] ) ) . the 3 site model provides the simplest example for such vertical chains : by calculating the @xmath139 field which is created ( so to say ) by these chains , we were able to determine the charge which is pumped during a cycle ( e.g. eq.([e_qmodel ] ) ) . the ( monopoles of the ) vertical chains have _ near field _ regions ( eq.([e9 ] ) ) . if the chains are well isolated in @xmath40 space , then there are also _ far field _ regions . the far field regions are defined as those where the thouless conductance is very small ( @xmath167 ) . pumping cycles that are contained in the far field region of a given chain lead to an approximately quantized pumping . it is important to realize that the existence of far field regions in @xmath40 space is associated with having a low dimensional system far away from the classical limit . in a quantized chaotic system it is unlikely to have @xmath167 along a pumping cycle . as we take the @xmath233 limit the vertical chains become very dense , and the far field regions disappear . in the subtle limiting case of open geometry we expect to get agreement with the @xmath1-matrix formula of bttiker prtre and thomas ( bpt ) @xcite . using the notations of the present paper the bpt formula for the current that comes out of ( say ) the right lead can be written as : @xmath234 where @xmath235 is the projector on the right lead channels . for @xmath236 the above reduces to the landauer formula . the details regarding the relation between the kubo formula and the bpt formula will be published in a separate paper @xcite . here we just note that the derivation is based on a generalization of the fisher - lee approach @xcite . finally it is important to remember that the theory of driven systems is the corner stone for the analysis of interaction between slow " and fast `` degrees of freedom . assume that that the @xmath237 are in fact dynamical variable , and that the conjugate momenta are @xmath238 . the standard textbook example is the study of diatomic molecules . in such case @xmath237 are the locations of the nuclei . the total hamiltonian is assumed to be of the general form @xmath239 where @xmath240 is the hamiltonian of the ' ' fast " degrees of freedom ( in the context of molecular physics these are the electrons ) . rather than using the standard basis , one can use the born - oppenheimer basis @xmath241 . then the hamiltonian can be written as @xmath242 where the interaction term is consistent with eq.([e_23 ] ) . thus it is evident that the theory of driven systems is a special limit of this problem , which is obtained if we treat the @xmath237 as classical variables . + + * acknowledgments : * it is my pleasure to thank yshai avishai ( ben - gurion university ) , yosi avron ( technion ) , thomas dittrich ( colombia ) , shmuel fishman ( technion ) , tsampikos kottos ( gottingen ) , and holger schantz ( gottingen ) for useful discussions . this research was supported by the israel science foundation ( grant no.11/02 ) , and by a grant from the gif , the german - israeli foundation for scientific research and development . in this appendix we present an elementary textbook - style derivation of the kubo formula . for notational simplicity we write the hamiltonian as . it is assumed that the system , in the absence of driving , is prepared in a stationary state @xmath243 . in the presence of driving we look for a first order solution @xmath244 . the equation for @xmath245 is : @xmath246 + if(t)[{\bm{\mathsf{v}}},\rho_0]\end{aligned}\ ] ] this equation can be re - written as @xmath247\end{aligned}\ ] ] where @xmath248 is the evolution operator which is generated by @xmath249 . the solution of the latter equation is @xmath250 \ f(t')dt'\end{aligned}\ ] ] where we use the usual definition of the interaction picture " operator @xmath251 . consider now the time dependence of the expectation value of an observable . disregarding the zero order contribution , the first order expression is @xmath252\right ) \ f(t')dt ' \\ \nonumber & = & \ \int^{t } \alpha(t - t ' ) \ f(t')dt'\end{aligned}\ ] ] where the response kernel @xmath253 is defined for @xmath254 as @xmath255\right ) \nonumber \\ \ & = & i\ { \mbox{trace}}\left([{\bm{\mathsf{f } } } , { \bm{\mathsf{v}}}(-\tau ) ] \rho_0\right ) \nonumber \\ \ & = & i \langle [ { \bm{\mathsf{f } } } , { \bm{\mathsf{v}}}(-\tau ) ] \rangle \nonumber \\ \ & = & i \langle [ { \bm{\mathsf{f}}}(\tau ) , { \bm{\mathsf{v } } } ] \rangle\end{aligned}\ ] ] we have used above the cyclic property of the trace operation ; the stationarity @xmath256 of the unperturbed state ; and the definition @xmath257 . in this appendix we would like to further illuminate the relation between the generalized susceptibility and the conductance matrix . the generalized susceptibility @xmath24 is the fourier transform of the causal response kernel @xmath23 . therefore it is an analytic function in the upper half of the complex @xmath258 plan , whose real and imaginary parts are related by hilbert transforms ( kramers - kronig relations ) : @xmath259 = \int_{-\infty}^{\infty } \frac{{\mbox{im}}[\chi^{kj}(\omega')]}{\omega'-\omega } \ \frac{d\omega'}{\pi}\end{aligned}\ ] ] the imaginary part of @xmath24 is the sine transforms of @xmath23 , and therefore it is proportional to @xmath258 for small frequencies . consequently it is convenient to write the fourier transformed version of eq.([e_3 ] ) as @xmath260_{\omega } \ \sum_j \chi_0^{kj}(\omega ) [ x_j]_{\omega } -\mu^{kj}(\omega ) [ \dot{x}_j]_{\omega}\end{aligned}\ ] ] where the dissipation coefficient is defined as @xmath261}{\omega } = \int_0^{\infty } \alpha^{kj}(\tau ) \frac{\sin(\omega\tau)}{\omega}d\tau\end{aligned}\ ] ] in this paper we ignore the first term in eq.([e3 ] ) which signify the non - dissipative in - phase response . rather we put the emphasis on the dc limit " ( @xmath262 ) of the second term . thus the conductance matrix is just a synonym for the term dissipation coefficient " . however , conductance " is a better ( less misleading ) terminology : it does not have the ( wrong ) connotation of being specifically associated with dissipation , and consequently it is less confusing to say that it contains a ( non - dissipative ) adiabatic component . for systems where time reversal symmetry is broken due to the presence of a magnetic field @xmath263 , the response kernel , and consequently the generalized susceptibility and the conductance matrix satisfies the onsager reciprocity relations @xmath264 \ \alpha^{ji}(\tau , { \cal b } ) \\ \chi^{ij}(\omega , -{\cal b } ) \ & = & \ [ \pm ] \ \chi^{ji}(\omega , { \cal b } ) \\ { g}^{ij } ( -{\cal b } ) \ & = & \ [ \pm ] \ { g}^{ji}({\cal b})\end{aligned}\ ] ] where the plus ( minus ) applies if the signs of @xmath265 and @xmath266 transform ( not ) in the same way under time reversal . these reciprocity relations follow from the kubo formula ( eq.([e_kubo ] ) , using @xmath267 k^{ij}(\tau,{\cal b } ) $ ] , together with the trivial identity @xmath268 . in section . 9 we discuss the implications of the reciprocity relations in the context of pumping . the functions @xmath270 and @xmath83 are the expectation values of hermitian operators . therefore they are real functions . it follows that the real part of their fourier transform is a symmetric function with respect to @xmath258 , while the imaginary part of their fourier transform is anti symmetric with respect to @xmath258 . by definition they satisfy @xmath271 and @xmath272 . it is convenient to regard them as the real and imaginary parts of one complex function @xmath273 . namely , @xmath274 it is possible to express the decomposition @xmath275 in terms of @xmath84 . using the definition eq.([e_33 ] ) we get : @xmath276}{\omega^2 } \frac{d\omega}{2\pi } + \left[\frac{1}{2}\frac{{\mbox{im}}[\tilde{k}^{ij}(\omega)]}{\omega}\right]_{\omega{=}0}\end{aligned}\ ] ] the first term is antisymmetric with respect to its indexes , and is identified as @xmath277 . the second term is symmetric with respect to its indexes , and is identified as @xmath92 . the last step in the above derivation involves the following identity that hold for any real function @xmath278 @xmath279}{\omega^2 } -\pi{\mbox{im}}[\tilde{f}(\omega)]\delta'(\omega)\right ) = { -}\int_{-\infty}^{\infty}\frac{{\mbox{re}}[\tilde{f}(\omega)]}{\omega^2 } \frac{d\omega}{2\pi } + \left[\frac{1}{2}\frac{{\mbox{im}}[\tilde{f}(\omega)]}{\omega}\right]_{\omega{=}0}\end{aligned}\ ] ] note that @xmath280 $ ] is the sine transform of @xmath278 , and therefore it is proportional to @xmath258 is the limit of small frequencies . it is of practical value to re - derive eq.([eb4 ] ) by writing @xmath273 using the energies @xmath64 and the matrix elements @xmath281 . then we can get from it straightforwardly ( using the definitions ) all the other expressions . namely : @xmath282 \ \delta'(e_m - e_n ) \\ \label{eb10 } { b}^{ij } \ & = & \ 2\hbar \sum_n f(e_n ) \sum_{m(\ne n ) } \frac{{\mbox{im}}\left [ f^i_{nm}f^j_{mn}\right ] } { ( e_m - e_n)^2}\end{aligned}\ ] ] one observes that the expression for @xmath277 coincides with the adiabatic transport result eq.([e_30 ] ) . alternatively this identification can be obtained by expressing the sum in eq.([eb10 ] ) as an integral , getting form it the first term in eq.([eb4 ] ) : @xmath283}{\omega^2}\frac{d\omega}{2\pi } \ \ = \ \ \frac{2}{\hbar}\int_{-\infty}^{\infty } \frac{{\mbox{im}}[\tilde{c}^{ij}(\omega)]-\frac{\hbar}{2}{\mbox{re } } [ \tilde{k}^{ij}(\omega ) ] } { \omega^2}\frac{d\omega}{2\pi } \ \ = \ \ -\int_{-\infty}^{\infty } \frac{{\mbox{re}}[\tilde{k}^{ij}(\omega)]}{\omega^2}\frac{d\omega}{2\pi}\end{aligned}\ ] ] we can use the following manipulation in order to relate @xmath84 to @xmath95 , @xmath286 where we use the notation @xmath287 . the third line differs from the second line by permutation of the dummy summation indexes , while the fourth line is the sum of the second and the third lines divided by 2 . in the last equality we assume small @xmath258 . if the levels are very dense , then we can replace the summation by integration , leading to the relation : @xmath288 where @xmath289 and @xmath290 are microcanonically smoothed functions . since this equality hold for any smoothed @xmath291 , it follows that the following relation holds ( in the limit @xmath26 ) : @xmath292\end{aligned}\ ] ] if we do not assume small @xmath258 , but instead assume canonical state , then a variation on the last steps in eq.([ec1 ] ) , using the fact that @xmath293 is an odd function , leads to the relation @xmath294 upon substitution of the above expressions in the kubo formula for @xmath92 , one obtains the fluctuation - dissipation relation . + + ( 1,0)490 + + the illuminating derivation of eq.([e43 ] ) is based on the observation that energy absorption is related to having diffusion in energy space . let us assume that the probability distribution @xmath295 of the energy satisfies the following diffusion equation : @xmath296 the energy of the system is @xmath297 . it follows that the rate of energy absorption is @xmath298 for a microcanonical preparation we get @xmath299\end{aligned}\ ] ] this diffusion - dissipation relation reduces immediately to the fluctuation - dissipation relation if we assume that the diffusion in energy space due to the driving is given by @xmath300 thus it is clear that a theory for linear response should establish that there is a diffusion process in energy space due to the driving , and that the diffusion coefficient is given by eq.([e49 ] ) . more importantly , this approach also allows treating cases where the expression for @xmath301 is non - perturbative , while the diffusion - dissipation relation eq.([e48 ] ) still holds ! a full exposition ( and further reference ) for this route of derivation can be found in @xcite . here we shall give just the classical derivation of eq.([e49 ] ) , which is extremely simple . we start with the identity @xmath302 assuming ( for presentation purpose ) that the rates @xmath303 are constant numbers , it follows that energy changes are related to the fluctuating @xmath304 as follows : @xmath305 squaring this expression , and performing microcanonical averaging over initial conditions we obtain : @xmath306 where @xmath307 is the correlation function . for very short times this equation implies `` ballistic '' spreading ( @xmath308 ) while on _ intermediate time scales _ it leads to diffusive spreading , where @xmath309 the latter result assumes a short correlation time . this is also the reason that the integration over @xmath310 can be extended form @xmath311 to @xmath312 . hence we get eq.([e49 ] ) . we note that for long times the systems deviates significantly from the initial microcanonical preparation . hence , for long times , one should justify the use of the diffusion equation ( [ ed1 ] ) . this leads to the classical slowness condition which is discussed in ref.@xcite . d. cohen in _ new directions in quantum chaos _ , proceedings of the international school of physics enrico fermi " , course cxliii , edited by g. casati , i. guarneri and u. smilansky , ( ios press , amsterdam 2000 ) . illustration of a ring system ( a ) . the shape of the ring is controlled by some parameters @xmath5 and @xmath6 . the flux through the ring is @xmath313 . a system with equivalent topology , and abstraction of the model are presented in ( b ) and ( c ) . the dot " can be represented by an @xmath1 matrix that depends on @xmath5 and @xmath6 . in ( d ) also the flux @xmath10 is regarded as a parameter of the dot . if we cut the wire in ( d ) we get the open two lead geometry of ( e ) . if we put many such units in series we get the period system in ( f ) . + fig2 . schematic illustration of quantum pumping in a closed wire - dot system . the net charge via the third level ( thick solid line on the right ) is vanishingly small : as the dot potential is lowered an electron is taken from the left side ( first avoided crossing ) , and then emitted back to the left side ( second avoided crossing ) . assuming that the bias is inverted before the dot potential is raised back , only the second level carry a net charge @xmath183 . . the first component of the @xmath139 field for a particle in the middle level of the 3 site lattice model . it is plotted as a function of the dot potential @xmath202 . the other parameters are @xmath314 , and @xmath315 , while @xmath316 for the thick line and @xmath317 for the thin line . in the limit @xmath318 , all the charge that is transfered from the left side into the dot during the first avoided crossing , is emitted back into the left side during the second avoided crossing . inset : the eigenenergies @xmath319 for the @xmath316 calculation .

we present a library of empirical stellar spectra created using spectra from the sloan digital sky survey s baryon oscillation spectroscopic survey ( boss ) . the templates cover spectral types o5 through l3 , are binned by metallicity from -2.0 dex through + 1.0 dex and are separated into main sequence ( dwarf ) stars and giant stars . with recently developed m dwarf metallicity indicators , we are able to extend the metallicity bins down through the spectral subtype m8 , making this the first empirical library with this degree of temperature _ and _ metallicity coverage . the wavelength coverage for the templates is from 3650 through 10200 at a resolution better than r . using the templates , we identify trends in color space with metallicity and surface gravity , which will be useful for analyzing large data sets from upcoming missions like lsst . along with the templates , we are releasing a code for automatically ( and/or visually ) identifying the spectral type and metallicity of a star .

1702.06957

empirical stellar spectral libraries are crucial for many areas of astronomical research that include , but are not limited to , simple assignment of stellar spectral types by - eye , more complicated machine learning spectral typing , and modeling the spectral energy distributions and stellar populations of galaxies . the determination of stellar properties such as surface gravity , metallicity and effective temperature is often completed by comparisons to empirical templates with known parameters , or modeled spectra . stellar template libraries are also an important teaching resource , from examples of stellar spectra for introductory classes , to detailed radiative transfer at the graduate level . with the advent of the large synoptic survey telescope ( lsst ; * ) and many other large surveys , photometric data sets of unprecedented size will be available to astronomers . lsst will not include a spectrograph , and so the characterization of stellar parameters must be based entirely on the available photometric data . in preparation for lsst , @xcite created a machine learning technique , which was taught to determine metallicities solely from photometric colors , using sloan digital sky survey ( sdss ) stellar spectra with log @xmath1 , [ fe / h ] , and @xmath2 measurements from the segue stellar parameters pipeline ( sspp ; * ? ? ? * ) . the pipeline , however , is only quoted to be accurate for spectral types f through mid - k ( outside this temperature range may be accurate but requires further testing ) , excluding a large portion of stellar parameter space . in the modern age of these large surveys , data sets will be so large that individual analyses will be infeasible , and machine learning or other statistical techniques will become increasingly important . to accurately train machines , datasets with known properties that cover the entire range of parameter space is needed . empirical spectral libraries have a rich history in stellar astronomy ( e.g. , an atlas of digital spectra of cool stars : @xcite ; pickles : @xcite ; elodie : @xcite ) , but each has limitations , and none of them include a range of m star properties . because of the complex molecular rich atmospheres of low - temperature stars , estimates of log @xmath1 and [ fe / h ] have only become accurate within the last few years ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and were therefore unavailable when these libraries were assembled . the elodie library is also limited by its small wavelength coverage ( 3900 6800 ) , which does not include many of the longer - wavelength features necessary for studying low - mass stars metallicities and surface gravities ( e.g. , na i at 8200 ) . numerous groups have used sdss to create partial libraries of specific effective temperature regions because of the wide wavelength coverage and vast number of spectroscopic and photometric observations available . in the solar mass regime ( 4,500 7,500 k ) , @xcite created the sloan extension for galactic exploration and understanding ( segue ) sspp and estimated @xmath2 , log @xmath1 , and [ fe / h ] for over 100,000 stars . in the low - mass regime , @xcite first assembled a sample of m , l and t dwarf single object spectra . @xcite improved upon the sample by creating co - added templates for active and non - active low - mass stars ( m0-l0 ) , and @xcite extended the templates through the spectral subtype of l6 . @xcite expanded the library further by compiling co - added templates of low - mass ( and low - metallicity ) subdwarfs . however , there is no single source library that covers parameter space in the low - mass end ( m dwarf metallicities and surface gravities ) through the high - mass regime . synthetic template libraries are also used widely throughout astronomy and are extremely useful tools , since the parameters like log @xmath1 , [ fe / h ] , and @xmath2 are model inputs and can span a wider parameter space than any empirical template library . many of these complete model libraries are available to the public ( e.g. , kurucz : @xcite ; bt settl : @xcite ) . however , the models must use simplifications such as plane parallel , local thermal equilibrium , etc . , to create these synthetic spectra @xcite . synthetic stellar spectra are also limited because of incomplete lists of line opacities , and a lack of knowledge of input parameters for pressure broadening models @xcite . this is especially problematic for the complex , molecular - rich atmospheres of low - mass stars , where the model spectra do not exactly match the observed spectra across all wavelengths . empirical libraries are therefore still extremely important both to constrain models and to use in regimes where the models are limited . with a new , complete stellar template library , spectral typing and determinations of other important stellar properties will improve . in addition , the templates can be used in more complicated stellar population modeling . stellar population synthesis models have long been used as a tool to assemble model galaxies from the co - addition of individual stellar spectra ( e.g. , * ? ? ? * ; * ? ? ? many studies have used these model galaxies to determine basic galactic properties such as mass - to - light ratios ( m / l ) and to refine the tully - fisher relation ( e.g. , * ? ? ? recently , this method has been used to study the initial mass function ( imf ) in distant galaxies and has led to a growing body of evidence that the imf is not constant as previously thought , but can change depending on the environment ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the debate over the form of the imf is concentrated in the low - mass end of the stellar sequence ( @xmath3 ) , where estimates of its form produce different numbers of low - mass stars that span at least an order of magnitude . to disentangle the low - mass stars from the spectrum of an elliptical galaxy , whose light is completely dominated by red giants , @xcite examined absorption features ( na i and feh ) that are prominent in low - mass , main - sequence stars and inferred the abundance of low - mass stars , and hence an imf . a leading hypothesis for the physical parameter that best correlates with changes in the imf is metallicity . to explore this scenario further , we need a template library with high quality low - mass stars that include metallicity variations and a wavelength coverage that includes these dwarf sensitive features ( na i at 8200 and feh at 9919 ) . in this paper , we present a library of empirical stellar spectra for spectral types o5-l3 . our library separates out luminosity classes ( dwarf and giant ) for spectral types a0 through m8 , and contains metallicity bins for spectral types a3 through m8 . we do not include white dwarfs in our sample ; several catalogs of sdss white dwarfs can be found in other studies ( e.g. , * ? ? ? the templates were created by the co - addition of individual stellar spectra from sdss s baryon oscillation spectroscopic survey ( boss ) . boss contains over 500,000 well - calibrated stellar spectra spanning wavelengths 3,600 10,400 at a resolution of @xmath4 @xcite . this stellar template library spans stellar parameter space and contains wavelength coverage that is not available from other empirical libraries . in addition , because each object has photometry associated with it , we can use this library to identify empirical photometric relations among spectral type , metallicity , surface gravity and color . along with the templates , we are also releasing a revised version the hammer " spectral type facility @xcite , dubbed pyhammer " . the original code assigns an automatic spectral type by measuring a variety of spectral lines / features and performing a least - squares minimization . the code also allows the user to complete visual spectral fitting . we have rewritten the code in python using our templates for comparison , and augmented the code to assign a metallicity , in addition to a spectral type . we also include a gui for visual spectral typing . in the following sections , we present the methods for determining the spectral type , the radial velocity ( section [ rv ] ) , the surface gravity ( section [ logg ] ) , and the metallicity ( section [ metallicity ] ) . we then describe the co - addition process ( section [ coadd ] ) . next , we describe the results from the spectroscopic data of the templates ( section [ results : spec ] ) , and the photometric data ( section [ results : phot ] ) . finally we draw conclusions for our templates in section [ conclusions ] . we also include an appendix , which describes the pyhammer " spectral typing code released along with this paper . we selected and co - added spectra in bins of metallicity ( [ fe / h ] ) , surface gravity and spectral type to create our empirical catalog . since no universal pipeline or technique exists for determining metallicities and surface gravities over the full range of stellar temperatures , we used a combination of techniques to extract this information from individual spectra , dependent on their spectral type . we separated the spectra into the following six spectral classification categories , each of which have a different method for the determination of parameters : o and b stars ; a stars ; f , g and early k stars ; late k and early m stars ; and late m stars . our catalog contains templates for early - type l stars as well , but we did not separate by surface gravity or metallicity for the l subclasses because the majority of the metal sensitive and surface gravity sensitive features are in the infrared . @xcite assembled a library of ultracool dwarfs , which uses both original sdss spectra and newer boss spectra , and has been spectroscopically classified ( using the original hammer " program ) . we therefore selected all of the spectra taken with boss , and used those for our l dwarf templates . to ensure the quality of all of the spectra and photometry , we applied some basic quality cuts within our initial sdss casjobs query . we measured the signal - to - noise in each of the five photometric bands . we included the object if the median value of the five signal - to - noise values ( from each band ) was greater than ten . we also required that the spectra to be classified as a star ( not a galaxy or qso ) , and to be taken with the boss spectrograph . we required that the errors in the photometry be less than 0.1 mag in each individual band that is included in the final co - added photometry , and that the photometry in that band was not flagged for being deblended , containing a cosmic ray , or saturated . finally , we required that the extinction in the _ r_-band be less than 0.25 mag for the object to be included , which we found effectively removed any spectra that were visibly altered by extinction . once the spectra were selected , we visually classified each of the spectra using the hammer " spectral type facility @xcite . we then calculated a radial velocity ( section [ rv ] ) to shift each spectrum into its rest frame before further spectral analysis and/or co - addition . we separated the spectra into metallicity ( section [ metallicity ] ) and surface gravity ( section [ logg ] ) bins when applicable . there are some holes in our template coverage across parameter space , where we could not find boss spectra at all the metallicity / spectral type / surface gravity values . once segregated , each spectrum was cataloged and co - added to create empirical spectral templates ( section [ coadd ] ) . for the o , b , and a stars , we used the spectroscopic radial velocities from the sdss spectroscopic reduction pipeline @xcite . the spectro1d pipeline performs two estimates of radial velocities : a cross - correlation with stellar templates from sdss commissioning spectra with known radial velocities , and a comparison of emission lines to common galaxy and quasar emission lines . the final radial velocity is chosen from the method with the highest confidence level . the quoted accuracy is about 1015 km @xmath5 @xcite . stars with effective temperatures between 4,500 7,500k ( @xmath6k4-f0 ) fall within the temperature range where @xcite quote that the sspp produces reliable stellar parameters ( e.g. , metallicities , surface gravities , effective temperatures , and radial velocities ) . we used the radial velocities determined by the sspp for these intermediate temperature stars . the sspp determined the radial velocity shifts by cross correlating the templates with the elodie templates @xcite , a method which was shown to be more reliable than the original sdss spectroscopic reduction pipeline in @xcite . however , the elodie library is not complete , and if no matching template existed , the sspp adopted the original radial velocity determined by the spectro1d pipeline . the radial velocities determined by cross correlation to the elodie templates have reported uncertainties of 59 km @xmath5 @xcite . to determine the radial velocities for the lowest - temperature stars , we chose to perform our own cross correlation with the m dwarf template library from @xcite , for spectral subclasses k5 through m9 , and @xcite for the l dwarfs . the cross correlation method examines three regions of the spectrum ( 50006000 , 60007000 , and 80009000 ) , and for each region performs a cross correlation . the minimum of the cross correlation function is recorded , and a sigma clipped median of the measurements from all the regions is reported as the radial velocity . we calculated an uncertainty from the standard deviation of all the individual radial velocity measurements ( in each region ) added in quadrature with the precision at which radial velocities can be calculated from boss s resolution . we take this radial velocity precision to be @xmath7 km @xmath5 as determined by @xcite in a comparison with high precision radial velocity measurements of hyades cluster members . in the comparison with the hyades cluster , @xcite found radial velocities that were on average 4 km @xmath5 more precise than the standard sdss pipeline s radial velocities . for the templates with a signal - to - noise ratio ( snr ) @xmath8 we find a median uncertainty of @xmath9 km @xmath5 . the radial velocity code used to compute these values will be available with the pyhammer code . further descriptions of the code , and tests of its accuracy are given in the appendix . for the highest temperature stars ( o and b ) , we did not separate by surface gravity . even though we do not distinguish between dwarf and giant stars , we can eliminate white dwarf contaminants from our sample , which are the vast majority of non - main - sequence stars in this temperature regime @xcite . to address this contamination , we cross matched our sample with the @xcite catalog of sdss white dwarfs to remove the majority of the contaminants . we also spectroscopically vetted our sample of white dwarfs during the visual spectral classification process , removing any sources with significant line broadening associated with the extreme surface gravities of white dwarfs . this white dwarf vetting process was also applied to a - type stars . according to stellar evolution models , giants do not spend much time in the a star color regime @xcite . in a uniform sample , we could assume that the majority of the a stars in our sample were main sequence stars and would not need to be separated by surface gravity . however , due to the sdss target selection , previous studies have shown that a significant portion of halo stars with a star colors are giants called blue horizontal branch stars ( bhb ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , making surface gravity estimates necessary . because a stars do not fall within the conservative temperature range for which @xcite state that the sspp parameters are accurate , @xcite independently determined the surface gravities for over 10,000 stars with a - star colors in sdss by measuring the width of the wings of the balmer absorption lines . stars with higher surface gravities should exhibit noticeably broader lines because of increased pressure broadening . @xcite then compared the results of their surface gravity values to the results reported by the sspp and found @xmath10 of stars fell into the same category ( dwarf or giant ) , thus extending the temperature range for which the sspp can be trusted to distinguish main - sequence from giant a - type stars . we therefore adopted the sspp method and separated our a stars into a main sequence and a giant bin , using the same criteria listed in @xcite , where a bhb giant had log @xmath11 and a main sequence star had log @xmath12 . for more information about how the sspp calculated surface gravities see section [ fgklogg ] . in figure [ f : alogg ] , which shows the distribution of a star surface gravities , we see two distinct populations of a - type stars . the distribution is very similar to the previous work on surface gravities of a stars by @xcite . figure [ f : alogg ] also validates our visual vetting process to remove white dwarfs , since there are no stars with surface gravities in the white dwarf range ( typically log @xmath13 ) in our sample . we can then be relatively confident that white dwarf contamination in our o and b stars is also minimal . distribution of all the stars selected ( and spectral typed ) as a stars . there are two distinct populations : one centered around log @xmath14 and one around log @xmath15 . the population with lower surface gravity is consistent with blue horizontal branch giants , while the higher surface gravity population is composed of main sequence stars and/or blue stragglers . the dashed line at log @xmath16 marks the distinction between main sequence stars ( to the right of the line ) and bhb giants ( to the left of the line ) . ] the sspp computes an extremely robust calculation of stellar parameters based on both sdss spectroscopy and photometry . reported surface gravities from the sspp use ten different methods , each of which determines a surface gravity estimate . most of these methods are well tested , and include : minimization of spectra and photometry compared to models @xcite , neural network approaches @xcite , spectral line fits as a function of broadband colors @xcite , and measurements of well known surface gravity sensitive lines ( ca i : 4226.7 , mgh : 5200 ; * ? ? ? * ) . after a surface gravity is returned by each of these methods , statistical outliers are removed from the sample and the final bi - weight average of the log @xmath1 of the star is returned @xcite . the sspp returns surface gravities to an accuracy of @xmath17(log @xmath18 dex within the spectral range of f , g , and early k stars @xcite . we split the spectra into two different surface gravity bins representative of giants and dwarfs . the padova stellar evolution tracks @xcite specify a temperature , surface gravity , and evolutionary phase ( e.g. , subgiant , giant , supergiant , white dwarf ) for each initial mass . using this information , we chose to call anything with @xmath19 log @xmath12 a main sequence dwarf star , and anything with log @xmath20 a giant star . we do not include any stars with log @xmath1 values between 3.2 and 3.8 , but keep this range as a buffer zone , because stars with different temperatures turn off the main sequence at slightly different log @xmath1 values . anything with log @xmath21 is considered a white dwarf . we did not separate into giants , subgiants and supergiants because any further separation resulted in too few spectra for each bin . to distinguish low - temperature dwarfs from giants we followed the steps in @xcite . since giants are intrinsically bright , and thus have to be at large distances in sdss , they will not exhibit observable proper motions . to isolate main sequence stars , we required that the proper motion in ra or dec be greater than two times the error in the proper motion , while the giant stars must have proper motions less than two times the error in the proper motion . this proper motion cut however , introduces a high percentage of extra - galactic source contamination in the giant category because elliptical galaxies spectra often resemble m stars ( since m giants are the main observable constituent ) , and also will not have any measurable proper motion . @xcite note this problem in isolating m giants , and use a @xmath22 color cut ( equation 5 in * ? ? ? * ) , which @xcite show effectively separates the two populations . main sequence and giant stars also occupy distinct locations on the @xmath23 color diagram @xcite . to use the ir color cuts ( equations 1 3 in * ? ? * ) that make use of the dwarf - giant distinction in ir color space , we matched all of our sdss targets with the corresponding targets from the two micron all - sky survey ( 2mass ) , which provided us with @xmath24 , @xmath25 , and @xmath26-band photometry for each object . @xcite conclude that using all of the m giant photometric and proper motion cuts returns approximately @xmath27 m giants ( contaminants include : dwarfs , galaxies , other sources ) . contamination in the m dwarf category should be minimal , since any giant close enough to have a measured proper motion would saturate the sdss detectors . to remove the remainder of the non - giant contaminants from our m giant category , we resorted to visual inspection of the spectra . we found a few carbon star and extragalactic sources , which could easily be identified spectroscopically . to spectroscopically distinguish dwarfs from giant stars , we focused on the gravity sensitive features , na i ( 8200 ) and the calcium triplet ( @xmath68600 ) . the na i doublet is extremely sensitive to gravity @xcite and is only strong in dwarf stars , while the calcium triplet is more prominent in giant stars @xcite . after our spectroscopic classification , we conservatively removed @xmath28 of the selected stars and we expect the vast majority of our remaining spectra are true m giants . we do not separate o and b - type stars by metallicity . the metallicity history of the disk of the milky way is well constrained in time @xcite , with recent measurements of metal content being roughly solar metallicity . since o and b stars have such short lifetimes ( @xmath29 million years ) and the majority of stars are formed within the disk , we can assume that the majority have metallicities ( [ fe / h ] ) between 0.0 and 0.5 dex . because a stars have significant lifetimes ( up to a couple billion years ) , we expect there to be a larger spread in metallicities than found in o and b stars . previous studies of a stars within sdss have also shown a much larger spread in metallicity than the metallicity histories of the milky way disk predict ( e.g. , * ? ? ? * ) , making it important to determine metallicity estimates for these stars . since the halo does not have ongoing star formation , we would expect the only low - metallicity stars to be low - mass stars , because a stars formed when the galaxy was on average lower - metallicity would have evolved off the main sequence . however , low - metallicity ( [ fe / h ] @xmath30 ) a stars with surface gravities comparable to main sequence stars ( dubbed blue stragglers " ) are observed in the halo with high frequencies @xcite . because a stars have effective temperatures @xmath31 k , and are therefore outside the quoted accuracy limits of the sspp , we tested the sspp metallicity outputs by comparing to synthetic stellar models . we used the phoenix models ( bt settl grid ; * ? ? ? * ) and compared different metallicities of the same temperature to determine the most metal sensitive lines . synthetic models of a stars are reliably accurate because their absorption profiles are relatively simple ( do not contain molecules or dust ) , yet they are not quite hot enough to be dominated by non - local thermodynamic equilibrium metal line - blanketing @xcite . figure [ f : adivision ] shows the ca ii k line ( 3933.7 ) , which is by far the most metal sensitive line and remains metal sensitive for temperatures from 7400 k 9800 k. above 9800 k , the effective temperature is so high that the ca ii k line is weak and the depth of the ca ii k absorption line is on the same order as the uncertainties , making metallicity estimates for a0 through a2 stars unreliable . = 0.5 for different spectral subtypes @xcite . the ca ii k line shows the largest dependence on metallicity for a stars . for the a8 ( top ) and a4 ( middle ) stars , there are significant changes between the metallicity bins . while the a1 ( bottom ) spectrum does show changes in ca ii k line depth , they are not significant enough changes for us to confidently separate out metallicity bins based on the equivalent width of the ca ii k line . ] we created metallicity estimates by determining the equivalent width of the ca ii k line for different metallicity bt settl models @xcite . we used models with a log @xmath1 value of @xmath32 . to determine the equivalent widths of the bt settl models , we first changed the resolution to match that of boss ( r=2000 ) . we processed all of the models through the hammer " spectral typing facility to accurately compare the effective temperature based models to the the spectral typed boss spectra . effective temperature and spectral type should be roughly one to one , but are not necessarily regularly spaced throughout the entire stellar spectral sequence . to construct the metallicity bins , we measured the equivalent width of ca ii k for stars with metallicities of [ fe / h ] = -2.0 through [ fe / h ] = + 0.5 ( with bin sizes of @xmath33 dex ) , for the equivalent model of each spectral subtype . we calculated a 5-sigma clipped average of the regions to the left ( 4010 4030 ) and to the right ( 3913 3933 ) of the feature to fit a continuum value . depending on the temperature of the model spectrum , we either defined a 12 region ( for stars with @xmath34 k ) or a 6 region ( for stars with @xmath35 k ) around the central wavelength of 3933.7 , to completely encase the absorption line . using the equivalent widths of the ca ii k lines measured from the model spectra , we identified a relation between metallicity and the strength of the ca ii k absorption line ( table [ t : astar ] ) . we then measured the equivalent width of the line using the same numerical integration technique on all the boss spectra , which were first shifted into their rest frame using the radial velocities ( see section [ rv ] ) . we then separated the spectra into metallicity bins using the equivalent width regions in table [ t : astar ] . c c c c c c l a3 & 0.2 & 0.35 & 0.56 & 0.74 & 0.92 + a4 & 0.36 & 0.56 & 0.87 & 1.10 & 1.50 + a6 & 0.44 & 0.63 & 1.00 & 1.30 & 1.66 + a7 & 0.53 & 0.84 & 1.35 & 1.71 & 2.00 + a8 & 0.62 & 1.05 & 1.39 & 2.19 & 2.63 + a9 & 0.74 & 1.24 & 1.97 & 2.62 & 3.09 + [ t : astar ] we compared our metallicity estimates to those from the sspp as both a sanity check , and as validation that the sspp can determine metallicity estimates passed the quoted upper temperature limit . we find that @xmath36 of the sspp estimates are within 1 bin of our estimates . we expect this type of spread within 1 bin since even with small uncertainties in [ fe / h ] ( @xmath37 dex ) , the spectra can be placed in an adjacent bin . we can conclude that the sspp and measurements of the ca ii k line are in relative agreement . with this validation , we continue to use our derived ca ii k metallicity bins , and are confident that these cuts represent real changes in metallicity . the metallicity estimate returned by the sspp uses a combination of 12 different methods , which include all the same spectral template fitting , neural network approaches , and photometry comparisons as mentioned in section [ fgklogg ] . however , for metallicity estimates , different lines that are known to be metal sensitive ( instead of gravity sensitive ) are used @xcite , including the ca ii h and k lines ( @xmath38 ) and the ca ii triplet ( @xmath39 ) . after determining metallicities in all these ways , statistical outliers were removed from the sample and the final bi - weight average of the metallicity of the star is returned @xcite in the same way as was done for the log @xmath1 values . this method is accurate to @xmath17([fe / h ] ) = 0.11 dex @xcite . low mass stars molecule rich atmospheres make them extremely difficult to model @xcite , and therefore we can not use the same method as the a stars ( comparison to models ) to measure their metallicities . recent work using m dwarfs with wide binary fgk companions have led to several new techniques for estimating m dwarf metallicities . @xcite uses @xmath40 absorption features in k5 through m4 dwarfs that are shown to be metal sensitive to estimate metallicities to a precision of @xmath41 dex for the metallicity range @xmath42[fe / h]@xmath43 . we employed the @xcite method and measured metallicities for over 3,000 individual m dwarfs ( no giants ) . each spectrum was separated into a metallicity bin : [ fe / h ] @xmath44 ; -1.0 @xmath45 [ fe / h ] @xmath46 ; @xmath47 [ fe / h ] @xmath48 ; @xmath49 [ fe / h ] @xmath50 ; and [ fe / h ] @xmath51 . the majority of methods for determining the metallicity for late - type m dwarfs ( m5-m8 ) use ir spectral features ( e.g. , * ? ? ? * ; * ? ? ? * ) , which are not within the boss spectral coverage . @xcite also showed that the spread in the @xmath52 color around the stellar locus in a @xmath53 color diagram is primarily due to metallicity . we obtained @xmath54 and @xmath26-band photometry by matching the sdss catalog with the 2mass catalog . we therefore separated the m dwarf spectra ( not giants ) into rough metallicity bins using ir color cuts equation 15 in @xcite , which have a precision between @xmath55 dex and @xmath33 dex . we created three metallicity bins : [ fe / h ] @xmath56 ; @xmath47 [ fe / h ] @xmath48 ; and [ fe / h ] @xmath57 . even though the uncertainties are relatively high , the overlap between the high and low metallicity bins should be minimal since the difference in metallicity between our low and high metallicity bin is greater than the 0.5 dex uncertainty stated by @xcite . the boss spectra are logarithmically spaced and in vacuum wavelengths . using the radial velocities determined as described above , we shifted all of the spectra into their rest frames . we used a similar method to @xcite , except our wavelength grid was spaced evenly in logarithmic space ( intervals of 5 km s@xmath58 ) instead of linearly spaced . the flux was then inserted ( without interpolation ) into the appropriate location on the flux grid corresponding to the shifted wavelength . the flux grids were then normalized to the flux at 8000 . the spectral resolution is increased in this process because the shifts in radial velocity can be measured to precisions of @xmath59 km s@xmath58 and were added into a grid with a spacing of 5 km s@xmath58 . the process is essentially combining many low resolution spectra into a higher resolution template , and is very similar to the drizzle " process used to combine astronomical images @xcite . during the co - addition , we also corrected for a spike in flux located around 5600 present in many of the spectra , caused by stitching the spectra from the red and blue boss cameras together . finally , we trimmed the grids at 3650 and 10200 , to avoid areas that were not complete after the radial velocity shifts . we also propagated the uncertainty reported by boss for each spectrum throughout this process and calculated the standard deviation of the co - added templates at each spectral channel for each template . c c c c c c l o5 & 10 & 1.015 & 1.013 & 0.892 & 0.912 + b6 & 5 & 0.986 & 1.019 & 0.834 & 1.029 + a6_-0.5_dwarf & 3 & 0.953 & 1.003 & 0.788 & 0.979 + f6_+0.0_dwarf & 29 & 0.648 & 0.965 & 0.881 & 0.985 + g6_-1.5_giant & 13 & 0.587 & 0.975 & 0.922 & 0.987 + k5_-0.5_dwarf & 11 & 0.588 & 0.781 & 0.954 & 0.991 + m4_+0.5_dwarf & 7 & 1.010 & 0.672 & 1.053 & 0.902 + l1 & 34 & 0.613 & 1.391 & 1.043 & 0.683 + [ t : spectra ] we present a subset of our 324 template spectra in table [ t : spectra ] ; the full table is available for download in the online journal . each entry displays the number of boss spectra co - added to create the template , along with measurements of 29 spectral indices . the spectral indices , which are ratios of the flux in the region of the absorption feature divided by the continuum flux , are the features used by pyhammer to estimate the spectral type and metallicity ( see section [ pyhammerestimates ] for details on the spectral indices ) . figures [ f : obaspec ] through [ f : mkspec ] display a subset of the template spectra . all of the spectra are available in fits formats by clicking on the link for the data behind the figures , and online in sdss formatted fits binary tables , or in standard fits format in a public dropbox folder . for those interested in the complete sample , all the individual boss spectra co - added for each template are available in a zenodo repository , organized by the spectral type , luminosity class , and metallicity , or by contacting the corresponding author . figures [ f : label_highmass ] and [ f : label_lowmass ] show one spectrum from each spectral class , with all of the prominent absorption features labeled . these figures will be especially useful in the classroom . the templates and associated figures can be used by introductory students first viewing stellar spectra , as well as the seasoned astronomer , who may not have seen high - quality spectra that span the entire optical wavelength and parameter space . figure [ f : metalspec ] shows an expanded view of the ca ii k and na i d metal sensitive features in different regions of the spectra ( 3933 and 5890 , respectively ) . the ca ii k line is prominent in high - temperature stars , while the na i d lines are almost absent . in low - temperature stars , the trend is the opposite . we find that the na i d lines are a useful metallicity indicator for f through early m - type stars , and the ca ii k line to be a useful metallicity indicator for a and f stars . the higher metallicity spectra have deeper absorption features ( larger equivalent widths ) for both of these features . in the bottom panel of figure [ f : metalspec ] , the lowest metallicity templates also show very few features outside of the neutral sodium d doublet and are virtually flat for the f6 and g5 spectra . the k3 and m1 spectra have very few low - metallicity spectra that are co - added to create the templates , and are therefore more noisy than their higher metallicity counterparts , yet still show a lack of real absorption features . figure [ f : dwarfgiantspec ] is similar to figure [ f : metalspec ] except it shows differences between dwarf and giant spectra for a range of spectral types . because g and k stars were selected with the sspp , we have metallicities for both the dwarf and the giant star , and therefore show the same metallicity and spectral type . the top panel of figure [ f : dwarfgiantspec ] shows the entire spectrum for both a dwarf and a giant spectrum for the spectral classes g5 , k3 and m0 . the continua ( even without features ) are slightly different , with more flux in giants coming from the red end , and more flux in the dwarfs coming from the blue end of the spectrum . the bottom panels show a closer look at three absorption features , which which are sensitive to surface gravity ( bottom left : mg b / mgh and na i d , bottom right : ca ii triplet ) . the mg b and mgh feature around 5100 is absent in metal poor giants , weaker in population i giants as compared to dwarfs , and prominent in dwarfs @xcite . the m giant does not include a metallicity estimate , but because the mg / mgh feature is absent , we can assume it is a sub - solar metallicity giant . the na i d lines are also often used as an age indicator since the feature is extremely strong in dwarfs and nearly non - existent in giants @xcite . alkali atoms in general are all extremely sensitive to density because the sole valence electron can easily be perturbed by small changes in pressure @xcite , making sodium transitions good surface gravity ( and age ) indicators . lastly , the equivalent width of the calcium triplet around 8600 is known to be larger in giants than in dwarfs @xcite , which our spectra show in the the bottom left panel . in our sample , we find low - metallicity k giants , but no high - metallicity k giants or late - type m giants ( of any metallicity ) . stellar evolution models show that low and high - metallicity giants occupy two different temperature regimes . stars with lower metallicity become hotter giants since they can not cool as efficiently , while higher metallicity giants can cool more quickly and therefore occupy the late - type m giant regime @xcite . with the stellar evolution models in mind , the lack of late - type giants is expected in the sdss data ; there is a strong bias in the metallicity of the giants we can observe with sdss . because of the sight lines observed in the sdss footprint , the most distant stars ( i.e. giants ) are located at high galactic latitudes . the stars located at large distances above the galactic plane are on average between 0.5 and 1.0 dex more metal poor than their closer counterparts @xcite . therefore , in sdss we do not expect to see many ( high metallicity ) m - type giants , which is validated by @xcite who find that less than 2% of stars redder than a spectral type of k5 in a magnitude limited sdss field were giants . along with these templates , we are releasing a computing product ( dubbed pyhammer " ) , which can assign both an automatic estimate of spectral type and metallicity , and/or be used to visually classify spectra . pyhammer uses the templates from this paper as comparisons to automatically determine the spectral type and an estimate of the metallicity by measuring prominent line indices and performing a weighted least squares minimization . the gui then allows the user to visually compare their spectra to the templates and visually determine the spectral type and metallicity . this code is based on the hammer " spectral typing facility @xcite , but has been updated to include metallicity information and now is in python ( instead of idl ) . we have also added a component to determine the radial velocity of the stars , and improved on the automatic guessing method . the pyhammer code is unique because it does not use synthetic spectra to determine metallicities and spectral types , and both has an automatic and visual or by - eye " component . details of the code ( including how to download and operate it ) can be found in the appendix . along with the spectra , we report average photometric colors for each template . table [ table : photometry ] shows a sample of our templates with the photometry information . the full table , which includes each template with all of the color information and associated uncertainties is available to download in ascii format in the online journal . the colors are averaged over each individual spectrum co - added in a template , after those with bad photometry are removed ( i.e. , flagged for being deblended , containing a cosmic ray , or saturated ) . the rms is the standard deviation of all of the colors for the individual spectra , and the @xmath17 is the propagated uncertainty in photometry provided by sdss . on average , the low metallicity templates have bluer colors than the high metallicity templates . a comparison with @xcite indicates that the colors are comparable . figure [ f : colorcolor_metal ] shows graphically the trend that low - metallicity stars are bluer on average . the overall trends ( ignoring metallicity ) are almost identical to those shown in @xcite , however when metallicity information is added , the spread in the main sequence seems to be almost entirely due to metallicity . the top two plots in figure [ f : colorcolor_metal ] show an especially significant distinction between the low and high - metallicity stars , which demonstrates that the _ u - g _ color is extremely useful for differentiating metallicities . this trend seems to be valid from f dwarfs through mid - type m dwarfs ( @xmath6m4 ) . the a stars seem to hold to the same trend , however the photometry is not as spread out in that region of color - space , making it extremely difficult to accurately distinguish metallicities . beyond the mid - type m dwarfs , the light emitted in the _ u_-band is too faint to report an accurate _ u_-band measurement . other studies of stars in sdss ( e.g. , * ? ? ? * ) have observed color trends with metallicity in the past , however metallicity information for low - mass stars was not available for those studies . figure [ f : colorcolor_dg ] shows three color - color diagrams demonstrating the color difference between luminosity classes . in general , all of the plots show the trend that giants emit more light in redder wavelengths than dwarfs . to ensure this trend is not an effect of extinction , we compared the average extinction of each template in each band ( provided by the sdss database ) . we find the difference in average extinction between the dwarf and giant templates to be smaller than the propagated errors in extinction provided by sdss , leading us to conclude that any significant change in color is not an effect of extinction and thus , a real physical effect . for the higher temperature stars ( f / g ) , we can only see the trend for bins of low metallicity because we do not have many templates for high - metallicity giants at high temperatures , and for low - temperature stars ( k / m ) we can only see the trend for high metallicities because we do not have many low - temperature , low - metallicity dwarf templates . however , with a complete set of data , we expect this trend to persist for all metallicity bins . we see the most separation between dwarf and giant stars for the lowest temperature stars ( k / m ) . k and m stars contain many large molecular features and absorption lines , so this prominent color difference can be attributed to the surface gravity affecting the strength of many of these lines . the higher temperature stars do not show any separation in most cases ( a - type , g / k - type ) , and a slight separation around @xmath60 ( f - type stars ) . c c c c c c c c c c c c c c o5v & - & -0.24 & 0.14 & 0.02 & -0.51 & 0.02 & 0.02 & -0.36 & 0.02 & 0.02 & -0.31 & 0.02 & 0.02 + o7v & - & -0.31 & 0.05 & 0.01 & -0.47 & 0.01 & 0.01 & -0.32 & 0.01 & 0.01 & -0.34 & 0.02 & 0.02 + a3v & + 1.0 & 1.2 & 0.04 & 0.02 & -0.02 & 0.05 & 0.02 & -0.08 & 0.01 & 0.02 & -0.05 & 0.04 & 0.03 + a3v & 0.0 & 1.17 & 0.01 & 0.01 & -0.07 & 0.03 & 0.01 & -0.12 & 0.02 & 0.01 & -0.07 & 0.02 & 0.01 + a3v & -0.5 & 1.12 & 0.04 & 0.02 & -0.09 & 0.02 & 0.01 & -0.1 & 0.02 & 0.01 & -0.09 & 0.02 & 0.02 + m0v & 1.0 & 2.84 & 0.05 & 0.04 & 1.29 & 0.05 & 0.01 & 0.57 & 0.03 & 0.01 & 0.3 & 0.02 & 0.01 + m0v & 0.5 & 2.59 & 0.03 & 0.01 & 1.28 & 0.01 & 0.0 & 0.58 & 0.01 & 0.0 & 0.35 & 0.01 & 0.0 + m0v & 0.0 & 2.46 & 0.03 & 0.01 & 1.26 & 0.01 & 0.0 & 0.56 & 0.01 & 0.0 & 0.33 & 0.01 & 0.0 + f0iii & -0.5 & 1.25 & 0.01 & 0.02 & 0.27 & 0.0 & 0.01 & 0.07 & 0.02 & 0.01 & 0.05 & 0.0 & 0.02 + f0iii & -1.0 & 1.21 & 0.03 & 0.01 & 0.2 & 0.03 & 0.01 & 0.05 & 0.02 & 0.01 & 0.01 & 0.01 & 0.01 + f0iii & -1.5 & 1.16 & 0.05 & 0.01 & 0.18 & 0.03 & 0.01 & 0.04 & 0.02 & 0.01 & 0.02 & 0.01 & 0.01 + f1iii & -1.0 & 1.29 & 0.0 & 0.02 & 0.13 & 0.0 & 0.01 & -0.04 & 0.0 & 0.01 & 0.01 & 0.0 & 0.02 + [ table : photometry ] using data taken with the sdss boss spectrograph , we have compiled a new empirical stellar template library . our template library : * covers spectral types o5 through l3 * includes dwarf and giant separation for spectral types a0 through m8 * contains metallicity [ fe / h ] bins for spectral types a3 through m8 * reports averaged photometric colors ( in sloan bands ) for all the co - added stars in each template , along with a propagated errors and standard deviations along with the templates , we have released the pyhammer code for assigning a spectral type and metallicity automatically ( or by visual inspection ) . this code is based on the hammer " spectral typing facility , written by @xcite , but includes metallicity information and is now written in python . the automatic spectral typing portion of code returns the exact spectral type we determined using the original hammer " code and metallicity we determined using the methods described in section [ metallicity ] over 50% of the time . the spread in the spectral type was 1.5 spectral sub - types , and a spread in the metallicity was 0.4 dex . visual spectral typing allows for direct comparison between input spectra and our empirical templates in an easy to use gui . the code is available on github . the library of empirical stellar spectra will be important for a wide range of research topics from extragalactic to galactic astronomy , planetary system stellar characterization , and even as an astronomical teaching tool . with large photometric surveys such as lsst , machine learning techniques will become increasingly important to quickly characterize large amounts of data . along with releasing our templates , we will provide lists of the individual boss spectra co - added to construct each template . this combination of information will be an ideal training set for machine learning , and can extend the work of @xcite on f , g , and k stars to both higher and lower mass stars . the templates also provide the necessary tool for characterizing stellar populations in other galaxies , especially for studies of the imf . scrutiny of the low - mass end of the imf , has led many people to suggest it changes form in different environments ( i.e. different metallicity environments ) . our catalog represents the first empirical template library with metallicity and surface gravity separation for low - mass ( m - type ) stars . the catalog and the new pyhammer " spectral typing facility will be a useful tool for the community as a whole . the authors would like to thank phil muirhead and chris theissen for reading this manuscript and providing much helpful feedback . the authors would also like to thank dylan morgan , chris theissen , conor robinson and philip phipps for help developing the pyhammer " code . a.a.w . also acknowledges the support of the nsf grants ast-1109273 and ast-1255568 along with research corporation for science advancement s cottrell scholarship . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is @xmath61 . the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max planck institute for astronomy ( mpia ) , the max planck institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . this publication makes use of data products from the two micron all - sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . this publication also 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 . in this appendix , we explain in detail the automatic , and visual spectral typing code , dubbed pyhammer " . the code automatically outputs best estimates for the radial velocity , spectral type , and metallicity ( when metallicity information is available : a3 - m8 ) . we also present the methods for determining these estimates , tests to determine the accuracy / precision of these methods , and the resulting accuracy for each parameter determined . there is also a visual , or by - eye " spectral typing feature , which allows the user to visually compare the input spectrum to any of the template spectra in a gui window . the pyhammer code is available on github and we have set up an email address for questions , comments and suggestions . the general procedure of the code is to first interpolate the input spectrum onto the same wavelength grid as our templates and convert to vacuum wavelengths ( if necessary ) to allow for direct comparison . we then measure 34 spectral indices and color regions given in table [ table : hammer ] ( all in vacuum wavelengths ) and make an initial estimate of the spectral type using these indices . with that spectral type estimate , we can determine the radial velocity by cross correlating the spectrum with the corresponding template . we then shift the input spectrum to its rest frame , and re - measure the 34 spectral indices to determine a more accurate estimate of the spectral type and metallicity . the radial velocity cross correlation method is based on an idl procedure ` xcorl ` @xcite , which was translated into python by @xcite . the cross correlation method examines three regions of the spectrum ( 5000 - 6000 , 6000 - 7000 , and 7000 - 8000 ) , and for each region performs a cross correlation . the shift that produces a minimum in the cross - correlation function is recorded , and a sigma - clipped median of the measurements from all the regions is reported as the radial velocity . to test the radial velocity code , we compared our measured radial velocities to previously measured radial velocities across the full spectral range . since there were three different methods ( one for o , b , and a - type stars , one for f , g , k - type stars , and one for m - type stars ) used for the initial radial velocity calculations , each with different uncertainties , we chose a sample of spectra in each of the temperature regimes with which to compare our radial velocities . for the f , g and k - type stars , we compared our radial velocities to radial velocities derived in the sspp , which are accurate to @xmath679 km s@xmath58 ( see section [ fgk : rv ] ) . to validate the radial velocities for the low - temperature stars , we compared the radial velocities of the @xcite sample of m dwarfs , derived by cross correlation with @xcite m dwarf templates , to our measured radial velocities . the m dwarf radial velocities reported in the sample ( boorv field ) are accurate to @xmath610 km s@xmath58 . for the higher temperature stars , we compared our radial velocities to the radial velocities derived from the sdss pipeline , which are accurate to 10 15 km s@xmath58 ( see section [ oba : rv ] ) , making this comparison the least accurate . the results are shown in figure [ f : rv ] . for all the comparisons the difference between the two radial velocities peaks around 0 km s@xmath58 . instead of computing a simple standard deviation , we recorded the value which encloses 34% of the data both above and below the median . we find this to be more representative than a standard deviation because the distributions are not true gaussians ( especially for the o and b - type stars ) . we find a median value of -1.2 km s@xmath58 between our measurements and the sspp values for the g dwarfs . the 1-sigma " spreads above and below the median are 7.9 km s@xmath58 and -10 km s@xmath58 respectively . for the low - temperature stars , we find a median of 2.2 km s@xmath58 , a 1-sigma " upper and lower bound of 10.8 km s@xmath58 and -7.1 km s@xmath58 respectively . the high - temperature stars have the lowest precision , with a median of -1.6 km s@xmath58 , an upper bound of 10.7 km s@xmath58 and a lower bound of -22.7 km s@xmath58 . the high - temperature distribution contains many outliers , which we believe is due to a combination of higher uncertainty in the sdss pipeline for high - mass stars than the 10 15 km s@xmath58 quoted values , and higher errors in our calculated radial velocities because of fewer absorption lines ( especially in o stars ) . our analysis of figure [ f : rv ] leads us to conclude that our radial velocity measurements have comparable uncertainties to those given by previous measurements for mid - temperature and low - temperature stars , with an uncertainty of @xmath62 km s@xmath58 . the radial velocities of the high - temperature stars are on par or slightly less accurate than previous measurements , with uncertainties on the order of 15 - 20 km s@xmath58 . to assign an initial estimate of the spectral type and metallicity , we compared the input spectrum to the template spectra in the following manner . we first measured 34 spectral indices ( table [ table : hammer ] ) from each individual spectrum that was used to create the templates . for all spectra of a given spectral type and metallicity ( i.e. , each template ) , we measured the weighted mean and variance of each index for all of the co - added spectra , where the weight is the variance in the index value due to uncertainties in the observed spectrum . we repeated this process for all spectral types and metallicities . to reduce computation time , the results of this procedure were saved to an external file and not repeated each time the code was run . we compared the input spectrum to each template by computing a chi - squared value that compares the spectral indices measured from the input spectrum to the weighted mean indices described above . the variance used in the chi - squared calculation is the variance among indices of multiple stars of the same template type as described above . we reported the spectral type and metallicity of the template that produces the minimum chi - squared as the initial estimate . we chose to use indices instead of doing a chi - squared minimization of the entire spectrum because spectral indices are normalization independent , robust against missing data , and take less computational time . we tested the automatic spectral typing by running the code on the individual spectra that were used in constructing the templates and compared the estimated metallicity and spectral type to the actual metallicity and spectral type we determined . the results are shown in figure [ f : spt_compare ] . both comparisons show good agreement , with over 50% of the estimates exactly the same as the determined spectral types and metallicities . the spread in both comparisons is also extremely small , with the standard deviation being 1.5 spectral subtypes in the spectral - type comparison , and 0.4 dex in the metallicity comparison . we conclude that even without the additional visual inspection , the automatic spectral typing and metallicity estimates are within one metallicity bin and one spectral subtype over 80% of the time . c c c l ca ii k & 3923.7 - 3943.7 & 3943.7 - 3953.7 + h@xmath63 & 4086.7 - 4116.7 & 4136.7 - 4176.7 + ca i 4227 & 4216.7 - 4236.7 & 4236.7 - 4256.7 + g - band & 4285.0 - 4315.0 & 4260.0 - 4285.0 + h@xmath64 & 4332.5 - 4347.5 & 4355.0 - 4370.0 + fe i 4383 & 4378.6 - 4388.6 & 4355.0 - 4370.0 + fe i 4404 & 4399.8 - 4409.8 & 4414.8 - 4424.8 + h@xmath65 & 4847.0 - 4877.0 & 4817.0 - 4847.0 + mg i & 5152.7 - 5192.7 & 5100.0 - 5150.0 + na d & 5880.0 - 5905.0 & 5910.0 - 5935.0 + ca i 6162 & 6150.0 - 6175.0 & 6120.0 - 6145.0 + h@xmath66 & 6548.0 - 6578.0 & 6583.0 - 6613.0 + cah2 & 6814.0 - 6845.0 & 7042.0 , 7046.0 + cah3 & 6960.0 - 6990.0 & 7042.0 - 7046.0 + tio5 & 7126.0 - 7135.0 & 7042.0 - 7046.0 + vo 7434 & 7430.0 - 7470.0 & 7550.0 - 7570.0 + vo 7445 & 7350.0 - 7400.0 , 0.5625 ; 7510.0 - 7560.0 , 0.4375 & 7420.0 - 7470.0 + vo - b & 7860.0 - 7880.0 , 0.5 ; 8080.0 - 8100.0 , 0.5 & 7960.0 - 8000.0 + vo 7912 & 7900.0 - 7980.0 & 8100.0 - 8150.0 + rb - b & 7922.6 - 7932.6 , 0.5 ; 7962.6 - 7972.6 , 0.5 & 7942.6 - 7952.6 + na i & 8177.0 - 8201.0 & 8151.0 - 8175.0 + tio8 & 8400.0 - 8415.0 & 8455.0 - 8470.0 + tio 8440 & 8440.0 - 8470.0 & 8400.0 - 8420.0 + cs - a & 8496.1 - 8506.1 , 0.5 ; 8536.1 - 8546.1 , 0.5 & 8516.1 - 8526.1 + ca ii 8498 & 8483.0 - 8513.0 & 8513.0 - 8543.0 + crh - a & 8580.0 - 8600.0 & 8621.0 - 8641.0 + ca ii 8662 & 8650.0 - 8675.0 & 8625.0 - 8650.0 + fe i 8689 & 8684.0 - 8694.0 & 8664.0 - 8674.0 + feh & 9880.0 - 10000.0 & 9820.0 - 9860.0 + color region 1 & 4550 - 4650 & 4160 - 4210 + color region 2 & 5700 - 5800 & 4160 - 4210 + color region 3 & 7480 - 7580 & 4160 - 4210 + color region 4 & 9100 - 9200 & 4160 - 4210 + color region 5 & 10100 - 10200 & 4160 - 4210 + detailed instructions on running pyhammer as well as more advanced features of the code are provided in the readme file in github , and in the ` help ' menu ( available when the visual classification guis are displayed ) . here , we will give a brief overview of how to run the code and a description of some of the features available . after starting the code ( type python pyhammer.py " on the command line ) , an initial gui window will appear , allowing the user to enter the name of ( or create ) the input and output files , specify a path to the spectra files , skip straight to the by - eye spectral typing ( as opposed to doing the automatic spectral typing first ) , and apply a signal - to - noise cut - off . skipping directly to the eye check stage should only be done if the automatic spectral typing has already been completed at an earlier time . if the user does not skip the automatic spectral typing , the program determines the radial velocity , metallicity and spectral type and writes these to the output file ( pyhammerresults.csv by default ) . if the user supplies a signal - to - noise cut - off , any spectrum with a median signal to noise ratio ( calculated simply as the median flux divided by the median uncertainty over the entire spectrum ) lower than the specified value is written to rejectspectra.csv . there is an example input file ( exampleinputfile.txt ) and a few spectra in the test_case folder to demonstrate how the input file should be set up , and to aid the first time user . a screenshot of the gui for the visual spectral typing is shown in figure [ f : gui ] . the gui allows the user to display the spectra and the templates on the same plot , allowing for direct comparisons . all of the normal matplotlib graphing buttons are functional , and shown in the top left corner of the gui in figure [ f : gui ] . by pushing the magnifying glass button , the user can zoom in on specific regions of the spectrum . while zoomed in the user can use the four sided arrow to scroll . finally , by pressing the home button , the graph will be taken to the original view . in the ` options ' menu at the top of the screen there are four different viewing options for the graph . the user can choose to display or not display the template rms ( in transparent blue ) . the user can also smooth the spectrum , and lock the smoothing since by default every new spectrum loaded will be unsmoothed . the smoothing option runs a simple boxcar smoothing over the input spectrum , which reduces noise and allows for easier comparisons for low signal to noise spectra . finally if the spectra are from sdss , the user can choose to remove the known false spike in the spectra at 5580 , created by the stitching together of the red end and blue end of the spectra . along with simply clicking through different templates on the bottom part of the gui shown in figure [ f : gui ] , there are many options designed to make visual spectral typing easier . the ` earlier ' and ` later ' buttons change the spectral subtype by one each time , where ` earlier ' scrolls to lower numbered subtypes ( e.g. , m4 to m3 or m0 to k7 ) , and the ` later ' to higher numbered subtypes . while the ` lower ' and ` higher ' buttons change the metallicity by one bin ( 0.5 dex ) with each click . there are also buttons with other options , which have been adopted from the original hammer code . the ` odd ' button allows the user to input their own note into the space where the spectral type would normally be stored . we have set a few standard ` odd ' spectral types like white dwarf ( wd ) , white dwarf+m dwarf binary ( wdm ) , carbon star , galaxy ( gal ) and unknown , but the user may also type anything he or she would like into this space . finally , the user can proceed to the next spectrum from their input list or return to a previous spectrum using the ` back ' , ` next ' buttons . each of these buttons also has a keyboard shortcut so the entire visual spectral typing can be done with only a keyboard ( to speed up the process ) . for more information on the keyboard shortcuts go to the ` help ' menu and click on the ` keys ' tab in the pyhammer help window .

we consider an exact reflection symmetry in neutrino sector realized at the gut scale in the context of the seesaw model with and without supersymmetry . assuming the two lighter heavy majorana neutrinos are degenerate at the gut scale , it is shown that the renormalization group ( rg ) evolution from the gut scale to the seesaw scale gives rise to breaking of the symmetry and a tiny splitting between two degenerate heavy majorana neutrino masses as well as small variations of the cp phases in , which are essential to achieve a successful leptogenesis . such small rg effects lead to tiny deviations of from the maximal value and the cp phase from imposed at the gut scale due to reflection symmetry . in our scenario , the required amount of the baryon asymmetry can be generated via so - called resonant-leptogenesis , in which the wash - out factor concerned with electron flavor plays a crucial role in reproducing a successful leptogenesis . we show that cp violation responsible for the generation of baryon asymmetry of our universe can be directly linked with cp violation measurable through neutrino oscillation as well as neutrino mixing angles and . we expect that , in addition to the reactor and long baseline neutrino experiments , the measurements for the supersymmetric parameter at future collider experiments would serve as an indirect test of our scenario of baryogenesis based on the reflection symmetry . + pacs : 14.60.pq , 11.30.fs , 11.10.hi , 98.80.cq , 13.35.hb

0811.1458

the present neutrino experimental data @xcite exhibit that the atmospheric neutrino deficit points toward a maximal mixing between the tau and muon neutrinos , however , the solar neutrino deficit favors a not - so - maximal mixing between the electron and muon neutrinos , @xmath11 , @xmath12 , where @xmath13 is the leptonic pmns mixing matrix . in addition , although we do not have yet any firm evidence for the neutrino oscillation arisen from the 1st and 3rd generation flavor mixing , there is a bound on the mixing element @xmath14 from chooz reactor experiment , @xmath15 @xcite . although neutrinos have gradually revealed their properties in various experiments since the historic super - kamiokande confirmation of neutrino oscillations @xcite , properties related to the leptonic cp violation are completely unknown yet . cp violations in the leptonic sector are also obligatory , if the matter and antimatter asymmetry of the universe is made through leptogenesis scenario @xcite in the seesaw models @xcite . recently , the so - called @xmath16-@xmath17 reflection symmetry @xcite has been imposed in light neutrino mass matrices so that the maximal atmospheric mixing @xmath18 as well as the maximal value for the dirac cp violating phase @xmath19 are predicted @xcite . under the @xmath16-@xmath17 reflection symmetry in the context of a seesaw model , it is supposed that the right - handed neutrinos transform into the charge conjugates of themselves , @xmath20 and the light left - handed neutrinos into @xmath21 @xcite with the corresponding phases @xmath22 . the effective mass matrix derived from the seesaw mechanism is invariant under the @xmath1 reflection symmetry , too . in the light of the cp violation from the neutrino oscillations @xcite , the @xmath1 reflection symmetry indicates that , for given values of mixing angles , the cp asymmetry @xmath23 is maximal . in this paper , we consider a seesaw model with the @xmath1 reflection symmetry . in contrast with other models concerned with the @xmath1 reflection symmetry , we impose the exact symmetry at the gut scale and consider renormalization group ( rg ) evolution @xcite on the neutrino dirac yukawa couplings , heavy majorana neutrino masses and cp phases by running from the gut scale to a seesaw scale . one of interesting issues concerned with neutrino models with the @xmath1 reflection symmetry is if baryogenesis can be successfully realized through leptogenesis or not . in fact , high energy cosmological cp violation responsible for baryogenesis vanishes as long as the @xmath1 reflection symmetry is kept exact and flavor effects associated with the charged leptons are not taken into account @xcite . however , it has been shown that flavor effects can lead to a successful leptogenesis for @xmath24 gev in the limit of the @xmath1 reflection symmetry as long as a partial lepton asymmetry associated with a lepton flavor does not vanish even if total lepton asymmetry is zero @xcite . according to our numerical estimates made in this paper , it is not true in some cases of heavy majorana neutrino mass spectrum with hierarchy structures like @xmath25 and @xmath26 . in such cases , as will be shown later , breaking of the @xmath1 reflection symmetry can lead to successful leptogenesis . instead of introducing _ ad - hoc _ soft symmetry breaking terms , in this paper , we consider a possibility that rg evolutions of neutrino parameters can beak the @xmath1 reflection symmetry and then examine if such breaking effects due to the rg evolutions can lead to successful baryogenesis . to do that , we simply assume that two lighter heavy majorana neutrino masses are exactly degenerate at the gut scale . if a mass splitting between two degenerate heavy majorana neutrinos due to the rg evolution comes out to be small , we expect that the lepton asymmetry can be resonantly enhanced @xcite . however , we show that even though the total lepton asymmetry can be resonantly enhanced , the magnitude of the total lepton asymmetry is not enough to achieve successful baryogenesis in the context of the standard model ( sm ) with the @xmath1 reflection symmetry broken by rg corrections . in general , the absolute magnitudes of the lepton asymmetries for each lepton flavor can be larger than that of the total lepton asymmetry . furthermore , since the interactions mediated by the tau and muon coupling are in thermal equilibrium at below the temperature @xmath27 gev , the processes which wash out lepton number are flavor dependent and thus the lepton asymmetries for each flavor should be treated separately with different wash - out factors . it has been known that flavor effects can enlarge the area of parameter space where leptogenesis can work . thus , it is meaningful to investigate if leptogenesis including flavor effects can be successfully realized in the case that leptogenesis without flavor effects does not work successfully . in this paper , we shall examine if the flavor effects on leptogenesis can help to achieve successful baryogenesis in the context of the sm with the @xmath1 reflection symmetry broken by rg corrections . in particular , we shall carefully discuss how the wash - out factors for each lepton flavor can be significant to achieve successful leptogenesis . in the supersymmetric seesaw model ( ssm ) , the baryon asymmetry of our universe can be achieved leptogenesis via the decays of heavy majorana neutrinos . one of interesting points is that the lepton asymmetry in the ssm can be enhanced in the case of large @xmath10 because it is proportional to the charged lepton yukawa couplings @xmath28 . thus , even for the parameter space where leptogenesis in the sm does not work , supersymmetry can make leptogenesis working . we also expect that since the rg evolutions in the ssm are different from those in the sm , the parameter space where leptogenesis can work must be different each other . in this paper , we shall study how the results concerned with leptogenesis in the sm can change when we supersymmetrize it . this paper is organized as follows : in sec . ii , we present a seesaw model reflecting @xmath1 reflecting symmetry at the gut scale in the context of both sm and ssm . in sec . iii , we discuss how we can obtain a rg - improved neutrino mass matrix . we show that the rg evolutions break the @xmath1 reflection symmetry and derive the deviations of low energy neutrino mixing angles and cp phase from their tree level values which can be parameterized in terms of a parameter concerned with the rg evolution . the discussion for rg evolution from the gut scale to low scale and useful formulae are given in appendix . in sec . iv , we show how successful leptogenesis can be radiatively induced in our scheme . here , we discuss how lepton flavor effects are important to achieve successful leptogenesis and carefully estimate the wash - out factors by taking flavor effects into account . numerical results and conclusion are given in sec . to begin with , let us consider the lagrangian of the lepton sector from which the seesaw mechanism works , @xmath29 where the family indices have been omitted and @xmath30 , @xmath31 stand for su(2 ) lepton doublet fields , charged lepton singlet fields and higgs scalar and singlet heavy majorana neutrino , respectively . in the above lagrangian , @xmath32 , @xmath33 and @xmath34 are the @xmath35 heavy majorana neutrino mass matrix , charged lepton and neutrino dirac yukawa matrices , respectively . after spontaneous electroweak symmetry breaking , the seesaw mechanism leads to the following effective light neutrino mass term , @xmath36 where @xmath37 is a vacuum expectation value of the higgs field with @xmath38 gev . let us impose the @xmath1 reflection symmetry for the neutrino sectors at the gut scale , which is the combined operation of @xmath1 flavor exchange in pmns mixing matrix and cp transformation on the leptonic sector @xcite , in the basis where both the charged lepton mass and heavy majorana mass matrices are diagonal . under the @xmath1 reflection symmetry , the right - handed neutrinos transform into the charge conjugates of themselves , @xmath20 and thus the elements of @xmath32 should be real . then , the neutrino dirac yukawa matrix and the heavy majorana neutrino mass matrix , at the gut scale , are taken to be @xmath39 where @xmath40 of the neutrino dirac yukawa matrix are all real . we assume that the two heavy majorana neutrinos are exactly degenerate in mass , which is much smaller than that of the right - handed neutrino @xmath41 , _ i.e. _ @xmath42 . it turns out , after our analysis , that in the light of seesaw mechanism with such a hierarchy of heavy majorana neutrinos the effects of the parameters @xmath43 and @xmath44 on low energy neutrino phenomenology and even leptogenesis are negligibly small . thus , for our convenience , we take @xmath45 and @xmath46 at the gut scale . introducing several parameters defined by the ratios among the parameters appeared in eq . ( [ input1 ] ) , as follows , @xmath47 the effective light neutrino mass matrix generated through seesaw mechanism can be written as @xmath48 where @xmath49 we see that the effective neutrino mass matrix @xmath50 reflects the @xmath1 reflection symmetry @xcite . here , we have not yet considered rg corrections to @xmath50 from the gut scale to a seesaw scale , so the form of @xmath50 given in eq . ( [ meff1 ] ) is regarded at tree level . it is not difficult to see that the seesaw model based on eq . ( [ input1 ] ) leads to the normal hierarchical light neutrino mass spectrum because we take diagonal form of heavy majorana neutrino mass matrix @xcite . thus , the rg effects from a seesaw scale to electroweak scale on @xmath50 as well as on the neutrino mixing matrix @xmath51 are expected to be very small . starting from the @xmath1 reflection symmetric forms of @xmath2 and @xmath52 , we shall show that the @xmath1 reflection symmetry is broken due to the rg running effects between the gut and a seesaw scale and they can lead to successful leptogenesis without being in conflict with experimental low energy constraints . concerned with cp violation , we notice from eq . ( [ meff1 ] ) that the cp phases @xmath53 coming from @xmath34 take part in low - energy cp violation . to see how the cp phases @xmath53 are correlated with low energy cp violation measurable through neutrino oscillations , let us consider cp violation parameter defined through jarlskog invariant @xcite @xmath54}{\delta m^{2}_{21}\delta m^{2}_{31}\delta m^{2}_{32 } } , \label{cp0 } \end{aligned}\ ] ] where @xmath55 and @xmath56 . from the matrix given in eq . ( [ meff1 ] ) it can be written as @xmath57\simeq m^{6}_{0}a\kappa\omega\sin\delta\phi_{12}+o(\eta^{-1 } ) , \end{aligned}\ ] ] where @xmath58 is the lengthy function composed of the parameters @xmath59 and @xmath60 . we see from the above equation that @xmath61 is correlated with high energy cp parameter @xmath62 . as long as @xmath63 and @xmath64 , @xmath61 have non - vanishing value , which would be a signal of cp violation . the combination of dirac neutrino yukawa matrices , which is relevant for leptogenesis , is given by @xmath65 from this , we find that the hermitian quantity @xmath66 in the limit of the @xmath1 reflection symmetry leads to @xmath67=0 $ ] and thus vanishing lepton asymmetry which is undesirable for a successful leptogenesis . to generate non - vanishing lepton asymmetry , both the mass degeneracy of the 1st and 2nd heavy majorana neutrinos and the @xmath1 reflection symmetric texture of @xmath34 in eq . ( [ input1 ] ) should be broken . in the ssm , the leptonic superpotential is given by @xmath68 where the family indices have been omitted and @xmath69 stands for the chiral super - multiplets of the @xmath70 doublet lepton fields , @xmath71 are the higgs doublet fields with hypercharge @xmath72 , @xmath73 and @xmath74 are the super - multiplet of the @xmath70 singlet neutrino and charged lepton field , respectively . after spontaneous electroweak symmetry breaking , the seesaw mechanism leads to a following effective light neutrino mass term , @xmath75 where @xmath76 . for our purpose , we take the forms of @xmath34 and @xmath32 in the ssm to be the same forms given in eq . ( [ meff1 ] ) . considering the rg effects in @xmath2 and @xmath77 , we obtain a rg improved effective neutrino mass matrix via the seesaw reconstruction at the decoupling scales of the heavy neutrino singlets . the effective mass matrix at low energies can be found by running @xmath78 from the decoupling scales to the electroweak scale @xmath79 . the procedures of the rg running and the useful formulae are given in appendix . however , in the case of the normal hierarchical neutrino mass spectrum of light neutrinos , the rg running effects from the decoupling scales of heavy majorana neutrinos to the electroweak scale are turned out to be negligibly small . ignoring the rg running effects from seesaw scale to the electroweak scale in eqs . ( [ seesaw1],[seesaw11 ] ) , we can obtain the rg improved effective neutrino mass matrix at low energy presented as follows , @xmath80 where the parameter @xmath81 corresponding to the rg correction is given as @xmath82 we recast eq . ( [ cmatrix ] ) with the transformation @xmath83 and @xmath84 , as @xmath85 where @xmath86 and @xmath87 it is quite interesting to notice that the size of @xmath81 is very small , which breaks the @xmath1 reflection symmetry slightly in the mass matrix @xmath50 , however , this tiny breaking effects can play a crucial role in linking between low energy neutrino data and leptogenesis , as will be shown later . this neutrino mass matrix is diagonalized by the pmns mixing matrix @xmath51 , @xmath88 $ ] , where @xmath89 indicates the mass eigenvalues of light majorana neutrinos . as a result of the rg effects , cp violation phase @xmath4 is shifted by a tiny amount , @xmath90 : @xmath91 here if the exact @xmath1 reflection symmetry is recovered , then @xmath92 and @xmath90 goes to zero . atmospheric mixing angle @xmath3 is also deviated from @xmath93 due to the rg corrections , and the shift is approximately written by @xmath94 the magnitude of the unknown angle @xmath9 , which will complete our knowledge of neutrino mixing , can be written as @xmath95 note here that @xmath96 can be directly linked with leptogenesis through @xmath97 defined in eq . ( [ leptocomp ] ) . solar neutrino mixing is also presented by @xmath98 the ratio between @xmath99 and @xmath100 can be approximated by the exact @xmath1 reflection parameter as @xmath101 which is independent of the rg correction . from eqs . ( [ deltacp1]-[sol1 ] ) , the jarlskog invariant @xmath61 can be approximated for @xmath102 by @xmath103 indicating that @xmath61 mainly depends on @xmath8 and @xmath9 , which is in turn proportional to @xmath104 defined in eq . ( [ cmatrix11 ] ) . let us consider the cp asymmetry generated by the decays of the heavy majorana neutrinos @xmath105 ( i=1 , 2 ) . in a basis where the right - handed majorana neutrino mass matrix is real and diagonal , the cp asymmetry generated through the interference between tree and one - loop diagrams for the decay of the heavy majorana neutrino @xmath105 is given , for each lepton flavor @xmath106 , by @xcite @xmath107}\\ & = & \frac{1}{8\pi(y_{\nu}y^{\dag}_{\nu})_{ii}}\sum_{j\neq i}{\rm im}\big\{(y_{\nu}y^{\dag}_{\nu})_{ij}(y_{\nu})_{i\alpha}(y_{\nu})^{\ast}_{j\alpha}\big\}g\big(\frac{m^{2}_{j}}{m^{2}_{i}}\big ) , \label{cpasym1 } \end{aligned}\ ] ] where the function @xmath108 is given by @xmath109~. \label{decayfunction } \end{aligned}\ ] ] here @xmath110 denotes a generation index and @xmath111 is the decay width of the @xmath110th - generation right - handed neutrino . note here that the flavor effects generated due to the term @xmath112 disappears in the summation of @xmath113 for all flavors . in order for @xmath114 to be non - vanishing , not only breaking of the degeneracy of right - handed neutrinos but also non - vanishing @xmath115 $ ] and/or @xmath116 $ ] are required at a seesaw scale @xmath117 . note here that @xmath136 , which is justified from eq . ( [ tan2alpha ] ) , and the sign of plus and minus in eq . ( [ flalepto1 ] ) correspond to the case of the decay of @xmath137 and @xmath138 , respectively . it is worthwhile to notice that since the angle @xmath139 given by eq . ( [ tan2alpha ] ) is limited by @xmath140 , @xmath141 ( order of 10 ) is always greater than @xmath142 ( order of 1 ) , as will be shown later . this implies that @xmath143 is dominant over @xmath144 because of @xmath145 . since the total cp asymmetries @xmath146 are obtained by summing @xmath130 over the lepton flavors @xmath139 , with the help of eqs . ( [ leptocomp],[tan2alpha],[deltam],[rgparameter ] ) , we can express the lepton asymmetries as follows ) can also be obtained directly by using eq . ( [ imreh ] ) . ] @xmath147 where the sign of minus and plus correspond to the case of the decay of @xmath137 and @xmath138 , respectively . as indicated in eq . ( [ flalepto ] ) , the cp asymmetries @xmath114 weakly depend on the heavy majorana neutrino scale @xmath117 . from eqs . ( [ flalepto1],[convlepto2 ] ) , we see that @xmath148 and @xmath149 . therefore , the cp asymmetry @xmath150 gets enhanced by @xmath151 due to flavor effects @xcite . below temperature @xmath152 gev , it is known that muon and tau charged lepton yukawa interactions are much faster than the hubble expansion parameter rendering the @xmath16 and @xmath17 yukawa couplings in equilibrium . then , the processes which wash out lepton number are flavor dependent and thus the lepton asymmetries for each flavor should be treated separately with different wash - out factors . once the initial values of @xmath114 are fixed , the final result of @xmath153 or @xmath154 can be obtained by solving a set of flavor - dependent boltzmann equations including the decay , inverse decay , and scattering processes as well as the non - perturbative sphaleron interaction . in order to estimate the wash - out effects , we introduce the parameters @xmath155 which are the wash - out factors mainly due to the inverse decay of the majorana neutrino @xmath105 into the lepton flavor @xmath156 @xcite . the explicit form of @xmath155 is given by @xmath157 where @xmath158 ev , and @xmath159 and @xmath160 denote the partial decay rate of the process @xmath161 and the hubble parameter at temperature @xmath162 , respectively . the decay rate of @xmath105 to the leptons with flavor @xmath163 is parameterized by @xmath164 and the trace @xmath165 coincides with the parameter @xmath166 . from eq . ( [ k - factor2 ] ) and eq . ( [ seesaw1 ] ) , the wash - out parameters are given as @xmath167 the final baryon asymmetry @xmath168 is then given by @xcite @xmath169~. \label{baryon3 } \end{aligned}\ ] ] notice that each cp asymmetry for a single flavor given in eq . ( [ baryon3 ] ) is weighted differently by the corresponding wash - out parameter given by eq . ( [ k - factor2 ] ) , and appears with different weight in the final formula for the baryon asymmetry @xcite . from our numerical estimate , we found that the wash - out factors @xmath170 except for @xmath171 are much greater than one indicating that corresponding lepton flavor asymmetries are strongly washed out , which is undesirable for successful leptogenesis . since @xmath172 can be much less than one and the lepton asymmetries given in eq . ( [ flalepto1 ] ) are the same order of magnitude , the lepton asymmetry for electron flavor in the weak wash - out regime leads to the dominant contribution to the lepton asymmetry , which can be enough to give rise to successful baryogenesis . thus , @xmath168 is approximately given by @xmath173 and the magnitudes of @xmath174 and @xmath171 are approximately @xmath175 in the weak wash - out regime ( @xmath176 ) , the lepton asymmetry @xmath177 generated through the decay of @xmath138 is given by @xcite @xmath178 therefore , the resulting baryon - to - photon ratio @xmath179 can be simply given as @xmath180 as will be shown later , the wash - out factor @xmath181 is proportional to the low energy mixing angle @xmath9 and there is a connection between @xmath179 and low energy neutrino observables such as @xmath9 and @xmath61 . on the other hand , the ratio of @xmath6 without lepton flavor effects compared to @xmath182 with lepton flavor effects @xcite is presented by @xmath183 thus , without taking lepton flavor effects into account , the prediction of @xmath6 is suppressed by @xmath184 , _ i.e. , _ @xmath185 orders of magnitude compared with @xmath182 , which is too small to give a successful leptogenesis . it is worthwhile to notice that the radiatively induced lepton asymmetries @xmath192 are proportional to @xmath193 , and thus for large @xmath10 it can be highly enhanced and proportional to @xmath194 . furthermore , it has an explicit dependence of the evolution scale @xmath129 . these two points are already mentioned in ref . taking lepton flavor effects into account , in this case the final baryon asymmetry is given as @xcite @xmath195~. \label{baryon3 m } \end{aligned}\ ] ] similar to the sm , below temperatures @xmath196 , muon and tau charged lepton yukawa interactions are much faster than the hubble expansion parameter rendering the @xmath16 and @xmath17 yukawa couplings in equilibrium . to obtain the final baryon asymmetry survived , we should consider the wash - out factors . similar to the sm , the magnitudes of @xmath197 except for @xmath198 are greater than one in our scenario . so , we will consider two cases depending on the magnitude of @xmath198 : ( i)@xmath199 and @xmath200 , ( ii ) @xmath201 . this case implies that the muon and tau lepton asymmetries are strongly washed out , whereas the electron asymmetry generated through the decay of @xmath120 is weakly washed out . the form of @xmath177 in the weak wash - out regime in the ssm is the same as that in the sm given by eq . ( [ weak ] ) . in this case , similar to the sm , the contribution of @xmath177 to @xmath168 is dominant over the others . thus , @xmath168 is approximately give by @xmath203 and the magnitudes of @xmath204 and @xmath171 are approximately @xmath205 the resulting baryon - to - photon ratio @xmath179 is simply given as @xmath206 just like the sm , one can see that there is a connection between baryon asymmetry and low energy neutrino observables . the ratio of @xmath6 without lepton flavor effects to @xmath182 with lepton flavor effects are simply given by @xcite @xmath207 from this result , we see that without taking lepton flavor effects into account the prediction of @xmath6 is suppressed by @xmath208 , compared with @xmath182 . however , on the contrary to the flavor independent leptogenesis in the sm , it is possible to get a right amount of baryon asymmetry without flavor effects for the case of large @xmath209 . in the strong wash - out regime @xmath211 , the lepton asymmetry for the flavor @xmath163 generated through the decay of @xmath105 is given by @xcite @xmath212 in this case , given the initial thermal abundance of @xmath105 and the condition for @xmath155 , the baryon asymmetry including lepton flavor effects is approximately given by @xcite @xmath213 here , we note that each lepton asymmetry @xmath113 are almost the same order of magnitudes , even though @xmath214 for large @xmath10 and @xmath150 for small @xmath10 are slightly dominant within a few factors . according to our numerical estimate , the typical sizes of the suppression factors @xmath215 are @xmath216 which indicates that the flavor dependent wash - out factors equally contribute to leptogenesis . using eq . ( [ kstrong ] ) , the resulting baryon - to - photon ratio @xmath179 can be simply approximated as an order of magnitude @xmath217 which represents that at least @xmath218 is needed for the flavored leptogenesis to successfully work ( see , fig . [ fig6 ] in sec . v ) . as a consequence , in the strong wash - out case , the resulting baryon asymmetry is too small to give a successful leptogenesis in the sm , whereas at least @xmath218 would be necessary to accommodate required baryon asymmetry in the ssm , and there exists also a connection between leptogenesis and low energy observable @xmath8 . note that the strong wash - out case only allow @xmath9 to be @xmath219 which is from the condition @xmath220 . in order to estimate the rg evolutions of neutrino dirac - yukawa matrix and heavy majorana neutrino masses from the gut scale to a seesaw scale , we numerically solve all the relevant rg equations presented in @xcite . in our numerical calculation of the rg evolutions , we first fix the values of two masses of heavy majorana neutrinos with hierarchy @xmath221 to be @xmath222 , and @xmath223 as inputs can be as light as @xmath224 gev in our scenario . ] . then we solve the rg equations by varying input values of all the parameter space @xmath225 given at the gut scale . then , finally we select the parameter space allowed by low energy neutrino experimental data . at present , we have five experimental data , which are taken as low energy constraints in our numerical analysis , given at @xmath226 by @xcite , @xmath227\leq2.8~,\quad\quad ~~7.1\leq\delta m^{2}_{\rm sol}[10^{-5}{\rm ev}^{2}]\leq8.3 , \label{exp bound } \end{aligned}\ ] ] where @xmath228 and @xmath229 . in addition , the current measurement of baryon asymmetry of our universe @xmath230 we take in our numerical calculation is given by @xcite @xmath231 using the results of the rg evolutions , we estimate baryon asymmetry for the parameter space constrained from the low energy experimental data . for this case of wash - out factors , we found that baryogenesis could be successfully implemented in the sm and the ssm through electron flavor dominant leptogenesis . in addition , we show that the successful leptogenesis could be linked with low energy observables ( @xmath9 and @xmath61 ) through the rg parameter @xmath133 given in eq . ( [ rgparameter ] ) as well as @xmath171 . in fig . [ fig1 ] , we show the scatter plots for the model parameters in the sm ( upper panel ) and the ssm ( @xmath232 ) ( lower panel ) constrained by the experimental data given in eq . ( [ exp bound ] ) : the figures in the left - hand side exhibit how the parameters @xmath233 , @xmath234 , @xmath235 and @xmath236 are correlated with the angle @xmath139 defined in eq . ( [ tan2alpha ] ) , respectively and the figures in the right - hand side represent how the difference between the phases @xmath237 and @xmath238 in eq . ( [ input1 ] ) , @xmath239 , depends on the angle @xmath139 . in fig . [ fig2 ] , we present how the wash - out factor @xmath198 ( fig . 2-a ) and cp violating observable @xmath61 ( fig . 2-b ) are correlated with the mixing angle @xmath9 . the circles , triangles , stars and spots in fig . 2 correspond to the cases of the sm , and the ssm with @xmath240 , @xmath232 and @xmath241 , respectively . from fig . 2-(a ) , we see that in the case @xmath242 ( weak wash - out regime ) the mixing angle @xmath9 is limited . even in the case of the ssm with @xmath243 , the upper limit of @xmath9 is about 9 degree . it is remarkable from fig . 2-(a ) that @xmath179 is expected to be enhanced through @xmath171 as @xmath9 gets lower . thus , this case with @xmath242 favors low values of @xmath9 . the vertical dotted lines in fig . 2-(b ) indicate the values of @xmath9 corresponding to best - fit value of baryon asymmetry of our universe , @xmath244 , when @xmath245 and 60 , respectively , whereas the vertical solid line corresponds to the case of the sm . as can be seen in fig . [ fig2]-(b ) , the prediction of @xmath61 for the ssm with a low @xmath10 can be measurable in the near future from long baseline neutrino oscillations . in fig . [ fig3 ] , we present the predictions of ( a ) @xmath179 and ( b ) @xmath153 for the sm ( triangle ) , the ssm with @xmath240 ( daggers ) , 10 ( asters ) and 60 ( circles ) as a function of the mixing angle @xmath9 . in the case of flavor dependent leptogenesis , the allowed values of @xmath182 prefers very low values of @xmath9 . the horizontal dotted lines and solid line in fig . [ fig3 ] ( also in figs . [ fig4 ] and fig . [ fig6 ] ) correspond to the current bounds on @xmath6 . we see from fig . [ fig3 ] that the prediction of @xmath179 in the ssm for a fixed @xmath9 is increased as @xmath10 increases . as explained with eq . ( [ ratiobaryonssm ] ) , we see from fig . [ fig3 ] that @xmath6 is suppressed by factor @xmath185 compared with @xmath182 because of the different wash - out factors presented in eq . ( [ k - em ] ) . we also see from fig . [ fig3 ] that a successful baryogenesis can be achieved via flavor dependent leptogenesis in both the sm and the ssm , whereas a successful baryogenesis via flavor independent leptogenesis is possible only for the ssm with large @xmath246 . for this case , we can obtain lower limit of @xmath10 from the current observation for @xmath153 based on eqs . ( [ cmatrix],[h],[deltam],[rgparameter ] ) as follows ; @xmath247\big)^{1/4}~. \end{aligned}\ ] ] imposing the best - fit value of observed baryon asymmetry , @xmath9 is determined to be @xmath248 for the sm and @xmath249 and @xmath250 for the ssm with @xmath245 and @xmath251 , respectively . for the case @xmath252 , the dominant contribution to @xmath179 comes from the @xmath7-leptogenesis @xmath253 , as shown in eq . ( [ baryonasymssm ] ) . in the case of the ssm , the value of @xmath9 corresponding to @xmath254 becomes lower as @xmath10 gets higher because @xmath171 is proportional to @xmath9 . from eqs . ( [ flalepto1],[baryonasymssm],[etabexp ] ) , we can roughly get the value of @xmath10 corresponding to @xmath254 as follows : @xmath255 - 1\big)^{1/2}~. \label{tanbwk } \end{aligned}\ ] ] in fig . [ fig4]-(a ) , we present how @xmath10 is correlated with the mixing angle @xmath9 , which reflects the above equation ( [ tanbwk ] ) . [ fig4]-(b ) shows how the baryon asymmetry @xmath182 is sensitive to the cp violating observable @xmath61 given by eq . ( [ cp1 ] ) . the triangles , daggers , asters and circles correspond to the cases of the sm and the ssm with @xmath256 , respectively . the value of @xmath182 corresponding to @xmath257 for the case of the sm leads to @xmath258 ( and @xmath259 ) , which can be measured in the future long - baseline neutrino oscillation experiments . from fig . [ fig4]-(b ) and fig . [ fig2]-(b ) , we see that the value of @xmath61 corresponding to @xmath257 for the case of the ssm becomes higher as @xmath10 gets smaller . thus , if future experiments can measure @xmath61 with a sensitivity better than 0.01 , the predictions of @xmath182 in the ssm with low @xmath260 could be probed . in the case of strong wash - out for the decay of @xmath261 , @xmath262 , it turns out that baryogenesis in the sm can not be successfully realized . however , as can be seen in eq . ( [ baryonasymst ] ) , flavor dependent leptogenesis in the ssm could successfully be implemented at least for @xmath218 as explained below . in fig . [ fig5 ] , we show the scatter plots for the model parameters in the ssm with @xmath232 constrained by the experimental data given in eq . ( [ exp bound ] ) : fig . 5-(a ) exhibits how the parameters @xmath233 , @xmath234 , @xmath235 and @xmath263 are correlated with the angle @xmath139 defined in eq . ( [ tan2alpha ] ) , respectively . 5-(b ) represents how @xmath264 depends on the angle @xmath139 . [ fig6 ] presents the predictions of baryon asymmetry for the allowed parameter regions presented in fig . [ fig5 ] as a function of ( a ) the solar mixing angle @xmath8 and ( b ) the chooz mixing angle @xmath9 . the results indicate that in the ssm flavor dependent leptogenesis can successfully work out for @xmath265 , whereas flavor independent leptogenesis does so only for the ssm with @xmath266 . from eqs . ( [ flalepto1],[etab],[etabexp ] ) , we can get the relation , @xmath267 where @xmath268 . the relation given by eq . ( [ tanbst ] ) indicates that there exits a correlation between @xmath10 and @xmath8 for a fixed value of @xmath6 . in particular , taking @xmath269 , we plot the correlation between @xmath10 and @xmath8 in fig . [ fig7]-(b ) . from our numerical estimate , we found that in order for the prediction of @xmath182 in the case of strong wash - out to be consistent with the right amount of baryon asymmetry given by eq . ( [ etabexp ] ) the value of @xmath10 should be @xmath270 thus , future measurements for @xmath10 and @xmath8 could serve as an indirect test for the scenario of baryogenesis considered in this paper . fig . [ fig7]-(b ) represents the correlation between @xmath61 and @xmath8 for a fixed value of @xmath271 . the results show that the best fit value of @xmath6 and the current measurement of @xmath8 favors @xmath272 which can be measurable in the upcoming long baseline neutrino oscillation experiments . as a summary , we have considered an exact @xmath0 reflection symmetry in neutrino sector realized at the gut scale in the context of the seesaw model . the exact @xmath0 reflection symmetry has been imposed in the basis where both the charged lepton mass and heavy majorana neutrino mass matrices are real and diagonal . we have assumed that the two lighter heavy majorana neutrinos are degenerate at the gut scale . it has been shown that the rg evolution from the gut scale to the seesaw scale gives rise to breaking of the @xmath1 symmetry and a tiny splitting between two degenerate heavy majorana neutrino masses as well as small variations of the cp phases in @xmath2 , which are essential to achieve a successful leptogenesis . such small rg effects lead to tiny deviations of @xmath3 from the maximal value and the cp phase @xmath4 from @xmath5 imposed at the gut scale due to @xmath1 reflection symmetry . in our scenario , the required amount of the baryon asymmetry @xmath6 could be generated via so - called resonant @xmath7-leptogenesis , in which the wash - out factor concerned with electron flavor plays a crucial role in reproducing a successful leptogenesis . and we have found that the magnitude of the electron leptogenesis is enhanced by @xmath185 orders due to the flavor effects in the sm , and enhanced further by a factor of @xmath273 for the ssm . a point deserved to notice is that cp violation responsible for the generation of baryon asymmetry of our universe comes from the breaking of both the two degenerate heavy majorana neutrinos and the cp phases @xmath53 in @xmath2 by rg evolutions , which corresponds to the tiny breakdown of @xmath1 reflection symmetry , and such small rg effects cause leptogenesis to directly be linked with the cp violation measurable through neutrino oscillation as well as neutrino mixing angles @xmath8 and @xmath9 . we expect that in addition to the reactor and long baseline neutrino experiments for precise measurements of neutrino mixing angles and cp violation , the measurements for the supersymmetric parameter @xmath10 at future collider experiments would serve as an indirect test of our scenario of baryogenesis based on the @xmath1 reflection symmetry . the radiative behavior of the heavy majorana neutrinos mass matrix @xmath280 is dictated by the following rg equations @xcite : @xmath281~,~~~\text{sm}\nonumber\\ \frac{d\textbf{m}_{r}}{dt } & = & 2[(\textbf{y}_\nu \textbf{y}_\nu^\dagger)\textbf{m}_r+\textbf{m}_r(\textbf{y}_\nu \textbf{y}_\nu^\dagger)^t]~,~~\text{ssm } \label{rg 1 } \end{aligned}\ ] ] where @xmath282 with an arbitrary renormalization scale @xmath283 . the rg equations for the dirac - yukawa neutrino matrix can be written as @xmath284~,~~~\text{sm}\nonumber\\ \frac{d\textbf{y}_{\nu}}{dt } & = & \textbf{y}_{\nu}[(t-3g^{2}_2-\frac{3}{5}g^{2}_{1})+\textbf{y}^{\dag}_{l}\textbf{y}_{l}+3\textbf{y}^{\dag}_{\nu}\textbf{y}_{\nu}~,~~~~~~~\text{ssm } \label{rg 2 } \end{aligned}\ ] ] where @xmath285 , @xmath286 and @xmath287 are the yukawa matrices for up - type quarks and charged leptons and @xmath288 are the @xmath70 and @xmath289 gauge coupling constants . let us first reformulate the rg equations ( [ rg 1 ] ) in the basis where @xmath32 is diagonal . since @xmath32 is symmetric , it can be diagonalized by a unitary matrix @xmath290 , @xmath291 as the structure of the mass matrix @xmath32 changes with the evolution of the scale , the unitary matrix @xmath290 depends on the scale , too . the rg evolution of the matrix @xmath292 can be written as @xmath293 where @xmath58 is an anti - hermitian matrix @xmath294 due to the unitary of @xmath290 . then , differentiating eq . ( [ rg 3 ] ) , we obtain @xmath295 v\}_{ij}~,~~\text{sm}\nonumber\\ \frac{dm_{i } \delta_{ij}}{dt } & = & a^t_{ij}m_j + m_i a_{ij } + 2\ { v^t[(\textbf{y}_\nu \textbf{y}_{\nu}^\dagger)\textbf{m}_r+\textbf{m}_r(\textbf{y}_\nu \textbf{y}_\nu ^\dagger)^t ] v\}_{ij}~,~\text{ssm}. \label{rg 4}\end{aligned}\ ] ] absorbing the unitary transformation into the dirac - yukawa coupling @xmath296 the real diagonal part of eq . ( [ rg 4 ] ) becomes @xmath297 the off diagonal part of eq . ( [ rg 4 ] ) leads to @xmath298 + i\frac{m_j - m_i}{m_j+m_i}{\rm im}[(y_\nu y_\nu^\dagger)_{ij}]~,~~~~~\text{sm}\nonumber\\ a_{ij}&= & 2\frac{m_j+m_i}{m_j - m_i}{\rm re}[(y_\nu y_\nu^\dagger)_{ij } ] + i2\frac{m_j - m_i}{m_j+m_i}{\rm im}[(y_\nu y_\nu^\dagger)_{ij}]~,~~\text{ssm}. \label{rg 6 } \end{aligned}\ ] ] the rg equations for @xmath2 in the basis where @xmath280 are diagonal is written as @xmath299+a^{t}y_{\nu}~,~~~\text{sm}\nonumber\\ \frac{dy_{\nu}}{dt } & = & y_{\nu}[(t-3g^{2}_2-\frac{3}{5}g^{2}_{1})+\textbf{y}^{\dag}_{l}\textbf{y}_{l}+3y^{\dag}_{\nu}y_{\nu})]+a^{t}y_{\nu}~,~~~~~~\text{ssm}. \label{rg 7 } \end{aligned}\ ] ] the rg equations for the quantity @xmath300 relevant for leptogenesis can be written as @xmath301 we see from eq . ( [ rg 6 ] ) that the real part of @xmath302 is singular when @xmath303 . the singularity in @xmath304 $ ] can be eliminated with the help of an appropriate rotation between degenerate heavy majorana neutrino states . such a rotation does not change any physics and it is equivalent to absorb the rotation matrix @xmath305 into the dirac - yukawa neutrino matrix @xmath2 , @xmath306 where the matrix @xmath305 , in our case , particularly rotating the 1st and 2rd generations of heavy majorana neutrinos can be parameterized as @xmath307 then , the singularity in the real part of @xmath302 is indeed removed when the rotation angle @xmath139 is taken to be satisfied with the condition , @xmath308=0~,~~\text{for any pair}~i , j ~\text{corresponding to}~m_{i}=m_{j}~ , \label{singularity } \end{aligned}\ ] ] for @xmath309 , which leads to @xmath310 with @xmath311 , we construct a parameter @xmath312 as follows ; @xmath313 where @xmath314 and the components of @xmath312 are given by @xmath315 it is then obvious from eq . ( [ h1 ] ) that @xmath316=0 $ ] and thus the singularity in @xmath317 does not appear . now , let us consider rg effects which may play an important role in successful leptogenesis . first , we parameterize the mass splitting of the degenerate heavy majorana neutrinos in terms of a parameter @xmath318 defined by @xmath319 which is governed by the following rg equations derived from eq . ( [ rg 5 ] ) , @xmath320 \simeq \frac{4h_{12}}{\sin2\alpha}~,~~~~\text{sm}\nonumber\\ \frac{d\delta_n}{dt } & = & 4(1-\delta_{n})[\widetilde{h}_{11}-\widetilde{h}_{22 } ] \simeq \frac{8h_{12}}{\sin2\alpha}~,~~~~\text{ssm}. \label{degeneracy1 } \end{aligned}\ ] ] the solutions of the rg equations ( [ degeneracy1 ] ) are approximately given by @xmath321 where we used the parameters defined in eq . ( [ input2 ] ) . next , we consider rg running of the dirac neutrino yukawa matrix from the gut scale to the seesaw scale , @xmath322 . since the rg evolution produces non - zero off - diagonal entries in @xmath77 , it has to be re - diagonalized by a unitary transformation , @xmath323 , which leads to the rotation , @xmath324 and @xmath325 . neglecting the corrections proportional to charged-@xmath16 and -@xmath7 yukawa coupling , @xmath326 and @xmath327 , we can obtain the rg improved dirac neutrino yukawa matrix given for the sm by @xmath328 and for the ssm by @xmath329 where @xmath330 , the components of @xmath331 are @xmath332 and the parameter @xmath133 presenting rg corrections is given by @xmath333 note here that the third low of @xmath331 is not changed because @xmath334 does not depend on @xmath335 . as will be shown later , the parameter @xmath133 is proportional to @xmath336 in the sm and @xmath337 in the ssm and it plays important roles in both leptogenesis and low - energy cp violation . at the seesaw scale , the combination of the yukawa - dirac coupling matrices and the mass matrix of right - handed heavy majorana neutrino , respectively , can be written as @xmath338 here the ( 1,2 ) and ( 2,1)-components of @xmath339 are the quantities radiatively generated by rg running . considering the structure of @xmath312 in eq . ( [ h1 ] ) , up to non - zero leading contributions in the right side of eq . ( [ rg 8 ] ) , the rg equations of @xmath340 are given for the sm by @xmath341}{dt}&\simeq & -3y^{2}_{\tau}{\rm re}[(\widetilde{y}_{\nu1\tau}\widetilde{y}^\ast_{\nu2\tau})]+{\rm re}[a_{21}](\widetilde{h}_{22}-\widetilde{h}_{11})~,\nonumber\\ \frac{d { \rm im}[\widetilde{h}_{12}]}{dt}&\simeq & -3y^{2}_{\tau}{\rm im}[(\widetilde{y}_{\nu1\tau}\widetilde{y}^\ast_{\nu2\tau})]~ , \end{aligned}\ ] ] and for the ssm by @xmath341}{dt}&\simeq & 2y^{2}_{\tau}{\rm re}[(\widetilde{y}_{\nu1\tau}\widetilde{y}^\ast_{\nu2\tau})]+{\rm re}[a_{21}](\widetilde{h}_{22}-\widetilde{h}_{11})~,\nonumber\\ \frac{d { \rm im}[\widetilde{h}_{12}]}{dt}&\simeq & 2y^{2}_{\tau}{\rm im}[(\widetilde{y}_{\nu1\tau}\widetilde{y}^\ast_{\nu2\tau})]~. \end{aligned}\ ] ] using eqs . ( [ rg 6],[degeneracy1 ] ) , radiatively generated @xmath340 is given approximately in terms of the parameters in eqs . ( [ h1 ] ) and ( [ h1 ] ) , for the sm , by @xmath342 & = & { \rm re}[\widetilde{h}_{21}]\simeq -\frac{3}{2}y_\tau^2b_3 ^ 2\big [ \frac{1}{2}(\kappa^2 -\omega^2)\sin2\alpha+\kappa\omega\cos 2\alpha\cos\delta\phi_{12}\big ] \cdot t~,\nonumber\\ { \rm im}[\widetilde{h}_{12 } ] & = & -{\rm im}[\widetilde{h}_{21}]\simeq 3y_\tau^2b_3 ^ 2\omega\kappa\sin\delta\phi_{12}\cdot t~ , \label{imreh ' } \end{aligned}\ ] ] and for the ssm by @xmath342 & = & { \rm re}[\widetilde{h}_{21}]\simeq y_\tau^2b_3 ^ 2\big [ \frac{1}{2}(\kappa^2 -\omega^2)\sin2\alpha+\kappa\omega\cos 2\alpha\cos\delta\phi_{12}\big ] \cdot t~,\nonumber\\ { \rm im}[\widetilde{h}_{12 } ] & = & -{\rm im}[\widetilde{h}_{21}]\simeq -2y_\tau^2b_3 ^ 2\omega\kappa\sin\delta\phi_{12}\cdot t~. \label{imreh'm } \end{aligned}\ ] ] note here that radiatively generated quantity @xmath343 $ ] is proportional to @xmath344 which is fixed by the angle @xmath139 in eq . ( [ tan2alpha ] ) due to the @xmath1 reflection symmetry , and related with low energy observables @xmath9 and @xmath61 in eqs . ( [ theta13],[cp1 ] ) . 99 # 1#2#3phys . b * # 1 * , ( # 3 ) # 2 # 1#2#3nucl . * b#1 * , ( # 3 ) # 2 # 1#2#3phys . * d#1 * , ( # 3 ) # 2 # 1#2#3phys . * # 1 * , ( # 3 ) # 2 # 1#2#3mod . * a#1 * , ( # 3 ) # 2 # 1#2#3phys . rep . * # 1 * , ( # 3 ) # 2 # 1#2#3science * # 1 * , ( # 3 ) # 2 # 1#2#3astrophys . j. * # 1 * , ( # 3 ) # 2 # 1#2#3eur . j. * c#1 * , ( # 3 ) # 2 # 1#2#3jhep * # 1 * , ( # 3 ) # 2 # 1#2#3j . * g#1 * , ( # 3 ) # 2 # 1#2#3int . j. mod . phys . * # 1 * , ( # 3 ) # 2 # 1#2#3prog . * # 1 * , ( # 3 ) # 2 y. fukuda _ et al . _ [ super - kamiokande collaboration ] , . m. fukugita and t. yanagida , phys . b * 174 * , ( 1986 ) 45 ; g. f. giudice _ et al . _ , b * 685 * ( 2004 ) 89 [ arxiv : hep - ph/0310123 ] ; w. buchmuller , p. di bari and m. plumacher , annals phys . * 315 * , ( 2005 ) 305 [ arxiv : hep - ph/0401240 ] ; a. pilaftsis and t. e. j. underwood , phys . d * 72 * , ( 2005 ) 113001 [ arxiv : hep - ph/0506107 ] . p. minkowski , phys . b * 67 * , ( 1977 ) 421 ; m. gell - mann , p. ramond and r. slansky , _ proceedings of the supergravity stony brook workshop _ , new york 1979 , eds . p. van nieuwenhuizen and d. freedman ; t. yanagida , _ proceedinds of the workshop on unified theories and baryon number in the universe _ , tsukuba , japan 1979 , ed.s a. sawada and a. sugamoto ; r. n. mohapatra , g. senjanovic , . i. aizawa , t. kitabayashi and m. yasue , phys . d * 72 * , ( 2005 ) 055014 [ arxiv : hep - ph/0504172 ] ; i. aizawa , t. kitabayashi and m. yasue , nucl . b * 728 * , ( 2005 ) 220 [ arxiv : hep - ph/0507332 ] ; z. z. xing , h. zhang and s. zhou , lett . b * 641 * , ( 2006 ) 189 [ arxiv : hep - ph/0607091 ] . a. de rujula , m. b. gavela and p. hernandez , nucl . b * 547 * , ( 1999 ) 21 [ arxiv : hep - ph/9811390 ] ; a. cervera , a. donini , m. b. gavela , j. j. gomez cadenas , p. hernandez , o. mena and s. rigolin , nucl . phys . b * 579 * , ( 2000 ) 17 [ erratum - ibid . b * 593 * , ( 2001 ) 731 ] [ arxiv : hep - ph/0002108 ] ; c. albright _ et al . _ , arxiv : hep - ex/0008064 ; k. dick , m. freund , m. lindner and a. romanino , nucl . phys . b * 562 * , ( 1999 ) 29 [ arxiv : hep - ph/9903308 ] ; v. d. barger , s. geer , r. raja and k. whisnant , phys . d * 63 * , ( 2001 ) 033002 [ arxiv : hep - ph/0007181 ] ; m. freund , p. huber and m. lindner , nucl . b * 585 * , ( 2000 ) 105 [ arxiv : hep - ph/0004085 ] ; k. hagiwara , n. okamura and k. i. senda , arxiv : hep - ph/0607255 . p. h. chankowski and z. pluciennik , phys . b * 316 * , ( 1993 ) 312 [ arxiv : hep - ph/9306333 ] ; k. s. babu , c. n. leung and j. t. pantaleone , phys . b * 319 * , ( 1993 ) 191 [ arxiv : hep - ph/9309223 ] ; m. tanimoto , phys . b * 360 * , ( 1995 ) 41 [ arxiv : hep - ph/9508247 ] ; n. haba and n. okamura , eur . j. c * 14 * , ( 2000 ) 347 [ arxiv : hep - ph/9906481 ] ; p. h. chankowski and s. pokorski , int . j. mod . a * 17 * , ( 2002 ) 575 [ arxiv : hep - ph/0110249 ] . j. a. casas , j. r. espinosa , a. ibarra and i. navarro , nucl . b * 573 * , ( 2000 ) 652 [ arxiv : hep - ph/9910420 ] ; j. a. casas , j. r. espinosa , a. ibarra and i. navarro , nucl . b * 569 * , ( 2000 ) 82 [ arxiv : hep - ph/9905381 ] . s. antusch , j. kersten , m. lindner and m. ratz , phys . b * 538 * , ( 2002 ) 87 [ arxiv : hep - ph/0203233 ] . a. pilaftsis , phys . d * 56 * , ( 1997 ) 5431 [ arxiv : hep - ph/9707235 ] ; a. pilaftsis and t. e. j. underwood , nucl . phys . b * 692 * , ( 2004 ) 303 [ arxiv : hep - ph/0309342 ] . l. covi , e. roulet and f. vissani , phys . lett . * b384 * , ( 1996 ) 169 ; a. pilaftsis , int . j. mod . a * 14 * , ( 1999 ) 1811 [ arxiv : hep - ph/9812256 ] . t. endoh , t. morozumi , z. xiong , prog . phys . * 111 * , ( 2004 ) 123 [ arxive : hep - ph/0308276 ] ; t. fujihara , s. kaneko , s. k. kang , d. kimura , t. morozumi , m. tanimoto , phys . rev.d * 72 * , ( 2005 ) 016006 [ arxive : hep - ph/0505076 ] ; a. abada , s. davidson , a. ibarra , f. x. josse - michaux , m. losada and a. riotto , jhep * 0609 * , ( 2006 ) 010 [ arxiv : hep - ph/0605281 ] ; s. blanchet and p. di bari , jcap * 0703 * , ( 2007 ) 018 [ arxiv : hep - ph/0607330 ] ; s. antusch , s. f. king and a. riotto , jcap * 0611 * , ( 2006 ) 011 [ arxiv : hep - ph/0609038 ] ; s. pascoli , s. t. petcov and a. riotto , phys . d * 75 * , ( 2007 ) 083511 [ arxiv : hep - ph/0609125 ] ; s. pascoli , s. t. petcov and a. riotto , nucl . b * 774 * , ( 2007 ) 1 [ arxiv : hep - ph/0611338 ] ; g. c. branco , r. gonzalez felipe and f. r. joaquim , phys . b * 645 * , ( 2007 ) 432 [ arxiv : hep - ph/0609297 ] ; g. c. branco , a. j. buras , s. jager , s. uhlig and a. weiler , jhep * 0709 * , ( 2007 ) 004 [ arxiv : hep - ph/0609067 ] . y. h. ahn , s. k. kang , c. s. kim and j. lee , phys . d * 77 * , ( 2008 ) 073009 [ arxiv : hep - ph/0711.1001 ] . a. abada , s. davidson , f. x. josse - michaux , m. losada and a. riotto , jcap * 0604 * , ( 2006 ) 004 [ arxiv : hep - ph/0601083 ] ; s. antusch , s. f. king and a. riotto , jcap * 0611 * , ( 2006 ) 011 [ arxiv : hep - ph/0609038 ] . a. pilaftsis , phys . lett . * 95 * , ( 2005 ) 081602 [ arxiv : hep - ph/0408103 ] ; a. pilaftsis and t. e. j. underwood , phys . rev . d * 72 * , ( 2005 ) 113001 [ arxiv : hep - ph/0506107 ] . wmap collaboration , d.n . _ , astrophys . j. suppl . * 148 * , ( 2003 ) 175 ; m. tegmark _ et al . _ , d * 69 * , ( 2004 ) 103501 ; c. l. bennett _ et al . _ , astrophys . j. suppl . * 148 * , ( 2003 ) 1 [ arxiv : astro - ph/0302207 ] .

a spin particle propagating in a de sitter background has a one parameter family of states which transform covariantly under the isometry group of the background . these states are the fermionic analogues of the-vacua for a scalar field . we shall show how using a point - source propagator for a fermion in an-state produces divergent perturbative corrections . these corrections can not be used to cancel similar divergences arising from scalar fields in bosonic-vacua since they have an incompatible dependence on the external momenta . the theory can be regularized by modifying the propagator to include an antipodal source .

hep-th0410229

while de sitter space shares the same number of isometries as minkowski space , field theories exhibit some surprising properties in this simplest of curved backgrounds . an immediate example is the enormous stretching of scales in de sitter space which naturally connects short distances in the past to large distances today . this rapid expansion is a familiar and very appealing feature of inflation @xcite . during the slow - roll regime of inflation , for which de sitter space is an idealization , quantum fluctuations grow exponentially large and eventually seed the large scale structure of the universe . depending upon the hubble scale and the duration of inflation , this large scale structure could have been determined by fluctuations which occurred at or well below the planck scale . most models of inflation produce far more than the necessary sixty @xmath2-folds necessary to solve the horizon problem and this connection between large scales and potentially planckian physics has been called the `` transplanckian problem '' of inflation @xcite , although it has recently been viewed as more of an opportunity , since it could allow the observation of physics well beyond experimentally accessible scales , most typically within an order of magnitude or two above the hubble scale during inflation . non - thermal features of the state for the field driving inflation tend to provide a more robust signal of these transplanckian effects @xcite . a further difference from minkowski space is the existence of a much richer family of invariant or covariant states in de sitter space . for a free scalar field in a de sitter background , the states invariant under the @xmath3 isometry group can be distinguished by a complex parameter @xmath1 , although only real values of @xmath1 correspond to @xmath4 invariant theories @xcite . these @xmath1-vacua are not the lowest energy eigenstates of a globally conserved hamiltonian , as is the case of the standard poincar ' e - invariant vacuum of minkowski space , since de sitter space does not admit a globally defined time - like killing vector . nevertheless , a unique element in this infinite family , the bunch - davies vacuum @xcite , can be selected by demanding that at short distances or as the curvature of the de sitter space is taken to zero the state should match with the vacuum of minkowski space . both of these features emphasize the need for understanding quantum field theory particularly the ideas of decoupling and renormalization in an expanding background starting from a non - standard state . for this purpose the @xmath1-vacua provide an ideal test case to study how these ideas are to be modified in such a setting since the high amount of symmetry of these non - thermal states allows them to be readily analyzed analytically . it was recently realized that for a scalar field in an @xmath1-state , a point source propagator does not produce a well behaved perturbation theory @xcite . one method for expressing this pathological behavior is to impose a cutoff @xmath5 on physical three - momentum of the theory . loop processes then diverge as @xmath6 in such a way as can not be cancelled by simple counterterms . for example , the one - loop correction to the self - energy in a @xmath7 theory diverges linearly with @xmath5 and the dependence of this divergent term on the external momentum does not match that of a @xmath8 counterterm @xcite . the resolution of these divergences came with the realization that the propagator should be modified for these states to be the green s function for _ two _ point sources @xcite . in this article , we examine the structure and the properties of a spin @xmath0 fermion in a de sitter background , which possesses its own one - parameter set of covariant states @xcite . one reason for doing so is to learn whether the double source construction for the scalar field can be circumvented by using fermionic loops to cancel the non - renormalizable divergences from bosonic loops . while a fermion loop correction to a scalar self - energy also diverges linearly with the cutoff @xmath5 , here we show that this divergence can not be cancelled by that of the bosonic loop , even allowing an arbitrary fine - tuning of the relative values of @xmath1 for the scalar and the fermion fields . as with a bosonic @xmath1-vacuum , the peculiar divergences in fermion loops arise from an inconsistency between the single - source propagator and the fermionic @xmath1-state . in some sense , the physical setting resembles a field theory where we have imposed boundary conditions along an initial time surface @xcite . there , we must also modify the propagator by adding an image source to obtain a consistent perturbation theory ; any new divergences that result from this modification only appear as counterterms restricted to the initial surface . the bulk theory is unchanged . in de sitter space , the inconsistency is also resolved by adding a new source term in the definition of the propagator when in an @xmath1-state . the remarkable property of de sitter space is that there exists a special point , the antipode , at which a source can be placed without breaking the @xmath3 symmetry properties of the state . for a fermion , the extra antipodal source entails some additional dirac structure . the next section derives the @xmath1-states for a spin @xmath0 dirac field in a de sitter background . the @xmath1-propagator for a point source is developed in sec . [ pointpropagation ] . we then show in sec . [ fermiloop ] that a theory with a yukawa coupling produces divergences in the one - loop corrections to the scalar propagator which can not be cancelled by adding simple counterterms to the lagrangian nor do they cancel divergences from analogous graphs where a scalar loop replaces the fermion . section [ antiprop ] shows that the these divergences can be avoided by modifying the propagator , adding an additional source at the antipode , resulting in a renormalizable theory . the final section concludes with comments on the relation between @xmath1-vacua and the problem of quantizing a theory with a specified initial state . the existence of the fermionic @xmath1-vacua was originally established in @xcite ; here we present the structure of these states in a conformally flat coordinate system . the action for a free massive fermion propagating in a general curved space - time is given by @xmath9 \psi \label{action}\ ] ] where the @xmath10 is the vierbein defining a locally flat frame , @xmath11 varying this action with respect to the fermion field yields the dirac equation , @xmath12 \psi = 0 . \label{dirac}\ ] ] in a curved background , the covariant derivative includes a term for the spin connection , @xmath13 , @xmath14 where @xmath15 . \label{sigmadef}\ ] ] in terms of the vierbein , the spin connection is given by @xmath16 a standard choice for coordinatizing de sitter space @xcite is provided by writing the metric in a conformally flat form , @xmath17 where @xmath18 $ ] and @xmath19 is the hubble constant ; we shall most often choose our units so that @xmath20 except later when comparing analogous fermionic and bosonic loops corrections . the spatial flatness of these coordinates permits a simple expression for the fields and green s functions for either a spatial or a momentum representation . these coordinates are also equivalent to the standard set used in inflation by defining @xmath21 . although they cover only half of de sitter space , indicated by the unshaded region of fig . [ penrose ] , the complementary patch is covered by an analogous set of coordinates given by taking @xmath22 with @xmath23 $ ] . ) cover the unshaded region and surfaces of constant @xmath24 are as shown.[penrose ] ] in these coordinates , the vierbein becomes ( with @xmath20 ) @xmath25 and the spin connection and the covariant derivative of a spin @xmath0 field are respectively @xmath26 \label{cfspinconnection}\ ] ] and @xmath27 , \label{cfd}\ ] ] so that the dirac equation is @xmath28 \psi + i \eta \vec\gamma \cdot \vec\nabla \psi - m \psi = 0 . \label{confdirac}\ ] ] since the metric is spatially flat , a general fermion @xmath29 is conveniently expanded in terms of positive and negative frequency modes , @xmath30 \nonumber\end{aligned}\ ] ] where the creation and annihilation operators obey the following anticommutation relations @xmath31 all other anticommutators vanish . the dirac equation can be expressed either as two coupled first order differential equations in terms of a pair of two - component spinors or as two uncoupled second order equations . of the two constants of integration , one is fixed the canonical equal - time anticommutation relation . the spinor field @xmath32 and its conjugate momentum , @xmath33 satisfy the relation , @xmath34 where the indices in this equation refer to the components of the dirac spinors . this condition fixes the normalization of the modes . inserting the mode expansion of eq . ( [ fermion ] ) into this relation and applying eq . ( [ commsforcbs ] ) , we find that the components of the modes should satisfy @xmath35^a [ u_{\vec k}^{(s)\dagger}]_b + [ v_{\vec k}^{(s)}]^a [ v_{\vec k}^{(s)\dagger}]_b \right ] = \eta^3 \delta^a_b . \label{sspin}\ ] ] the second constant of integration corresponds to specifying the vacuum state of the fermion . while it is not possible to define a globally conserved energy in de sitter space , with respect to which the vacuum is the lowest energy state , a standard vacuum can be selected by prescribing that at distances much shorter than the natural curvature length associated with de sitter space , the mode functions should match with the positive and negative frequency solutions defined for flat space . this prescription allows a theory in de sitter space effectively to inherit the renormalizability of the analogous theory in flat space . this prescription generalizes that used by bunch and davies @xcite to define the vacuum for a scalar field , so we shall also refer to this state as the `` bunch - davies vacuum . '' demanding that the leading time dependence of the modes should satisfy @xmath36 and @xmath37 at short distances completely fixes the mode functions , @xmath38 and @xmath39 in these expressions @xmath40 represents an arbitrary phase . the normalization of these mode spinors is consistent with eq . ( [ sspin ] ) since the hankel functions satisfy the following identity , @xmath41 the indices of the hankel functions are related to the mass of the fermion , @xmath42 the label @xmath43 refers to the helicity of the mode . the two - component spinors @xmath44 and @xmath45 are eigenvectors of the helicity operator , @xmath46 and they are related by @xmath47 further properties of these two - component spinors are included in the appendix . note that the negative frequency modes are the charge conjugates of the positive frequency modes , @xmath48 in the dirac representation , the charge conjugation operator is given by @xmath49 . we now define a new vacuum @xmath50 which is annihilated by the operators @xmath51 and @xmath52 , given by a bogolubov transformation of the bunch - davies operators , @xmath53 \nonumber \\ c_{\vec k}^{\alpha(s ) } & = & n_\alpha \left [ c_{\vec k}^{(s ) } + e^{\alpha^ * } b_{-\vec k}^{(s)\dagger } \right ] \label{matransform}\end{aligned}\ ] ] with @xmath54 the fermionic @xmath1-vacuum is then defined to be the state such that @xmath55 this bogolubov transformation is the fermionic analogue @xcite of the transformation introduced by mottola @xcite and allen @xcite to define the @xmath1-vacuum for a scalar field . the ma transform also induces a transformation of the mode functions , @xmath56 in terms of the conformally flat patch , the @xmath1 modes assume the form @xmath57\ , \varphi_{\hat k}^{(s ) } \cr is [ h_{\nu-1}^{(2)}(k\eta ) - is e^\alpha e^{i\theta } h_{1-\nu}^{(1)}(k\eta)]\ , \varphi_{\hat k}^{(s ) } \cr } \nonumber\end{aligned}\ ] ] and @xmath58\ , \chi_{\hat k}^{(s ) } \cr -is [ h_{1-\nu}^{(1)}(k\eta ) - is e^{\alpha^ * } e^{-i\theta } h_{\nu-1}^{(2)}(k\eta)]\ , \chi_{\hat k}^{(s ) } \cr } , \nonumber\end{aligned}\ ] ] respectively . note that the @xmath1 modes are also charge conjugates of each other , @xmath59 for simplicity , we include an @xmath1 index for the mode functions associated with the fermionic @xmath1 state so that modes functions written without an index always refer to the bunch - davies ( @xmath60 ) state . the correct prescription for defining the propagator depends on the state being considered . this dependence is very familiar in systems with boundary conditions ; for example , in flat space , if we were to choose neumann boundary conditions at some initial time ( so that time derivatives vanish there ) , the ordinary free - field propagator is not consistent with these conditions since the @xmath61-functions that enforce the time - ordering are not consistent with neumann boundary conditions . by adding a fictitious image source at the same position but with the opposite displacement in time as the physical source , we obtain a consistent structure for the time - ordering in the propagator @xcite . the extra source encodes the propagation of the initial state information . in de sitter space , the enhanced family of @xmath3 covariant states , which is linked with the existence of antipodal pairs of points , makes the construction of the propagator more subtle than in flat space where only the unique vacuum state transforms consistently with the poincar ' e invariance of the background . in this section we show how the most nave generalization of the flat space propagator ,- propagators used in this section from those used in the next , we write the former using script characters , e.g. @xmath62 , while the latter will be written as @xmath63 . these latter are the green s functions associated with two sources . green s functions without an explicit @xmath1 index are those for the bunch - davies limit , which is the same function in either case . ] @xmath64 { \cal s}^f_\alpha(x , y ) = { \delta^4(x - y)\over\sqrt{-g(x)}}\ , { \bf 1 } , \label{badfeynman}\ ] ] is only appropriate for the bunch - davies state . what emerges when we study the loop corrections in an @xmath1-state is an inconsistency in this definition . it is reflected in a dependence on antipodal points on the left side of the equation which is absent from the right side . we begin by separating the propagator @xmath65 in eq . ( [ badfeynman ] ) into two - point functions , @xmath66 where @xmath67 with @xmath68 and @xmath69 . the momentum representation of the two - point functions then corresponds to the appropriate sum over products of spinors , @xmath70 the sums over the dirac spinors can be more compactly expressed in terms of the corresponding spin sum for the bunch - davies state , with some terms evaluated at antipodal points . using the spinor modes in eqs . ( [ umodes][vmodes ] ) , we have @xmath71 \nonumber \\ & & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\ ! s_{\vec k}^<(\eta,\eta ' ) \nonumber \\ & = & { i\pi k\over 4 } e^{m\pi } ( \eta\eta')^2 \nonumber \\ & & \times \left [ \matrix { { \scriptstyle h_{-\nu}^{(1)}(k\eta ) h_{\nu-1}^{(2)}(k\eta ' ) \cdot { \bf 1 } } & { \scriptstyle - h_{-\nu}^{(1)}(k\eta ) h_\nu^{(2)}(k\eta ' ) \cdot i\hat k\cdot\vec\sigma } \cr { \scriptstyle - h_{1-\nu}^{(1)}(k\eta ) h_{\nu-1}^{(2)}(k\eta ' ) \cdot i\hat k\cdot\vec\sigma } & { \scriptstyle - h_{1-\nu}^{(1)}(k\eta ) h_\nu^{(2)}(k\eta ' ) \cdot { \bf 1 } } \cr } \right ] , \nonumber \end{aligned}\ ] ] where the lack of an @xmath1 label indicates the values for the bunch - davies limit . if we note that @xmath72 and define the following dirac operator @xmath73 , \label{calmdef}\ ] ] then we can formally write the dirac structure of the @xmath74 part of the propagator as @xmath75 while that of the @xmath76 part is @xmath77 recall that in conformally flat coordinates the antipodes associated with points @xmath78 and @xmath79 can be formally written as @xmath80 and @xmath81 . by defining the fourier transform of the operator @xmath82 to be @xmath83 , we can write the position - space representation of the two - point functions as @xmath84 from the expression for @xmath82 in eq . ( [ calmdef ] ) , note that @xmath85 is an anti - hermitian operator . in these expressions , it is understood that the operator @xmath82 ( @xmath86 ) acts on the nearest argument of the two - point function to the right ( left ) . when written in terms of antipodal coordinates , we can begin to see how loop corrections from fermions , based upon the @xmath1 propagator defined in eq . ( [ feynman ] ) , lead to exactly the same new divergences as arose in the @xmath1-vacua of a scalar field in de sitter space . in the large momentum limit , @xmath87 , the two - point functions simplify considerably since the hankel functions become proportional to exponentials ; for example , the leading behavior of the bunch - davies spin sums are given by @xmath88 \label{uvusum}\ ] ] and @xmath89 . \label{uvvsum}\ ] ] therefore , the fourier components of the @xmath90 two - point function in the high momentum limit reduce to @xmath91 \nonumber \\ & & \qquad + e^{\alpha+\alpha^ * } e^{ik(\eta-\eta ' ) } \gamma_5\gamma^0 [ k\gamma^0 - \vec k\cdot\vec\gamma ] \gamma_5\gamma^0 \nonumber \\ & & \qquad - e^\alpha e^{i\theta } e^{ik(\eta+\eta ' ) } \gamma_5\gamma^0 [ k\gamma^0 - \vec k\cdot\vec\gamma ] \nonumber \\ & & \qquad - e^{\alpha^ * } e^{-i\theta } e^{-ik(\eta+\eta ' ) } [ k\gamma^0 - \vec k\cdot\vec\gamma ] \gamma_5\gamma^0 \bigr\ } \nonumber\end{aligned}\ ] ] while for the @xmath92 two - point function , @xmath93 \nonumber \\ & & \qquad + e^{\alpha+\alpha^ * } e^{-ik(\eta-\eta ' ) } \gamma_5\gamma^0 [ k\gamma^0 + \vec k\cdot\vec\gamma ] \gamma_5\gamma^0 \nonumber \\ & & \qquad - e^{\alpha^ * } e^{-i\theta } e^{-ik(\eta+\eta ' ) } \gamma_5\gamma^0 [ k\gamma^0 + \vec k\cdot\vec\gamma ] \nonumber \\ & & \qquad - e^\alpha e^{i\theta } e^{ik(\eta+\eta ' ) } [ k\gamma^0 + \vec k\cdot\vec\gamma ] \gamma_5\gamma^0 \bigr\ } . \nonumber \end{aligned}\ ] ] different signs appear in the @xmath94 $ ] factors since we have chosen the same sign for the three momentum in eq . ( [ greensdef ] ) in both cases when defining the fourier modes . in the bunch - davies limit , the @xmath61-functions in the propagator keep the momentum dependence in the exponentials from cancelling between different lines in a loop . for example , in a loop consisting of two propagators through which momenta @xmath95 and @xmath96 flow respectively , the @xmath61-functions only allow the products @xmath97 and @xmath98 to appear . from eqs ( [ apropfuvg][apropfuvl ] ) , these products are proportional respectively to @xmath99 and @xmath100 in the @xmath101 limit . since the @xmath102-dependent parts of the phases do not cancel , it is possible to define a consistent @xmath103 prescription that renders the integral over @xmath102 finite . the appearance of all possible phases in the @xmath1 case immediately indicates that the time ordering inherited from the bunch - davies limit will always produce some phase cancellation . as an example , the product @xmath104 also occurs in the @xmath1 case , since the @xmath61-function structure is identical ; but in this product now occur terms of the proportional to @xmath105 , @xmath106 and their complex conjugates . in the @xmath107 limit , the @xmath102-dependent part of these phases cancels so that the loop integral will divergence if sufficiently many powers of @xmath102 appear in the loop integrand . stated differently , these loops contain pinched singularities @xcite . this phase cancellation is a first indication that in using a single @xmath108-function source we have not constructed a green s function which is compatible with the underlying structure of the state . we next establish that the pathologies which arise when we use this propagator do not fortuitously cancel among various divergent terms in an actual perturbative correction . to understand these pathologies better , we examine a simple one - loop graph in a theory with a scalar and a spinor interacting through a yukawa coupling , @xmath109 \psi + { \textstyle{1\over 2 } } \partial_\mu\phi\partial^\mu\phi - { \textstyle{1\over 2 } } \mu^2\phi^2 - g \phi\bar\psi\psi + \cdots . \label{yukcoup}\ ] ] later we include as well a scalar self - interaction term , @xmath110 to be able to compare the corrections from a fermion and a scalar loop to the self - energy of the scalar field , as shown in fig . [ fbloop ] . , and a cubic self - coupling , @xmath111 . [ fbloop ] ] since de sitter space lacks a well - defined @xmath112-matrix @xcite , we evaluate the self - energy corrections to the scalar using the schwinger - keldysh formalism @xcite which essentially corresponds to time - evolving both the `` in '' and `` out '' states of a matrix element . this evolution is accomplished by formally doubling the interactions of the theory , writing the interacting part of the hamiltonian as @xmath113 , \label{hyukcoup}\ ] ] with an analogous field doubling for the scalar self - interaction . in essence the first term represents the effect of time - evolving the `` in '' state while the second results from time - evolving the `` out '' state . we assume the system is initially in an general @xmath1-state , @xmath114 where @xmath115 is the parameter for the bosonic vacuum and @xmath1 is the parameter for the fermionic state . when evaluating the contraction of two fields , there are four possibilities for each field depending upon whether the states are labeled with a @xmath116 . the important feature of this formal field doubling is that the time - ordering is such that the arguments of the @xmath117 fields always occur after and _ in the opposite order _ as those of the @xmath118 fields . thus the four possible contractions of two spinors yield four propagators , @xmath119 notice that the time - ordering of a contraction of two @xmath117 fields , @xmath120 , is the opposite that of a contraction of two @xmath118 fields , @xmath121 . the structure of the contractions of scalar fields is analogous . a more extensive discussion of the schwinger - keldysh formalism as applied to de sitter space is contained in @xcite . there we show that that time - evolution of the matrix element of an operator @xmath122 from initial @xmath123 state at @xmath124 is given by @xmath125 - h_i[\psi^-,\phi^- ] \right ] } \bigr ) \langle \tilde\alpha , \alpha | t \bigl ( e^{-i\int_{\eta_0}^\infty d\eta ' \left [ h_i[\psi^+,\phi^+ ] - h_i[\psi^-,\phi^- ] \right ] } \bigr ) } . \label{inoutmatrix}\ ] ] in a generic fermionic @xmath1-state , the self - energy correction to the scalar propagator contains some terms which diverge linearly in terms of a cutoff @xmath5 imposed on the integral of the loop three - momentum . if we define the fourier transform of the fermion loop correction in fig . [ fbloop ] as @xmath126 then the linearly divergent term is @xmath127 \int_{\eta_0}^{\eta ' } { d\eta_2\over\eta_2 } \left [ { \cal g}_{\tilde\alpha , k}^>(\eta',\eta_2 ) - { \cal g}_{\tilde\alpha , k}^<(\eta',\eta_2 ) \right ] \label{sefermion } \\ & & \qquad\quad \times\biggl [ { \sin k(\eta_1-\eta_2 ) - k(\eta_1-\eta_2)\cos k(\eta_1-\eta_2 ) \over ( \eta_1-\eta_2)^3 } + { \sin k(\eta_1+\eta_2 ) - k(\eta_1+\eta_2)\cos k(\eta_1+\eta_2 ) \over ( \eta_1+\eta_2)^3 } \biggr ] + \cdots . \nonumber \end{aligned}\ ] ] the functions @xmath128 are the fourier transforms of the wightman functions for the scalar field , @xmath129 here we have only explicitly written the new divergent part for the @xmath1-vacuum although the full self - energy correction contains other finite terms as well as some divergent terms which can be cancelled by a mass counterterm . these latter terms survive in the bunch - davies limit and correspond to the usual need to renormalize the theory , which occurs even in flat space . an unrenormalized interacting field theory in flat space typically has divergent corrections . the theory remains predictive since these divergences can be removed by defining rescaled fields and couplings , in terms of which the perturbative corrections are small and finite for small couplings . this process is possible since the divergent parts of diagrams come from the region , in position space , where a loop shrinks to a point . thus such divergences are cancelled with local counterterms . the divergences in the @xmath1-states are quite different . if we shrink the loop in fig . [ fbloop ] to a point , it might appear that the divergence in eq . ( [ sefermion ] ) could be cancelled by the insertion of a mass counterterm , @xmath130 however , the resulting leading correction to the self - energy from a mass counterterm yields instead @xmath131 \left [ { \cal g}_{\tilde\alpha , k}^>(\eta',\eta_1 ) - { \cal g}_{\tilde\alpha , k}^<(\eta',\eta_1 ) \right ] \label{sedm } \\ & & + \int_{\eta_0}^\eta { d\eta_1\over\eta_1 ^ 4 } \left [ { \cal g}_{\tilde\alpha , k}^>(\eta,\eta_1 ) - { \cal g}_{\tilde\alpha , k}^<(\eta,\eta_1 ) \right ] { \cal g}_{\tilde\alpha , k}^<(\eta',\eta_1 ) + \int_{\eta_0}^{\eta ' } { d\eta_1\over\eta_1 ^ 4 } { \cal g}_{\tilde\alpha , k}^<(\eta,\eta_1 ) \left [ { \cal g}_{\tilde\alpha , k}^>(\eta',\eta_1 ) - { \cal g}_{\tilde\alpha , k}^<(\eta',\eta_1 ) \right ] \biggr\ } . \nonumber \end{aligned}\ ] ] the important feature of this contribution is that its dependence on the external momentum , @xmath132 , differs from that of the divergence from the fermion loop . this difference means that there is no constant choice for @xmath133 which removes the divergence . for comparison , the loop correction from a cubic scalar vertex @xmath134 is @xcite @xmath135 \int_{\eta_0}^{\eta_1 } { d\eta_2\over\eta_2 ^ 2 } \left [ { \cal g}_{\tilde\alpha , k}^>(\eta,\eta_2 ) - { \cal g}_{\tilde\alpha , k}^<(\eta,\eta_2 ) \right ] \nonumber \\ & & \qquad\qquad\qquad\qquad\qquad \times \biggl [ { \sin k(\eta_1-\eta_2)\over\eta_1-\eta_2 } + { \sin k(\eta_1+\eta_2)\over\eta_1+\eta_2 } \biggr ] + \cdots \label{sescalar}\end{aligned}\ ] ] where we have let the scalar field be in an @xmath115-vacuum . @xmath136 is the bosonic normalization , @xmath137 in order to be able to compare the self - energy corrections from bosonic and fermionic loops , we have restored the hubble scale @xmath19 in these expressions . here also the @xmath138-dependence prevents the possibility of using the @xmath139-loop in eq . ( [ sescalar ] ) to cancel the divergence from the original @xmath32-loop in eq . ( [ sefermion ] ) . the divergences in the loop corrections for a field in an @xmath1-state are not renormalizable in the sense that they can not be cancelled by a counterterm which contributes through an analogous graph with the loop shrunk to a point . in the case of the standard scalar @xmath1-vacuum , these diverges are removed , not by modifying the interaction part of the lagrangian , but rather by altering the propagator . this approach emerges naturally from a generalized time - ordering prescription @xcite or by regarding the @xmath1-states as squeezed states @xcite . in either case the propagator is given by the sum of two bunch - davies propagators , @xmath140 corresponding to placing sources at @xmath141 and @xmath142 , weighted appropriately . here @xmath143 is the bunch - davies propagator for a scalar field , @xmath144 the values for @xmath145 and @xmath146 differ slightly among the various prescriptions but they agree for @xmath4-invariant ( real @xmath115 ) theories @xcite . the removal of the divergent terms does not depend on the detailed form for these coefficients . in this section , we generalize this prescription to the fermionic @xmath1-states . the origin of the divergences of the last section lay essentially in an inconsistency in the non - local features of the two - point functions in eqs . ( [ aprop ] ) and the time - ordering prescription given in eq . ( [ feynman ] ) . the time - ordering was chosen so that the propagator was the green s function associated with a single point source , which was appropriate for the standard vacuum choice . however , for a non - standard choice for the initial state , the propagator must be appropriately modified to be compatible with that state . in this section we describe this modification , which leads to an essentially unique prescription for obtaining a renormalizable theory in an @xmath1-state . consider a propagator of the following form , @xmath147 we shall demonstrate that this propagator yields no new non - renormalizable loop divergences such as occurred in the previous section . the operator @xmath83 again corresponds to the fourier transform of the dirac operator @xmath82 defined in eq . ( [ calmdef ] ) and is always understood to act on the dirac index of the spinor that depends on the antipodal coordinate . the part of the propagator which depends on the antipode @xmath148 is given explicitly by @xmath149 with @xmath150 \qquad \label{monpsi}\end{aligned}\ ] ] in conformally flat coordinates . the structure of this @xmath1-propagator is essentially unique since the terms @xmath151 and @xmath152 are related to those already present , @xmath153 here we have used that antipodal times have the opposite ordering , @xmath154 . thus , once we have chosen the time - ordering of the individual terms within the @xmath1 propagator to be consistent with that of the associated euclidean two - point functions , eq . ( [ afeyndef ] ) represents the most general structure . note that since @xmath83 is anti - hermitian , the sign which would have appeared on the right side of the first line in eq . ( [ antipodereduce ] ) is cancelled . the free field generating functional , @xmath155 = { \int { \cal d}\psi { \cal d}\bar\psi\ , e^{i \int d^4x\ , \sqrt{-g } \left\ { { \cal l}_0(x ) + \bar\xi(x ) \psi(x ) + \bar\psi(x ) \xi(x ) \right\ } } \over \int { \cal d}\psi { \cal d}\bar\psi\ , e^{i \int d^4x\ , \sqrt{-g } { \cal l}_0(x ) } } , \label{freegenfunc}\ ] ] is constructed so that a functional derivative with respect to each of the sources yields the correct propagator , @xmath156 \bigl [ i { \textstyle{\delta\over\delta\bar\xi(y ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } & = & \langle \alpha | t \bigl ( \psi(x ) \bar\psi(y ) \bigr ) | \alpha \rangle \nonumber \\ & = & -i s_\alpha^f(x , y ) . \label{saffromw}\end{aligned}\ ] ] since the propagator in this theory contains both the @xmath78 and its antipode @xmath148 , the source should couple the field at both points which is represented in the generating functional by the spinor field @xmath157 , @xmath158 which through a change of coordinates can be written as coupling the spinor @xmath29 to sources @xmath159 and @xmath160 . we assume that @xmath161 and @xmath162 are real . extracting the propagator from @xmath163 $ ] , we obtain eq . ( [ afeyndef ] ) when @xmath164 the free generating functional can also be written in a form that is quadratic in the sources . if we shift the fermion field by replacing @xmath29 with @xmath165 the generating functional becomes @xmath155 = e^{-i \int d^4x\ , \sqrt{-g}\int d^4y\ , \sqrt{-g}\ , \bar\xi(x ) s_\alpha^f(x , y ) \xi(y ) } . \label{freegenfuncxx}\ ] ] the generating functional for an interacting theory is given by extending the lagrangian in the path integral to include interactions , @xmath166 = { \int { \cal d}\psi { \cal d}\bar\psi\ , e^{i \int d^4x\ , \sqrt{-g } \left\ { { \cal l}(x ) + \bar\xi(x ) \psi(x ) + \bar\psi(x ) \xi(x ) \right\ } } \over \int { \cal d}\psi { \cal d}\bar\psi\ , e^{i \int d^4x\ , \sqrt{-g } { \cal l}(x ) } } . \label{genfunc}\ ] ] for illustration , we examine a general yukawa interaction of the form @xmath167 + \bar\psi v_3(x ) \psi . \label{yukact}\ ] ] for generality , we have coupled a complete set of fermion bilinears constructed from the local field @xmath29 and the non - local field @xmath157 given in eq . ( [ psidef ] ) to potentials @xmath168 , @xmath169 and @xmath170 which could represent other fields present in the theory and can include a non - trivial dirac structure as well . we have included a general set of interactions , including terms which contain an antipodal dependence , since the sources @xmath171 and @xmath172 have already introduced some antipodal non - locality into the theory . the fields that appear in the interactions , whether a @xmath29 or @xmath157 field , determine the corresponding functional derivative required to rewrite @xmath173 $ ] in terms of @xmath163 $ ] . since both @xmath174 and @xmath175 couple to the source @xmath176 , the functional derivative used to extract a factor of @xmath174 is constructed remove the antipodal term , @xmath177 ; \label{xidef}\ ] ] thus , for example , @xmath178 a simple functional derivative with respect to @xmath176 , extracts instead the linear combination @xmath157 , @xmath179 therefore , for the yukawa interactions given in eq . ( [ yukact ] ) the corresponding generating functional is @xmath166 = % { e^{i\int d^4x\ , \sqrt{-g}\ , \left\ { \left [ -i { \delta\over\delta\xi(x ) } \right ] v_1(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] + \left [ \left [ -i { \delta\over\delta\xi(x ) } \right ] v_2(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] + { \rm h.c . } \right ] + \left [ -i { \delta\over\delta\xi(x ) } \right ] v_3(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] \right\ } } w_0[\xi,\bar\xi]\over e^{i\int d^4x\ , \sqrt{-g}\ , \left\ { \left [ -i { \delta\over\delta\xi(x ) } \right ] v_1(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] + \left [ \left [ -i { \delta\over\delta\xi(x ) } \right ] v_2(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] + { \rm h.c . } \right ] + \left [ -i { \delta\over\delta\xi(x ) } \right ] v_3(x ) \left [ i { \delta\over\delta\overline\xi(x ) } \right ] \right\ } } w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } } . \label{genfuncderv}\ ] ] the appearance of the different types of interactions leads to a richer propagator structure depending upon whether the contraction is between an internal or an external vertex and the the type of interaction . an @xmath180-point function with @xmath181 external @xmath32 lines and @xmath182 external @xmath183 lines is given by @xmath184 \prod_{j=1}^{\bar n}\bigl [ i { \textstyle{\delta\over\delta\bar\xi(y_j ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } \qquad \label{npoint}\end{aligned}\ ] ] which also implicitly generalizes the time - ordering used in the @xmath1-vacuum to products of more than two fields . thus an external point is always associated with a @xmath185 or a @xmath186 derivative . the functional derivative for an internal vertex , however , can be either of these as well as @xmath187 or @xmath188 depending upon the form of the interaction , as indicated in eq . ( [ genfuncderv ] ) . a theory whose interactions only contain the field @xmath157 will only contain @xmath1 propagators , defined in eq . ( [ afeyndef ] ) . but a theory with only local interactions , depending solely on @xmath29 , or mixed interactions , as in the term coupled to @xmath169 in eq . ( [ yukact ] ) , will contain several sets of propagators either a purely bunch - davies propagator , @xmath189 \bigl [ i { \textstyle{\delta\over\delta\xi(y ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } , \label{bdpropw}\ ] ] a mixed propagator , @xmath190 \nonumber \\ & = & \bigl [ -i { \textstyle{\delta\over\delta\bar\xi(x ) } } \bigr ] \bigl [ i { \textstyle{\delta\over\delta\xi(y ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } \nonumber \\ & = & \bigl [ -i { \textstyle{\delta\over\delta\bar\xi(x ) } } \bigr ] \bigl [ i { \textstyle{\delta\over\delta\xi(y ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } \qquad \label{mpropw}\end{aligned}\ ] ] or a genuine @xmath1-propagator , @xmath191 \bigl [ i { \textstyle{\delta\over\delta\xi(y ) } } \bigr ] w_0[\xi,\bar\xi ] \bigr|_{\xi=\bar\xi=0 } ; \label{apropw}\ ] ] here the @xmath1-propagator is that given in eq . ( [ afeyndef ] ) and not the propagator corresponding to a single point source , as discussed in sec . [ pointpropagation ] . one of the consequences of eq . ( [ bdpropw ] ) is that in a theory with only local interactions , so that @xmath157 does not appear in @xmath192 , automatically has the same divergence structure in loop corrections as a theory in the bunch - davies vacuum . consequently , the counterterms and the renormalization is identical to the usual vacuum state . the only source of @xmath1-dependence is that which occurs in the external legs which will be mixed propagators , eq . ( [ mpropw ] ) . a more interesting scenario , which more closely resembles the real - time ordering prescription applied to a scalar theory in @xcite , is to consider a theory constructed only out of the linear combination @xmath157 . in this case , all of the fermion propagators in a diagram are @xmath1-propagators and the renormalization differs slightly from that of the bunch - davies limit @xcite . in the next section we shall show that even with the additional antipodal terms in the double source @xmath1 , no phase cancellation occurs in the uv region of the loop momentum which produced the non - renormalizable divergences seen in the previous section . as an example we examine a theory with a non - local yukawa interaction of the special form , @xmath193 for complete generality , we have coupled the linear combination @xmath157 which yields only @xmath1-propagators in the diagrams to a similar linear combination for the scalar field @xcite , @xmath194^{1/2}\ , \phi(x ) \nonumber \\ & & + { 1\over\sqrt{2 } } { b_{\tilde\alpha}\over \left [ a_{\tilde\alpha } + \sqrt{a_{\tilde\alpha}^2 - b_{\tilde\alpha}^2 } \right]^{1/2 } } \ , \phi(x_a ) , \label{scalarlin}\end{aligned}\ ] ] constructed so that the scalar lines in a graph correspond to the bosonic @xmath1-propagator of eq . ( [ doublescalar ] ) . as before , each field can be in a different invariant state , as we have denoted by writing different labels for the scalar , @xmath115 , and spinor , @xmath1 , states . the leading correction to the scalar propagator is given by the one - loop graph shown in fig.[sig1def ] . . [ sig1def ] ] the only new element of for the schwinger - keldysh approach is to state how the additional antipodal propagators affect the structure of a contraction between @xmath195 fermions , @xmath196 note that the additional terms can be written with either coordinate evaluated at the antipode using the relation @xmath197 which follows directly from eqs . ( [ bfsdef][calmdef ] ) . with these forms for the propagators , we find @xmath198 [ \tilde g_k^>(\eta',\eta_2 ) l_k^{{\scriptscriptstyle > , < } } ( \eta_1,\eta_2 ) - \tilde g_k^<(\eta',\eta_2 ) l_k^{{\scriptscriptstyle < , > } } ( \eta_1,\eta_2 ) ] \nonumber \\ % & & + ig^2 \int_{\eta_0}^{\eta ' } { d\eta_2\over\eta_2 ^ 4 } \int_{\eta_0}^{\min(\eta,\eta_2 ) } { d\eta_1\over\eta_1 ^ 4 } [ \tilde g_k^<(\eta,\eta_1 ) l_k^{{\scriptscriptstyle > , < } } ( \eta_1,\eta_2 ) - \tilde g_k^>(\eta,\eta_1 ) l_k^{{\scriptscriptstyle < , > } } ( \eta_1,\eta_2 ) ] [ \tilde g_k^>(\eta',\eta_2 ) - \tilde g_k^<(\eta',\eta_2 ) ] \nonumber \\ % & & - ig^2 \int_{\eta'}^\eta { d\eta_1\over\eta_1 ^ 4 } \int_{\eta'}^{\eta_1 } { d\eta_2\over\eta_2 ^ 4 } [ \tilde g_k^>(\eta,\eta_1 ) - \tilde g_k^<(\eta,\eta_1 ) ] \tilde g_k^<(\eta',\eta_2 ) [ l_k^{{\scriptscriptstyle > , < } } ( \eta_1,\eta_2 ) - l_k^{{\scriptscriptstyle < , > } } ( \eta_1,\eta_2 ) ] \nonumber \\ % & & + ig^2 \int_\eta^{\eta ' } { d\eta_2\over\eta_2 ^ 4 } \int_\eta^{\eta_2 } { d\eta_1\over\eta_1 ^ 4 } \tilde g_k^<(\eta,\eta_1 ) [ \tilde g_k^>(\eta',\eta_2 ) - \tilde g_k^<(\eta',\eta_2 ) ] [ l_k^{{\scriptscriptstyle > , < } } ( \eta_1,\eta_2 ) - l_k^{{\scriptscriptstyle < , > } } ( \eta_1,\eta_2 ) ] . % \label{loopyuka}\end{aligned}\ ] ] in this expression the fermionic loop integrals have been written as @xmath199 \nonumber \\ l_k^{{\scriptscriptstyle < , > } } ( \eta_1,\eta_2 ) & = & \int d^3\vec p\ , { \rm tr } \bigl [ \tilde s_{\vec p}^<(\eta_1,\eta_2 ) \tilde s_{\vec p-\vec k}^>(\eta_2,\eta_1 ) \bigr ] \qquad \label{loopgtlt}\end{aligned}\ ] ] and we have grouped together the two - point functions to include both ordinary and antipodal pieces ; for the fermions we have @xmath200 while for the scalar field , @xmath201 in the ultraviolet region of the the loop integral , the leading @xmath102-dependence of either integral in eq . ( [ loopgtlt ] ) is the same , @xmath202 \nonumber \\ & \to & - ( \eta_1\eta_2)^3 \bigl [ 1 + \hat p \cdot \widehat{p - k } \bigr ] \nonumber \\ & & \bigl [ a_\alpha^2 e^{-i[p+|\vec p-\vec k|](\eta_1-\eta_2 ) } - b_\alpha^2 e^{i[p+|\vec p-\vec k|](\eta_1+\eta_2 ) } \bigr ] \nonumber \\ & & + \cdots \label{uvstildes}\end{aligned}\ ] ] note that there is no longer any phase cancellation in the far uv of the loop integral , such as occurred in the previous section . in fact , the second term will never present any new divergences since @xmath203 and it is always possible to define an @xmath103 prescription which renders the @xmath204-term finite . the remaining term , proportional to @xmath205 , has exactly the same structure as the loop divergence in the bunch - davies vacuum . it can be renormalized therefore in the same way as in the standard vacuum through a wave - function and mass renormalization . while the construction of the two - source propagator arose from a need to remove the non - renormalizable divergences that appeared for the single - source propagator , it can be viewed from the more general vantage of the need to write the propagator consistently with the initial state . this perspective should apply even if the state we are using is well behaved in the ultraviolet approaching the bunch - davies state at short distances but dramatically different at longer distances , such as the `` truncated @xmath1-vacua '' @xcite . it would be useful to learn whether using the construction here leads to a different estimate for the size of transplanckian effects in the cosmic microwave background . the existence of fermionic @xmath1-states can lead to more important role for fermions in inflation than is usually assumed . for example , when the inflaton is coupled to a fermion in a non - thermal state , the loop corrections can be significantly larger than would be expected from the bunch - davies vacuum @xcite . moreover , since this effect only depends on the state of the field in the loop , it can occur even when the inflaton is itself in the bunch - davies vacuum . the presence of many potential fermionic degrees of freedom in the standard model allows for further enhancement , depending upon what fraction of the fermions are in a non - thermal state during inflation . it is important to note that if the @xmath1-vacua are to be applied to model of inflation , then the model must allow for their decay once the inflationary phase has ended . just as with the scalar @xmath1-vacuum , the value of @xmath1 today for a fermion is severely constrained by cosmic ray measurements . the values of @xmath1 during inflation and during a possible de sitter phase today , however , do not need to be the same . the @xmath1-states are only the vacua of a pure de sitter background ; the de sitter symmetry is clearly broken in our universe . nothing therefore precludes the decay of @xmath1 to an observationally acceptable value after inflation has ended @xcite . while interesting in itself , the @xmath1-vacuum is perhaps even more valuable as an illustration of the new elements required to describe the propagation of a field in a state other than the standard vacuum . one of these new elements is that the propagator is modified . in the @xmath1 case , it becomes the green s function associated with two sources : one corresponds to the physical source while the other is located at its antipode . the antipodal source encodes the effect of the background state on the propagation of a fermion very analogously to how a fictitious image charge encodes the effects of a conducting plate on the propagation of a charge in classical electrodynamics . ultimately we would like to extend this formalism to a generic initial state in a general robertson - walker background . the enormous stretching of scales during inflation would suggest that the power spectrum of the cosmic microwave background should be sensitive to the details of a field theory in the deeply ultra - violet region . however , the observed spectrum is still completely consistent with having the inflaton and all other fields which could appreciably affect it , such as the fermions discussed here in the bunch - davies vacuum . what is needed to reconcile these two observations is an extension of the principle of decoupling to rapidly expanding backgrounds . in flat space , we have a well defined prescription for understanding the corrections from a more fundamental theory valid at high energies on the effective theories we apply at low energies @xcite . similarly , we need to consider the size of the error we make in choosing a bunch - davies state when the actual short distance properties of the state are different . to do so requires being able to renormalize the theory in that state . by examining the @xmath1-vacuum , we have learned that the propagator must be modified to obtain the correctly renormalized theory . more generally , while the @xmath1-states respect the background space - time symmetries , an arbitrary initial state does not . this lessened symmetry means that we must also renormalize the initial state @xcite . when such a program has been completed , we shall be able to understand the extent to which our ignorance of the short distance details of the state can be exchanged for a scale - dependent , renormalized initial state and , more importantly , the size of the corrections we can expect to observe . this work was supported in part by doe grant de - fg03 - 91-er40682 . i have greatly benefitted from conversations with rich holman and matt martin . throughout this article , we work in the dirac representation for the @xmath206-matrices , @xmath207 the two - component positive and negative helicity eigenspinors are denoted respectively by @xmath44 and @xmath45 . their normalizations are fixed to satisfy @xmath208^i [ \varphi^{(s)\dagger}_{\hat k}]_j & = & \delta^i_j \nonumber \\ \sum_s [ \chi^{(s)}_{\hat k}]^i [ \chi^{(s)\dagger}_{\hat k}]_j & = & \delta^i_j \label{spinnorm}\end{aligned}\ ] ] and @xmath209^i [ \varphi^{(s)\dagger}_{\hat k}]_j & = & [ \hat k\cdot \vec\sigma]^i_j \nonumber \\ \sum_s s[\chi^{(s)}_{\hat k}]^i [ \chi^{(s)\dagger}_{\hat k}]_j & = & - [ \hat k\cdot \vec\sigma]^i_j \label{dspinnorm}\end{aligned}\ ] ] where the indices @xmath210 and @xmath211 refer to the components of the spinors . if in addition we would like these spinors to obey @xmath212 , then this condition along with eqs . ( [ omegadef][spinnorm ] ) is enough to fix the form of the helicity eigenspinors , @xmath213 where @xmath214 note that these two - component spinors satisfy @xmath215^ * \nonumber \\ \chi^{(-)}_{\hat k } & = & \varphi^{(-)}_{-\hat k } = -i \sigma^2 [ \varphi^{(-)}_{\hat k}]^ * . \label{hspinorsrels}\end{aligned}\ ] ] later we shall use the following spinor product relation , @xmath216 which vanishes when @xmath217 . a. linde , _ particle physics and inflationary cosmology _ ( harwood academic , chur , switzerland , 1990 ) ; e. w. kolb and m. s. turner , _ the early universe _ ( addison - wesley , reading , ma , 1990 ) ; a. liddle and d. lyth , _ cosmological inflation and large - scale structure _ ( cambridge university press , cambridge , england , 2000 ) . r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 64 * , 103502 ( 2001 ) ; r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 67 * , 063508 ( 2003 ) ; g. shiu and i. wasserman , phys . b * 536 * , 1 ( 2002 ) ; r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 66 * , 023518 ( 2002 ) . j. c. niemeyer , phys . d * 63 * , 123502 ( 2001 ) ; a. kempf , phys . d * 63 * , 083514 ( 2001 ) ; j. c. niemeyer and r. parentani , phys . d * 64 * , 101301 ( 2001 ) ; a. kempf and j. c. niemeyer , phys . d * 64 * , 103501 ( 2001 ) ; a. a. starobinsky , pisma zh . fiz . * 73 * , 415 ( 2001 ) [ jetp lett . * 73 * , 371 ( 2001 ) ] ; l. hui and w. h. kinney , phys . d * 65 * , 103507 ( 2002 ) ; s. shankaranarayanan , class . * 20 * , 75 ( 2003 ) ; s. f. hassan and m. s. sloth , nucl . phys . b * 674 * , 434 ( 2003 ) ; k. goldstein and d. a. lowe , phys . d * 67 * , 063502 ( 2003 ) ; v. bozza , m. giovannini and g. veneziano , jcap * 0305 * , 001 ( 2003 ) ; g. l. alberghi , r. casadio and a. tronconi , phys . b * 579 * , 1 ( 2004 ) ; j. martin and r. brandenberger , phys . d * 68 * , 063513 ( 2003 ) ; r. h. brandenberger and j. martin , int . j. mod . a * 17 * , 3663 ( 2002 ) . c. p. burgess , j. m. cline , f. lemieux and r. holman , jhep * 0302 * , 048 ( 2003 ) [ hep - th/0210233 ] ; c. p. burgess , j. m. cline and r. holman , jcap * 0310 * , 004 ( 2003 ) [ hep - th/0306079 ] ; c. p. burgess , j. cline , f. lemieux and r. holman , astro - ph/0306236 . h. georgi , `` weak interactions and modern particle theory , '' ( benjamin / cummings , menlo park , california , 1984 ) ; h. georgi , `` effective field theory , '' ann . nucl . part . sci . * 43 * , 209 ( 1993 ) ; i. z. rothstein , `` tasi lectures on effective field theories , '' hep - ph/0308266 .

long - time - integrated quantities in stochastic processes , in or out of equilibrium , usually exhibit rare but huge fluctuations . work or heat production is such a quantity , of which the probability distribution function displays an exponential decay characterized by the large deviation function ( ldf ) . the ldf is often deduced from the cumulant generating function through the inverse fourier transformation . the saddle - point integration method is a powerful technique to obtain the asymptotic results in the fourier integral , but a special care should be taken when the saddle point is located near a singularity of the integrand . in this paper , we present a modified saddle - point method to handle such a difficulty efficiently . we investigate the dissipated and injected heat production in equilibration processes with various initial conditions , as an example , where the generating functions contain branch - cut singularities as well as power - law ones . exploiting the new modified saddle - point integrations , we obtain the leading finite - time corrections for the ldf s , which are confirmed by numerical results .

1307.1556

detailed balance is satisfied in equilibrium and gives rise to the boltzmann distribution , which is a well established basis for equilibrium statistical mechanics . on the other hand , nonequilibrium is characterized by the breakage of detailed balance and in turn there appears irreversibility in dynamics . a typical consequence is the existence of nonzero current in state space . it has been noticed that nonzero current accompanies an incessant production of work , hence heat and entropy @xcite-@xcite , each of which satisfies the fluctuation theorem ( ft ) given at specific initial distributions @xcite-@xcite . such time - integrated quantities exhibit rare but huge fluctuations which are prominent in small systems . the large deviation function ( ldf ) is the characteristic function that contains all the information regarding complicated fluctuations in the long - time limit and has been nowadays one of main issues in nonequilibrium statistical mechanics @xcite-@xcite . for a time - integrated quantity @xmath0 produced from time @xmath1 to @xmath2 , the ldf @xmath3 for its average production rate @xmath4 is defined as @xmath5 where @xmath6 is the probability density function ( pdf ) of rate @xmath7 for @xmath0 produced up to time @xmath8 . it provides an essential information on the asymptotic property of fluctuations in the long - time limit @xcite . experimental or numerical confirmation for a theoretically obtained ldf is a very difficult task because the ldf tail is determined by extremely rare events . van zon and cohen @xcite studied heat production of a brownian particle trapped in the harmonic potential moving with a constant velocity and found that the heat production pdf exhibits a deviation from the conventional ft in the tail region . their numerical simulation data , however , did not seem to show good accordance with the theoretical ldf due to an insufficient number of samples . there were also experimental attempts in the electric circuit and mechanical pendulum setups @xcite . however , it also seemed not clear that the experimental data are fully consistent with the theoretical estimates in the tail region . therefore , it is desirable to calculate the finite - time correction of the ldf so as to confirm the validity of the theory from the finite - time data in numerical or experimental tests . the cumulant generating function associated with @xmath6 is defined as @xmath9 in most cases @xcite , it is easier to calculate the generating function than the pdf directly . then @xmath6 can be deduced by the inverse fourier transform of the generating function . the corresponding fourier integral can be estimated for large @xmath8 as @xmath10 where @xmath11 is factorized into the exponential term contributed to @xmath12 and the leftover to @xmath13 for large @xmath8 . the integral path @xmath0 is chosen as the steepest descent contour passing through the saddle point , which is usually taken as the solution of @xmath14 with @xmath15 . we call this saddle point as a _ conventional _ saddle point , denoted by @xmath16 . the gaussian integration for equation ( [ inverse_fourier ] ) near @xmath16 , which will be called the conventional saddle - point method , yields @xmath17 where @xmath18 and @xmath19 is an angle between the steepest descent path and the horizontal axis at @xmath20 . when there is no singularity in the prefactor @xmath13 , the above result leads to the correct ldf @xmath3 in equation ( [ ldf_definition ] ) and its finite - time correction @xcite . however , there are many examples where the prefactor has a power - law type singularity @xcite . if the prefactor has a singularity at @xmath21 and @xmath22 passes through @xmath23 at @xmath24 as @xmath7 varies , the conventional saddle - point method gives rise to the @xmath19-function type divergence in the pdf at @xmath24 , which is physically unreasonable . this problem was carefully treated in @xcite based on @xcite and @xcite when the prefactor has a simple or square - root pole , respectively . however , in those works , the conventional saddle points were still used to construct the steepest descent contour , thus the calculation of the ldf near @xmath24 demands rather complicated algebra as well as composite deformed contours . in this study , we take a different saddle point , denoted by @xmath25 , which is the solution of @xmath26/d\lambda=0 $ ] @xcite . this saddle point is @xmath8-dependent , but never passes through the singularity . it approaches the singularity only asymptotically in the long - time limit . this feature simplifies the analysis to obtain the ldf as well as its finite - time correction . however , special care should be taken to calculate the integral near the _ modified _ saddle point @xmath27 . when @xmath27 approaches the singular point asymptotically , the integration along the steepest descent path near the modified saddle point becomes a _ non - gaussian _ integral , which means that the usual gaussian integration can not be performed as in the conventional saddle point method . this feature was not properly treated in @xcite , where the usual gaussian integration was used for the modified saddle point . therefore , we develop a saddle - point integration method to treat a non - gaussian integration near this _ modified _ saddle point @xmath25 , especially when @xmath25 asymptotically approaches the singular point . to illustrate our method explicitly , we revisited the equilibration process @xcite as an example . in this case , the prefactor @xmath13 has a square - root singularity with a branch cut . however , our modified method is applicable to general power - law type singularity ( see equation ( [ alpha - singularity ] ) ) and it would be straightforward to generalize to any type of singularities including an essential singularity . we show all mathematical details of our modified saddle - point method to obtain the ldf s and their leading finite - time corrections . this paper is organized as follows . in sec . 2 , we introduce the modified saddle point method and discuss its advantage over the conventional method . in sec . 3 , the equilibration process is introduced in brief . in sec . 4 , we calculate the ldf s for the dissipated and injected powers of heat in the long - time limit . in sec . 5 , detailed calculation results are presented for finite - time corrections of the ldf s . we also perform numerical simulations to confirm our results . numerical data are in excellent agreement with the analytic results . finally , we summarize our work in sec . 6 . in the appendix a , the details of the modified saddle - point method are presented . consider the case when the prefactor has a singularity such that @xmath28 then , equation ( [ inverse_fourier ] ) becomes @xmath29 where @xmath30 and @xmath12 are analytic functions of @xmath31 . suppose that the conventional saddle point @xmath22 satisfying @xmath32 , passes through @xmath23 at @xmath24 , i.e. @xmath33 . the modified saddle point @xmath25 is determined by the equation , @xmath34 . here , the last term is always negligible for large @xmath8 , thus can be ignored . however , the second term can become comparable to the first term when the saddle point is in the vicinity of @xmath23 , and thus , should be taken into account . hence the saddle point equation becomes @xmath35 in contrast to @xmath22 , the solution @xmath25 does not pass through @xmath23 but asymptotically approaches it for large @xmath8 . in order to illustrate this feature clearly in an example , we assume that @xmath23 and @xmath22 are located on a real axis and @xmath36 is a real and monotonically increasing function of @xmath31 on a real axis . figs . [ modified_saddle_point](a ) , [ modified_saddle_point](b ) , and [ modified_saddle_point](c ) show the plots for @xmath37 , @xmath38 , and @xmath39 $ ] versus real @xmath31 for @xmath40 , @xmath41 , and @xmath42 cases , respectively . as shown in the figures , there are always two solutions satisfying @xmath43 ; @xmath44 and @xmath45 which are located on the left and right side of @xmath23 respectively , due to the singularity even though there is only one solution @xmath46 for @xmath47 . furthermore , @xmath44 and @xmath45 can not pass through @xmath23 as @xmath7 ( or @xmath46 ) varies , while @xmath22 can . instead , @xmath44 and @xmath45 asymptotically approach @xmath23 or @xmath22 for large @xmath8 depending on the location of @xmath22 : @xmath48 , @xmath44 , and @xmath45 when ( a ) @xmath49 , ( b ) @xmath41 , and ( c ) @xmath50 . the horizontal axis represents @xmath31 , and the various lines represent @xmath37 ( thick line ) , @xmath38 ( dot - dashed line ) , and @xmath51 ( dashed line ) , respectively . ] these modified saddle points , @xmath44 and @xmath45 , make the integration of equation ( [ new_saddle_integration ] ) much simpler compared to the case using a conventional one @xmath46 . for general non - integer @xmath52 , non - analytic branch cuts appear in the complex plane of @xmath31 , which becomes a nuisance because the integration path should be chosen not to cross them . with the two modified saddle points , it is always possible to choose one of them , of which the steepest descent path does not cross the branch cuts . in contrast , this is not always possible with a single conventional saddle point . this is a big advantage of our modified saddle point method over the conventional one . when @xmath46 is located on a branch cut . ( b ) contour path for a modified saddle point @xmath45 . ] consider the case of @xmath42 in fig . [ modified_saddle_point](c ) with a branch cut on a real axis for @xmath53 as in fig . [ advantage ] . with the choice of @xmath45 as the saddle point , one can construct the steepest descent path not crossing the branch cut in fig . [ advantage](b ) . hence , the leading contribution of equation ( [ new_saddle_integration ] ) is simply obtained as the integrand evaluated at @xmath54 : @xmath55\simeq \exp \left [ \tau h(\lambda_+^*;\varepsilon)\right]$ ] . note that @xmath56 vanishes not exponentially , but only algebraically in time @xmath8 ( see appendix a ) . from equation ( [ two_solutions ] ) , we find that the ldf becomes in the infinite - time limit @xmath57 here , the interesting feature is the ` saddle - point fixation ' at the singular point @xmath23 for @xmath58 , which is a natural and straightforward consequence in our scheme . equation ( [ ldf0 ] ) can be also obtained by using the conventional saddle point scheme . however , in the conventional saddle point scheme , the saddle point lies at the branch cut and nontrivial explanation is needed to derive the same result @xcite . this advantage also applies to the case of integer @xmath59 . our method has a clear advantage in obtaining a finite - time correction for @xmath6 when @xmath42 . in the conventional approach @xcite , one needs to construct the detoured path composed of @xmath60 , @xmath61 , @xmath62 , and @xmath63 segments as in fig . [ advantage](a ) , in order to avoid the branch cut . therefore , integrations for all four segments are needed to obtain a finite - time correction . however , in our scheme ( fig . [ advantage](b ) ) , we need to calculate only one saddle point integration near @xmath45 , which makes the formulation much simpler ( see equation ( [ app_int_c3 ] ) ) . in the following sections , we investigate the equilibration process as an example to reveal these advantages more explicitly . consider a brownian particle which is initially in equilibrium with a heat bath a at temperature @xmath64 . at @xmath65 , the thermal contact is abruptly switched to the heat bath b at temperature @xmath66 and is maintained forever . for @xmath67 , the motion of the brownian particle with unit mass is described by the langevin equation @xmath68 where @xmath69 is the velocity of the particle and @xmath70 is the dissipative coefficient . gaussian white noise @xmath71 satisfies @xmath72 and @xmath73 , where the fluctuation - dissipation relation holds as @xmath74 . here , we set the boltzmann constant @xmath75 for convenience . when @xmath76 , the system will remain in the same initial equilibrium state . for @xmath77 , however , the system goes through a relaxation process toward equilibrium with the heat bath b. note that the initial velocity distribution @xmath78 of the particle at @xmath1 is given as @xmath79 where @xmath80 is the ratio of the two temperatures . the total heat in the equilibration process is not an extensive quantity in time because it ceases to be produced as the system approaches equilibrium . however , decomposed into two _ partial _ heats such as dissipated one @xmath81 and injected one @xmath82 , then each is accumulated incessantly in time . multiplying @xmath69 to equation ( [ langevin ] ) and integrate it over time @xmath83 from @xmath84 to @xmath8 , we find @xmath85 where @xmath86 , which is the energy difference between the final and initial time , and partial heats are defined as @xmath87 the fluctuation nature of @xmath88 and @xmath82 is quantified by their pdf s , @xmath89 and @xmath90 , respectively . @xmath91 and @xmath92 increase linearly in time for sufficiently large @xmath8 , while their difference @xmath93 is proportional to @xmath94 that is bounded . therefore , in the long - time limit , our conventional wisdom may lead us to expect that the pdf s of @xmath88 and @xmath82 will lose all initial memory , thus become independent of @xmath95 . however , it has been noticed in various examples @xcite that the effect of initial conditions can remain in the tail of the pdf s ( rare - event region ) even in the infinite - time limit . this can be understood intuitively in general as follows : a large amount of heat with respect to its mean can be produced from a relaxation ( decay ) dynamics of highly energetic particles , which is an exponentially rare event . highly energetic particles can be generated from a given initial ensemble at the beginning , which are also exponentially rare but become a source for a large amount of heat in the long - time limit by losing most of their energy by decay dynamics . this initial - condition - dependent rare events certainly affect the ( exponentially small ) tail of the pdf s even in the infinite - time limit . however , there is another source for highly energetic particles generated by the heat bath , which is also exponentially rare . these two rare events compete each other and sometimes a sharp nontrivial threshold for the initial condition ( @xmath95 ) appears with regard to the initial - condition dependence of the pdf tail shape or the ldf @xcite . in the following sections , we will explicitly calculate the pdf s and analyze their ldf s with finite - time corrections . the pdf of each heat production @xmath96 is expected to exhibit a large deviation nature , @xmath97 $ ] for large @xmath8 . then , it is convenient to express the ldf as a function of heat production rate , i.e. , power . here , the dissipated power is defined as @xmath98 and the injected power as @xmath99 . note that @xmath100 as @xmath88 is always positive by definition in equation ( [ heats ] ) , while @xmath101 can take any value . to calculate @xmath103 , it is convenient to compute first its generating function @xmath104 equation ( [ generatingf ] ) can be exactly calculated by using the standard path integral method @xcite with the initial distribution as in equation ( [ initial_dist ] ) . the result is given by @xcite @xmath105 where @xmath106 and @xmath107 . then , the pdf of the dissipated power can be obtained from its inverse fourier transform as @xmath108 where the dimensionless dissipated power is defined as @xmath109 . the leading contribution of the above integration can be obtained by using the saddle point method in the large-@xmath8 limit . however , care should be taken when there is a singularity characterized by @xmath110 this singular point is in fact a _ point connected to the branch cut . for convenience , we choose the branch cut lying on the real-@xmath111 axis for @xmath112 . the branch point location depends on @xmath95 in general as shown in figures [ branch](a ) and [ branch](b ) . first , consider the case with @xmath113 . for @xmath114 , @xmath115 is positive real , so @xmath116 for large @xmath8 . here , @xmath117 is also positive , so there is no branch point satisfying @xmath118 . however , for @xmath119 , @xmath115 becomes pure imaginary and @xmath120 with @xmath121 . one can find many solutions satisfying @xmath118 such as @xmath122 ( @xmath123 ) for large @xmath8 , equivalently @xmath124 $ ] . so there are infinitely many branching points with vanishingly small intervals between them . as we choose the branch cut for @xmath125 , there appear infinitely many patches of branch cuts with vanishingly small sizes in the negative real axis for @xmath126 . note that all branch points have no @xmath95 dependence and approach @xmath127 from below in the large-@xmath8 limit . second , for @xmath128 , @xmath117 can be negative for @xmath114 . in this range of @xmath129 , @xmath115 is positive real and @xmath130 for large @xmath8 . therefore , the branch point is the solution for @xmath131 , equivalently @xmath132 , which is negative and greater than @xmath133 . for @xmath119 , we have a similar branch - cut structure to the case for @xmath134 . on the complex @xmath129 plane for @xmath134 and @xmath128 , respectively . ( c ) and ( d ) show the branch - cut structure of @xmath135 for @xmath136 and @xmath137 , respectively . wiggled lines denote branch cuts . a head of an each arrow locates at its asymptotic value of the respective branch point as @xmath138 . ] as discussed in section [ sec2 ] , the new _ modified _ saddle point never passes through the largest branch point ( singularity of the prefactor ) and its asymptotic location critically depends on the relative position of the conventional saddle point and the largest branch point , see equation ( [ two_solutions ] ) . therefore , only the location of the largest branch point is relevant in the saddle point integration . to summarize , the generating function for the dissipated power has infinitely many branch points with a square - root singularity on the negative real axis and the largest branch point @xmath139 is located asymptotically at @xmath140 the branch cut is located to the left of the largest branch point on the real axis , so it does not come into play for the saddle point calculation because the modified saddle point always sits outside of the branch cut . the ldf can be obtained by applying the saddle point approximation to equation ( [ p_d_integration ] ) . we look for a modified saddle point in the region of @xmath114 ( real positive @xmath115 ) . then , equation ( [ f_d ] ) becomes @xmath141 for large @xmath8 . from equation ( [ p_d_integration ] ) , the modified saddle point @xmath142 satisfies the following equation : @xmath143\right|_{\widetilde\lambda=\widetilde{\lambda}_d^ * } = 0 , \label{saddle_point_eq}\ ] ] where @xmath144 first , consider the case of @xmath134 . the conventional saddle point @xmath145 is obtained from @xmath146 , leading to @xmath147 which is always larger than the branch point @xmath148 for any @xmath149 , similar to fig . [ modified_saddle_point](a ) . the modified saddle point @xmath150 can be obtained from equation ( [ saddle_point_eq ] ) , which approaches the conventional one @xmath145 asymptotically as @xmath151 . with no singularity in the vicinity of the saddle point , the conventional saddle - point integration yields the ldf in the long - time limit ( see equation ( [ ldf0 ] ) ) for @xmath134 as @xmath152 as @xmath153 for @xmath154 , the ldf is defined only for @xmath155 . this ldf has no initial - temperature or @xmath95 dependence but is determined only by the heat bath properties ( @xmath70 and @xmath156 ) . thus , we call equation ( [ characteristic ] ) the heat - bath characteristic curve ( hbcc ) . for @xmath128 , the branch point is located at @xmath157 . the conventional saddle point @xmath145 is larger than the branch point only when @xmath158 , and the other way around when @xmath159 . therefore , we expect that the ldf is determined by the conventional saddle point for the former case as @xmath160 , but by the branch point for the latter case as @xmath161 . to summarize , we find the ldf for @xmath128 as @xmath162 & { \rm for}\quad \widetilde{\varepsilon}_d > ( 1 - 2\beta)^{-1 } ~. \end{array } \right . \label{dissipated_ldf2}\end{aligned}\ ] ] note that the ldf is identical to the hbcc for small @xmath149 , but the initial condition ( @xmath95 ) dependence shows up in the rare - event region with large @xmath149 when @xmath128 . as discussed in section [ model ] , we may expect that the initial - condition dependence remains in the long - time limit when highly energetic particles are prepared in the initial ensemble at high temperature ( small @xmath95 ) . nontheless , it is remarkable to see the sharp and finite threshold ( @xmath163 ) regarding to the existence of the everlasting initial memory in the ldf . we will investigate the finite - time correction to the ldf in section [ ftc ] and appendix a. in this subsection , we calculate the ldf of the injected power @xmath99 . the calculation method is similar to that for the dissipated power . the generating function of the injected power is given by @xcite @xmath164^{-1/2 } . \label{pi_i}\end{aligned}\ ] ] the pdf of the dimensionless injected power @xmath165 with @xmath166 , can be obtained from the inverse fourier transform of @xmath167 as in equation ( [ p_d_integration ] ) . the branch points are determined by the equation @xmath168 which has two relevant solutions : one is on the positive real axis , @xmath169 , and the other is on the negative real axis @xmath170 . for @xmath171 , then @xmath172 and @xmath173 $ ] for large @xmath8 . it is easy to check that @xmath174 in the @xmath175 limit and @xmath176 for @xmath177 . so it is natural to introduce a branch cut on the real-@xmath129 axis to the right of the branch point for @xmath177 . the negative branch point behaves in a little complicated way as in the case of the dissipated power . for @xmath136 , @xmath178 in equation ( [ f_i ] ) is positive for @xmath179 . then , @xmath180 is always positive and there is no solution satisfying @xmath181 within that range of @xmath129 . for @xmath119 , @xmath182 with @xmath183 , of which the roots are given by @xmath122 ( @xmath123 ) for large @xmath8 , equivalently @xmath184 $ ] , which is independent of @xmath95 . for @xmath137 , @xmath178 can be negative for @xmath179 , then equation ( [ f_i ] ) has a solution in this range , given by @xmath185 . to summarize , the two relevant branch points are located asymptotically at @xmath186 @xmath187 which are shown in figures [ branch](c ) and [ branch](d ) . the branch cut is located for @xmath177 and @xmath188 on the real axis . now we look for a modified saddle point in between these two branch points . in this region , @xmath115 is always positive and @xmath173 $ ] for large @xmath8 . then the modified saddle - point @xmath189 satisfies the equation @xmath190\right|_{\widetilde\lambda=\widetilde{\lambda}_i^ * } = 0 , \label{saddle_point_eq1}\ ] ] where @xmath191 . for @xmath136 , the conventional saddle point @xmath192 is obtained from @xmath193 , leading to @xmath194 which is always larger than @xmath195 for any @xmath196 , but may increase and go through @xmath174 at @xmath197 as @xmath196 decreases . therefore , we expect that the ldf is given as @xmath198 for @xmath199 , and @xmath200 for @xmath201 . so the ldf is for @xmath136 as @xmath202 , & { \rm for}\quad \widetilde{\varepsilon}_i < ( 1 + 2\sqrt\beta)^{-1 } \\ h_1(\widetilde{\varepsilon}_i ) & { \rm for}\quad \widetilde{\varepsilon}_i > ( 1 + 2\sqrt\beta)^{-1 } ~. \end{array } \right . \label{injected_ldf1}\end{aligned}\ ] ] for @xmath137 , we can do a similar analysis to get @xmath203 for @xmath204 , @xmath198 for @xmath205 , and @xmath206 for @xmath201 . so the ldf is for @xmath137 as @xmath207 , & \left ( \widetilde{\varepsilon}_i > \frac{1}{1 - 2\sqrt{\beta } } \right ) \end{array } \right . \label{injected_ldf2}\end{aligned}\ ] ] the non - analyticity in the negative tail comes from a trivial constraint that the energy loss of a particle is bounded by its initial energy @xcite . however , the positive tail exhibits again the sharp and finite threshold but at the different value of @xmath208 . in general , the threshold value should vary with the quantity interested . and @xmath209 ( @xmath134 ) , respectively . the solid line is the hbcc , equation ( [ characteristic ] ) . panels ( c ) and ( d ) show the ldf of the dissipated power for @xmath210 and @xmath211 ( @xmath128 ) , respectively . the solid line denotes equation ( [ dissipated_ldf2 ] ) . in the above four plots , numerical results are presented as open circles in red color and the calculated ldf s with finite - time corrections , given in equations ( [ dissipated_ldf1_finite ] ) and ( [ dissipated_ldf2_finite ] ) , as thick dotted line in blue color , both obtained at @xmath212 . in panels ( c ) and ( d ) , the dot - dashed line presents the hbcc for comparison . ] we performed numerical simulations for the stochastic differential equation ( [ langevin ] ) using @xmath213 to confirm the ldf s in equations ( [ characteristic ] ) , ( [ dissipated_ldf2 ] ) , ( [ injected_ldf1 ] ) , and ( [ injected_ldf2 ] ) numerically . figures [ finiteldfs_dissipated](a ) and [ finiteldfs_dissipated](b ) show the calculated ldf s of the dissipated power for @xmath214 cases ( black solid line ) , and the numerically obtained @xmath215 ( red open circles ) at @xmath216 . as shown in the figures , all the simulation results are close to the hbcc and independent of @xmath95 , as expected from equation ( [ characteristic ] ) . figures [ finiteldfs_dissipated](c ) and [ finiteldfs_dissipated](d ) show the ldf s of the dissipated power for @xmath217 cases ( black solid line ) and the numerically obtained @xmath215 ( red open circles ) at @xmath216 . contrary to the @xmath134 cases , the numerical results depends on @xmath95 for @xmath218 , as expected from equation ( [ dissipated_ldf2 ] ) . there appear noticeable deviations of numerical data from the calculated ldf curves due to finite - time effect , which are resolved in section 5 . and @xmath209 ( @xmath136 ) , respectively . the solid line plots equation ( [ injected_ldf1 ] ) . panels ( c ) and ( d ) show the ldf of the injected power for @xmath211 and @xmath219 ( @xmath137 ) , respectively . the solid line plots equation ( [ injected_ldf2 ] ) . in the above four plots , numerical results are presented as open circles in red color and the calculated ldf s with finite - time corrections at @xmath212 , given in equations ( [ injected_ldf1_finite ] ) and ( [ injected_ldf2_finite ] ) , as thick dotted line in blue color . in panels ( c ) and ( d ) , we plot the hbcc as the dot - dashed line for comparison . ] figure [ finiteldfs_injected](a ) and [ finiteldfs_injected](b ) show the analytical ( black solid line ) and numerical ( red open circles ) ldf s of the injected power for @xmath136 at @xmath216 . in these figures , the @xmath95-dependent ldf for @xmath220 does not appear simply because the region is outside of the plot range . figure [ finiteldfs_injected](c ) and [ finiteldfs_injected](d ) show the analytical ( black solid line ) and numerical ( red open circles ) ldf s of the injected power for @xmath137 . in these cases , the three regions are clearly seen , as expected from equation ( [ injected_ldf2 ] ) . similar deviations are seen due to finite - time effects . in this section , we calculate the ldf s for the dissipated and injected powers with finite - time corrections up to the first order in @xmath221 from the pdf integrations given by ( see equation ( [ p_d_integration ] ) ) @xmath222 where @xmath223 is @xmath224 ( dissipated ) or @xmath225 ( injected ) , and @xmath226 is its generating function of equation ( [ pi_d ] ) or ( [ pi_i ] ) . factorizing @xmath227 into the exponential term and the leftover as in equation ( [ inverse_fourier ] ) , we write the above pdf integration as @xmath228 where @xmath229 with real @xmath230 . we employ the modified saddle - point integration to evaluate the pdf s and their finite - time corrections in appendix a , where the prefactor function @xmath231 has a power - law singularity like in equation ( [ alpha - singularity ] ) . in the following subsections , we present the results for various cases , especially focussing on the closed form of the ldf s in the scaling regimes . we use the notations @xmath232 ( branch point ) , @xmath233 ( conventional saddle point ) , and @xmath234 ( modified saddle point ) with their relative distances @xmath235 and @xmath236 . the prefactor reads @xmath237^{-1/2 } = \frac{\sqrt{2}\gamma \tau}{4 \pi i}\left[\frac { 2\beta\eta}{\left(\eta+2\beta-1 \right)(\eta+1)}\right]^{1/2}$ ] with @xmath230 . in subsection [ ss - dp ] , we found @xmath238 and @xmath239 , leading to @xmath240 for all @xmath149 . so we can apply the conventional saddle point integration result in equation ( [ app_conventional_finite_int ] ) . using @xmath241 , @xmath242^{-1/2}$ ] , and equation ( [ characteristic ] ) for @xmath243 , then the ldf becomes @xmath244 where @xmath245}~\right].\ ] ] the above equation with @xmath212 is presented in figs . [ finiteldfs_dissipated](a ) and [ finiteldfs_dissipated](b ) for @xmath246 and @xmath247 cases , respectively . they are in excellent accord with the numerical data . in subsection [ ss - dp ] , we found that the branch point location depends on @xmath95 , given by @xmath248 with the same @xmath249 as above . note that @xmath240 for @xmath250 , while @xmath251 otherwise . the prefactor is written as @xmath252 where the analytic function @xmath253 is given by @xmath254^{1/2 } \label{g_lambda1}\ ] ] with @xmath230 . for @xmath255 , we use equation ( [ app_int_c3 ] ) with @xmath256 , @xmath257 , @xmath258 , @xmath259 in equation ( [ dissipated_ldf2 ] ) , and @xmath260 in equation ( [ app_delta_z_3 ] ) to generate the accurate numerical data . for @xmath250 , we can use equation ( [ app_int_c32 ] ) with @xmath261^{1/2}$ ] , @xmath241 , @xmath243 in equation ( [ dissipated_ldf2 ] ) , and @xmath262 . the ldf with a finite - time correction in the scaling regimes can be obtained from equation ( [ app_i_tot1 ] ) as @xmath263 where @xmath264 , \nonumber \\ r_3 ( \widetilde{\varepsilon}_d ) & = & \frac{1}{2 } \ln \left[\frac{\gamma \tau\beta(1 - 2\beta)^2 } { \pi ( 1-\beta)\left(\widetilde{\varepsilon}_d - 1/(1 - 2\beta)\right)}~\right ] . \nonumber\end{aligned}\ ] ] the above equation with @xmath212 is compared with the numerical data in figs . [ finiteldfs_dissipated](c ) and [ finiteldfs_dissipated](d ) for @xmath210 and @xmath265 cases , respectively . as shown in the figures , they show excellent agreement with numerical data . the prefactor reads @xmath266^{-1/2}=\frac{\sqrt{2}\gamma \tau}{4 \pi i ( \eta+1)}\left[\frac { 8\beta\eta}{4\beta-(\eta-1)^2 } \right]^{1/2}$ ] . as seen in subsection [ ss - ip ] , two branch points are located at @xmath268 and @xmath269 , respectively . as @xmath270 , @xmath271 for all @xmath196 , but @xmath272 changes its sign at @xmath273 . thus , we can write the prefactor @xmath274 as @xmath275 where the analytic function @xmath276 is given by @xmath277^{{1}/{2}}.\ ] ] for @xmath278 , we use the slightly modified version of equation ( [ app_int_c3 ] ) by replacing @xmath38 by @xmath279 and @xmath280 by @xmath281 , because the modified saddle point is located to the left of the branching point . in order to generate the accurate numerical data , we take @xmath282 $ ] , @xmath283 , @xmath284 , @xmath285 in equation ( [ injected_ldf1 ] ) , and @xmath286 in equation ( [ app_delta_z_3 ] ) . for @xmath287 , we also use the slightly modified version of equation ( [ app_int_c32 ] ) by replacing @xmath280 and @xmath288 by @xmath281 and @xmath289 , respectively . we take @xmath290^{1/2}$ ] , @xmath291 , @xmath292 in equation ( [ injected_ldf1 ] ) , and @xmath293 . the ldf with a finite - time correction can be obtained from equation ( [ app_i_tot1 ] ) with @xmath294 and @xmath295 as @xmath296 where @xmath297 , \nonumber \\ r_5 ( \widetilde{\varepsilon}_i ) & = & \frac{1}{2}\ln \left[\frac{(\gamma \tau)^{3/2}\sqrt\beta(1 + 2\sqrt\beta)^{7/2 } \gamma^2 ( 5/4)}{\pi^2 ( 1+\sqrt\beta)^2}~\right ] , \nonumber \\ r_6 ( \widetilde{\varepsilon}_i ) & = & \frac{1}{2}\ln\left [ \frac{\gamma \tau ( 4 \beta)}{\pi ( \widetilde{\varepsilon}_i+1)^2 [ ( 4\beta-1){\widetilde{\varepsilon}_i}^2 + 2\widetilde{\varepsilon}_i-1]}~\right ] , \nonumber\end{aligned}\ ] ] the above equation with @xmath212 is compared with numerical data in figs . [ finiteldfs_injected](a ) and [ finiteldfs_injected](b ) for @xmath246 and @xmath298 , respectively . as seen in subsection [ ss - ip ] , two branch points are located at @xmath300 and @xmath269 , respectively . note that @xmath170 is now larger than @xmath301 and the conventional saddle point @xmath192 passes over to the left side of the branch point @xmath170 for @xmath302 . thus , the ldf should be the same as in equation ( [ injected_ldf1_finite ] ) for @xmath303 or equivalently @xmath304 , but one needs to focus on the singularity behavior of the prefactor near @xmath305 to analyze the ldf for @xmath306 . we can rewrite the prefactor @xmath274 as @xmath307 where the analytic function @xmath308 is given by @xmath309^{{1}/{2}}.\ ] ] for @xmath310 , we use equation ( [ app_int_c3 ] ) with @xmath311 $ ] , @xmath312 , @xmath313 , @xmath314 in equation ( [ injected_ldf2 ] ) , and @xmath315 in equation ( [ app_delta_z_3 ] ) . then , the ldf with a finite - time correction becomes @xmath316 where @xmath317 , \nonumber \\ r_8 ( \widetilde{\varepsilon}_i ) & = & \frac{1}{2 } \ln\left [ \frac{\gamma \tau\sqrt{\beta } ( 1 - 2\sqrt{\beta})^2 } { 2\pi ( 1-\sqrt{\beta})^2\left(\widetilde{\varepsilon}_i - 1/(1 - 2\sqrt{\beta})\right)}~ \right ] . \nonumber\end{aligned}\ ] ] in figs . [ finiteldfs_injected](c ) and [ finiteldfs_injected](d ) , the above equation with @xmath212 is compared with numerical data for @xmath211 and @xmath318 , respectively . in this study , we introduced a new method to manipulate a saddle - point integral when a saddle point is in the vicinity of a singular point characterized by a power - law singularity as @xmath319 with @xmath52 . instead of the integration around the conventional saddle point , we choose a modified saddle point , which takes into account the singularity of the prefactor in the long - time limit . the main feature of the modified saddle point is that it does not pass through the singularity as the parameter varies , but asymptotically approaches it for the long - time limit , while the conventional saddle point passes through it , leading to a rather complicated integration . this property results in the so - called ` saddle - point fixation ' which simplifies the analysis to obtain a leading - order term as well as finite - time corrections of the integral . to obtain leading finite - time corrections , one should do the integral near the modified saddle point , which turns out to be non - gaussian in general , especially when the modified saddle point is asymptotically close to the singular point . however , it can be still explicitly written in a specific non - gaussian integral form , which can be evaluated numerically with very high precision . in general , there exist three different scaling regimes depending on the relative position of the conventional saddle point and the singularity , where the non - gaussian integral can be done analytically . to explicitly show the mathematical convenience of our method , we investigated the pdf s of the dissipated and injected heats for a brownian particle whose initial state is not in equilibrium with a thermal bath . the pdf of the power @xmath320 of each heat can be found from the inverse fourier transform of the generating function . the corresponding generating function has a square - root singularity with @xmath321 . by using our method , we obtained the ldf s and its finite - time corrections for each heat . we found that there are sharp transitions of the ldf s depending on the ratio of the initial and heat - bath temperatures . from the numerical simulations , we also confirmed that our method gives correct finite - time corrections . the mathematical finding in this paper is not specific to the square - root singularity , but applies to general @xmath59 . recently , we have investigated the problem where the harmonic potential is pulled by a constant speed and the particle is prepared with the initial distribution at the temperature different from the heat bath temperature . the motion in @xmath224 space dimensions gives rise to the pole of order @xmath322 in the generating function . we have observed the similar asymptotic behavior of the saddle - point and found the pdf , hence the ldf , by using the modified saddle - point integral technique developed here @xcite . the pdf obtained from the experimental and simulation data accumulated for a long time contains strong statistical errors , especially in the tail of rare events . therefore , it is a desirable task to find the ldf with finite - time correction analytically . the ldf with finite - time corrections obtained from the modified saddle - point integral technique shows an excellent agreement with the simulation data . we expect that our method would be useful for many similar problems with a singular generating function , which appears frequently in various non - equilibrium phenomena @xcite . this research was supported by the nrf grant no . 2011- 35b - c00014(j.s.l . ) , 2013r1a1a2011079(c.k . ) , and 2013r1a1a2a10009722(h.p . ) . consider the following integral : @xmath323 where @xmath8 is a large positive number , @xmath324 is an analytic function of @xmath325 and , @xmath59 is a positive number . note that @xmath321 for the equilibration process we studied in the main text . for convenience , we consider that the singularity at @xmath326 lies on the negative real axis of @xmath325 and the branch cut is chosen to its left side on the real axis as shown in fig . [ integration_schematic ] . it would be straightforward to apply our procedure to other general cases . we assume that @xmath327 is an analytic function of @xmath325 at given @xmath7 and takes a real value for real @xmath325 . the conventional saddle point , @xmath328 , is assumed to be located on the real axis , satisfying @xmath329 with @xmath330 . as @xmath328 is a function of @xmath7 , thus its location can be varied and pass through the branch point @xmath331 as @xmath7 varies ( fig . [ integration_schematic ] ) . on the other hand , the modified saddle point @xmath332 satisfies ( equation ( [ modified_s ] ) ) @xmath333 as discussed in sec . [ sec2 ] and fig . [ modified_saddle_point ] , there are always two solutions satisfying the above equation and we choose the saddle point which are located on the right side of @xmath331 . then , it is easy to show that @xmath334 as well as @xmath335 from equation ( [ two_solutions ] ) and fig . [ modified_saddle_point ] . moreover , for large @xmath8 , @xmath332 asymptotically approaches @xmath328 when @xmath336 , while it approaches @xmath331 otherwise . , the branch cut ( wiggled line ) , the conventional saddle point @xmath337 , and the modified saddle point @xmath338 . the two - headed arrow lines denote the range of @xmath328 ( thick line ) and @xmath332 ( dashed line ) , respectively , as @xmath7 varies . ] for convenience , we define @xmath339 as an alternative variable for @xmath7 and also @xmath340 and @xmath341 as asymptotically vanishing positive quantities for @xmath342 and @xmath343 , respectively . we investigate how these small quantities vanish as a function of @xmath8 , in order to calculate the finite - time correction of the ldf , @xmath344 , up to the order of @xmath221 . first , consider the case @xmath342 . expansion of equation ( [ appendix_saddle_eq ] ) near @xmath328 with small @xmath345 yields @xmath346 where the @xmath7 dependence is dropped for simplicity and afterwards . the positive solution of the above quadratic equation is @xmath347 which leads to two different scaling behaviors as @xmath348 second , consider the case @xmath349 . expansion of equation ( [ appendix_saddle_eq ] ) near @xmath350 with small @xmath280 yields @xmath351 its positive solution is @xmath352 where @xmath353@xmath354 is approximately equal to @xmath288 for small @xmath355 . again , two scaling regimes are found as @xmath356 summarizing equations ( [ app_delta_z_2 ] ) and ( [ app_delta_z_3 ] ) , small quantities @xmath345 and @xmath280 are categorized into three regimes as follows : @xmath357 we calculate equation ( [ app_int ] ) by the steepest descent method along a contour @xmath60 passing through the modified saddle point @xmath332 ( fig . [ contours_change ] ) . by expanding @xmath359 near @xmath332 up to the second order of @xmath360 , we get @xmath361 where @xmath362 for the modified saddle - point . the denominator of the integrand was not expanded on purpose because its expansion series does not converge when @xmath280 is small enough as in equation ( [ app_delta_z_tot ] ) for @xmath363 . for equation ( [ app_int_expansion_modified ] ) . ( c ) contour paths @xmath61 and @xmath62 . ] it is convenient to use a variable @xmath364 . then , the above integral is rewritten as @xmath365 \int_{c_2 } du~ { u^{-\alpha}}\nonumber \\ & \times & \exp \left [ \tau h'(z^ * ) \delta z_b ( u-1 ) + \frac{\tau}{2 } h''(z^*)(\delta z_b)^2 ( u-1)^2 \right]~ , \label{arrange1}\end{aligned}\ ] ] where the contour @xmath61 is given in figure [ contours_change](c ) of the complex @xmath366 plane . this integral can be more simplified by replacing the contour @xmath61 by @xmath62 in figure [ contours_change ] , which comprises of the vertical sector @xmath367 with @xmath368 for @xmath369 and the small half - circular sector @xmath370 with fixed @xmath371 and @xmath372 . the half - circular sector @xmath370 is introduced to avoid the singularity of the integrand at @xmath373 . later , we will take the @xmath374 limit , for simplicity . now , consider the case @xmath349 first , where @xmath280 is small . by expanding @xmath375 and @xmath376 near @xmath331 up to the order of @xmath377 , the integral is simplified as @xmath378 \nonumber \\ & \times & \int_{c_3 } du~ { u^{-\alpha}}\exp \left [ \tau h'(z_b ) \delta z_b u + \frac{\tau}{2 } h''(z_b)(\delta z_b)^2 u^2 \right]~. \label{app_int_c2}\end{aligned}\ ] ] this is a non - gaussian integral , which may not be expressed in a closed form in general . the contribution to the integral from the vertical sector @xmath367 with @xmath368 for @xmath369 can be written as @xmath379 \nonumber \\ & \times & \int_{\rho}^{\infty } dy~y^{-\alpha } \cos\left[\theta_\alpha+\theta_b(y)\right ] e^{-\frac{\tau}{2 } h''(z_b ) ( \delta z_b)^2 y^2 } , \label{app_int_c3}\end{aligned}\ ] ] where @xmath380 and @xmath381 . the second contribution to the integral from the half - circular sector @xmath370 with small @xmath371 and @xmath372 is given as @xmath382 \nonumber \\ & \times & \rho^{1-\alpha } \left[\frac{\cos\theta_\alpha}{1-\alpha } - \theta_b(\rho ) \frac{\sin\theta_\alpha}{2-\alpha } + \cdots\right ] , \label{app_int_c3_2}\end{aligned}\ ] ] up to the order of @xmath383 . the integral @xmath384 should be independent of @xmath371 . note that @xmath385 vanishes in the @xmath374 limit for @xmath386 , so @xmath387 . for general @xmath388 , both @xmath389 and @xmath385 diverge , but the divergent terms cancel out exactly , leading to a convergent value for the integral @xmath358 . this cancelation can be seen easily by integrating equation ( [ app_int_c3 ] ) by parts . for @xmath390 , the sum of @xmath391 and @xmath392 yields that @xmath393 \nonumber \\ & \times & \int_{0}^{\infty } dy~\frac{y^{1-\alpha}}{\alpha-1 } \frac{d}{dy } \left[\cos\left[\theta_\alpha+\theta_b(y)\right ] e^{-\frac{\tau}{2 } h''(z_b ) ( \delta z_b)^2 y^2}\right ] . \label{app_int_c3_3}\end{aligned}\ ] ] at @xmath394 , @xmath395 in the integrand should be replaced by @xmath396 . for a larger value of @xmath59 , we only need to perform the appropriate number of successive integrations of equation ( [ app_int_c3 ] ) by parts . now , consider the case @xmath397 where @xmath345 is small . by expanding @xmath375 and @xmath376 near @xmath398 up to the order of @xmath399 , the integral becomes @xmath400 \int_{c_3 } du~ u^{-\alpha}\nonumber \\ & \times & \exp \left [ -\tau h''(z_0^ * ) \delta z_b ( \delta z ) u + \frac{\tau}{2 } h''(z_0^*)(\delta z_b)^2 u^2 \right ] . \label{app_int_case2}\end{aligned}\ ] ] then , the two contributions to the integral are written as @xmath401 \nonumber \\ & \times & \int_{\rho}^{\infty } dy~y^{-\alpha } \cos\left[\theta_\alpha + \theta_0(y)\right ] e^{-\frac{\tau}{2 } h''(z_0^ * ) ( \delta z_b)^2 y^2 } , \label{app_int_c32}\\ i^{(2)}(\varepsilon ) & = & 2i g(z_0^ * ) ( \delta z_b)^{1-\alpha } \exp\left[\tau h(z_0^*)+\frac{\tau}{2 } h''(z_0^*)(\delta z)^2\right]\nonumber \\ & \times & \rho^{1-\alpha } \left[\frac{\cos\theta_\alpha}{1-\alpha } - \theta_0(\rho ) \frac{\sin\theta_\alpha}{2-\alpha } + \cdots\right ] , \label{app_int_c33}\end{aligned}\ ] ] where @xmath402 . again , for @xmath386 , @xmath387 , while a similar calculation as above can be done for @xmath403 . note that equations ( [ app_int_c3])-([app_int_c3_3 ] ) and ( [ app_int_case2])-([app_int_c33 ] ) become identical for small @xmath355 , which implies that the integral @xmath358 also has only three different scaling regimes , similar to the small quantities @xmath345 and @xmath280 in equation ( [ app_delta_z_tot ] ) . our results together with equations ( [ app_delta_z_0 ] ) and ( [ app_delta_z ] ) for @xmath345 and @xmath280 provide us to calculate @xmath358 or the ldf s numerically with a very high precision even in the rare - event regime ( large @xmath404 ) . as an example , we numerically calculated the integrals of the dissipated power for @xmath128 studied in the main text , where @xmath321 , @xmath405 ( @xmath406 ) , @xmath407 , and @xmath324 is given by equation ( [ g_lambda1 ] ) . as shown in fig . [ app_full_range_ldf_curves ] , the numerical integrations show the perfect match with the numerical simulation data at @xmath408 , obtained from direct integration of the langevin equation ( [ langevin ] ) . with finite - time corrections at @xmath408 with ( a ) @xmath409 and ( b ) @xmath410 . red - open circles , blue - dashed lines , and green solid lines denote numerical simulation data , piece - wise analytic curves ( closed form in [ closed ] ) , and full - range analytic curves ( integral form ) , respectively . ] we assume that @xmath411 in this subsection , for simplicity . generalization to general @xmath59 is straightforward . first , consider the case @xmath412 . from equation ( [ app_delta_z_tot ] ) , we find @xmath413 , so @xmath414 . in equation ( [ app_int_c32 ] ) , the integral can be done easily by noting that @xmath415\approx \frac{1}{2 } ( \pm i)^\alpha\sqrt{\pi/\tau}$ ] for large @xmath8 , which leads to @xmath416^{1/2 } \frac{g(z_0^ * ) } { ( \delta z)^\alpha } e^{\tau h(z_0^*)}.\label{app_conventional_finite_int}\ ] ] as expected , we recover the result by using the conventional saddle point method in equation ( [ conventionalsp ] ) , which should be correct for any positive @xmath59 . second , consider the case @xmath417 , where @xmath418 from equation ( [ app_delta_z_tot ] ) , so @xmath288 can be ignored . in this case , @xmath419 becomes negligible for large @xmath8 . in addition , @xmath420 from equation ( [ app_delta_z_tot ] ) . then , equations ( [ app_int_c3 ] ) and ( [ app_int_c32 ] ) are simplified as the following identical formula in the @xmath374 limit ; @xmath421^{\frac{1-\alpha}{2 } } } ~e^{\tau h(z_b)}. \label{app_near_zb_result1}\end{aligned}\ ] ] for the last equality , @xmath422 is used . finally , consider the case @xmath423 , where @xmath424 from equation ( [ app_delta_z_tot ] ) . in this case , @xmath425 and @xmath426 is negligible . then , equation ( [ app_int_c3 ] ) in the @xmath374 limit is simplified as @xmath427^{1-\alpha}}~ e^{\tau h(z_b ) } \int_0^\infty dy ~y^{-\alpha } \cos \left(\theta_\alpha -\alpha y\right ) \nonumber \\ & = & i \frac{2 \gamma(1-\alpha ) \sin ( \alpha \pi ) g(z_b)}{\left[\tau h'(z_b)\right]^{1-\alpha } } e^{\tau h(z_b ) } \label{app_near_zb_result2}\end{aligned}\ ] ] for the last equality , @xmath428 is used . summarizing equations . ( [ app_conventional_finite_int ] ) , ( [ app_near_zb_result1 ] ) , and ( [ app_near_zb_result2 ] ) for @xmath321 , @xmath358 becomes @xmath429^{1/4 } } ~g(z_b ) ~e^{\tau h(z_b ) } , & |\delta z| \sqrt{\tau } \ll 1\\ i \frac{2 \sqrt{\pi } } { \sqrt{\tau h'(z_b ) } } ~g(z_b ) ~e^{\tau h(z_b ) } , & \delta z \sqrt{\tau } \ll -1 \\ \end{array } \right . \label{app_i_tot1}\ ] ] where @xmath430 . the above equation corresponds to the result in reference @xcite . when the branch cut is located to its right side of the branch point @xmath331 with a singularity like @xmath431 in equation ( [ app_int ] ) , we get the same equation for @xmath358 except for changing @xmath38 to @xmath279 and @xmath288 to @xmath432 . 10 crooks g e , 1999 _ phys . e _ * 60 * 2721 evans d j , cohen e g d and morriss g p , 1993 _ phys . lett . _ * 71 * 2401 gallavotti g and cohen e g d , 1995 _ phys . lett . _ * 74 * 2694 jarzynski c , 1997 _ phys . rev . lett . _ * 78 * 2690 ; 2004 _ phys . lett . _ * 92 * 230602 kurchan j , 1998 _ j. phys . a : math . gen . _ * 31 * 3719 lebowitz j l and spohn h , 1999 _ j. stat . phys . _ * 95 * 333 dewar r c , 2005 _ j. phys . a : math . gen . _ * 38 * l371 hatano t and sasa s , 2001 _ phys . lett . _ * 86 * , 3463 noh j d and park j -m , 2012 _ phys . lett . _ * 108 * , 240603 evans d j and searles d j , 2002 _ adv . phys . _ * 51 * 1529 jarzynski c , 1997 _ phys . rev . lett . _ * 78 * 2690 wang g m , sevick e m , mittag e , searles d j and evans d j , 1997 _ phys . * 89 * 050601 andrieux d , gaspard p , monnai t and tasaki s , 2009 _ new j. phys . _ * 11 * 043014 talkner p and hnggi p , 2007 _ j. phys . a : math . _ * 40 * f569 kwon c , noh j d and park h , 2011 _ phys . e _ * 83 * 061145 van zon r and cohen e g d , 2003 _ phys . lett . _ * 91 * 110601 ; 2004 _ phys . rev . * 69 * 056121 derrida b , 2007 _ j. stat . mech . : theory exp . _ p07023 seifert u , 2005 _ phys . * 95 * 040602 esposito m , 2010 _ phys . * 104 * 090601 spinney r e and ford i j , 2012 _ phys lett . _ * 108 * 170603 garca - garca r , lecomte v , kolton a b and domnguez , 2012 _ j. stat . mech . : theory exp . _ p02009 tom t and de oliveira m j , 2012 _ phys . _ * 108 * 020601 farago j , 2002 _ j. stat . phys . _ * 107 * 781 ; 2004 _ physica a _ * 331 * , 69 puglish a , rondoni l , and vulpiani a , 2006 _ j. stat . mech . : theory exp . _ p08010 harris r j , rkos a , and schtz g m , 2006 _ europhs . lett . _ * 75 * 227 visco p , 2006 _ j. stat . mech . : theory exp . _ p06006 van zon r , ciliberto s and cohen e g d , 2004 _ phys lett . _ * 92 * 130601 ; garnier n and ciliberto s , 2005 _ phys . e _ * 71 * 060101(r ) ; ciliberto s , joubaud s and petrosyan a , 2010 _ j. stat . mech : theory exp . _ p12003 sabhapandit s , 2011 _ epl _ * 96 * 20005 ; 2012 _ phys . rev . e _ * 85 * 021108 the generating function or equation ( [ fourier_transform ] ) is properly defined only in a region of convergence on the complex @xmath31 plane , so all singular points of the generating function must be located at the boundary or the outside of the convergence region . thus , the phrase that we are using in this paper such as ` the generating function has a singularity ' makes sense only when the analytic continuation of the generating function is taken into account . engel a , 2009 _ phys . * 8**0 021120 nickelsen d and engel a , 2011 _ eur . phys . j. b. _ * * 8**2 207 noh j d , kwon c and park h , 2013 preprint _ arxiv_:1301.1806 kwon c , noh j d and park h , 2013 preprint _ arxiv _ : 1303.6742 touchette h , harris r j and tailleur j , 2010 _ phys . e _ * * 8**1 030101 lee j s , kwon c and park h , 2013 _ phys . * 87 * 020104(r ) wiegel f w , _ introduction to path - integral methods in physics and polymer science _ ( world scientific , 1986 ) kim k , kwon c and park h , in preparation

this paper presents an exploration of two fundamental scaling relations of spiral galaxies , the luminosity - rotation speed ( or tully - fisher ; tf ) relation , and the size - luminosity ( sl ) relation , and the dependences of their scatter , at red and infrared bands . we verify that the observed virial relations of disk galaxies are given by and using distance - redshift surveys of high surface brightness ( hsb ) and low surface brightness ( lsb ) non - interacting galaxies . these results imply that the galaxy surface brightness . the collected surveys provide accurate-band luminosities , disk scale lengths , circular velocities , and , in some cases , color . various issues regarding the scatter of scaling relations in blue to near - infrared bands are being re - examined with accurate kron luminosities , effective radii , and colors from the 2mass database . we derive the first extensive-band tf and sl relations . at a given infrared luminosity , the tf velocity residuals are correlated with infrared color , which in turn is determined by the variations in galaxy formation ages and dark halo concentrations ; these residuals are fully independent of surface brightness and other tested galaxy observables . we verify that tf relations ( tfrs ) for various morphological types have different zero points , but a common slope , such that early - type ( redder ) disks rotate faster than later - type ( bluer ) systems of the same luminosity . the morphological type dependence of the tfr is a direct consequence of the more fundamental scatter dependence on color , which itself is related to the star formation history of a galaxy . the scatter of the sl relation is mostly dominated by surface brightness , but color also plays a small role . the observed systematic variations of disk size with color , for a given luminosity , are weaker than expected , perhaps as a result of uncertainty in disk scale length measurements . the tf and sl residuals for hsb and lsb galaxies are weakly correlated with , in agreement with the earlier claim by courteau & rix ( 1999 , apj , 513 , 561 ) , and . the former result suggests that spiral disks of all color , brightness and barredness may be , on average , dominated by dark matter even in their inner parts ( the so - called `` sub - maximal disk '' solution at disk scale lengths ) . the observed scaling relations of disk galaxies are likely the result of the simplest scenario of galaxy formation in the standard cosmological picture , but their color dependence remains a challenge to hierarchical structure formation models . these relations and their derivatives are analysed in terms of , and serve as stringent constraints for , galaxy formation models in our companion paper .

astro-ph0310440

understanding the origin and nature of galaxy scaling relations is a fundamental quest of any successful theory of galaxy formation . the success of a particular theory will be judged by its ability to predict the slope , scatter , and zero - point of any robust galaxy scaling relation at any particular wavelength . some observed scaling relations in spiral galaxies , based on their size , luminosity , and rotation speed , can be reproduced _ individually _ to fairly good accuracy by invoking galaxy formation models that include virial equilibrium after dissipational collapse of spherical cold dark matter ( cdm ) halos and angular momentum conservation ( e.g. mo , mao , & white 1998 , hereafter mmw98 ; van den bosch 1998 , 2000 , hereafter collectively as vdb00 ; navarro & steinmetz 2000 , hereafter ns00 ; firmani & avila - reese 2000 ; hereafter far00 ) . one of the most firmly established empirical scaling relation of disk galaxies is the tully - fisher relation ( tfr ; tully & fisher 1977 ) ; a tight correlation between the total luminosity and the rotation speed of a disk galaxy . however , to date , no single cdm - based model of galaxy formation can _ simultaneously _ reproduce the slope , zero - point , scatter , and color trends of the tfr , match the shape and normalization of the luminosity function , and explain the sizes , colors , and metallicity of disk galaxies ( see , e.g , . vdb00 ; bell et al . 2003 ) . in addition , simultaneously accounting for the mass and angular momentum distribution of spiral galaxies in a gas dynamical context remains a major challenge for hierarchical formation models ( navarro & white 1994 ; bullock et al . 2001b ; van den bosch et al . 2002b ) . a complete theory of galaxy scaling relations awaits a fuller understanding of structure forming mechanisms and evolutionary processes ( e.g. star formation , merging , feedback , and cooling prescriptions ) in galaxies ( see somerville & primack 1999 for a comparison of `` recipes '' among competing theories . ) likewise , the fine - tuning of these galaxy formation and evolutionary models demands a careful examination of empirical scaling relations of galaxies . in order to set up a framework for the study of galaxy scaling relations , we examine the correlations of parameters related by the virial theorem , @xmath9 , for bright galaxies ( i.e. where feedback effects are minimal ) . we consider three fundamental observables for each disk galaxy : the total luminosity @xmath10 , the stellar scale length @xmath11 of the exponential disk , and the observed circular velocity @xmath12 . the stellar mass , @xmath13 , can be estimated from the luminosity by assuming a stellar mass - to - light ratio , @xmath14 . the size - luminosity ( sl ) relation of galaxy disks is also expressed as @xmath15 , where @xmath16 is the surface brightness . key to mapping fundamental dynamical trends in spiral galaxies , the measurement of tf and sl relations and detection of their correlated residuals require velocity amplitudes measured at a suitably chosen radius representative of the flat part of resolved rotation curves , red / infrared magnitudes to minimize extinction and population effects , accurate disk scale lengths , and , ideally , color terms with a broad baseline ( e.g. @xmath17 ) to test for @xmath18 variations in the stellar population and extinction effects . our goal in this paper is thus to assemble such a data base . a study of scaling relations in irregular and spiral galaxies by salpeter & hoffman ( 1996 ; hereafter sh96 ) yielded the correlations @xmath19 , where @xmath20 , @xmath21 , and @xmath22 , are the and dynamical masses , and a characteristic radius , respectively . the blue luminosities , as used in that study , are notoriously sensitive to dust extinction and stellar population effects and dynamical effects can not be simply isolated . a new study of scaling relations in the ( near-)infrared would provide more robust dynamical constraints to galaxy formation models . while the tfr has been examined at nearly all optical - ir wavelengths ( e.g. strauss & willick 1995 ; verheijen 2001 , hereafter v01 ) , comparatively few multi - wavelength analyses of the sl relation ( slr ) of spiral galaxies have been done so far ( sh96 ; mmw98 ; shen et al . this is partly because the accurate disk scale lengths needed to calibrate the slr , at any wavelength , have only recently become available for large databases ( e.g. courteau 1996 ; dale et al . 1999 ; macarthur et al . 2003 , hereafter mch03 ) . the recent availability of light profile decompositions for very large galaxy databases ( e.g. sloan digital sky survey and two micron all - sky survey ; ) heralds a new era for the study of galaxy scaling relations ( e.g. shen et al . we will return to considerations about the slr in . the remainder of this section focuses on the tfr and the dependences of its scatter on galaxy observables . the empirical tfr is expressed as l v^a_obs [ eq : irtf ] with the near - ir log - slope @xmath23 ( willick et al . 1997 ; giovanelli et al . 1997 , hereafter g97 ; courteau et al . 2000 ; v01 ) . reported values of the log - slope @xmath24 range from 2.8 in the blue to 4.0 in the infrared ( willick et al . 1997 ; tully & pierce 2000 ; v01 ) for both high and low surface brightness galaxies ( zwaan et al . 1995 ; v01 ) . log - slopes steeper than @xmath25 in the infrared typically result from small samples and excessive pruning on the basis of idealized morphology or kinematics , a narrow range of inclinations , redshift cutoffs , etc . ( bernstein et al . 1994 ; v01 ; kannappan , fabricant , & franx 2002 , hereafter kff02 ) . the slope , scatter , and zero - point of blue tfrs are predominantly dominated by stellar population and dust extinction effects ( e.g. aaronson & mould 1983 ; c97 ; g97 ; willick et al . 1997 ; tully & pierce 2000 ) and on the techniques used to recover the major observables and fitting for fundamental relations ( e.g. strauss & willick 1995 ; c97 ; v01 ; bell & de jong 2001 ; kff02 ) . because we are mainly interested in masses , rather than luminosities , we do not concern ourselves with tfrs and other scaling relations measured at blue wavelengths . we show in appendix a that the fundamental form of the tfr , based on dynamical principles only , is given by @xmath26 , that is @xmath27 . the modern interpretation of the tfr is that of a correlation between the total baryonic mass of a galaxy , inferred via its infrared luminosity and a stellar mass - to - light ratio and total gas mass ( + he ) , and its total mass inferred from the asymptotic circular velocity of the galaxy disk ( mcgaugh et al . 2000 ; bell & de jong 2001 ; v01 ) . the `` baryonic '' tfr is expressed as _ bary v^a_bary_obs [ eq : bartf ] since disk gas mass fractions typically increase with decreasing luminosity ( i.e. , mcgaugh & de blok 1997 ) , one typically has that @xmath28 . while the log - slope of the tfr can be reproduced fairly well by most cdm - based structure formation models ( e.g. mmw98 ; vdb00 ; ns00 ) , the predicted scatter can be large compared to the inferred `` cosmic '' scatter of @xmath29 mag in red / infrared bands ( willick et al . 1996 ; v01 ) and interpretations about its dependence differ ( see below ) . besides the basic understanding of the slopes of galaxy scaling relations , the dependence of their scatter has also been addressed by many , especially for the tfr ( e.g. aaronson & mould 1983 ; giraud 1986 ; rhee 1996 ; willick et al . 1997 ; kff02 ) , and can be used to set stringent constraints on structure formation models ( courteau & rix 1999 , hereafter cr99 ; heavens & jimenez 1999 ; far00 ; ns00 ; v01 ; buchalter , jimenez , & kamionkowski 2001 ; shen , mo , & shu 2002 ) . while various trends in the scatter of the blue tfr have been reported in the past , no correlations of the infrared tf residuals with inclination , size , concentration , gas fraction , or far infrared luminosity have thus far been reported ( aaronson & mould 1983 ; v01 ) . the dependence of near - ir tfr scatter on color and surface brightness is however still a matter of contention that we discuss in . the slr scatter is also addressed in . the study of scaling relations in galaxies has benefited from the two - pronged application of the tfr for the purposes of : ( i ) estimating relative distances to measure deviations from the mean hubble flow ( see , e.g. , strauss & willick 1995 and the reviews in the `` cosmic flows 1999 '' proceedings by courteau , strauss , & willick 1999 ) ; and ( ii ) testing galaxy formation and evolution models ( dalcanton , spergel & summers 1997 ; mmw98 ; vdb00 ) . the philosophy of sample selection and calibration differs in both cases . for cosmic flow analyses , the calibration and science samples must be pruned mostly on the basis of morphology and visual appearance in order to minimize systematic errors and thus ensure the smallest possible magnitude ( distance ) error . since tf scatter depends strongly on the slope of the tf , it is found that the combination of steepness , magnitude errors , extinction correction , and sky stability favors red ( @xmath30 ) bands for smallest distance errors and cosmic flows applications ( courteau 1997 , hereafter c97 ; v01 ) . the accuracy of bulk flow solutions also depends on the size of the sample . in order to collect large enough samples , tf calibrations for flow studies rely mostly on line widths or rotation curves that can be collected relatively quickly on modest aperture telescopes . these rotation measures typically sample the disk rotation out to 2 to 3 disk scale lengths ( c97 ) . by contrast , use of the tfr as a test bed for galaxy formation models requires the widest range of morphological types , to sample all structural properties , and that extinction and stellar population effects be minimized to isolate genuine dynamical correlations . the nearly dust - insensitive @xmath31-band is thus the one of choice for such applications ( v01)-band rather than at @xmath31 ; the @xmath32-band tfr is hardest to control due to airglow fluctuations ( jarrett et al . 2003 ) . ] . rotation velocities are preferably extracted from fully resolved rotation curves obtained using aperture synthesis maps that sample the disk rotation out to 4 to 5 disk scale lengths . ideally , the study of scaling relations should rely on homogeneous samples assembled with the very purpose of testing for broad structural and dynamical differences ; at the moment , we must contend with the more finely pruned heterogeneous samples of late - type spirals that have been collected during the last decade mostly for flow studies . these data , which include near - infrared luminosities , and sometimes colors , for large samples of galaxies , still enable us to characterize the dependence of scaling relations and examine various constraints of structure formation models , in ways hitherto unsettled . in courteau et al . ( 2003 ) , we showed that barred and unbarred galaxies have similar physical properties and that they share the same tfr . as a natural extension of this study and cr99 , in we use the extensive all - sky distance - redshift catalogs presented in to characterize the scatter of the tf and sl relations in terms of galaxy observables for a broad suite of galaxy types . in we bolster the notions of surface brightness independence and color dependence of the tfr , as well as that of surface brightness dependence for the slr , for disk galaxies . this prompts a renewed examination of the correlated scatter study of cr99 which favored high dark - to - luminous mass fractions in galaxy disk s interiors . in we develop a simple analytical interpretation for the origin of the galaxy virial relations , such as the tf and sl relations , in a cosmological setting . the analysis of the scatter in the scaling relations , and a discussion about the implications of our results in terms of existing galaxy formation models , are presented in our companion paper ( s. courteau et al . 2004a , in preparation ; hereafter paper ii ) . [ sec : data ] we consider four samples for which accurate galaxy observables , including crucial rotational velocities , are available . these include ( a ) the large @xmath3-band survey of galaxy distances for bright field spirals in the southern sky by mathewson , ford , & buchhorn ( 1992 ; hereafter `` * mat * '' ) ; the all - sky @xmath3-band tf surveys by ( b ) dale et al . ( 1999 ; hereafter `` * scii * '' ) and ( c ) courteau et al . ( 2000 ; 2004b , in preparation ; hereafter `` * shellflow * '' ) ; and ( d ) the multi - band @xmath33 tf survey of ursa major cluster galaxies by tully et al . ( 1996 ) and v01 ( hereafter `` * uma * '' ) . the first three tf surveys were originally designed to map the convergence of the velocity field on @xmath34 mpc scales , with the scii and shellflow studies paying special attention to tf calibration errors between different observatories . by design , these surveys favor late - type galaxies with inclinations on the sky greater than 32 ; most have @xmath35 . properties for each sample are given in table 1 . these include in col . ( 2 ) , the number of galaxies from the original database that have a full complement of useful observables ; col . ( 3 ) , the nature of the sampled galaxies ( cluster or field ; the predominantly `` field '' surveys of mat and shellflow include a small fraction of cluster galaxies ) ; col . ( 4 ) , the digital photometric coverage . @xmath36-band magnitudes for all the mat galaxies were extracted from the rc3 ( de vaucouleurs et al . we have _ sloan digital sky survey _ ( york et al . 2000 , hereafter sdss ) @xmath37 colors for 39 scii galaxies ( see appendix b ) and @xmath4 kron magnitudes and colors from the _ two micron all - sky survey _ ( skrutskie et al . 1997 ; hereafter 2mass ) for the brightest @xmath38 scii galaxies ; col . ( 5 ) the magnitude or diameter limits of the original catalog ; col . ( 6 ) , the redshift limits of the survey in kms^-1km@xmath39s@xmath40 . for uma , we list the adopted cluster distance in mpc ; and in col . ( 7 ) , the rotation measure , either extracted from line mapping or spatially resolved rotation curves . while mat gives both and rotation measures we use only the sub - sample of galaxies with available , which is largest . our results for the mat sample do not depend upon this particular choice of rotation measures . for scii , rotation velocities were obtained using either long - slit spectra or line profiles . these two rotation measures are comparable as long as the rotation curve is flat beyond 2 disk scale lengths and the surface density drops rapidly in the outer disk ( c97 ) . rotation velocities for the shellflow sample were measured at @xmath41 from resolved rotation curves . for the uma database , we use the `` '' sample of 44 galaxies with line widths ( v01 ) . disk scale length measurements were computed in somewhat different ways for each sample . for the mat galaxies , we have used the two - dimensional bulge - disk ( b / d ) decompositions of mat @xmath3-band images by byun & freeman ( 1995 ) . these decompositions assumed a de vaucouleurs bulge profile which is not ideal for late - type galaxies ( mch03 ) and therefore the disk central surface brightnesses are likely biased low . the disk scale lengths are less affected , in a relative sense due to the broader baseline of the disk modeled as an exponential profile , and deemed adequate for the purpose of our study ( see ) . scii disk scale lengths were obtained by fitting a straight line to the exponential part of the @xmath3-band surface brightness profile from @xmath42 @xmath3- to @xmath43 @xmath3- ( the so - called `` marking the disk '' technique ) and corrected for projection effects according to @xmath44 $ ] , where @xmath24 and @xmath45 are the semi - major and semi - minor axes of the disk ( dale et al . 1999 ; g97 ) . the correction for projection is small and somewhat uncertain but the choice of observed or deprojected scale lengths for the scii galaxies does not affect our final results . shellflow scale lengths were extracted from one - dimensional b / d decompositions of azimuthally - averaged @xmath3-band surface brightness profiles ( mch03 ; s. courteau et al . 2004 ; in preparation ) . these fits account for a srsic bulge and an exponential disk . disk scale lengths for the uma sample were measured from a `` marking the disk '' technique but the fit baseline is unspecified and erratic fits are reported in tully et al . the latter could be due to inclusion of bulge light in the disk fit ( scale lengths biased low ) for the brighter galaxies and sky domination ( scale lengths biased high ) for the fainter galaxies . the latter signature is detected when we compare the scale length measurements for the four samples in . physical parameters are computed using h@xmath46=72 km sec@xmath47 mpc@xmath48 . this matches the revised distance to the uma cluster from 15.5 mpc to 18.6 mpc , based on new cepheid distances from the hst key project ( tully & pierce 2000 ) . the exact choice of distance scale does not affect our conclusions so long as it is consistent for all samples in order to put absolute scales on equal footing ( luminosities in @xmath49 and scale lengths in kpc ) . hubble types for all but the scii galaxies were obtained from the heterogeneous nasa extragalactic data ( ned ) base . most scii galaxies were not classified in ned and their morphological types were determined in a homogeneous fashion via a combination of eye - ball examination , b / d ratio and/or concentration index ( dale et al . interacting and disturbed galaxies were rejected in all samples . together , the mat , scii , shellflow , and uma samples combine for a total of 1,750 separate entries ; some repeat measurements within a given sample exist but all observations are independent . [ sec : virial ] we consider the projected distribution in each of the planes defined by a pair of the three log virial variables , @xmath50 , @xmath51 , and @xmath52 ; these are shown in figs . [ fig : scii_3rel][fig : uma_3rel ] for the scii , mat , shellflow , and uma galaxies . the axis limits are the same in all figures for trend matching between each sample . different symbols identify the full range of spiral hubble types , as a proxy for accurate optical colors which are not available for the full mat and scii samples . the solid lines correspond to one dimensional linear regressions to ( a ) @xmath53 on @xmath54 , ( b ) @xmath55 on @xmath54 , and ( c ) @xmath55 on @xmath53 , given the measurement errors in the first variable in each case when available ( or otherwise using unweighted fits ) . the fits for the scii sample are used as fiducials against the other , less complete , samples and are shown as a dotted line on the virial projections for each sample . for each relation , we fit the functional form @xmath56 , where @xmath57 and @xmath58 stand for any combination of @xmath53 , @xmath55 , and @xmath52 ; the upper bar denotes the median value of each variable . to achieve a robust fit to the data @xmath59 $ ] , we varied @xmath60 to minimize the data@xmath61model deviation ( rather than the squared difference ) . we repeat the fits for all the log quantities keeping either the total luminosity , stellar scale length , or maximum velocity fixed . this fitting technique follows the treament in cr99 . we have also computed standard least - squares fits which yield similar results . log - slopes and the pearson linear correlation coefficients @xmath62 for each parameter combination are reported in table 2 ( we do not concern ourselves with calibration zero - point differences in this paper ) . the regressions were performed using the full range of luminosities ( when the regression was on @xmath63 ) and velocities ( when the regression was on @xmath51 ) . it has been proposed that the scaling relations may be different for the fainter , lsb galaxies ( e.g. , kauffmann et al . 2003 ; shen et al . 2003 ) , where supernova feedback effects are likely to be important ( e.g. , dekel & silk 1986 ; dekel & woo 2003 ) . such a departure is , however , not apparent in the samples ( scii and mat ) that most closely probe the lsb regime ( see figs . [ fig : scii_3rel ] & [ fig : mat_3rel ] ) . the agreement in table 2 between the log - slopes for different samples is relatively good , though differences exist . all samples are well matched by a tfr , @xmath64 , with log - slope @xmath65 . our tf fits are also a close match to those published by the original authors ( as @xmath66 ) , even though the minimization techniques can be quite different . the pearson correlation coefficient of the distribution ( not considering the errors ) averages @xmath67 for the tfr and the conditional distribution of @xmath12 at a given @xmath10 is roughly log - normal . the correlations for the sl and size - velocity ( sv ) relations are weaker , averaging @xmath68 . the size distributions are less robust than the tfr owing to a number of factors such as , ( a ) small range of galaxy sizes in some samples ( shellflow , uma ) , ( b ) lack of a uniform and universal definition of scale length ( mch03 ) , and ( c ) intrinsic scatter due to the natural dispersion of the spin parameter , @xmath69 ( ) . whether scale lengths are measured from b / d decompositions or `` marked '' over a specified range of surface brightnesses ( while omitting the bulge region or not ) can yield scale differences greater than 20% ( mch03 ) . the overlap between the four samples is too small ( in some cases non - existant ) to calibrate such systematic errors . thus it is difficult to determine how much of the lower correlation coefficients in the sl and sv relations is due to genuine dynamical processes and@xmath70or inadequate scale lengths . in spite of potential pathologies with the mat scale lengths ( see ) , the scalings for this sample and scii are comfortably close . for the sake of uniformity , we will base our final analysis of the virial relations of disk galaxies on the scii sample alone , being the most extensive and robust sample considered here . furthermore , unlike our results for some of the other , less complete and thus noisier , data sets our fits for scii are self - consistent , within the errors . if @xmath64 and @xmath71 and @xmath72 , then self - consistency requires that @xmath73 . this is indeed the case for scii , our most complete sample . in summary , we find the following mean power - law scaling relations at @xmath3-band for spiral galaxies : l_i^0.31 , l_i^0.34 , ^1.00 . [ eq : scaling ] the uncertainties in the log - slopes for the scii sample , based on measurement errors , are @xmath74 , @xmath74 , and @xmath75 , respectively . shows the combined tf and sl relations for the four samples . this operation formally requires that the magnitudes , rotation velocities , and scale lengths be perfectly homogeneous . however , zero - pointing differences exist between the various samples thus preventing a direct merging of all samples into one large homogeneous catalog for our study of scaling relations . the rest of our investigation thus relies on the separate examination of each data set . [ sec : color ] the morphological type dependence of the tfr is highlighted in figs . [ fig : scii_3rel][fig : uma_3rel ] by different point types and colors . this dependence , first noted by roberts ( 1978 ) , and revisited by aaronson & mould ( 1983 ) , rubin et al . ( 1985 ) , giraud ( 1986 ) , pierce & tully ( 1988 ) and g97 to name a few goes in the general sense of early - type galaxies rotating faster than later - types at a given optical luminosity . this trend is confirmed for each sample in the top panels of figs . [ fig : scii_3rel][fig : uma_3rel ] where the early - type and late - type spirals , represented by red stellated points and black filled triangles , are located above and below the fiducial fits , respectivelyor just the nature of s0 galaxies . ] . recent optical tf relations of nearby spirals ( kff02 ) and of distant spirals ( b. weiner et al . 2003 [ deep project ] , in preparation ) covering a wide range of morphologies support the hubble - type dependence of the optical tfr , up to at least @xmath76 . figs . [ fig : tfres_shell ] & [ fig : tfres_scii ] show histograms of the luminosity residuals from the mean tf relation as a function of hubble type for the shellflow and scii samples ( similar results are obtained for mat and uma ) . by selection , the mean tf fit is dominated by sb - sc galaxies . for the shellflow sample , early and late - type spirals differ by 0.08 dex and -0.25 dex from the mean tfr respectively . this is accentuated in the scii sample ( fig . [ fig : tfres_scii ] ) with tf luminosity residuals of + 0.17 dex and -0.33 dex for early and late - types , respectively . dale et al . ( 1999 ) applied a morphological - type correction to their galaxy magnitudes for the construction of their tf template . their corrections to total @xmath3-band magnitudes for t@xmath77sab , sb , and t@xmath78sbc are -0.27 , -0.11 , and 0.00 mag , respectively , which is somewhat larger than the ones we find for their data . while hubble - type classification depends on the bandpass of selection , a morphological - type dependence of the tfr , even at near - infrared bands , is undeniable . the morphology dependence of the sv and sl relations is weaker , as one would expect based on the relatively larger errors of disk scale lengths compared to circular velocity measurements . the distributions of scale lengths in the sl diagrams show a slight excess for bluer ( later type ) galaxies over the mean at a given luminosity , suggesting that bluer galaxies are bigger than the mean . prejudice from theoretical models ( e.g. far00 ; vdb00 ) and our own simple scenario of disk formation ( ) would have us favor a scenario where redder objects are both faster rotators and more compact , but observational evidence for the latter is weak . this could be the result of inhomogeneous disk scale length measurements , and especially for the fainter systems for which sky errors are more conspicuous . the slight excess of disk scale lengths over the mean at the faint end of the slr could be understood if sky subraction was systematically under - estimated . in that respect alone , an extensive program to measure accurate disk scale lengths and half - light radii for a large collection of spiral galaxies based on one unique reduction method and high signal - to - noise data is badly needed . because morphological classification is a subjective measure , we look at more objective observables to determine what physical parameters drive the tfr scatter . [ fig : tfmag1 ] and [ fig : tfmag2 ] display the tfr scatter in magnitude against five galaxy observables for the four samples . besides the rc3 and sdss colors extracted for the mat and scii samples respectively ( see appendix b for scii ) , the data were all collected from the original literature . the concentration indices ( cis ) for shellflow and scii galaxies correspond to the ratio of radii containing 75 and 25 percent of the total light ( ci72 ) . cis for uma galaxies were computed at radii containing 80 and 20 percent of total light ( ci82 ; tully et al . 1996 ) . note that all exponential disks without a bulge have the same value of ci72=2.8 and ci82=3.6 , regardless of their total mass or scale length . for the mat sample , we substituted missing ci data with b / d ratios . looking at figs . [ fig : tfmag1 ] & [ fig : tfmag2 ] , we see that the tfr scatter ( in magnitudes ) is clearly independent of central surface brightness , @xmath79 , physical disk scale length , @xmath11 , and concentration index . color and morphological types , on the other hand , are both correlated with tf magnitude residuals . given that color and morphological type are themselves strongly correlated ( see ) , this is consistent with a picture in which color is the fundamental driver of scatter in the tf relation . shows that the tightest correlation with morphological type is indeed with color . also shows that velocity depends more strongly upon morphological type ( color ) than disk size , an important observation that we return to in and paper ii . the correlation of morphological types with concentration index is poorest , highlighting the relative inadequacy of the ci test to discriminate between different spiral hubble types . according to v01 , all correlations of tf residuals with galaxy observables detected in blue bands , including color , vanish in the infrared . this important conclusion however rests on the study of a small data sample ( less than 44 galaxies ) and deserves closer attention . contrary to findings in aaronson & mould ( 1983 ) and v01 , rubin et al . ( 1985 ) found that the morphological type dependence of the tfr was reduced , but not eliminated , at infrared @xmath32 luminosities - band . such a dependence could be an artifact of rotational speed measurements with optical spectra . the shape of optical rotation curves is often dominated by the inner rising part of an early - type spiral or the slowly rising part of a later - type spiral of the same luminosity . however , the presence or absence of a bulge or bar has a weaker effect on the overall extended shape of a resolved rotation curve . thus the asymptotic rotational speed measured from synthesis maps or velocity fields from say , 2 disk scale lengths to the last detected velocity point ( with acceptable signal - to - noise ) , will not be subject to the vagaries of the inner rotation curve in the same way as optical rotation curve measurements ( c97 , v01 ) . so long as velocity measurements are made beyond @xmath80 , the rotation speed should be unaffected by a bulge or bar ( cr99 ; courteau et al . 2003 ) , and the trends detected by rubin et al . and ourselves are robust . ] . thus color is potentially another fundamental parameter of galaxy structure and formation , in addition to size , luminosity and rotation speed . until recently , the ccd luminosity measurements in large galaxy samples with available line widths were only collected in a single band ( typically @xmath62 or @xmath3 ) , thus thwarting any systematic test for color dependence . recent scatter studies based on the small , pruned , uma galaxy sample ( heavens & jimenez 1999 ; v01 ) can now be duplicated for much larger samples , as we do here . the recent @xmath22-band tf study of the nearby field galaxy survey ( jansen et al . 2000 ) by kff02 identified @xmath81 color and equivalent width as the main drivers of the tfr scatter . color and equivalent width are both tributary of star formation histories though the former depends both on the _ integrated _ and instantaneous star formation while the latter is a function of the current star formation rate alone . taken together with our results , it thus seems that star formation history is the fundamental driver of the tf scatter ( e.g. heavens & jimenez 1999 ) . one expects the tf scatter in bluer bands to be more sensitive to contributions of instantaneous star formation activity , while at ( near-)infrared bands the scatter mainly reflects the convolved star formation histories . in order to verify that the basic trends in the scatter of the galaxy virial ( tf and sl ) relations are a manifestation of star formation _ histories _ , and thus persist at infrared bands , we need to expand our data base as we do below . [ sec : irtf ] the advent of large - scale infrared surveys such as 2mass provides us with reliable @xmath4 luminosities , effective radii , and colors to construct ( nearly ) dust - free scaling relations . with a typical surface brightness limit of @xmath82 mag arcsec@xmath83 , the 2mass luminosity profiles are a full two magnitudes shallower than the typical @xmath3-band profiles in our samples or the sdss brightness profiles . yet , they yield scaling relations that are as tight as the ones derived at @xmath3-band . we now restrict our ir - extended analysis to the scii sample for which the greatest range of luminosities and rotation speeds is found and thus for which the most reliable scaling relations can be derived . because of the 2mass magnitude limit of 13.5 @xmath31-mag ( e.g. bell et al . 2003b ) , only the @xmath84 brightest scii galaxies have measured kron magnitudes . figs . [ fig:2mass_reff ] to [ fig:2mass_jmk ] show the 2mass @xmath5-band scaling relations for scii galaxies plotted against @xmath85 , @xmath86 , and @xmath87 colors . in the @xmath32-band one is notoriously more sensitive to airglow fluctuations ( e.g. , jarrett et al . 2003 ) , making this photometric band less attractive for tf studies . the @xmath31-band tfr is very similar to that in the @xmath5-band , but since the latter yielded the smallest tf scatter , we focus on the @xmath5-band in what follows . the scatter in the sl and sv relations is significantly reduced compared to that in the @xmath3-band ( cf . table 2 ) . note , however , that the size parameter in the @xmath5-band is an effective radius rather than a disk scale length ( which was not available from the 2mass data release ) . therefore , it is unclear whether this reduction in scatter reflects the usage of a more robust and rigorous galaxy size measure , or whether it is related to stellar population differences . as reported in table 2 , the ( scii ) @xmath3 and @xmath5 tfrs have nearly the same log - slopes . for the scii sample with 2mass data , the infrared scaling relations are : l_j^0.30 , l_j^0.41 , ^1.36 . [ eq : irscaling ] while the log - slopes for our sl and sv relations in figs . [ fig:2mass_reff ] to [ fig:2mass_jmk ] were computed with respect to @xmath85 , they can be compared directly to @xmath11 since , for a pure exponential disk , @xmath88 . however , the @xmath5-band sl and sv log - slopes are significantly steeper than at @xmath3 , possibly resulting from biases in the measurement of disk scale lengths versus effective radii . the former is sensitive to the disk fitting baseline while the other depends on the strength of the bulge - to - disk ratio . in order to verify this , we would need both types of radii measurements ( either at @xmath3-band or in the infrared ) which are not currently available ; the assembly of such a data base is under way . low and intermediate - redshift ( @xmath89 ) studies of sv relations , based on data from courteau ( 1996 ; c97 ) and vogt et al . ( 1997 ) and as reported in mao , mo , & white ( 1998 ) , have sv log - slopes of @xmath90 and @xmath91 respectively , in good agreement with our @xmath3-band measurements . dynamical considerations ( ) also suggest a log - slope of unity for the sv relation . more work needs to be done to ascertain the cause of the steep infrared size - dependent scaling relations . figs . [ fig:2mass_reff][fig:2mass_jmk ] confirm that the tf scatter is dominated by color and nearly independent of size and surface brightness . previous reports of tf residual correlations with surface brightness may have generated confusion . willick ( 1999 ) reported a correlation of @xmath3-band tf residuals with surface brightness and compactness for his `` lp10k '' survey of distant cluster galaxies using a moments fitting method to determine the exponential disk parameters even in the presence of irregularities in the galaxy light profiles . in an attempt to alleviate subjective fitting boundaries , willick s computation used the entire surface brightness profile , including the bulge , and thereby biasing scale length and central surface brightness measurements . applying his procedure , we can reproduce the putative surface brightness dependence of the tfr while proper bulge - to - disk ( b / d ) fitting techniques ( mch03 ) find none . the absence of surface brightness dependence of the tfr scatter has been verified by sprayberry et al . ( 1995 ) , zwaan et al . ( 1995 ) , and v01 , and suggests a near - constancy of near - infrared ( nir ) @xmath92 ratio for high surface brightness ( hsb ) and low surface brightness ( lsb ) galaxies of comparable total luminosities . in other words , if the surface brightness decreases , so do @xmath12 and @xmath10 in such a way that the tfr is independent of surface brightness . thus , hsb and lsb galaxies belong to the same tfr , albeit with lsb galaxies showing a greater spread at lower luminosities . any tf offset between low and high surface brightness noticed in the @xmath36-band can be explained by mass - to - light ratio differences between the relevant stellar populations and , according to v01 , the offset disappears at @xmath31 . the slr shows a very strong dependence on @xmath93 ( as expected since @xmath94 ) , and a weaker dependence on color . we see from that bluer galaxies are on average larger than the mean , at a given luminosity , but we can not state with any confidence that redder galaxies have smaller characteristic sizes , since these scatter evenly about the mean line . [ sec : cr03 ] the previous section identified color and surface brightness as the primary source of scatter in the tf and sl relations , respectively . in the spirit of cr99 , we now examine correlations of the residuals from the mean virial relations in each sample . following , we define the residuals for each object @xmath95 as @xmath96 . fig . [ fig:3res ] shows residual correlations for combinations of @xmath97 , @xmath98 , and @xmath99 for each sample based on the @xmath3-band catalogs . the colored types have the same morphological dependence as in figs . [ fig : scii_3rel][fig : uma_3rel ] . the correlation residuals for the scii sample with 2mass @xmath5-band data are presented in ; the point types are a function of @xmath87 color . as in cr99 , we apply a robust non - parametric test by rotating the set of residuals @xmath100 by various angles @xmath101 to get @xmath102 and then applying a spearman rank test ( press et al . 1992 , 14.6 ) for correlations between the quantities @xmath103 . the acceptable range of @xmath104 can be calculated from the range of angles @xmath101 for which the @xmath105 are _ not _ significantly correlated . the correlation slopes and their associated errors are reported in table 3 and are shown at the bottom of each panel in . the linear correlation coefficients , @xmath62 , which provides another means of assessing the strength of the correlation between each residual , are also shown at the top left corner of each panel . the slopes and correlation coefficients shown at the bottom of are representative of the full data set ( including all colors ) . there is close agreement in the correlated residual solutions of all samples , and including the 2mass scii data set , with the general solution : @xmath106 these slopes are an average of the values for shellflow , mat , and the @xmath3 and @xmath5-band scii samples . besides uma , the slopes for the tf / sl residuals for all samples are negative ( anti - correlated ) and low . for the three large samples and at all colors , the tf / sl residual correlation is statistically different from @xmath107 , though only weakly . furthermore , redder galaxies ( top left corner of ) lie above the null line whereas blue galaxies ( bottom left ) lie below . note also that galaxies of all morphological types ( and barredness , not shown here ) , scatter normally about the zero line . far00 have also computed the tf / sl residuals for the uma sample ( v01 ; see lower left panel of ) and find no correlation for both for lsb and hsb galaxies , as do we for that sample ( see table 3 ) . that result is however heavily weighted by the small size of the sample . the weak and strong surface brightness dependences of the tf and sl relations , respectively , have been used to infer the ratio of baryonic to dark mass in the inner parts of galaxies ( cr99 ) . if the gradient of nir mass - to - light ratios is self - similar among galaxies of different brightnesses , cr99 conjectured that maximal disks , the radius of maximum disk circular speed ( sackett 1997 ) . @xmath11 is the scale length of the disk . ] should be offset from the mean tf and sl relations in such a way that the fractional deviations , @xmath108 and @xmath109 , from the mean relations , @xmath110 and @xmath111 , should be strongly correlated . the case of @xmath112 would be expected for pure ( maximal ) exponential stellar disks ( i.e. , without a dark matter halo ) . uncorrelated tf / sl residuals might then suggest that disk galaxies ( of all colors , types , and barredness ) are , on average , sub - maximal kms^-1km@xmath39s@xmath40 may harbor disks with high mass fractions , in agreement with the maximum disk hypothesis ( athanassoula , bosma , & papaioannou 1987 ; kranz , slyz , & rix 2003 ) . ] and@xmath70or that a fine - tuning of halo and disk parameters washes out any surface brightness correlation in the tfr . using the samples of bright late - type spiral galaxies of courteau ( 1992 ) and mat , with available @xmath22 and @xmath3-band luminosity profile and rotation measures , cr99 found a tentative correlation between the tf residuals ( measured in velocity space ) and the slr residuals . the tf / sl fractional deviations were found to be weakly correlated with @xmath113 . with our expanded study , we now find @xmath114 , suggesting that the disk mass plays only a small role in setting the total velocity at @xmath115 . according to the galaxy contraction models of cr99 , the residual correlation @xmath116 corresponds to @xmath117 , indicating that spiral galaxies , independent of barredness or type , would be dominated by dark matter with @xmath118 within @xmath119 . this result agrees with analytical models of cosmological structure formation ( dalcanton et al . 1997 ; mmw98 ; ns00 ; vdb00 ; cole et al . 2000 ) , whereby dark halos dominate the potential even in the inner parts of galaxies , and with independent measurements of galaxy dynamics ( e.g. fuchs 2001 ; trott & webster 2002 ; kranz , slyz , & rix 2003 ) . this result does not preclude the possibility of a stronger disk at radii smaller than @xmath119 . with @xmath120 , the sl residuals ( middle column of figs . [ fig:3res][fig:2mass_3res ] ) are also weakly correlated . a positive correlation may be real but not of fundamental interest . consider a model where v(r ) versus l(r ) is a strict relation with no scatter , while r is a scattered quantity about the relation r(l ) . then the scatter in both v(r ) and l(r ) comes solely from the scatter in r , and of course the residuals will be correlated , both determined by the value of r for a specific galaxy . thus it is the tfr between v and l that gives rise to the v(r)-l(r ) correlation . in a similar way , the weak sl correlation between r and l gives rise to a ( weak ) correlation between the residuals of r(v ) and l(v ) . models to explain the numerical value of the correlation residuals are presented in paper ii . the vl residual distributions ( third column of figs . [ fig:3res][fig:2mass_3res ] ) are tightly correlated with @xmath121 ( a little less at @xmath5 ) and relatively small slope errors ; indeed , this relation is simply the tfr recast in differential form . figs . [ fig : dvdr ] and [ fig:2massdvdr ] are expanded versions of the @xmath122 relations for the mat , scii , shellflow , and uma samples with @xmath3-band data , and the scii sample with @xmath5-band data , respectively . again , the correlated residuals are fully inconsistent with the prediction for a pure stellar exponential disk ( short - dashed line ) , @xmath123 at @xmath124 ( cr99 ) . the solid and long - dashed lines show the best fits from our spearman rank test . the lower panels in figs . [ fig : dvdr ] and [ fig:2massdvdr ] show tf and sl residuals with color . trends of @xmath97 and @xmath98 with color residuals are shown in the right lower and central panels ; the residual fits for the scii+2mass data are shown as solid and long - dashed lines in . velocity residuals correlate tightly with color whereas size residuals are noisier and statistically consistent with zero . these trends match those seen in cr99 for the courteau - faber and mat samples and ought to be reproduced by evolutionary models of disk galaxies ( e.g. far00 ; vdb00 ) . if we account for the observed trends with color , and the simplest assumptions about single stellar population models , we find that the mean observed value of @xmath125 yields an _ upper _ limit to the contribution of disk stars to the rotation curve at @xmath126 ( cr99 ) . the linear regressions of @xmath12 on @xmath10 and @xmath11 on @xmath10 , mostly for the @xmath3-band results , yield the following average tf , sl , and sv relations : l^0.31 , l^0.33 , ^1.00 . [ eq : ave_scaling ] within these relations , at a given luminosity , we found that redder spirals tend to rotate faster and show hints of being more compact than bluer spirals . in the spirit of the spherical - collapse model ( gunn & gott 1972 ) one can define the virial radius of a collapsed , virialized gravitational body as the radius inside of which the average density is a factor @xmath127 times the average background density ( the value of @xmath127 depends on redshift and cosmology ; see e.g. , eke , cole & frenk 1996 ; bryan & norman 1998 ) . it is straightforward to show ( see appendix a ) that with this definition the virial quantities @xmath128 , @xmath129 ( defined as the total mass within @xmath128 ) , and @xmath130 ( defined as the circular velocity at @xmath128 ) are related according to @xmath131 and @xmath132 . therefore , as long as @xmath133 , @xmath134 , and @xmath135 are roughly constant ( i.e. , are not significantly correlated with either of the virial quantities ) , these virial relations are all that is required to explain the observed scaling relations for disk galaxies . to realize the implications of this in terms of galaxy formation theory , we write [ ratone ] / l = here , @xmath136 is defined as the disk mass - to - light ratio , @xmath137 is the baryonic mass fraction of the universe ( @xmath138 for a @xmath139cdm cosmology ) , and @xmath140 is the galaxy formation efficiency that describes what fraction of the baryonic mass inside the halo ultimately ends up in the disk . the virial relation @xmath131 thus turns into the observed tfr @xmath27 if , and only if , _ tf ( ) ^3 is not correlated with @xmath12 and reveals an amount of scatter in agreement with that observed in the tf relation . in the standard model for disk formation , set forth by fall & efstathiou ( 1980 ) , dalcanton et al . ( 1997 ) and mmw98 , the relation between @xmath141 and @xmath128 derives from the acquisition and conservation of ( specific ) angular momentum by the baryons , when cooling to form a centrifugally supported disk . the total angular momentum of a virialized system is conveniently expressed by the dimensionless spin parameter ( bullock et al . for an exponential disk embedded in a dark matter halo , the total angular momentum is given by j_d = 2 _ 0^ ( r ) v_c(r ) r^2 dr = m_d r_d f_v with @xmath142 and @xmath143 the surface density and circular velocity of the disk , and f_v = _ 0^ e^-u u^2 v_c(r_d u ) du ( mmw98 ) a dimensionless number . if disk formation occured under conservation of specific angular momentum , as generally assumed , one therefore obtains that [ rattwo ] r_d = f_v ( mmw98 ) . thus , for the virial relation @xmath144 to turn into the observed sv relation @xmath145 , one requires that _ sv ( ) is not correlated with @xmath12 and has a scatter around its mean value in agreement with that in the observed sv relation . similarly , in order to relate the virial relation between @xmath128 and @xmath129 to the observed sl relation , the quantity _ sl ( ) ^1/3 can not be correlated with galaxy luminosity , and has to reveal an amount of scatter that matches that in the sl relation observed . thus , the challenge for galaxy formation theories is to understand how to meet the constraints on @xmath146 , @xmath147 , and @xmath148 derived here . that this is not a trivial matter becomes clear if one takes into consideration that ( i ) both @xmath140 and @xmath149 depend extremely sensitively on the efficiencies of cooling , star formation and feedback , and are thus expected to be strongly correlated with @xmath129 and @xmath69 ( see e.g. , van den bosch 2002a ) , and ( ii ) that both @xmath150 and @xmath151 depend , in a convoluted way , on @xmath69 , @xmath140 , and on halo concentration , which itself depends on halo mass ( e.g. , navarro , frenk & white 1997 ) . an additional challenge for galaxy formation models is to understand the dependence of the scaling relations on disk color . for instance , reproducing that color is the prime driver of scatter in the tf relation , requires that @xmath152 is correlated with color . in hierarchical structure formation scenarios , small mass haloes form earlier than their more massive counterparts . since gas cooling is very efficient in low mass haloes at high redshift , these models generally predict that fainter galaxies ( living in low mass haloes ) are older , and therefore redder , than their more luminous counterparts . this is opposite to the observed trend , and has long been known to be an important problem for galaxy formation in cdm comogonies ( see bell et al . 2003a and macarthur et al . 2003b for recent discussions ) . because of this problem , simple disk formation models within this cdm framework naturally predict that the slope of the tf should be steeper in bluer bands ( e.g. , van den bosch 2002a ) , again opposite to what has been observed . for the more massive galaxies to be older / redder , one could adjust the feedback prescription in order to vary the amount of gas accreted by spiral galaxies as a function of time . however , the energetic requirements for this to occur may exceed the energy reservoir in the stellar winds and supernovae and would require the action of an agn ( bell et al . the challenge is to realize this such that the constraints on @xmath146 , @xmath147 , and @xmath153 are not violated . in summary , we can conclude that the observed scaling relations of disk galaxies are predominantly a straightforward result of the basic dynamics of halo formation in the standard cosmological picture and the simple , standard scenario of disk formation . they do , however , impose some interesting constraints on galaxy formation models , which we address in more detail in paper ii . by assembling some of the most extensive existing databases of galaxy structural parameters , and adding new infrared luminosity and effective radii from 2mass , we have been able to confirm or infer the following major observational results : * the scaling relations of spiral galaxies , obtained by linear regressions of @xmath12 on @xmath10 and @xmath11 on @xmath10 , are @xmath154 + this implies that the surface brightness @xmath155 . these scaling relations seem to reflect the simplest possible model for disk galaxy formation based on angular momentum conservation of the baryons inside virialized dark matter halos . * the residuals of @xmath110 in the tfr are correlated with color , and uncorrelated with effective surface brightness . the residuals of @xmath111 are correlated with surface brightness and marginally correlated with color . * early - type ( redder ) disk galaxies rotate faster than later ( bluer ) types at a given luminosity . the effective radii of blue galaxies are marginally larger than the mean at a given luminosity , while the sizes of red galaxies do not show a significant systematic deviation from the mean . * the tf / sl residuals are weakly anti - correlated with @xmath6 . this confirms the results of cr99 , suggesting that most spiral disks are sub - maximal . * the sv / lv residuals are weakly correlated with @xmath156 . while the observed scaling relations of disk galaxies appear to be a straightforward result , the lack of correlation between the residuals from these relations of @xmath12 and @xmath11 at a given @xmath10 coupled with their color dependence is more puzzling . in paper ii , we argue that the small scatter in the tfr is consistent with being a reflection of the natural scatter in halo concentration @xmath157 , as measured in cosmological simulations . this scatter is small because it is independent of the scatter in halo spin . other explanations for the small size of the tfr scatter have invoked star formation and feedback processes ( eisenstein & loeb 1996 ; silk 1997 ; heavens & jimenez 1999 ; far00 ) , the small dispersion in halo and@xmath70or disk collapse times ( heavens & jimenez 1999 ; vdb00 ) , and dynamical response of the halo to the disk assembly ( ns00 ) . whether these , and other interpretations for the small scatter in the tfr , are all fully independent is still a matter of contention that we revisit , along with the interpretation of the major observational results above , in paper ii . we acknowledge useful conversations and suggestions from eric bell , hans - walter rix , and marc verheijen . dan zucker kindly helped with the extraction of petrosian magnitudes from the sdss / dr1 database . aaron dutton is also thanked for experimenting with morphological classification of 2mass images . his work demonstrated the difficulty of assigning unambiguous eye - ball morphologies to infrared galaxy images . sc wishes to acknowledge his colleagues on the shellflow team ( marc postman , david schlegel , and michael strauss ) for permission to use previously unpublished results . sc and lam acknowledge financial support from the national science and engineering council of canada . ad acknowledges support by the us - israel bi - national science foundation grant 98 - 00217 , the german - israel science foundation grant i-629 - 62.14/1999 , and nasa atp grant nag5 - 8218 . sc would also like to thank the max - planck institut f " ur astronomie in heidelberg and the max - planck institut f " ur astrophysik in munich for their hospitality while part of this paper was developed . this research has made use of ( i ) 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 , as well as nasa s astrophysics data system ; ( ii ) , the _ two micron all sky survey _ , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation ; and ( @xmath158 ) the _ sloan digital sky survey _ ( sdss ) . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , los alamos national laboratory , the max planck institut fr astronomie ( mpia ) , the max planck institut fr astrophysik ( mpa ) , new mexico state university , university of pittsburgh , princeton university , the united states naval observatory , and the university of washington . [ sec : appendixa ] below , we develop the theory for the tfr ( see also mmw98 ; ns00 ) , and determine the fundamental nature of the tfr . this is because two simple predictions for the tfr have been used in the past : one with a log slope of @xmath159 ( @xmath160 in magnitudes ) , and the other with a log - slope of @xmath161 ( @xmath162 in magnitudes ) . here we compare both predictions and show that they are related to one another . this discussion replicates , but giving more detail . let us consider a dark - matter halo of mass @xmath129 reaching virial equilibrium at a time corresponding to the cosmological expansion factor @xmath163 . we assume that the halo properties at that time are relevant for the properties of the stellar component that has formed in this halo . for a body that is dominated by gravitational forces and is approximately in steady - state , the virial radius @xmath128 is defined in the spirit of the spherical - collapse model ( gunn & gott 1972 ) by a given density contrast @xmath127 relative to the mean universal density at the time of collapse ( where mass shells are infalling for the first time ) , namely by @xmath164 . at early times , when @xmath165 , the relevant density contrast is @xmath166 , while for the standard @xmath139cdm cosmology ( with @xmath167 and @xmath168 today ) it rises to @xmath169 today ( eke et al . 1998 ; peacock 1999 ) . we ignore the weak redshift dependence of the @xmath127 factor in the following expressions the change is roughly @xmath170 , which implies that @xmath24 in the following expressions should be replaced by @xmath171 . this is a weak effect , which becomes even weaker at higher redshifts where the bulk of galaxy formation occurs . ] . if disks are embedded in dark halos we can define the total mass of the system ( dark plus baryonic matter ) as = ^2 g. [ eq : totmass ] here @xmath172 is the gravitational constant , and @xmath128 is defined as the radius inside of which the average density of the system is @xmath173 times the critical density of the universe , which at a given @xmath174 is given by _ crit = 3 h(z)^2 8 g. [ eq : critdens ] one can thus write = 10 h(z ) . [ eq : r_vir ] here @xmath130 is simply the circular velocity at @xmath128 , and @xmath175 is the hubble constant at the virialization redshift @xmath174 of the halo . note that @xmath176 , for an einstein - de sitter cosmology relevant at high redshifts , where @xmath177 is loosely referred to as the time of formation of the halo . the three _ virial _ quantities at @xmath24 thus define a one - parameter family : @xmath178 . we now define the disk mass - to - light ratio , [ eq : upsilond ] and write = . [ eq : galformeff ] here @xmath137 is the baryonic mass fraction of the universe ( @xmath179 for a @xmath139cdm model ) , and @xmath140 is the galaxy formation efficiency that describes what fraction of the baryonic mass inside the halo ultimately ends up in the disk . depending on the efficiencies of cooling , star formation and feedback , this can vary anywhere from zero to unity ( see e.g. , van den bosch 2002a ) combining equations ( [ eq : totmass ] ) to ( [ eq : galformeff ] ) yields [ eq : tfthree ] = 10 g h(z ) ^3 . this is a tfr with log - slope @xmath159 ( see also dalcanton et al . 1997 ; white 1997 ; mmw98 ; van den bosch 1998 ; syer , mao , & mo 1999 ) . in contrast , several authors in the past have predicted that the tfr should have a log - slope of @xmath161 ( e.g. , sargent et al . 1977 ; aaronson , huchra , & mould 1979 ; salucci , frenk , & persic 1993 ; sprayberry et al . 1995 ; zwaan et al . 1995 ; stil & israel 1998 ) . their argument goes as follows : assume that the disk surface brightness is described by an exponential function of the form [ eq : surf_disk ] ( r ) = _ 0 ( -r/ ) , with @xmath11 the stellar scale length of the disk . the total luminosity of the disk is [ eq : lum_disk ] = 2 _ 0 ^2 . assuming that one measures the rotation velocity @xmath12 at a radius @xmath180 , and that the gas and stars in the disk move on circular orbits , [ eq : vel_obs ] ^2 = g m(r ) r , with @xmath181 the total mass within radius @xmath62 . if we define the total mass - to - light ratio [ eq : upshat ] ( r ) = ( r ) + ( r ) we can combine equations ( [ eq : lum_disk ] ) , ( [ eq : vel_obs ] ) and ( [ eq : upshat ] ) to obtain [ eq : tffour ] = 1 2 g^2 ^4 this is a tfr with a log - slope of @xmath161 , and as long as one measures the rotation velocity at a constant number of scale lengths , @xmath182 , the scatter is determined purely by the variation in @xmath183 . this was emphasized by zwaan et al . ( 1995 ) , who , from the finding that both hsb and lsb spirals follow the same tfr , concluded that lsb galaxies must have much larger values of @xmath184 than their high surface brightness counterparts . note that this `` conspiracy '' does not apply for the tfr of , for which the zero - point is independent of the disk s surface brightness . the tfrs in eqs . ( [ eq : tfthree ] ) and ( [ eq : tffour ] ) thus predict a different log - slope . the cause of this apparent paradox lies in the fact that the surface brightness , @xmath185 ( mmw98 ) , and reduces to . thus , the two theoretical predictions of the tfr , with log - slopes @xmath159 and @xmath161 , are directly related , but only the derivation for is strictly based on dynamics and it is the most fundamental of the two . [ sec : appendixb ] accurate optical colors based on digital imaging is available only for the shellflow ( @xmath186 ) and uma ( @xmath187 ) samples . the mat and scii galaxy magnitudes were imaged only at @xmath3-band . in cr99 , we computed a @xmath188 color term for mat galaxies using rc3 b - band fluxes but rc3 magnitudes are inherently uncertain with @xmath189 mag . with the release of sdss / dr1 data ( york et al . 2000 ; abazajian et al . 2003 ) , we can extract @xmath190 and @xmath95 magnitudes ( as proxy for @xmath191 and @xmath3 magnitudes ) for galaxies that overlap with shellflow and scii galaxies . we do this for shellflow galaxies in order to map the sdss system to the cousins system , upon which shellflow magnitudes are based . a final transformation can then be applied to scii magnitudes . we compare in figs . [ fig : shellsloan ] & [ fig : sciisloan ] instrumental petrosian magnitudes from the sdss first data release against instrumental magnitudes for 31 shellflow and 39 scii galaxies . the comparisons avoid any differences due to extinction and k - corrections . besides a zero - point offset , the @xmath190 and @xmath95 petrosian magnitudes scale linearly with the @xmath191 and @xmath3 cousins magnitudes in shellflow and scii . we compute transformations for the sdss petrosian magnitudes into cousins magnitudes for the scii galaxies that overlap with the sdss / dr1 database . the instrumental `` @xmath191 '' magnitudes for scii were corrected the same way as the @xmath3-band magnitudes but using an extinction dependence @xmath192 , instead of @xmath193 , and the poggianti ( 1997 ) formulation for the @xmath191-band k - correction . lcccccc mat & 845 & field & @xmath3 & @xmath194 17 & @xmath195 kms^-1km@xmath39s@xmath40 & + scii & 521 & cluster & @xmath3 & @xmath196 & [ 500019,000 ] & , + shell_flow _ & 340 & field & @xmath197 & @xmath198 & [ 4500,7000 ] & + uma & 44 & cluster & @xmath199 & @xmath200 & 18.6 mpc & + lrrrr @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 + @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 + @xmath232 & @xmath233 & @xmath234 & @xmath235 & @xmath236 + & & & & + @xmath237 & & @xmath238 & & + @xmath227 & & @xmath239 & & + @xmath240 & & @xmath241 & &

we propose two new observables to measure the charge asymmetry in hadronic top - quark pair production in association with a hard jet . the _ incline asymmetry _ , based on the inclination between the planes of initial- and final - state momenta , probes the charge asymmetry in the quark - antiquark channel . compared to the hitherto investigated rapidity asymmetries , the incline asymmetry provides improved access to the partonic charge asymmetry at both the tevatron and the lhc . the _ energy asymmetry _ , based on the energy difference between top and antitop quarks , for the first time allows us to probe the charge asymmetry in the quark - gluon channel at the lhc . in quantum chromodynamics , asymmetries of up to at the leading order are achievable with appropriate cuts . top - pair plus jet production thus has the potential to become the discovery channel of the charge asymmetry in proton - proton collisions . mitp/13 - 005 + pitt - pacc-1304 * top - quark charge asymmetry goes forward : + two new observables for hadron colliders * stefan berge and susanne westhoff + + +

1305.3272

the top quark is often suspected to play a special role among the quarks of the standard model ( sm ) . its strong coupling to the higgs boson suggests that the top quark may be involved in the mechanism of electroweak symmetry breaking and/or the generation of fermion masses . thanks to the recent progress in top - quark physics on both the theoretical and the experimental side , those fundamental issues can now be addressed at an unprecedented level of precision . to date , top - quark observables at hadron colliders have passed the various tests of top - quark properties in the sm . one exception is the top - quark charge asymmetry , which has been measured as a forward - backward asymmetry in inclusive top - antitop production at the tevatron @xcite . it exceeds the sm prediction by about three standard deviations . the limited set of data recorded during the runtime of the tevatron makes it difficult to draw a conclusion on the origin of this discrepancy . at the lhc , the charge asymmetry in top - quark pair production is buried under the large charge - symmetric background from partonic gluon - gluon ( @xmath3 ) initial states . measurements of the charge asymmetry by the atlas and cms collaborations do not indicate a deviation from the sm prediction , but they are affected by large uncertainties @xcite . novel information on the top quark s properties can be obtained by investigating top - quark pair production in association with a hard jet . beyond the standard model , @xmath4 production provides a wide playground for sensitive probes of new physics , which recently started to be explored @xcite . in this work , we will focus on the charge asymmetry in quantum chromodynamics ( qcd ) . the presence of an additional jet allows us to investigate qcd effects at leading order ( lo ) , which contribute to inclusive top - pair production at next - to - leading order ( nlo ) . in inclusive @xmath5 production , the charge asymmetry arises at nlo from virtual and real gluon radiation @xcite . the same real gluon contributions generate the asymmetry in @xmath4 production at lo . while the completion of nnlo qcd calculations for the inclusive @xmath5 asymmetry is still ongoing , nlo corrections to the asymmetry in @xmath4 have been calculated and found to be sizeable @xcite . the effects of top - quark decay and parton showers have been investigated in detail @xcite . the sm prediction of the charge asymmetry in @xmath4 thus has a solid basis , ready to face experimental results . at the tevatron , the production rate of @xmath4 final states is relatively small due to the limited phase space @xcite . with higher collision energies at the lhc , @xmath5 events are produced abundantly in association with at least one hard jet @xcite . from a statistical point of view , the measurement of a charge asymmetry in @xmath4 production is therefore within reach . a detailed understanding of the jet kinematics in @xmath4 production is crucial in order to control the infrared and collinear behavior of the process . as a result of the enhanced symmetric cross section in the collinear limit , the normalized charge asymmetry is maximized if the jet is emitted perpendicular to the beam axis in the partonic center - of - mass ( cm ) frame @xcite . this feature facilitates the construction of a collinear - safe observable by setting an experimental cut on the jet rapidity in the parton frame . the jet handle can be further exploited to probe the charge asymmetry in @xmath4 production through observables that are not accessible in inclusive @xmath5 production . the construction and investigation of such observables is the purpose of this work . we aim at providing optimal access to the charge asymmetry in qcd in both the partonic quark - antiquark ( @xmath6 ) channel and the quark - gluon ( @xmath7 ) channel of @xmath4 production . to this end , we explore the relation between the momenta of the top quarks and the jet in the final state . we show that the angular correlation between the light quarks inside the proton and the top quarks in the final state is appropriately described by the inclination between the planes spanned by the light - quark and ( gluon-)jet momenta and by the top - quark and jet momenta , respectively . the corresponding observable , called _ incline asymmetry _ , probes the charge asymmetry in the @xmath6 channel . it is superior to the previously studied forward - backward or rapidity asymmetries , which do not take the jet kinematics into account . the incline asymmetry can be formulated for both proton - antiproton collisions at the tevatron and proton - proton collisions at the lhc . due to the relation between the particles four - momenta in the final state , the scattering angles of the top and antitop quarks with respect to the jet direction are connected to their energies . this relation proves to be particularly useful in probing the charge asymmetry in the @xmath7 channel , where the jet emerges from a quark at the parton level . the resulting _ energy asymmetry _ measures the charge asymmetry in terms of the difference between the top- and antitop - quark energies in the partonic cm frame . it is tailored to the lhc , where the @xmath7 parton luminosity is sizeable , such that the relative @xmath3 background is smaller than for the asymmetries in the @xmath6 channel . for the first time , we have an observable at hand that allows a measurement of the charge asymmetry in the @xmath7 channel . it will be explained in detail why the energy asymmetry in the @xmath7 channel is sizeable , while the previously considered forward - backward or rapidity asymmetries are tiny . this paper is organized as follows . in section [ sec : asymmetry ] , we perform a comprehensive analysis of the charge asymmetry in @xmath4 production at the parton level . we start by discussing the kinematics in the @xmath6 channel at an analytic level ( section [ subsec : kinematics ] ) . on this basis , we derive the incline and energy asymmetries for the @xmath6 and @xmath7 channels ( sections [ subsec : aphi ] and [ subsec : aenergy ] ) . the incline asymmetry is compared to the forward - backward asymmetry in section [ subsec : afb ] . a summary of the new asymmetries and their kinematic features concludes the partonic analysis ( section [ subsec : asparton - summary ] ) . in section [ sec : colliders ] , we study the incline and energy asymmetries at hadron colliders . at the tevatron , the incline asymmetry is a large quantity and is expected to be accessible with the full data set ( section [ subsec : iatevatron ] ) . the energy asymmetry can not be observed at the tevatron . at the lhc , both the incline and the energy asymmetry can be explored ( sections [ subsec : ialhc ] and [ subsec : ealhc ] ) . due to the large @xmath3 background , suitable cuts are indispensable to the observation of a sizeable asymmetry . we comment on the prospects to measure the new observables at the lhc with @xmath8 tev data and provide predictions for the @xmath9 tev run , where a high significance can be reached with a larger data set . we conclude in section [ sec : conclusions ] . in appendix [ app : boosted ] , we discuss the relation between the asymmetries in the @xmath6 and @xmath7 channels . the results allow us to understand the magnitude of the respective observables in different channels . the definition of a charge asymmetry in @xmath4 production crucially depends on the jet kinematics in the final state . momentum conservation requires that the three - momenta of the final - state particles in the partonic process @xmath10 sum to zero in the partonic cm frame , @xmath11 . the top , antitop and jet ( @xmath12 ) momenta @xmath13 , @xmath14 , and @xmath15 thus lie in one plane , which features a certain inclination with respect to the plane spanned by the momenta of the incoming partons and the jet , @xmath16 , @xmath17 , and @xmath15 . below , we will derive an asymmetry in terms of this inclination , named incline asymmetry . this new observable is particularly useful for probing the charge asymmetry in @xmath18 in the partonic cm frame . the incline asymmetry is essentially free from the impact of the final - state gluon direction on the angular correlation between the light quarks and the top and antitop quarks . in principle , the incline asymmetry can also be defined for the process @xmath19 by boosting into a reference frame in which the initial state is antisymmetric under charge conjugation . in addition to the incline asymmetry , the final state @xmath4 allows us to probe the charge asymmetry via an energy asymmetry , which is based on the difference between the top and antitop energies . in the process @xmath18 , the incline asymmetry and the energy asymmetry complement each other , as they probe independent parts of the differential charge asymmetry . in @xmath19 , the energy asymmetry is particularly well suited to explore the charge asymmetry in the partonic cm frame . remarkably , the energy asymmetry is the first observable that provides access to the charge asymmetry in the @xmath7 channel at the lhc . in this section , we will derive and discuss the incline asymmetry and the energy asymmetry at the parton level . we will also compare these new asymmetries with the previously investigated forward - backward asymmetry in terms of the top - quark scattering angle off the incoming quark . the jet angular distribution shows that the incline asymmetry equals the forward - backward asymmetry in its maximum , which is reached for central jet emission @xcite . for all other scattering directions of the hard jet , the incline asymmetry is larger than the forward - backward asymmetry . to describe the parton kinematics in a way that captures the relevant features of the incline and energy asymmetries , we use the parameterization from @xcite . in this parameterization , the three - body final state in the process @xmath20 is described by five variables in the partonic cm system : the top - quark and jet energies @xmath21 and @xmath22 , the jet scattering angle @xmath23 off the direction of the incoming parton @xmath24 , and the inclination angle @xmath25 between the planes spanned by the three - momenta @xmath26 and @xmath27 . the fifth degree of freedom , an azimuthal rotation of the final state around the beam axis , is irrelevant for our purposes . the kinematical setup is visualized in figure [ fig : kinematics ] . the four - momenta of the partons are given by @xmath28 where @xmath29 is the partonic cm energy and @xmath30 is the top - quark mass . the jet scattering angle @xmath23 and the inclination angle @xmath25 fix the position of the final - state plane @xmath31 with respect to the plane @xmath32 . they are defined by corresponds to @xmath33 in @xcite . ] @xmath34\,,\\\nonumber \cos\varphi & = & \vec{n}_{13}\cdot\vec{n}_{t3}\,,\qquad\qquad\quad \varphi \in [ 0,2\pi]\ , , \end{aligned}\ ] ] with the normal vectors of the planes @xmath32 and @xmath31 , @xmath35 and @xmath36 . the angle @xmath37 between the top - quark and jet momenta is fixed in terms of @xmath21 , @xmath22 and @xmath29 by the relation @xmath38\ , . \end{aligned}\ ] ] the range of the energies @xmath21 and @xmath22 is independent from @xmath23 and @xmath25 and given in @xcite . the angle @xmath39 between the antitop - quark and jet momenta is obtained by changing @xmath21 to the antitop - quark energy @xmath40 in ( [ eq : cosxi ] ) , @xmath41\ , . \end{aligned}\ ] ] for @xmath4 production at hadron colliders , the three partonic processes in qcd at lo are the quark - antiquark channel @xmath18 , where @xmath42 , the quark - gluon channel @xmath19 , where @xmath43 , and the gluon - gluon channel @xmath44 . the @xmath3 channel does not contribute to the charge asymmetry , but is an important background at the lhc . the parameterization introduced above allows us to control the infrared and collinear behavior of the partonic cross sections in the limits @xmath45 and @xmath46 . for later purposes , it is important to bear in mind that the charge asymmetry in the @xmath6 channel has an infrared divergence , whereas the @xmath7 contribution is finite for @xmath45 . the symmetric cross section for @xmath4 production , which normalizes the asymmetry , exhibits both soft and collinear divergences . the size of the observable charge asymmetry thus depends significantly on @xmath22 and @xmath23 , and on the cuts being introduced to regularize the divergences . the differential charge - symmetric cross section @xmath47 and the charge - asymmetric cross section @xmath48 at the parton level are defined as @xmath49 since the @xmath6 initial state is antisymmetric under charge conjugation and the jet angular distribution is symmetric , the analytic expression of the charge asymmetries is most transparent in this channel . we will thus primarily focus on the process @xmath50 , when we introduce the new asymmetries at the parton level . in the @xmath7 channel , the initial state is neither symmetric nor antisymmetric under charge conjugation . furthermore , the jet distribution in @xmath51 is not forward - backward symmetric , but the jet is preferentially emitted in the direction of the incoming quark . as we aim at examining the features of the charge asymmetry in terms of top and antitop energies , we substitute the jet energy @xmath22 with the antitop energy @xmath52 in our parameterization in ( [ eq : parametrization ] ) . in terms of the inclination angle @xmath25 , the jet scattering angle @xmath23 , and the top and antitop energies @xmath21 and @xmath52 , the differential charge asymmetry in @xmath50 exhibits the following structure , @xmath53\,\cos\varphi\\ & & \hspace{4 cm } { } + \,[n_2 + \cos^2\varphi\,n_2^{\varphi}]\,\sin\theta_j\cos\theta_j\big\}\,,\nonumber \end{aligned}\ ] ] with the @xmath54 color factors @xmath55 and @xmath56 , and the coefficients @xmath57 given by @xmath58\\ & & { } - 4\sqrt{s}(e_t+e_{\bar t})[s - m_t^2 + e_t^2 + e_{\bar t}^2 + ( e_t+e_{\bar t})^2(1-m_t^2/s ) ] + s[s + 2(e_t^2+e_{\bar t}^2)]\big\}\,,\nonumber\\ n_1^j & = & \frac{16}{d}\,\frac{c}{(\sqrt{s}-e_t - e_{\bar t})^3}\,\big\{-2(e_t + e_{\bar t})^2[3s - 6m_t^2 + e_t^2 + e_{\bar t}^2]\nonumber\\ & & { } + 4\sqrt{s}(e_t+e_{\bar t})[s - 3m_t^2 + ( e_t+e_{\bar t})^2(1-m_t^2/s ) ] - s[s - 4m_t^2]\big\}\,,\nonumber\\ n_1^{\varphi } & = & \frac{16}{d}\,\frac{c^{3}}{(\sqrt{s}-e_t - e_{\bar t})^3}\,,\nonumber\\ n_2 & = & \frac{64}{d}\,m_t^2\,(e_t - e_{\bar t})\,,\qquad n_2^{\varphi } \,=\ , \frac{32}{d}\,\frac{c^2}{(\sqrt{s}-e_t - e_{\bar t})^3}\,(e_t^2 - e_{\bar t}^2)\,,\nonumber \end{aligned}\ ] ] where @xmath59(\sqrt{s } - 2e_t)(\sqrt{s}-2e_{\bar t})\,,\nonumber\\ c & = & 2\,\sqrt{-(e_t+e_{\bar t})^2(s+m_t^2 ) + \sqrt{s}(e_t + e_{\bar t})[s + 2m_t^2 + 2e_te_{\bar t } ] - s[s/4+m_t^2+e_te_{\bar t}]}\,.\nonumber \end{aligned}\ ] ] the first term of the distribution @xmath60 in ( [ eq : sigmaa ] ) is antisymmetric in @xmath61 , whereas the second term is antisymmetric in @xmath62 . an asymmetry in terms of @xmath61 , the incline asymmetry , thus probes the set of coefficients @xmath63 . while @xmath64 are symmetric under @xmath65 , the coefficients @xmath66 are antisymmetric under @xmath65 . an asymmetry in terms of the energy difference @xmath67 , the energy asymmetry , thus probes the complementary part of the differential asymmetry via @xmath68 . however , in the @xmath6 channel , the energy asymmetry vanishes when integrated over @xmath23 . one has to construct a two - fold asymmetry in terms of @xmath69 and @xmath23 to obtain a non - vanishing observable . in the @xmath6 channel , the asymmetries in @xmath61 and @xmath69 provide complementary information about the charge - asymmetric cross section @xmath70 in ( [ eq : sigmaa ] ) . in the @xmath7 channel , this complementarity is lifted , because the jet originates from a quark rather than from a gluon . in the @xmath6 channel , the incline asymmetry probes the charge asymmetry @xmath70 via the inclination angle @xmath25 between the planes @xmath71 and @xmath72 . we define the differential partonic incline asymmetry in terms of the jet scattering angle as @xmath73 the integrated partonic incline asymmetry is then given by @xmath74\ , , \end{aligned}\ ] ] where @xmath75 is the total partonic cross section . here and in the following sections , the jet distribution @xmath76 is denoted by @xmath48 under the integrand , and the variables @xmath25 , @xmath77 , and @xmath52 are implicitly integrated over . the coefficients @xmath78 , @xmath79 and @xmath80 are obtained from ( [ eq : coefficients ] ) by integrating @xmath64 over @xmath21 and @xmath52 . the prefactors are due to the integration over @xmath25 and @xmath23 . as we observe in ( [ eq : aphi ] ) , the incline asymmetry @xmath81 is sensitive to a combination of @xmath82 , namely to that part of the differential charge asymmetry which is symmetric in the top - antitop energy difference @xmath69 . these different contributions to @xmath81 are displayed in the left panel of figure [ fig : ni ] as functions of the jet energy @xmath22 for a partonic cm energy of @xmath83 . , the dependence on @xmath52 has been substituted with @xmath84 , and the remaining dependence on @xmath21 has been integrated out . ] we observe that all coefficients @xmath78 diverge in the infrared limit @xmath45 . the main contribution to the asymmetry @xmath81 thus stems from soft gluon emission . numerically , the coefficient @xmath78 is larger than @xmath79 and @xmath80 , which partially cancel in the combination @xmath85 that enters @xmath81 . from ( [ eq : sigmaa ] ) , it is apparent that the contribution of @xmath78 to the asymmetry is independent from @xmath23 . due to this behavior , the incline asymmetry is insensitive to the jet scattering angle . the incline asymmetry therefore provides optimal access to the differential charge asymmetry @xmath86 , independently of the jet direction . in the @xmath7 channel , the incline asymmetry as defined in ( [ eq : sigmaa - phi ] ) and ( [ eq : aphi ] ) is based on the inclination between the planes @xmath87 and @xmath88 . the inclination angle @xmath25 is now defined with respect to the @xmath89-jet momentum and therefore inappropriate to measure the angular correlation between the quarks in the initial and final states . a definition of the incline asymmetry with respect to the gluon momentum @xmath90 is not possible in the partonic cm frame , because in general @xmath13 , @xmath14 and @xmath90 do not lie in one plane . by boosting into the @xmath5 rest frame , the incline asymmetry in the @xmath7 channel can be defined on the basis of a charge - antisymmetric @xmath5 initial state , in analogy to the @xmath6 channel . we will only briefly comment on this possibility in appendix [ app : boosted ] , since the energy asymmetry will prove more useful as an observable of the charge asymmetry in the @xmath7 channel . the energy asymmetry is based on the difference between the top and antitop quarks energies , @xmath67 . we define the differential energy asymmetry as a function of the jet angle by @xmath91 the energy difference @xmath69 is connected to the kinematics in the final - state plane @xmath71 by energy and momentum conservation . in particular , the angles @xmath37 and @xmath39 of the top and antitop quarks with respect to the jet momentum , defined in ( [ eq : cosxi ] ) and ( [ eq : cosxibar ] ) , are fixed by the energies @xmath21 and @xmath40 . for @xmath92 , one has @xmath93 , and for @xmath94 , one has @xmath95 . the energy asymmetry @xmath96 can therefore be interpreted as an asymmetry of the top and antitop scattering angles with respect to the jet direction , @xmath97\ , . \end{aligned}\ ] ] in appendix [ app : boosted ] , we exploit this correspondence and relate the energy asymmetry in the @xmath6 channel to the top - angle asymmetry in the @xmath7 channel and vice versa . this relation allows us to understand the magnitude of the asymmetries in the respective channels . for the @xmath6 channel , the energy - asymmetric part of @xmath60 is encoded in the coefficients @xmath68 given in ( [ eq : sigmaa ] ) and ( [ eq : coefficients ] ) . it accompanies an asymmetry in the ( gluon-)jet scattering angle @xmath23 . the energy asymmetry @xmath96 is thus antisymmetric under @xmath98 and vanishes when integrated over @xmath99 $ ] . to construct a non - vanishing observable , one needs to consider a two - fold asymmetry in terms of @xmath69 and @xmath23 . we thus define the partonic energy asymmetry for the process @xmath100 as @xmath101\ , . \end{aligned}\ ] ] the coefficients @xmath102 are obtained from ( [ eq : coefficients ] ) by integrating @xmath68 over @xmath21 and @xmath103 . the factors @xmath104 and @xmath105 in front of @xmath106 and @xmath107 , respectively , stem from the integration over @xmath25 . by comparing the coefficients @xmath108 in ( [ eq : aphi ] ) and ( [ eq : energy - asymmetry - qqb ] ) , it becomes once more apparent that the energy asymmetry @xmath109 is independent from and complementary to the incline asymmetry @xmath81 . in order to examine the kinematic features of @xmath109 in the @xmath110 channel , we study the coefficients @xmath111 as functions of @xmath112 and @xmath67 . the dependence of @xmath113 and @xmath107 on the jet energy @xmath22 is shown in the right panel of figure [ fig : ni ] . since @xmath113 is finite for @xmath114 , the energy asymmetry is governed by @xmath107 , unless @xmath22 is close to its maximum . by comparing the scales of the coefficients @xmath82 and @xmath102 in figure [ fig : ni ] , the energy asymmetry @xmath109 is expected to be much smaller in magnitude than the incline asymmetry @xmath81 . the dependence of the dominant coefficient @xmath107 on @xmath69 is displayed in figure [ fig : n2phide ] , left , for fixed values of the jet energy , @xmath115 ( plain ) and @xmath116 ( dashed ) . the range of @xmath69 is limited for small @xmath22 due to momentum conservation in the final state . within this range , the maximum of @xmath107 is reached for large @xmath69 , corresponding to @xmath117 for not too large @xmath22 . in the @xmath7 channel , the jet distribution @xmath96 in ( [ eq : sigmaa - energy ] ) is not antisymmetric under @xmath118\theta_j$ ] , because the initial state is not charge - antisymmetric as in the @xmath6 channel . the differential energy asymmetry in ( [ eq : sigmaa - energy ] ) can therefore simply be integrated over the jet scattering angle @xmath23 . the partonic energy asymmetry for @xmath51 is then given by @xmath119 importantly , neither the jet direction nor the initial - state quark direction enter the definition of the charge asymmetry @xmath120 . this feature will be of great use to construct an observable at the lhc . the dependence of the energy asymmetry on @xmath69 is different for the @xmath6 and @xmath7 channels . in figure [ fig : n2phide ] , right , we show the absolute differential asymmetries @xmath121 and @xmath122 for lower cuts on the jet energy , @xmath123 ( plain ) and @xmath124 ( dashed ) . the dependence on the jet angle @xmath23 has been integrated out . since the asymmetry in the @xmath6 channel has an infrared divergence for @xmath45 , the main contribution to the asymmetry stems from small jet energies close to the cut . the range of @xmath69 is therefore limited to small values @xmath125 ( see figure [ fig : n2phide ] , left ) , such that @xmath126 depends strongly on the jet energy cut . on the contrary , in the @xmath7 channel , which is finite for @xmath45 , the distribution @xmath127 is rather flat . a jet energy cut has only a minor effect on the integrated asymmetry @xmath128 . these kinematic features will be important for the definition of cuts for the hadronic observables . the incline asymmetry @xmath81 in the @xmath6 channel is kinematically connected to the asymmetry @xmath129 in terms of the top - quark scattering angle @xmath130 . the latter has been measured as a forward - backward asymmetry in inclusive @xmath5 production at the tevatron and extensively studied in the literature . we compare both asymmetries in @xmath4 production and demonstrate that @xmath81 is superior to @xmath129 in its sensitivity to the differential cross section @xmath60 . the partonic top - angle asymmetry is defined by , as well as to @xmath19 . ] @xmath131 with the jet distribution @xmath132 in inclusive @xmath5 production , the jet direction is not detected . in this case , the top - angle asymmetry @xmath133 in the @xmath6 channel is equal to the forward - backward asymmetry @xmath134 0 ) - \hat{\sigma}(\cos\theta_t < 0)$ ] . in @xmath4 production , this relation holds true as long as the jet direction is integrated over , because the jet distribution in @xmath50 is symmetric . the differential distributions @xmath135 and @xmath136 are equal only for @xmath137 . for fixed angles @xmath138 , the forward - backward asymmetry @xmath139 of the top quark is to a significant extent due to momentum conservation in the final state , which blurs the charge asymmetry . channel , there is no relation between @xmath133 and @xmath139 . the top - quark forward - backward asymmetry originates in large part from the non - symmetric jet distribution . ] in @xmath4 production , we will thus refer to the top - angle asymmetry as defined in ( [ eq : top - angle - as ] ) and ( [ eq : top - angle - distribution ] ) . this definition is free from kinematical asymmetries caused by the jet direction . at the hadron level , the top - angle asymmetry is equal to the rapidity asymmetry defined later in ( [ eq : afbt ] ) , which has been measured in inclusive @xmath5 production by the tevatron experiments . the top- and antitop - quark angles @xmath130 and @xmath140 in the partonic cm frame can be expressed in terms of the variables introduced in ( [ eq : parametrization ] ) as @xmath141\,,\\\nonumber \cos\theta_{\bar t } & = & \sin\theta_j\cos\varphi\sin\bar{\xi } + \cos\theta_j\cos\bar{\xi}\,,\qquad\ \,\theta_{\bar t } \in [ 0,\pi]\ , . \end{aligned}\ ] ] in figure [ fig : as - parton - qq ] , left , we display the partonic incline asymmetry @xmath142 ( plain curve ) and the top - angle asymmetry @xmath133 ( dashed curve ) for the @xmath6 channel as functions of @xmath23 . we notice three characteristic features : 1 ) the incline asymmetry @xmath142 is largely independent from the jet direction . 2 ) the top - angle asymmetry @xmath133 reaches its maximum for central jet scattering , @xmath137 , where it equals the incline asymmetry . 3 ) the top - angle asymmetry vanishes for @xmath143 . let us discuss these features one by one . + 1 ) the flatness of the distribution @xmath144 has already been discussed in section [ subsec : aphi ] in terms of the coefficients @xmath82 . it is due to the fact that the inclination between the top quark and the incoming quark is defined perpendicular to the jet direction . the impact of the jet kinematics on the direction of the top and antitop quarks , which tends to wash out the asymmetry , is thereby largely eliminated . the incline asymmetry is thus at its maximum over ( almost ) the entire range of @xmath23 . + 2 ) to demonstrate that the incline asymmetry is equal to the top - angle ( or forward - backward ) asymmetry if the jet is emitted perpendicular to the beam axis , we consider both observables for @xmath137 . in this limit , ( [ eq : ctphi ] ) reduces to @xmath145 the variables @xmath146 and @xmath61 are now equal up to a factor @xmath147 , which is fixed by the energies @xmath21 and @xmath52 ( see ( [ eq : cosxi ] ) ) and is positive over its range @xmath148 $ ] . a forward - scattered top quark ( @xmath149 ) thus corresponds to a positive inclination @xmath150 , whereas a backward - scattered top ( @xmath151 ) induces a negative inclination @xmath152 . the antitop quark is emitted in the respective other hemisphere . this implies that for @xmath137 the incline asymmetry and the forward - backward asymmetry are equal , once the integration over @xmath61 and @xmath146 is performed , @xmath153 away from the region of central jet emission , the top - angle asymmetry @xmath129 is reduced . for @xmath154 , the dependence of @xmath146 and @xmath155 in ( [ eq : ctphi ] ) on @xmath37 and @xmath39 relaxes the correlation between the signs of @xmath61 and @xmath156 . for instance , events with a positive inclination @xmath150 can correspond to constellations with either @xmath157 or @xmath158 . the latter reduce the top - angle asymmetry @xmath129 with respect to the incline asymmetry @xmath81 . + 3 ) the top - angle asymmetry vanishes if the jet is emitted along the beam axis , corresponding to @xmath159 . the top - quark scattering angle is now given by @xmath160 for a forward jet ( @xmath161 ) , the variables @xmath146 and @xmath162 are equal ; for a backward jet ( @xmath163 ) , they are opposite in sign . at this limit , the correlation between @xmath130 and @xmath25 is lost . an inclination of the plane @xmath71 with respect to @xmath72 corresponds to an azimuthal rotation of the top quark around the beam axis , @xmath164 , which is orthogonal to the top - quark scattering angle @xmath130 . the differential charge asymmetry takes the simple form @xmath165 since @xmath146 and @xmath155 can be expressed in terms of @xmath21 and @xmath52 via @xmath162 and @xmath166 , the asymmetry @xmath129 obviously vanishes for @xmath143 , once the integration over @xmath167 $ ] is performed . in other words , the top - angle asymmetry vanishes for jet emission in the beam direction , because the top and antitop angles @xmath130 and @xmath140 are orthogonal to the inclination angle @xmath25 . let us conclude our analysis by comparing the different asymmetries at the parton level in the @xmath6 and @xmath7 channels . in figures [ fig : as - parton - qq ] and [ fig : as - parton - qg ] , the incline asymmetry ( plain ) , energy asymmetry ( dotted ) and top - angle asymmetry ( dashed ) are shown as functions of the jet scattering angle @xmath23 in the partonic cm frame . to visualize the effect of the normalization to the symmetric cross section @xmath168 , we display both the charge - asymmetric cross section @xmath76 ( left panel ) and the normalized asymmetry @xmath169 ( right panel ) . in the @xmath6 channel , all asymmetries are either symmetric or antisymmetric with respect to @xmath170 , because the jet distribution is forward - backward symmetric . since the symmetric cross section is enhanced in the collinear region , the normalized asymmetries are suppressed for @xmath171 . the suppression is particularly strong for the incline asymmetry @xmath81 , where the charge - antisymmetric cross section @xmath172 is largely constant in @xmath23 . for the energy asymmetry @xmath109 and the top - angle asymmetry @xmath129 , where per definition @xmath173 and @xmath174 decrease for @xmath175 , the shape distortion by the normalization is less significant . the energy asymmetry is much smaller than the incline asymmetry and the top - angle asymmetry . its maximum amounts to @xmath176 for @xmath177 and @xmath178 for @xmath179 and @xmath180 . as has been discussed in section [ subsec : afb ] , the incline and top - angle asymmetries are equal at their maximum at @xmath137 , where @xmath181 . for @xmath182 , @xmath81 is enhanced with respect to @xmath129 over the entire range of the jet scattering angle . the incline asymmetry is therefore superior to the top - angle asymmetry as a probe of the charge asymmetry in @xmath50 . the maximum of the asymmetries at the parton level serves as a guideline for the upper bound on the expected hadronic observables . the size of the effective observables at the hadron level can be significantly reduced , dependent on the amount of charge - symmetric background and on the cuts imposed on the signal . in the @xmath7 channel , the jet distribution of the asymmetries is neither symmetric nor antisymmetric , due to the asymmetric kinematics of the process @xmath183 . since the quark - jet is preferentially emitted in the direction of the incident quark , the antisymmetric cross section @xmath76 is enhanced , but finite , for small angles @xmath23 . the enhancement of the symmetric cross section @xmath168 for @xmath184 , however , is stronger due to its collinear divergence . the maximum of the normalized asymmetry @xmath169 is thus reached for larger jet angles @xmath23 . the top - angle asymmetry in the @xmath7 channel is tiny . its maximum amounts to no more than @xmath185 at @xmath186 . the incline asymmetry exhibits a larger maximum of @xmath187 at @xmath188 . however , since the inclination angle @xmath25 in the @xmath7 cm frame is defined with respect to the quark - jet rather than the gluon , @xmath81 changes sign around @xmath189 , such that the integrated incline asymmetry is much smaller . the energy asymmetry is the appropriate observable of the charge asymmetry in the @xmath7 cm frame . its maximum is sizeable , @xmath190 , and occurs for jet angles close to @xmath137 . the latter feature is particularly useful , because ( as for the incline asymmetry ) it allows one to cut out the collinear divergence of the normalization @xmath75 by focusing on the region of central jet emission . we elaborate on the origin and size of the different asymmetries in the @xmath6 and @xmath7 channels in appendix [ app : boosted ] . in the following sections , we will mostly focus on the asymmetries @xmath81 for the @xmath6 channel and @xmath191 for the @xmath7 channel . in order to construct robust observables for hadron colliders , it is crucial to control the jet energy and angular distributions of the charge asymmetries . the purpose of our analysis in qcd at lo is to show the kinematical behavior of the asymmetries at a qualitative level . a numerically precise prediction requires qcd calculations beyond lo and the resummation of the logarithmic infrared and collinear divergences . this is particularly important , as hadronic observables will depend on experimental cuts on the jet energy and transverse momentum , which regularize the divergent regions of the phase space . a stronger cut on the jet scattering angle that projects on the central region around @xmath192 simultaneously enhances the asymmetries and ensures collinear - safe observables . at the lhc , further cuts on kinematic variables are needed to increase the sensitivity to the asymmetries . to this end , we investigate the dependence of @xmath81 and @xmath191 on the respective defining variables @xmath61 and @xmath69 . in figure [ fig : as - parton - decp ] , we show the partonic incline asymmetry @xmath81 in @xmath50 ( left panel ) and the energy asymmetry @xmath120 in @xmath51 ( right panel ) as functions of @xmath193 and @xmath194 . a cut on the transverse momentum of the jet , @xmath195 , has been applied to regularize the infrared and collinear divergences . plain curves correspond to observables integrated over the jet distribution , whereas dashed curves show the observables at their maxima , @xmath196 and @xmath197 . the asymmetries increase with @xmath193 and @xmath194 , respectively . this effect is maximized , if the jet is emitted perpendicular to the beam axis in the partonic cm frame , namely at @xmath198 . in this region , the cross section @xmath75 does not exhibit a collinear enhancement , such that the sensitivity of the normalized asymmetries @xmath199 to the properties of @xmath200 is higher than in the collinear regime around @xmath171 . the incline asymmetry @xmath81 in the @xmath6 channel increases almost linearly with @xmath193 , reflecting the fact that the differential asymmetric cross section @xmath48 is proportional to @xmath61 ( see ( [ eq : sigmaa ] ) ) . the maximum of @xmath81 is reached for @xmath201 , where the planes @xmath71 and @xmath72 coincide . this is also the region with the highest production rate of @xmath4 events . a lower cut on @xmath193 thus increases the asymmetry without too strong a reduction of the cross section . in hadron collisions at the tevatron and the lhc , the total cross section for @xmath4 production is dominated by processes with a partonic cm energy of @xmath202 . since the dependence of @xmath81 on @xmath29 is mild , the partonic results for @xmath83 from figure [ fig : as - parton - decp ] , left , already give a rough estimate of the incline asymmetry at hadron level . in particular , it can be observed that the incline asymmetry at the lhc can be significantly enhanced by suitably combined cuts on @xmath23 and @xmath193 . the situation is different for the energy asymmetry @xmath120 in the @xmath7 channel . for kinematical reasons , a large top - antitop energy difference @xmath194 implies a lower bound on the jet energy @xmath22 ( see figure [ fig : n2phide ] , left ) . a lower cut on @xmath194 thus enhances the energy asymmetry , but comes along with a significant reduction of the cross section . as is apparent in figure [ fig : as - parton - decp ] , right , the increase of @xmath120 with @xmath194 is much more pronounced for central jet emission around @xmath198 . a cut on the jet angle thus will be important to raise the energy asymmetry to an observable level at the lhc . with these partonic investigations at hand , we construct observables for the tevatron and lhc experiments that probe the charge asymmetry in both the @xmath6 and the @xmath7 channel . at the hadron level , the partonic charge asymmetries introduced in section [ sec : asymmetry ] need to be adjusted according to the proton - antiproton and proton - proton initial states , respectively . the hadronic cross sections @xmath203 are obtained from the partonic cross sections @xmath204 by a convolution with the parton distribution functions ( pdfs ) @xmath205 at the qcd factorization scale @xmath206 and the sum over all partonic initial states @xmath207 , @xmath208 here @xmath209 denotes the momentum fraction of the parton @xmath210 inside the nucleon @xmath211 , and @xmath29 and @xmath212 are the partonic and hadronic cm energies . for the quark - antiquark initial state @xmath213 , which dominates the top - quark pair production rate at the tevatron , the differential partonic charge - asymmetric cross section @xmath214 has been given in ( [ eq : sigmaa ] ) . at the lhc , the quark - gluon initial state @xmath215 will also be important . the boost of the partonic cm frame with respect to the laboratory frame will be expressed in terms of the rapidity of the top - antitop - jet system in the laboratory frame , @xmath216 . in our numerical analysis , the factorization scale is set equal to the top - quark mass , @xmath217 . all calculations are performed at lo qcd , using cteq6l1 pdfs @xcite and the corresponding value of the strong coupling constant , @xmath218 . the phase - space integration is performed numerically by means of the vegas monte carlo algorithm implemented in @xcite . unless stated otherwise , we will apply typical experimental cuts on the jet s transverse momentum , @xmath219 , and the jet rapidity in the laboratory frame , @xmath220 . for the tevatron analysis , we use @xmath195 , @xmath221 , and for the lhc we apply @xmath222 , @xmath223 , subsequently referred to as `` detector cuts '' . cuts on the jet angle @xmath23 will be expressed in terms of the partonic jet rapidity , @xmath224 the jet scattering angle in the parton frame can thus be accessed by measuring the difference of the rapidities @xmath220 and @xmath225 in the laboratory frame . proton - antiproton collisions at the tevatron primarily proceed through the partonic @xmath6 channel . for a cm energy of @xmath226 and detector cuts @xmath195 and @xmath221 , the contributions of the partonic states to the total cross section of @xmath4 production amount to @xmath227 ( @xmath6 ) , @xmath228 ( @xmath229 ) and @xmath230 ( @xmath3 ) at lo qcd . the feature of the @xmath6 initial state being antisymmetric under charge conjugation is thus largely preserved in proton - antiproton collisions at the hadron level . the hadronic incline asymmetry @xmath231 can be straightforwardly derived from the partonic asymmetry @xmath81 using ( [ eq : aphi ] ) and ( [ eq : hadronicxsec ] ) , @xmath232 in figure [ fig : aphi - tev ] , we display the incline asymmetry @xmath233 ( left panel , plain curve ) and the symmetric cross section @xmath234 ( right panel ) at the tevatron as functions of the jet angle @xmath23 . the maximum of @xmath235 is obtained for jets emitted perpendicular to the beam axis in the partonic cm frame , namely for @xmath137 . recall from section [ subsec : afb ] that the partonic asymmetry @xmath142 is largely independent from @xmath23 ( figure [ fig : as - parton - qq ] , left ) . the drop - off of the normalized incline asymmetry at the edges of the spectrum is mainly due to the collinear enhancement of the symmetric cross section @xmath234 for @xmath46 . a cut on @xmath236 implies a minimal jet energy of @xmath237 for @xmath238 and of @xmath239 for @xmath240 , suppressing the infrared enhancement of the cross section at the edges of the jet distribution . however , as the @xmath219 cut affects both @xmath241 and @xmath242 in a similar way , the effect on the normalized incline asymmetry @xmath231 is small . for comparison , we also show the top - angle asymmetry @xmath243 ( figure [ fig : aphi - tev ] , left , dashed ) , @xmath244 which corresponds to the asymmetry @xmath245 in terms of the top - antitop rapidity difference @xmath246 . independently from detector cuts and normalization effects , the definition of @xmath247 implies a drop - off of the asymmetry for @xmath175 ( see figure [ fig : as - parton - qq ] , left ) . therefore , @xmath231 is always larger than @xmath243 for any fixed jet angle @xmath248 . integrated over the spectrum , the total asymmetries amount to @xmath249 and @xmath250 for a cross section of @xmath251 . the enhancement amounts to @xmath252 for the applied detector cuts . in the region of central jet emission , the enhancement is less pronounced . + [ cols="^,^,^,^,^,^,^ " , ] at the lhc14 , higher luminosities allow us to set stronger cuts on the energy asymmetry and still achieve a good statistical significance . compared to lhc8 , the maximal asymmetry is only slightly reduced due to the larger @xmath3 background . in figure [ fig : aphi - ae - lhc14 ] , right , we display the energy asymmetry at the lhc14 for @xmath253 . an energy asymmetry of @xmath254 can be observed with a statistical significance of @xmath255 already for luminosities less than @xmath256 . with higher luminosities of about @xmath257 , it is possible to explore the region of large asymmetries around @xmath258 . from the statistical point of view , the energy asymmetry is thus the preferred observable to probe the charge asymmetry at the lhc14 . compared to the incline asymmetry , it is sizeable for strong cuts and therefore benefits even more from high statistics . in table [ tab : ae - lhc14-cuts ] , numerical examples are given for different sets of cuts . the three last columns show the maximum incline asymmetries for @xmath259 and @xmath260 obtained with @xmath261 . they amount to @xmath262 . the implicit constraints on the jet s transverse momentum @xmath219 from combined cuts on @xmath194 and @xmath263 are shown in line four . generically , the minimal @xmath219 at the lhc14 is higher than at the lhc8 , because the preferred cuts on @xmath194 and @xmath263 are much stronger . therefore , a strong detector cut on @xmath219 can be set without affecting the magnitude of the cross section and the asymmetry at the lhc14 . these examples demonstrate once more the good prospects of observing a sizeable energy asymmetry at the lhc14 . top - quark pair production in association with a hard jet provides a new way to access the charge asymmetry at hadron colliders . from a theoretical perspective , the charge asymmetry in @xmath4 production is complementary to the asymmetry in inclusive @xmath5 production , as both originate from the interplay of different contributions in qcd . from a phenomenological point of view , the additional jet is advantageous insofar as it allows us to construct new observables beyond the hitherto considered rapidity asymmetries . the discovery potential of the charge asymmetry via these new observables in @xmath4 production at the lhc is expected to be superior to inclusive @xmath5 production . to construct optimal observables , jet kinematics is crucial because it affects the top- and antitop - quark kinematics due to momentum conservation in the final state . furthermore , jet kinematics governs the cross section in the soft and collinear regime and thereby confines the observability of a charge asymmetry to specific phase space regions . in this work , we have presented a pair of new observables that respect the jet kinematics , the incline asymmetry and the energy asymmetry . these observables probe the charge asymmetry in @xmath4 production in both the @xmath6 and the @xmath7 channel . in the @xmath6 channel , the incline and the energy asymmetry are complementary , as they are sensitive to independent parts of the differential charge asymmetry . in the @xmath7 channel , this complementarity is lost because the @xmath7 initial state is neither symmetric nor antisymmetric under charge conjugation . an incline asymmetry can be defined in the @xmath5 rest frame . in practice , however , the energy asymmetry in the @xmath7 frame proves very useful . it is equivalent to an angular asymmetry of the top and the antitop quark with respect to the direction of quark - jet and therefore independent of the momentum constellation in the initial state . at the tevatron , the charge asymmetry in @xmath4 production is accessible via the incline asymmetry @xmath231 in the @xmath6 channel . due to the collinear enhancement of the cross section ( the symmetric background to the charge asymmetry ) , the incline asymmetry is maximized if the jet is emitted perpendicular to the beam axis in the @xmath6 cm frame . integrated over the jet angle , the incline asymmetry is sizeable , amounting to @xmath249 at lo . it can be increased further by focusing on the region of central jet emission with a cut on the partonic jet rapidity @xmath263 . the experimental sensitivity to the incline asymmetry is limited by the amount of data collected at the tevatron . yet , with an estimated efficiency of @xmath264 and an integrated luminosity of @xmath265 , a statistical significance of @xmath266 standard deviations is expected . the energy asymmetry , in turn , is tiny and therefore not observable at the tevatron . at the lhc , an observation of the incline asymmetry in the @xmath6 channel relies on knowing the direction of the incoming quark . the latter can be estimated by exploiting the boost of the final state , which reflects the large momentum fraction carried by the valence quark inside the proton . the resulting incline asymmetry @xmath267 , however , is strongly suppressed by the large charge - symmetric @xmath3 background . by combining cuts on three variables , one can efficiently raise @xmath267 to an observable level . a cut on the final - state boost @xmath268 enhances the fraction of @xmath6 events with respect to the @xmath3 background . the collinear region is suppressed as at the tevatron by selecting events with small jet rapidities @xmath263 . the charge - asymmetric cross section can eventually be enhanced by focusing on small inclination angles @xmath25 . at the lhc8 with the full data set of @xmath269 , a significance of @xmath270 standard deviations can be obtained for moderate cuts on the three above - mentioned variables , corresponding to @xmath271 . at the lhc14 with an increased luminosity of @xmath272 , an incline asymmetry of almost @xmath273 can be observed with a significance of @xmath274 standard deviations . the energy asymmetry is the first observable that allows us to probe the charge asymmetry in the @xmath7 channel at the lhc . exploring the @xmath7 channel is particularly useful , because the @xmath3 suppression of the asymmetry is much milder than for the @xmath6 channel . importantly , the definition of the energy asymmetry does not require the determination of the quark direction in the initial state . at the lhc8 , with suitable cuts on @xmath275 , @xmath263 , and on the top - antitop energy difference @xmath69 , the energy asymmetry reaches a maximum of @xmath276 at lo . this maximum is twice as large as for the incline asymmetry , reflecting the increased parton luminosity of @xmath7 with respect to @xmath6 initial states . the limited statistics at lhc8 confine the statistical significance to at most @xmath277 standard deviations , corresponding to an energy asymmetry of @xmath278 . at the lhc14 , the maximal asymmetry is slightly lowered to @xmath279 due to the increased @xmath3 background . an observation of this maximum asymmetry with a significance of @xmath280 standard deviations requires @xmath272 of data . obviously , the prospects to observe a large energy asymmetry at the lhc14 increase with the accumulation of more data . compared to the so far considered rapidity asymmetries , the new incline and energy asymmetries improve the accessibility of the charge asymmetry in @xmath4 production in two ways . the incline asymmetry can be considered as a refined rapidity asymmetry that takes the jet kinematics into account . both observables are equal for central jet emission . integrated over the jet angular distribution , the incline asymmetry is consistently larger than the rapidity asymmetry . the energy asymmetry in addition provides us with access to the hitherto unexplored charge asymmetry in the @xmath7 channel . we suggest a measurement of both the incline asymmetry as well as the energy asymmetry in @xmath4 production at the lhc , in order to obtain complementary results for the charge asymmetry in the @xmath6 and the @xmath7 channel . with the sizeable energy asymmetry , @xmath4 production is likely to become the discovery channel of the charge asymmetry at the lhc , given the limited prospects of the rapidity asymmetries in inclusive @xmath5 production . concerning the numerical predictions in this work , a caveat is in order . since nlo corrections to the charge asymmetry in @xmath4 production are important , an analysis beyond lo is necessary to provide robust qcd predictions for the new observables . the theory of @xmath4 production and top - quark decay at nlo is well understood from investigations of the rapidity asymmetries . to what extent the incline and energy asymmetries are affected by nlo corrections is an interesting open question . we thank rohini godbole , kirill melnikov , and markus schulze for interesting discussions and hubert spiesberger for helpful comments on the manuscript . this work was supported by the initiative and networking fund of the helmholtz association , contract ha-101 ( ` physics at the terascale ' ) , by the research center ` elementary forces and mathematical foundations ' of the johannes gutenberg - universitt mainz , and by the national science foundation , grant phy-1212635 . in this appendix , we demonstrate that the energy asymmetry and the top - angle asymmetry in the channels @xmath18 and @xmath19 are related by a boost into the @xmath5 rest frame . this connection is due to the fact that both channels are related by parton crossing @xmath281p_3 $ ] . and @xmath12 are assigned as in section [ subsec : kinematics ] . ] the energy asymmetry in either channel is equal to a forward - backward asymmetry of parton @xmath280 ( the jet ) in the @xmath5 rest frame . the top - angle asymmetry in the respective other channel is approximately equal to the forward - backward asymmetry of parton @xmath282 ( an incoming antiquark or gluon ) in the @xmath5 rest frame . based on this relation , it is possible to understand the magnitude of the respective asymmetries . it will become clear why in the @xmath6 channel the top - angle asymmetry is so much larger than the energy asymmetry , whereas the situation is reversed in the @xmath7 channel . we start with an observation made in section [ subsec : aenergy ] : the energy asymmetry @xmath128 as defined in ( [ eq : sigmaa - energy ] ) is ( up to a sign ) equivalent to an angular asymmetry @xmath283 of the top and antitop quarks with respect to the jet direction , defined in ( [ eq : energy - angular ] ) . we can also express @xmath283 as a rapidity asymmetry with respect to the jet momentum @xmath284 , @xmath285 where the top and antitop rapidities with respect to @xmath284 are given by @xmath286since the rapidity difference @xmath287 is invariant under boosts along @xmath284 and since @xmath288 , it is convenient to boost the system into the @xmath5 rest frame , where @xmath289 . in the @xmath5 rest frame , the rapidity asymmetry @xmath290 appears as a forward - backward asymmetry of the jet with respect to the top - quark direction , @xmath291 where @xmath292 denotes the scattering angle of parton @xmath280 with respect to the top - quark momentum in the @xmath5 rest frame . the energy asymmetry @xmath128 is thus ( up to a sign ) equal to the forward - backward asymmetry of the jet in the @xmath5 rest frame , @xmath293 . we illustrate this result in figure [ fig : a - diff_restframes ] , where we display the jet angular distributions of the asymmetries for the @xmath6 channel ( left ) and the @xmath7 channel ( right ) . the energy asymmetry @xmath128 in the partonic cm frame ( left , plain / right , dashed ) and the forward - backward asymmetry @xmath294 in the @xmath5 rest frame ( left , bold dotted / right , fine dotted ) coincide . the relation @xmath295 holds in both the @xmath6 and the @xmath7 channel , since it is independent from the partonic origin of the jet , but only based on kinematic relations between the final - state momenta . whether the energy asymmetry in the partonic cm frame or the forward - backward asymmetry in the @xmath5 rest frame is more appropriate for a measurement remains to be examined on an experimental level . we will now establish the relation between the energy asymmetry @xmath128 and the top - angle asymmetry @xmath133 in different parton channels by using the forward - backward asymmetry in the @xmath5 rest frame . the top - angle asymmetry , as defined in ( [ eq : top - angle - as ] ) and ( [ eq : top - angle - distribution ] ) , can be considered as the asymmetry of the top- and antitop - quark rapidities with respect to the incident parton @xmath296 . due to the invariance under boosts along @xmath90 , this rapidity asymmetry is closely related to the forward - backward asymmetry of parton @xmath282 in the @xmath5 rest frame , @xmath297 . the asymmetry @xmath297 is defined as in ( [ eq : fb - as - ttb ] ) by changing @xmath298 . the difference between @xmath133 and @xmath297 is due to the imbalance of the top and antitop s transverse momenta with respect to the beam axis , caused by the presence of the jet in the final state . disregarding this small difference , we can eventually express the energy asymmetry in one channel and the top - angle asymmetry in the other channel in terms of forward - backward asymmetries in the @xmath5 rest frame , which are related by parton crossing , @xmath299 these results are visualized in figure [ fig : a - diff_restframes ] . the energy asymmetry @xmath128 in the @xmath6 channel ( left , plain ) and the top - angle asymmetry @xmath133 in the @xmath7 channel ( right , plain ) can both be related to the forward - backward asymmetry of the gluon in the @xmath5 rest frame ( bold dotted ) . the smallness of the top - angle asymmetry in the @xmath7 channel has been mentioned previously for inclusive top - pair production in the literature . it can now be ascribed to the small energy asymmetry in the @xmath6 channel , since both asymmetries are related by parton crossing . the different shapes of the jet angular distributions are due to the kinematic configurations : in the @xmath6 channel , @xmath23 is the angle of the gluon - jet with respect to the incident quark , whereas in the @xmath7 channel , it is the angle of the quark - jet . in particular , the energy asymmetry in the @xmath6 channel is antisymmetric and changes its sign at @xmath137 . this is not the case for the top - angle asymmetry in the @xmath7 channel , where the angle between the incident quark and the gluon is fixed to @xmath105 . the magnitudes of @xmath300 and @xmath301 differ mainly because of the soft enhancement for @xmath45 in the @xmath6 channel , which is absent in the @xmath7 channel . the same connection holds true for the energy asymmetry in the @xmath7 channel ( right , dashed ) and the top - angle asymmetry in the @xmath6 channel ( left , dashed ) . both asymmetries can be traced back to the forward - backward asymmetry of the quark in the @xmath5 rest frame ( fine dotted ) . thus , the sizeable energy asymmetry in the @xmath7 frame is due to the fact that the top - angle asymmetry in the @xmath6 channel ( which is connected to the incline asymmetry ) provides the dominant contribution to the charge asymmetry . these findings confirm that the incline asymmetry in the @xmath6 channel and the energy asymmetry in the @xmath7 channel capture most of the differential charge asymmetry in both channels . one might wonder , why we derived the correspondence between the energy asymmetry and the top - angle asymmetry , rather than the incline asymmetry . as was mentioned at the end of section [ subsec : kinematics ] , the incline asymmetry and the energy asymmetry form a `` perfect match '' , in the sense that they probe complementary parts of the charge asymmetry in the @xmath6 channel . it is not possible to translate this complementarity to the @xmath7 channel . a lorentz boost into the @xmath302 rest frame , which would mimic a @xmath6 initial state , is kinematically prohibited . the incline asymmetry can be defined in the @xmath5 rest frame via @xmath303 , where @xmath304 is the inclination angle between the plane spanned by the top quark and the incoming gluon and the plane spanned by the quark - jet and the incoming gluon . due to the asymmetric jet kinematics in the @xmath7 channel , the jet distribution of @xmath305 is not flat as in the @xmath6 channel , but peaks in the forward direction . the magnitude of @xmath305 is similar to the energy asymmetry @xmath128 in the @xmath7 channel . we consider the energy asymmetry a more suitable observable than the incline asymmetry in the @xmath5 rest frame , because the former does not rely on the collinear region . furthermore , the energy asymmetry can be easily formulated in the partonic cm frame and does not rely on the direction of the incoming gluon , contrary to the incline asymmetry in the @xmath5 rest frame . t. aaltonen _ et al . _ [ cdf collaboration ] , arxiv:1211.1003 [ hep - ex ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . d * 84 * , 112005 ( 2011 ) [ arxiv:1107.4995 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , eur . j. c * 72 * , 2039 ( 2012 ) [ arxiv:1203.4211 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , atlas - conf-2012 - 057 . s. berge and s. westhoff , d * 86 * , 094036 ( 2012 ) [ arxiv:1208.4104 [ hep - ph ] ] . j. h. khn and g. rodrigo , phys . d * 59 * , 054017 ( 1999 ) [ hep - ph/9807420 ] . s. dittmaier , p. uwer and s. weinzierl , phys . * 98 * , 262002 ( 2007 ) [ hep - ph/0703120 [ hep - ph ] ] . s. dittmaier , p. uwer and s. weinzierl , eur . j. c * 59 * , 625 ( 2009 ) [ arxiv:0810.0452 [ hep - ph ] ] . k. melnikov and m. schulze , nucl . b * 840 * , 129 ( 2010 ) [ arxiv:1004.3284 [ hep - ph ] ] . k. melnikov , a. scharf and m. schulze , phys . d * 85 * , 054002 ( 2012 ) [ arxiv:1111.4991 [ hep - ph ] ] . a. kardos , c. papadopoulos and z. trocsanyi , phys . b * 705 * , 76 ( 2011 ) [ arxiv:1101.2672 [ hep - ph ] ] . s. alioli , s .- o . moch and p. uwer , jhep * 1201 * , 137 ( 2012 ) [ arxiv:1110.5251 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , cdf public note 9850 ( 2009 ) . t. hahn and m. perez - victoria , comput . commun . * 118 * , 153 ( 1999 ) [ hep - ph/9807565 ] . parameterization in user s guide to formcalc 3.1 , available from . j. pumplin , d. r. stump , j. huston , h. l. lai , p. m. nadolsky and w. k. tung , jhep * 0207 * , 012 ( 2002 ) [ hep - ph/0201195 ] .

we emphasize the sizeable effects of absorption on high - energy ` soft ' processes , and , hence , the necessity to include multi - pomeron - pomeron interactions in the usual multi - channel eikonal description . we present a model which includes a complete set of the multi - pomeron vertices and which accounts for the diffusion in both , the impact parameter and , of the parton during its evolution in rapidity . we tune the model to the available data for soft processes in the cern - isr to tevatron energy range . we make predictions for ` soft ' observables at the lhc . plus 2 mm minus 2 mm 23.0 cm 17.0 cm -1.0 in -42pt ippp/08/92 + dcpt/08/184 + + * soft processes at the lhc * + m.g . ryskin , a.d . martin and v.a . khoze + institute for particle physics phenomenology , university of durham , durham , dh1 3le + petersburg nuclear physics institute , gatchina , st . petersburg , 188300 , russia

0812.2407

there are three main reasons for revisiting soft @xmath4 high energy interactions at this time . * a. * this paper is concerned with the description of the high energy behaviour of `` soft '' observables such as @xmath5 particle multiplicities etc . in terms of basic physics . this physics predated qcd and is sometimes regarded as the dark age of strong interactions . however , it is unfair to call this the dark age . we had a successful description of these processes in terms of the exchange of regge trajectories linked to particle states in the crossed channels @xcite . the dominant exchange at high energy is the pomeron , and we have gribov s reggeon calculus @xcite to account for the multi - pomeron contributions . however the available data did not reach high enough energy to distinguish between the different scenarios @xcite for the high - energy behaviour of the interaction amplitude @xcite . in the ` weak coupling ' scenario the total cross section @xmath6 , and in order not to violate unitarity , and to satisfy the inequality @xmath7 the triple - pomeron vertex must vanish when @xmath8 , that is the triple - pomeron coupling @xmath9 . in this case , the large logarithm coming from the integration over the mass of the system produced in diffractive dissociation ( @xmath10 ) is compensated by the small value of the mean momentum transferred through the pomeron , @xmath11 . on the other hand , in the ` strong coupling ' scenario , where @xmath12 with @xmath13 , the inequality ( [ eq:1 ] ) is provided by a small value of the rapidity gap survival factor @xmath14 which decreases with energy . the present diffractive data are better described within the ` strong coupling ' approach @xcite , and in this paper we shall give predictions for the lhc for this scenario . however , the possibility of the ` weak coupling ' scenario is not completely excluded yet . therefore , it is quite important to study the different channels of diffractive dissociation at the lhc in order to reach a final conclusion and to fix the parameters of the model for high - energy soft interactions . so the first motivation is the intrinsic interest in obtaining a reliable , self - consistent model for soft interactions , which may be illuminated by data from the lhc @xcite . * b. * in turn , obtaining a reliable model will be of great value for predictions of the gross features of soft interactions . in particular , it is essential for understanding the structure of the underlying events at the lhc . * c. * the third reason for studying soft interactions arises because it may not be an easy task to identify the production of a new object at the lhc when it is accompanied by hundreds of other particles emitted in the same event . for the detailed study of the new object , @xmath15 , it may be better to select the few , very clean , events with the large rapidity gaps ( lrg ) on either side of the new object , see , for example , @xcite . that is to observe the exclusive process @xmath16 . in such a central exclusive process ( cep ) the mass of @xmath15 can be measured with very good accuracy ( @xmath17 gev ) by the missing - mass method by detecting the outgoing very forward protons . moreover , a specific @xmath18 selection rule @xcite reduces the background and also greatly simplifies the spin - parity analysis of @xmath15 . however , the cep cross section is strongly suppressed by the small survival factor , @xmath19 , of the rapidity gaps . thus we need a reliable model of soft interactions to evaluate the corresponding value of @xmath14 @xcite . moreover , it is important to have a model which contains @xmath20-channel components of different size in order to evaluate the possible effects of the ` soft - hard factorisation ' breaking . this is the subject of the following paper @xcite . first , we briefly recall the relevant features of the single - channel eikonal model . that is , we focus on elastic unitarity . at high energy the position of the fast particle in the impact parameter , @xmath21 , plane is to a good approximation frozen during the interaction , since the value of @xmath21 is fixed by the orbital angular momentum @xmath22 of the incoming hadron . there is no mixture between the partial wave amplitudes with different @xmath23 . the well known solution of the elastic unitarity equation , 2 imt_el(s , b ) = |t_el(s , b)|^2 + g_inel(s , b ) , [ eq : a1 ] may be written in terms of the phase shift @xmath24 as @xmath25 the presence of inelastic channels , given by @xmath26 in ( [ eq : a1 ] ) , leads to the phase @xmath24 having an imaginary part . that is , @xmath24 becomes a complex number . moreover , at high energies we know that re@xmath27im@xmath28 is small . now , in the framework of the eikonal model , the elastic amplitude , t_el = i(1-e^-/2 ) is obtained by the sum of regge - exchange diagrams , which is equivalent to the iteration of the elastic unitarity equation , ( [ eq : a1 ] ) , as shown in fig . [ fig : aa ] . in other words , @xmath29-channel elastic unitarity gives t_el(s , b ) & = & 1-e^-/2 [ eq : elastamp ] + _ el(s , b ) & = & ( 1-e^-/2)^2 , [ eq : el ] + _ inel(s , b ) & = & 1- e^- , [ eq : inel ] where @xmath30 is called the opacity ( optical density ) or eikonal is called the eikonal . ] ; @xmath31 plays the role of @xmath32 . it is the fourier transform of the two - particle ( @xmath29-channel ) _ irreducible _ amplitude , @xmath33 . that is and @xmath34 to denote vectors in the transverse plane . ] @xmath35 where @xmath36 , and where the amplitude is normalized by the relation @xmath37 . from ( [ eq : inel ] ) , we see that @xmath38 is the probability that no inelastic scattering occurs at impact parameter @xmath21 . after summing over @xmath21 ( that is , all partial waves ) we obtain the total , elastic and inelastic cross sections _ tot & = & 2d^2b imt_el(s , b ) = 2d^2b ( 1-e^-/2 ) [ eq : ot ] + _ el & = & d^2b|t_el(s , b)|^2 = d^2b ( 1-e^-/2)^2 + _ inel & = & d^2b = d^2b ( 1-e^- ) . below we neglect the imaginary part of @xmath39 , apart from the contribution of secondary reggeons to high - mass diffractive dissociation . at high energies , the ratio re@xmath40 is small , and can be evaluated via a dispersion relation . so much for elastic diffraction . now we turn to inelastic diffraction , which is a consequence of the _ internal structure _ of hadrons . besides the pure elastic two - particle intermediate states shown in fig . [ fig : aa ] , there is the possibility of proton excitation , @xmath41 . as a rule such excitations are not included in the opacity @xmath39 , but are treated separately . this is simplest to describe at high energies , where the lifetime of the fluctuations of the fast proton is large , @xmath42 , and during these time intervals the corresponding fock states can be considered as ` frozen ' . each constituent of the proton can undergo scattering and thus destroy the coherence of the fluctuations . as a consequence , the outgoing superposition of states will be different from the incident particle , and will most likely contain multiparticle states , so we will have _ inelastic _ , as well as elastic , diffraction . to discuss inelastic diffraction , it is convenient to follow good and walker @xcite , and to introduce states @xmath43 which diagonalize the @xmath44 matrix . such eigenstates only undergo elastic scattering . since there are no off - diagonal transitions , _ j|t|_k= 0 jk , a state @xmath45 can not diffractively dissociate into a state @xmath46 . we have noted that this is not , in general , true for hadronic states , which are not eigenstates of the @xmath47-matrix , that is of @xmath44 . to account for the internal structure of the hadronic states , we have to enlarge the set of intermediate states , from just the single elastic channel , and to introduce a multichannel eikonal . we will consider such an example below , but first let us express the cross section in terms of the probability amplitudes @xmath48 of the hadronic process proceeding via the various diffractive eigenstates describes the probability that the diffractive eigenstate @xmath43 is not absorbed in the interaction . later we will see that the rapidity gap survival factors , @xmath14 , can be described in terms of such eikonal exponents . ] @xmath43 . let us denote the orthogonal matrix which diagonalizes @xmath49 by @xmath50 , so that [ eq : b3 ] imt = at^da^t _ j |t^d| _ k = t^d_k _ jk . [ eq : diag ] now consider the diffractive dissociation of an arbitrary incoming state [ eq : b4 ] | j = _ k a_jk | _ k . the elastic scattering amplitude for this state satisfies [ eq : b5 ] j |im t| j = _ k @xmath51 and where the brackets of @xmath52 mean that we take the average of @xmath53 over the initial probability distribution of diffractive eigenstates . after the diffractive scattering described by @xmath54 , the final state @xmath55 will , in general , be a different superposition of eigenstates from that of @xmath56 , which was shown in ( [ eq : b4 ] ) . at high energies we may neglect the real parts of the diffractive amplitudes . then , for cross sections at a given impact parameter @xmath21 , we have [ eq : b6 ] & = & 2 j |t| j = 2 _ k |a_jk|^2 t^d_k = 2 t^d + & & + & = & | j |t| j |^2 = ( _ k |a_jk|^2 t^d_k ) ^2 = t^d ^2 + & & + & = & _ k | _ k |t| j |^2 = _ k |a_jk|^2 ( t^d_k)^2 = ( t^d)^2 . it follows that the cross section for the single diffractive dissociation of a proton , [ eq : b7 ] = ( t^d)^2 - t^d ^2 , is given by the statistical dispersion in the absorption probabilities of the diffractive eigenstates . here the average is taken over the components @xmath45 of the incoming proton which dissociates . if the averages are taken over the components of both of the incoming particles , then in ( [ eq : b7 ] ) we must introduce a second index on @xmath53 , that is @xmath57 , and sum over @xmath45 and @xmath58 . in this case the sum is the cross section for single and double dissociation . at first sight , enlarging the number of eigenstates @xmath59 we may include even high - mass proton dissociation . however here we face the problem of double counting when the partons originating from dissociation of the beam and ` target ' initial protons overlap in rapidities . for this reason high - mass dissociation is usually described by `` enhanced '' multi - pomeron diagrams . the first and simplest is the triple - pomeron graph , see fig . [ fig:2lad ] below . the total and elastic proton - proton cross sections are usually described in terms of an eikonal model , which automatically satisfies @xmath29-channel elastic unitarity . to account for the possibility of excitation of the initial proton , that is for two - particle intermediate states with the proton replaced by @xmath60 , we use the good - walker formalism @xcite . already at tevatron energies the absorptive correction to the elastic amplitude , due to elastic eikonal rescattering , is not negligible ; it is about @xmath61% in comparison with the simple one pomeron exchange . after accounting for low - mass proton excitations ( that is @xmath60 s in the intermediate states ) the correction becomes twice larger ( that is , about @xmath62% ) . indeed , the possibility of proton excitation means that we have to include additional inelastic channels which were not accounted for in the irredicible amplitude @xmath15 of ( [ eq : append9 ] ) . this enlarges the probability of absorption for the elastic channel , that is the effective opacity @xmath39 . in terms of the good - walker formalism , the stronger absorption follows from the inequality with @xmath63 . it follows that @xmath64 which , since the second term is negative , is less than @xmath65 . ] _ ke^-_k < _ ke^-_k , where we average over the diffractive ( good - walker ) eigenstates . next , in order to describe high - mass diffractive dissociation , @xmath66 , we have to include an extra factor of 2 from the agk cutting rules @xcite . thus , the absorptive effects in the triple - regge domain are expected to be quite large . the previous triple - regge analyses ( see , for example , @xcite ) did not allow for absorptive corrections and the resulting triple - regge couplings must be regarded , not as bare vertices , but as effective couplings embodying the absorptive effects @xcite . since the inelastic cross section ( and , therefore , the absorptive corrections ) expected at the lhc are more than twice as large as that observed at fixed - target and cern - isr energies , the old triple - regge vertices can not be used to predict the diffractive cross sections at the lhc . thus , it is necessary to perform a new triple - regge analysis that includes the absorptive effects explicitly . such an analysis has recently been performed @xcite using the fixed - target fnal , cern - isr and tevatron data that are available in the triple - regge region . the ` @xmath67 ' , ` @xmath68 ' , ` @xmath69 ' , ` @xmath70 ' and @xmath71 contributions , were included assuming either the ` strong ' or ` weak ' coupling scenarios for the behaviour of the triple - pomeron vertex . to account for the absorptive corrections a two - channel ( good - walker ) eikonal model was used , which describes well the total , @xmath72 , and elastic , @xmath73 , @xmath4 and @xmath74 cross sections . in the ` strong ' coupling case , a good @xmath75dof=167/(210 - 8)=0.83 was obtained . in comparison with the old triple - regge analysis @xcite , a twice larger relative contribution of the ` @xmath68 ' term was found . this is mainly due to the inclusion of the higher - energy tevatron data in the analysis . since the absorptive effects are included explicitly , the extracted values of the triple - reggeon vertices are now much closer to the _ bare _ triple - regge couplings . in particular , the value @xmath76 is consistent with a reasonable extrapolation of the perturbative bfkl pomeron vertex to the low scale region @xcite ; here @xmath77 is the pomeron - proton coupling . note also that these values of the ` @xmath67 ' and ` @xmath68 ' vertices allow a good description of the hera data @xcite on inelastic @xmath78 photoproduction , @xmath79 , where the screening corrections are rather small . the ` weak ' coupling scenario leads to a larger @xmath75dof=1.4 and to a worse description of the @xmath80 process at the lowest values of @xmath20 . at the lhc energy the ` weak ' coupling fit predicts about 3 times smaller inclusive cross section @xmath81 at @xmath82 and low @xmath20 in comparison with that predicted in the ` strong ' coupling case . note that the effects due to the triple - pomeron vertex ( [ eq:2 ] ) are rather large . indeed , the contribution caused by such vertices is enhanced by the logarithmically large phase space available in rapidity . in particular , the total cross section of high - mass dissociation is roughly with @xmath83 . ] of the form _ sd = ~ s _ el , where @xmath84 reflects the suppression of high - mass dissociation in comparison with elastic scattering and the ln@xmath29 factor comes from the integration @xmath85 . thus actually we deal with the parameter @xmath86 at collider energies . for each fixed rapidity interval the probability of high - mass dissociation ( or , in other words , the contribution due to the triple - pomeron vertex ) is relatively small . however the cumulative effect in the complete interaction amplitude is enhanced by the large phase space available in rapidity . as a consequence , the contribution of the corresponding , so - called ` enhanced ' , diagrams , with a few vertices , is not negligible . moreover , we can not expect that more complicated multi - pomeron interactions , driven by the @xmath87 vertices , which describe the transition of @xmath88 to @xmath89 pomerons of fig . [ fig : gnm ] , will not affect the final result . it looks more reasonable to assume that @xmath90 than to assume that @xmath91 for any @xmath92 . thus we need a model which accounts for the possibility of multi - pomeron interactions ( with arbitrary @xmath88 and @xmath89 ) . in this paper we extend and develop the _ partonic _ approach of ref . @xcite . while the eikonal formalism describes the rescattering of the incoming fast particles , the enhanced multi - pomeron diagrams represent the rescattering of the intermediate partons in the ladder ( feynman diagram ) which describes the pomeron - exchange amplitude . .,height=113 ] indeed , we start with the generation of ladder - type structure of ` elastic ' bare pomeron exchange amplitude . it may be generated by the evolution equation @xcite ( in rapidity , @xmath93 , space ) @xmath94 where @xmath21 is the two - dimensional vector in impact parameter space . @xmath95 is the probability to emit new intermediate partons ( denoted @xmath96 ) within unit rapidity interval ; it is analogous to the splitting function of dglap evolution . the impact parameter of parton @xmath96 is not frozen in the evolution . at each step @xmath21 can be changed by a constant amount @xmath97 in any direction , leading to the diffusion represented by the second term on the right - hand side of ( [ eq:3 ] ) where @xmath98 plays the role of the diffusion coefficient @xcite . the evolution is shown symbolically in fig . [ fig : dt ] . the solution of ( [ eq:3 ] ) is ( y , b)=_0(y - b^2/4 y)/4 y. [ eq : yb ] it represents the opacity ( at point @xmath99 ) , corresponding to the incoming particle placed at @xmath100 and @xmath101 . it may be helpful to explain why ( [ eq:3 ] ) was written in terms of the opacity @xmath39 . first we note that the discontinuity of the amplitude generated by ( [ eq:3 ] ) does not contain a two - particle @xmath29-channel intermediate state ; it corresponds to a pure inelastic high multiplicity process , see fig . [ fig : mult ] . due to elastic unitarity , ( [ eq : a1 ] ) this inelastic interaction leads to elastic @xmath4 scattering . thus we have to put the solution of ( [ eq:3 ] ) into the eikonal formulae of ( [ eq : elastamp])-([eq : inel ] ) , as it does not contain two - particle @xmath29-channel states . in momentum space the solution ( [ eq : yb ] ) corresponds to the amplitude a(s , t ) = a_0 s^1++ t. that is to the bare pomeron exchange amplitude , where the pomeron trajectory has intercept @xmath102 and slope @xmath98 . a multi - pomeron enhanced contribution arises from the absorption of intermediate @xmath29-channel partons @xmath96 during the evolution of @xmath39 in @xmath93 . the simplest example is the triple - pomeron diagram in which parton @xmath96 undergoes an extra rescattering with the target parton @xmath45 , as shown in fig . [ fig:2lad ] . allowing for many rescatterings , we have to sum over different numbers of ladders between partons @xmath96 and @xmath45 . assuming an eikonal form for the multi - pomeron - proton vertex , it is natural to replace ( [ eq:3 ] ) by = e^-_k/2(+)_k(y , b ) [ eq : teik ] where the ` opacity ' @xmath103 describes the transparency of the target @xmath45 . as we are dealing with the elastic _ amplitude _ we use @xmath104 and not @xmath105 . the coefficient @xmath84 arises since parton @xmath96 will have a different absorption cross section from that of eigenstate @xmath58 . naively , we may assume that the beam @xmath58 contains a number @xmath106 of partons . the factor @xmath104 generates multi - pomeron vertices of the form @xmath107 where @xmath87 is defined in fig . [ fig : gnm ] . even though @xmath108 , the role of the factor @xmath104 is not negligible , since the suppression effect is accumulated throughout the evolution . for instance , if @xmath109 the full absorptive correction is given by the product @xmath110 , where the small value of @xmath84 is compensated by the large rapidity interval @xmath111 . in terms of regge diagrams , ( [ eq : teik ] ) sums up the system of fan diagrams in which any number @xmath89 of `` lower '' pomerons couples to a fan vertex @xmath112 . such multi - pomeron diagrams are called `` enhanced '' , since their contribution is enhanced , in comparison with the eikonal diagrams , by the large available rapidity interval @xmath111 . in order to include the rescattering with the beam @xmath58 we replace ( [ eq : teik ] ) by @xmath113 the final term in the exponent is the opacity of the beam @xmath58 , which depends on the rapidity interval @xmath114 , with @xmath115 . the equation for the opacity @xmath116 has an analogous form @xmath117 in which we now evolve in the opposite direction starting from the boundary condition @xmath118 at @xmath119 . recall that the fit to the data in the triple - reggeon domain indicated a very small ( consistent with zero ) @xmath20-slope of all the triple - reggeon vertices @xcite . thus , as the size of the multi - reggeon vertices are negligible in comparison with the size of the incoming hadron , we may write the absorptive corrections ( that is , the exponential factors on the right - hand - side of ( [ eq:4],[eq:5 ] ) ) such that the opacities @xmath120 are taken at the same point in the impact parameter plane @xmath21 . since the intermediate parton may be absorbed by the interaction with the particles ( partons ) from the wave function of both the beam or target hadron , we now need to solve the two equations , ( [ eq:4 ] ) and ( [ eq:5 ] ) . this is done iteratively . moreover , note that the opacities @xmath39 now depend on two vectors in impact parameter space - the separation @xmath121 between the position of the intermediate parton @xmath96 and the beam hadron , and the separation @xmath122 between @xmath96 and the target hadron . the argument @xmath21 in ( [ eq:4],[eq:5 ] ) now symbolically denotes both @xmath121 and @xmath122 . the resulting solution @xmath123 is then used in the eikonal formulae for the elastic amplitude , giving t_el(b)=1-(-(b = b_1-b_2)/2 ) . a more detailed description of the amplitude , and the cross sections of the different diffractive processes can be found in ref . @xcite . as mentioned above , as in @xcite , we use three diffractive components in the @xmath29-channel . in other words , we use a 3-channel eikonal for the rescattering of fast particles . the transverse size squared of each eigenstate is proportional to the corresponding absorptive cross section ; @xmath124 . that is , we assume that the parton density at the origin is the same for each eigenstate . the shape of the pomeron - nucleon vertex is parametrised by the form factor @xmath125 , whose fourier transform , @xmath126 , plays the role of the initial conditions for @xmath127 . however , now , we allow for a non - zero slope ( @xmath128 ) of the ( bare ) pomeron trajectory . a major development , of the model of @xcite , is that we use four different @xmath20-channel states , which we label @xmath50 : one for the secondary reggeon ( @xmath129 ) trajectory and three pomeron states ( @xmath130 ) to mimic the bfkl diffusion in the logarithm of parton transverse momentum , @xmath0 @xcite . to be precise , since the bfkl pomeron @xcite is not a pole in the complex @xmath46-plane , but a branch cut , we approximate the cut by three @xmath20-channel states of a different size . the typical values of @xmath131 in each of the three states is about @xmath132 gev , @xmath133 gev and @xmath134 gev . thus the system of evolution equations ( [ eq:4],[eq:5 ] ) is replaced by and @xmath103 depend on both subscripts @xmath58 and @xmath45 . here we keep only one subscript to distinguish the parent hadron for each active parton ( gluon ) . ] @xmath135/2 } \left(\delta^a+\alpha'_a\frac{d^2}{d^2b_1}\right)\omega^a_k(y,{\mathbf b}_1,{\mathbf b}_2 ) + v_{aa'}\omega^{a'}_k , \label{eq:6}\ ] ] @xmath136/2 } \left(\delta^a+\alpha'_a\frac{d^2}{d^2b_2}\right)\omega^a_i(y',{\mathbf b}_1,{\mathbf b}_2 ) + v_{aa'}\omega^{a'}_i , \label{eq:7}\ ] ] where @xmath137 and @xmath138 for @xmath139 , while for the secondary reggeon , @xmath140 , which is built of quarks , we take @xmath141 and @xmath142 , so that the last term @xmath143 is diagonal with @xmath144 to account for the spin @xmath145 nature of quarks . the key parameters which drive the evolution are the intercepts @xmath95 and the slopes @xmath98 . in general , each component @xmath50 may have different values of @xmath146 and @xmath147 . we discuss the values in section 5 . in the exponents , the opacities @xmath148 ( @xmath149 ) are actually the sum of the opacities @xmath150 ( @xmath151 ) with corresponding coefficients . namely @xmath152 @xmath153 @xmath154 @xmath155 we chose @xmath156 since , at the leading order , the probability of interaction of two components of different size ( @xmath131 ) is proportional to the ratio @xmath157 . we take @xmath158 since the third ( smallest size ) component collects _ all _ the higher @xmath131 contributions , and therefore here the mean value of @xmath131 is larger . for the screening of the secondary reggeon by the pomeron we take just the colour factor @xmath159 , as we assume that the secondary reggeon is composed of a @xmath20-channel quark - antiquark pair . finally the factors @xmath160 and @xmath161 were tuned to give a reasonable reproduction of the secondary reggeon contributions to the available @xmath162 data . thus , the value of the product @xmath163 is not large . ] . the transition factors @xmath164 between the different @xmath20-channel components are fixed by the properties of the bfkl equation . the only non - zero factors , apart from @xmath144 , are @xmath165 where @xmath166 @xmath167 is the density of partons emitted in the rapidity evolution of the @xmath20-channel component @xmath50 . the remaining transition factors were set to zero . that is v_p_3p_1=v_p1p_3=0 , v_ra=v_ar=0 , and v_aa=0 for a = a. for each good - walker @xmath29-channel component ( @xmath168 ) , the initial conditions are fixed by the parton ( matter ) distribution in the corresponding diffractive eigenstate @xmath169 where @xmath170 . the parametrisation @xmath171 was used for the pomeron , while for the secondary reggeon we chose the gaussian form @xmath172 . the relative couplings ( and the corresponding size ) of the components were taken to ( a ) reproduce the cross section of low - mass dissociation measured at the cern - isr @xcite , and ( b ) to make all three components quite different from each other ; @xmath173 , @xmath174 , @xmath175 . all the coefficients in the decomposition @xmath176 are taken to be @xmath177 ( with @xmath178 ) . to avoid possible double counting , and to fix the boundary between the ` low ' and ` high ' mass dissociation , we introduce a threshold @xmath179 in rapidity for the actual start of the evolution of ( [ eq:6],[eq:7 ] ) . that is we start the evolution at @xmath180 and not at @xmath101 . hence the available rapidity interval becomes @xmath181 . proton excitation ( dissociation ) which covers a rapidity interval larger than @xmath182 ( i.e. @xmath183 ) is called ` high - mass dissociation ' . it is natural to separate the different contributions in terms of rapidity , since in qcd the interference between the different diagrams for gluon radiation leads to angular ( rapidity ) ordering of emitted gluons , at least to leading log accuracy , but also in the form @xmath184 , where @xmath185 denotes the position of the edge of the rapidity gap . this may be possible using the forward shower counters proposed in @xcite . ] . note that the initial condition ( [ eq : inom ] ) is only valid for the secondary reggeon and for the large size pomeron component ( @xmath186 ) . for the smaller size pomeron components we use ^p_2_i(y=0,b)=^p_1_i(y=0,b)v__pp and ^p_3_i(y=0,b)=^p_1_i(y=0,b ) v__ppv__pp. of a high energy interaction.,height=151 ] to calculate the elastic amplitude we need the @xmath29-channel two - particle _ irreducible _ amplitudes for the scattering of the various diffractive eigenstates @xmath58 and @xmath45 , for given separations @xmath187 between the incoming protons . these are given by @xmath188 where @xmath189 , see fig . [ fig : fik ] . note that there is no integral gives the multiplicity . ] over @xmath93 . the convolution may be calculated at any rapidity @xmath93 , leading to the same result . given this effective ` @xmath190 eikonal ' , we can calculate the cross sections ( analogously to ( [ eq : elastamp])-([eq : inel ] ) ) . we obtain @xmath191 @xmath192 ^ 2 \label{eq : dsel}\ ] ] where @xmath193 , and @xmath194 ^ 2 \ ; . \label{eq : sel}\ ] ] for low - mass excitation of the beam proton we obtain @xmath195 @xmath196 \left[\int d^2be^{i{\mathbf q}_t\cdot{\mathbf b } } \left(1-e^{f_{ik}(y,{\mathbf b})/2 } \right)\right]\ ; , \label{eq : delsd}\ ] ] which has the symbolic structure shown in fig . [ fig : elsd ] . ) for low - mass dissociation of the ` beam ' diffractive eigenstate @xmath58.,height=75 ] strictly speaking , we may need a different diagonalisation matrix @xmath50 of ( [ eq : diag ] ) for the different @xmath20-channel exchanges . however , if the main difference between the diffractive eigenstates is due to the size and the impact parameter structure of the state , which is frozen for a fast hadron during the interaction , then it is justified to use the same eigenstates for any @xmath20-channel exchange , @xmath197 . the expression for the high - mass excitation is more complicated . the cross section for beam particle diffractive dissociation ( with the gap up to @xmath93 ) can be written using ( [ eq : elastamp])-([eq : inel ] ) . diffractive dissociation may be considered as the elastic scattering of intermediate parton @xmath96 caused by its absorption on the target , which is described by the factor @xmath198 . thus , in each impact parameter point @xmath34 the cross section for single dissociation is proportional to ( i ) the elastic @xmath199 cross section @xmath200 ; ( ii ) to the probability to find the intermediate parton @xmath96 in the interval @xmath201 , that is @xmath202 ; ( iii ) to the amplitude @xmath116 of the parton @xmath96-beam interaction ; ( iv ) to the gap survival factor @xmath203 ( @xmath189 ) . the resulting cross section reads @xmath204 @xmath205 where @xmath121 ( @xmath122 ) are the coordinates in the impact parameter plane with respect to the beam ( target ) hadron . the normalisation factor @xmath206 is specified in ( [ eq : sdf ] ) . the gap survival probability ) we should not put the whole irreducible amplitude @xmath207 , but , instead , part of it ; since the contribution from the processes with a gap in the same ( or a larger ) rapidity interval does not change qualitatively the structure of the diffractive dissociation event . in the present computations we neglect this effect . this means that actually the gap survival probabilities , and the true cross sections of diffractive dissociation , should be a bit larger . ] @xmath208 the symbolic structure of ( [ eq : sd ] ) , for high - mass single dissociation , is shown in fig . [ fig : sdm ] . ) for high - mass dissociation of the ` beam ' diffractive eigenstate @xmath58.,height=113 ] accounting for the different good - walker eigenstates and the different states in @xmath20-channel we obtain @xmath209 @xmath210 where the parton density @xmath211 is defined by eq.([eq : rho ] ) and , neglecting the secondary reggeon contribution , the elastic @xmath199 amplitude @xmath212 for the secondary reggeon , the real part may be large : note that @xmath213 . to allow for this , and since , that besides the @xmath214-trajectory with vacuum quantum numbers , there exists also @xmath215 and @xmath216 exchange , we enlarge the contribution due to @xmath217 by increasing the values of the effective triple - reggeon couplings @xmath218 as compared to those coming from the absorptive opacities ( [ eq : ombar ] ) . so we use ( [ eq : t ] ) for @xmath219 or @xmath220 . on the other hand , for @xmath186 or @xmath129 we use , respectively , @xmath221 and @xmath222 where @xmath223 . we take @xmath224 to reproduce the available data in the cern - isr to tevatron energy range . the slope of the diffractive dissociation cross section , @xmath225/dt$ ] at @xmath226 , can be calculated as the mean value of @xmath227 the separation of the intermediate parton @xmath96 from the target hadron @xmath228 @xmath229^{-1}\ , . \ ] ] central exclusive diffractive ( ced ) production of a system with mass squared @xmath230 with the large rapidity gaps either side , which is sometimes called the double - pomeron - exchange ( dpe ) process , has a cross section given by @xmath231 where @xmath232 @xmath233 are the probability amplitudes for elastic scattering of the intermediate parton @xmath96 ( @xmath234 ) on the beam ( target ) eigenstate @xmath58 ( @xmath45 ) . the coordinates of parton @xmath96 ( @xmath234 ) are @xmath121 and @xmath235 ( @xmath236 and @xmath122 ) with respect to the beam and the target proton respectively ; that is , @xmath237 . the momentum fractions ( @xmath238 ) of the incoming protons , transferred across the gaps , are @xmath239 and @xmath240 . the gap survival factor @xmath241 is given by ( [ eq : s ] ) . the amplitude of the interaction of partons @xmath96 and @xmath234 , @xmath242 , is obtained by the solution of the evolution ( [ eq:6 ] ) , which _ starts _ from the initial condition @xmath243 . that is , it _ starts _ from one parton at rapidity @xmath244 placed at coordinate @xmath122 in @xmath20-channel state @xmath50 , and _ finishes _ at the point @xmath245 in the state @xmath246 ; note @xmath247 . after the usual solution of ( [ eq:6],[eq:7 ] ) , the evolution ( [ eq:6 ] ) was performed in the known `` background '' fields @xmath248 to account for the absorption of intermediate partons . clearly , the number of parameters in our model is too large to perform a straightforward @xmath249 fit of the data . instead , we fix the majority of the parameters at reasonable values and demonstrate that such a model can reproduce all the features of the available data on diffractive cross sections , @xmath250 . before we give the values of the parameters , we show in fig . [ fig : dsdt ] the quality of the description of the data for the _ elastic differential _ cross section . [ t ] dependence of the elastic @xmath4 cross section . the dashed and dotted lines are the contributions from the elastic scattering of the largest size ( @xmath251 ) and the smallest size ( @xmath252 ) components.,height=453 ] we also present in fig . [ fig : dsdt ] the prediction for differential elastic cross section at the lhc energy @xmath253 tev . recall that we are using a three - channel eikonal . that is @xmath254 . it is interesting to note that the contribution to the cross section arising from the scattering of the two large - size eigenstates , @xmath255 , already has a diffractive dip at @xmath256 gev@xmath257 . however , after the contributions from all possible combinations @xmath258 are summed up , the prediction has no dip up to @xmath259 gev@xmath257 . note , also , that the pomeron and secondary reggeon couplings to the proton were taken to have the forms ( t ) = _ 0e^d_2 t/(1-t / d_1)^2 , ( t)=_re^d_r t. the values of the parameters that we use are d_2=0.15 gev^-2 , d_1=1.5 gev^2 , d_r=1 gev^-2 . the non - zero value of @xmath260 is simply to provide good convergence and accuracy of the fourier transform . the parameter @xmath261 controls the @xmath20 behaviour of the elastic cross section , while @xmath262 is responsible for the @xmath20 slope of diffractive dissociation at relatively low @xmath263 , where the cross section is dominated by the @xmath69 triple - reggeon term . the relative size of this contribution , as compared to that due to @xmath67 , was tuned by choosing @xmath160 and @xmath264 . in order to describe the data , the couplings were found to be @xmath265 mb and @xmath266 mb . since we choose a relatively simple @xmath20 dependence for the reggeon - proton couplings @xmath267 , our model is only reliable over a restricted @xmath20 interval , @xmath268 . note that in this domain , the real part of pomeron exchange , and a possible odderon exchange contribution , would give only very small effects . to describe the high energy behavior of the _ total _ cross section , we take @xmath269 for each of the three components of the pomeron . these pomeron intercepts are consistent with resummed nll bfkl , which gives @xmath270 practically independent of the scale @xmath131 @xcite . the slopes of the pomeron trajectories are driven by the transverse momentum associated with the particular component @xmath50 . in fact , we have @xmath271 . we find the data require @xmath272 gev@xmath273 for the large - size pomeron component , so we put @xmath274 gev@xmath273 for the second component and @xmath275 for the smallest - size component . for the secondary reggeon trajectory we take @xmath276 gev@xmath273 , and @xmath277 . the ` bare ' value is a bit larger than @xmath145 , since the final effective intercept is reduceed by the absorptive corrections included in the evolution equation . the description of the total cross section data are shown in fig . [ fig : sum4](a ) . the screening corrections arising from the ` enhanced ' multi - pomeron diagrams , that is from the high - mass dissociation , slow down the growth of the cross section with energy . thus , the model predicts a relatively low total cross section at the lhc @xmath278 mb . [ t ] cross sections and the cross sections of dissociation to a fixed @xmath279 state ( c ) ; ( d ) the parton multiplicity ( solid lines ) and the number of colour tubes ( dashed ) produced by the pomeron components of different size.,height=566 ] recall that the couplings of the good - walker eigenstates @xmath58 were specified by @xmath280 . the values @xmath173 , @xmath174 and @xmath175 were chosen so as to reproduce the _ low - mass dissociation _ cross section @xmath281 mb at the cern - isr energy ( see also the discussion in @xcite ) . ] @xcite . the value of the parameter @xmath84 , which controls the cross section of _ high - mass dissociation _ in the small @xmath282 ( that is , large @xmath93 ) region , was found to be @xmath283 . the dependence of the cross section for high - mass dissociation , @xmath284 , on @xmath285 is compared with the tevatron cdf data @xcite in fig . [ fig : dsy ] . recall that in our model we have not included pion exchange . since the @xmath71 term is essential at large @xmath282 , we have included the corresponding contribution using the parameters obtained in @xcite . the results without the @xmath71 term are shown by the dashed lines . we also show in fig . [ fig : dsy](a ) ( by the dotted line at small @xmath282 ) the prediction for the lhc energy . [ b ] for @xmath286 at @xmath287 1800 gev and 546 gev @xcite . the dashed lines are the predictions without the @xmath71 contribution . the dotted curve at small @xmath282 is the prediction for the lhc.,height=302 ] above , we introduced the different types of data mentioning the parameters that they mainly constrain . of course , in practice , these parameter values are used to describe all the ` soft ' data simultaneously . the energy behaviour of the cross sections are shown in table 1 and fig . [ fig : sum4 ] . fig . [ fig : sum4 ] also shows the energy behaviour of the multiplicities of the secondaries produced by the @xmath20-channel pomeron components of different sizes ; we will discuss the multiplicity distributions in some detail in section 6 . .cross sections ( in mb ) versus collider energy ( in tev ) . [ cols="<,^,^,^,^,^",options="header " , ] the values of @xmath288 quoted in table 1 look , at first sight , too large , when compared with the value @xmath289 mb given by cdf @xcite . however the cdf value does not include the secondary reggeon ( @xmath69 ) contribution , denoted as a ` non - diffractive ' component of @xmath290 mb . moreover , the trigger used to select the diffractive dissociation events rejects part of the low - mass proton excitations . taking these absences into account , there is no contradiction between the model prediction and the cdf data . furthermore , note that in the region where the cdf detector efficiency and resolution are good , our model gives an excellent description of the measured data , see fig . [ fig : dsy ] . it is interesting to note , that after tuning the parameters to describe all the available ` soft ' data , the model satisfies the finite energy sum rules @xcite to good accuracy . indeed , we can switch off the low - mass dissociation , putting the same couplings for each diffractive eigenstate ( @xmath291 ) , and replace the low - mass excitations for @xmath292 by the triple- and multi - regge contributions ; we started the evolution ( [ eq:6 ] ) , ( [ eq:7 ] ) at @xmath293 in order not to generate low - mass dissociation via triple- and multi - regee contributions and to avoid double counting . ] . keeping all the other parameters as before , we then obtain @xmath29473 mb ( 93 mb ) and @xmath288=13.6 mb ( 20.1 mb ) for the tevatron ( lhc ) energies , these values are close to those in table 1 . in principle , it is straightforward , although computer intensive , to use the model to calculate the cross section for double dissociation , @xmath295 . we do not show the values here . the values will be similar to those in table 2 of ref . @xcite ; footnote 25 of that paper shows that the values of @xmath295 are in excellent agreement with the tevatron data . the cross sections for the central exclusive diffractive ( ced ) production at the lhc energy @xmath296 tev are shown in fig . [ fig : ced ] for those @xmath282 intervals which can be studied by the totem and fp420 detectors . here we mean the soft ced production of a state with the mass given by @xmath230 separated from the incoming protons by two large rapidity gaps . the calculation is described in section 4.5 . the cross section integrated over the @xmath297 ( @xmath298 ) intervals ( for both @xmath299 and @xmath300 ) is predicted in the calculation of the amplitude @xmath301 . then there is no diffusion in impact parameter space and the integral over @xmath235 disapears ; since @xmath302 . however , to correct the final result we smear out the resulting amplitude , which allows for the larger gap survival probability at larger @xmath21 . in this way we retain reasonable ( @xmath303% ) accuracy of the computations . ] to be 53 ( 16 ) @xmath304b . the major contribution comes from pure soft interactions . for @xmath305 , more than 80% of the cross section is due to the large size component of the pomeron ; and more than half for larger values of @xmath282 . note that , in the ced calculations , we did not include the @xmath71 contribution . thus , actually , the expected cross section will be larger for @xmath282 values that are not too small , see fig . [ fig : dsy ] . [ h ] s are the momentum fractions of the incoming protons transferred across the rapidity gaps on either side of the centrally produced system of mass @xmath306.,height=377 ] note that the resulting ced cross section is about twice larger than that expected from the naive factorization formula @xmath307 this is due to the fact that each single dissociation contains a gap survival factor @xmath308 and therefore the r.h.s . of ( [ eq : fac ] ) is proportional to @xmath309 while the l.h.s . contains @xmath14 once only . on the other hand , for central exclusive production the typical values of the impact parameter @xmath21 are smaller ; so we have a smaller gap survival factor @xmath310 . this partly compensates the additional power of @xmath14 in the r.h.s . of ( [ eq : fac ] ) , and as a result the violation of the factorization shown in ( [ eq : fac ] ) is not so strong . we will discuss rapidity gap survival in central exclusive production in detail in the following paper @xcite . as we have a detailed model for high energy soft processes , it would appear to be possible to predict the multiplicity distribution at the lhc . however , although some general features can be predicted , it is not so simple to make a quantitative prediction . we address the problem below . recall that in the evolution equations for the _ amplitude _ , given in ( [ eq:6 ] ) , ( [ eq:7 ] ) , we include the absorptive factor @xmath311 , and not @xmath312 . that is we work with the forward amplitude @xmath313 , which at each step of the evolution ( in rapidity @xmath93 ) includes all possible processes - both elastic and inelastic interactions with cross sections @xmath314 and @xmath315 ; where @xmath316 , see eqs . ( [ eq : elastamp])-([eq : inel ] ) . as usual , inelastic processes include both single - ladder exchange , as well as multiple interactions with a larger density of secondary partons . the situation is similar to the rescattering of a fast hadron in a heavy nucleus . that is , in such an eikonal approach the probability , @xmath317 , of events with parton multiplicity @xmath206 times larger than that in a single ladder , is given by @xmath318 in the multi - channel case the opacity @xmath39 should be replaced by @xmath319 . unfortunately , we can not use this probability @xmath320 literally to describe the multiplicity distributions of secondary _ hadrons_. in particular a non - negligible fraction of the final hadrons may be produced via the fragmentation of minijets . these processes are beyond the ` pure soft ' approach used in the present paper . therefore , below , we discuss the multiplicity distribution only at the partonic level . the mean number of the ( @xmath20-channel ) ladders of the type @xmath50 produced in the collision of @xmath58 and @xmath45 good - walker eigenstates can be calculated as @xmath321 where @xmath322 $ ] and @xmath323 . recall that @xmath324 and @xmath220 are the large , intermediate and small size components of the pomeron respectively . after averaging over the impact parameter @xmath34 and the diffractive eigenstates @xmath168 of the incoming protons , we obtain @xmath325 this quantity may be considered as the mean number of colour tubes of type @xmath50 produced in the proton - proton interaction . note that the value of @xmath326 does not depend on the rapidity @xmath93 . to obtain the number of partons created by the ladder ` @xmath50 ' at rapidity @xmath93 , we have to include the parton density @xmath327 of ( [ eq : rho ] ) in the numerator of ( [ eq : nc ] ) . that is @xmath328 the results are shown in fig . [ fig : sum4](d ) . the main growth in multiplicity , as we go from tevatron to lhc energies , is due to the small size ( ` qcd ' ) pomeron component , which produces particles with typically @xmath329 gev . there is essentially no growth in multiplicity at small @xmath330 . this simply confirms the trend that has been observed through the cern - isr to tevatron energy range , see the data points in fig . [ fig : inclmult ] . [ t ] ranges correspond to the large- , intermediate- and small - size components of the pomeron.,height=680 ] in other words , starting with the same intercepts ( @xmath331 ) the large size component contribution after the absorptive correction becomes practically flat , while the small size contribution , which is much less affected by the absorption , continues to grow with energy . as mentioned above , such a behaviour is consistent with the experiment ( see fig . [ fig : inclmult ] ) where the density of a low @xmath131 secondaries is practically saturated while probability to produce a hadron with a large ( say , more then 5 gev ) transverse momentum grows with the initial energy . unfortunately we can not identify each ` parton ' with a pion . at a large @xmath131 it is probable that , after the hadronisation , a parton forms a gluon jet of pions . however , at low @xmath131 this is not evident ; it is hard to say what a gluon ` jet ' becomes at a low @xmath131 . nevertheless , in order to compare the model with the data we assume , that after hadronisation , each parton from pomeron components @xmath332 and @xmath220 gives three charged pions with @xmath333 and @xmath334 gev respectively . in this way we may estimate the inclusive cross section , at the three values of @xmath330 , using @xmath335 where @xmath72 allows for normalisation and @xmath336 accounts for the size of the phase space occupied by the particles from component @xmath58 . to obtain a qualitative feel for the expected behaviour , we show our predictions at the tevatron and lhc energies in fig.[fig : inclmult ] , where the horizontal lines indicate the typical @xmath330 interval associated with each pomeron component . recall , that in our ` soft ' model , we never use the _ value _ of the pomeron @xmath131 explicitly . the characteristic parameters actually used in the computations are the _ ratios _ @xmath337 . the horizontal lines reflect the @xmath330 intervals covered by the various components arising from the scale choices . some features of fig.[fig : inclmult ] are clear . first , although there is some freedom in assigning the overall scale , nevertheless , it appears that the scale choice made in the figure agrees satisfactorily with the tevatron data @xcite . second , as compared to the tevatron , the lhc distribution is more enhanced at large @xmath330 . the enhancement is a factor of 2.6 for the ` qcd ' small - size component of the pomeron , whereas it is only 1.25 for the ` soft ' pomeron component . new triple - regge analyses @xcite , which include absorptive effects , found that the triple - pomeron coupling is rather large ( @xmath338 with @xmath339 ) . thus , in order to obtain reliable predictions for diffractive processes at the lhc , it is necessary to have a model of ` soft ' high - energy processes which includes multi - pomeron interactions . here we have presented such a model , tuned to the existing ` soft ' data , which , in principle , is capable of predicting the basic features of high - energy soft @xmath4 interactions . the absorption of intermediate partons is described by conventional @xmath340-type factors . this corresponds to a coupling @xmath341 of the @xmath342 pomeron vertices . briefly , the model has multi - components in both the @xmath29- and @xmath20-channels . the former are based on a three - channel eikonal approach , together with the inclusion of multi - pomeron diagrams , so that both low- and high - mass diffractive dissociation are well described . predictions for the lhc are given . a novel feature of the model is the inclusion of different @xmath20-channel exchanges , which allows for small- , intermediate- and large - size components of the exchanged pomeron , each with a _ intercept @xmath343 . for the large - size component , the slope of the trajectory is @xmath344 gev@xmath273 . the large - size pomeron component is heavily screened by the effect of ` enhanced ' multi - pomeron diagrams , associated with high - mass dissociation . this leads , among other things , to the effective `` saturation '' of the low @xmath330 particle density , and to a slow growth of the total cross section . indeed , the model predicts a relatively low total cross section at the lhc @xmath278 mb . on the other hand , the small - size component of the pomeron is weakly screened , leading to an anticipated growth of the particle multiplicity at large @xmath345 gev ) at the lhc . thus the model has the possibility to embody a smooth matching of the perturbative qcd pomeron to the ` soft ' pomeron . we emphasized that a reliable model of soft interactions is essential in order to predict the rates of diffractive processes at the lhc . in particular , we used the model to calculate the rapidity gap survival factors , including the effects of _ both _ eikonal and enhanced rescattering . this is the subject of the following paper @xcite . we thank aliosha kaidalov and risto orava for useful discussions . mgr thanks the ippp at the university of durham for hospitality . the work was supported by grant rfbr 07 - 02 - 00023 , by the russian state grant rsgss-3628.2008.2 . see , for example , + p.d.b . collins , regge theory and high energy physics ( cambridge univ . press , 1977 ) + m.m . block , phys . rept . * 436 * , 71 ( 2006 ) ; + r. fiore _ et al . _ , arxiv:0810.2902 [ hep - ph ] . gribov , sov . jetp * 26 * , 414 ( 1968 ) . gribov and a.a . migdal , sov . * 8 * , 583 ( 1969 ) . gribov and a.a . migdal , sov . jetp * 28 * , 784 ( 1969 ) . m. g. ryskin , a. d. martin and v. a. khoze , arxiv : hep - ph/0506272 . v. a. khoze , a. d. martin and m. g. ryskin , arxiv:0810.3324 [ hep - ph ] . luna , v.a . khoze , a.d . martin and m.g . ryskin , eur . j. * c59 * , 1 ( 2009 ) . a. de roeck , v.a . khoze , a.d . martin , r. orava and m.g . ryskin , eur . j. c * 25 * , 391 ( 2002 ) . m. albrow and a. rostovtsev , arxiv : hep - ph/0009336 ; + m. g. albrow _ et al . _ [ fp420 r and d collaboration ] , arxiv:0806.0302 [ hep - ex ] . khoze , a.d . martin and m.g . ryskin , eur . j. * c19 * , 477 ( 2001 ) , erratum * c20 * , 599 ( 2001 ) ; arxiv : hep - ph/0006005 . abramovsky , v.n . gribov and o.v . kancheli , sov . . phys . * 18 * , 308 ( 1973 ) . r.d . field and g.c . fox , nucl . b80 * , 367 ( 1974 ) ; + a.b . kaidalov , v.a . khoze , yu.f . pirogov and n.l . ter - isaakyan , phys . lett . * b45 * 471 ( 1974 ) ; + for a review see a.b . kaidalov , phys . rep . * 50 * , 157 ( 1979 ) . a. capella , j. kaplan and j. tran thanh van , nucl . b * 105 * , 333 ( 1976 ) . j. bartels , m.g . ryskin and g.p . vacca , eur . j. * c27 * , 101 ( 2003 ) . zeus collaboration : abstract 549 , int . europhysics conf . on hep , aachen , july 2003 . ryskin , a.d . martin and v.a . khoze , eur . j. * c54 * , 199 ( 2008 ) . d. amati , a. stanghellini and s. fubini , nouvo cim . * 26 * , 896 ( 1962 ) . e.l . feinberg and d.s . chernavski , usp . nauk * 82 * , 41 ( 1964 ) ; + v.n . gribov , sov . * 9 * , 369 ( 1969 ) ; + v.n . gribov , in _ gauge theories and quark confinement _ , phasis , moscow , 2002 , p.3 . lipatov , sov . jetp * 63 * , 904 ( 1986 ) . v.s . fadin , e.a . kuraev , and l.n . lipatov , phys . b * 60 * , 50 ( 1975 ) ; + e.a . kuraev , l.n . lipatov , and v.s . fadin , zh . . fiz . * 71 * , 840 ( 1976 ) [ sov . jetp * 44 * , 443 ( 1976 ) ] ; _ ibid . _ * 72 * , 377 ( 1977 ) [ * 45 * , 199 ( 1977 ) ] ; + i.i . balitsky and l.n . lipatov , yad . * 28 * , 1597 ( 1978 ) [ sov . j. nucl . phys . * 28 * , 822 ( 1978 ) ] . l. baksay _ et al . _ , phys . lett . * b53 * 484 ( 1975 ) ; + r. webb _ et al . _ , phys . * b55 * 331 ( 1975 ) ; + l. baksay _ et al . _ , phys . b61 * 405 ( 1976 ) ; + h. de kerret _ et al . _ , phys . * b63 * 477 ( 1976 ) ; + g.c . mantovani _ et al . _ , phys . * b64 * 471 ( 1976 ) . m. albrow _ et al . _ , arxiv:0811.0120 [ hep - ex ] . fadin and l.n . lipatov , phys . lett . * b429 * , 127 ( 1998 ) ; + g. camici and m. ciafaloni , phys . lett . * b430 * , 349 ( 1998 ) ; + g.p . salam , jhep * 9807 * , 019 ( 1998 ) , act . phys . pol . * b30 * , 3679 ( 1999 ) ; + m. ciafaloni , d. colferai and g.p . salam , phys . lett . * b452 * , 372 ( 1999 ) , phys . rev . * d60 * , 114036 ( 1999 ) .

we present new calculations of rosseland and planck gaseous mean opacities relevant to the atmospheres of giant planets and ultracool dwarfs . such calculations are used in modeling the atmospheres , interiors , formation , and evolution of these objects . our calculations are an expansion of those presented in freedman et al . ( 2008 ) to include lower pressures , finer temperature resolution , and also the higher metallicities most relevant for giant planet atmospheres . span 1bar to 300 bar , and 75 k to 4000 k , in a nearly square grid . opacities at metallicities from solar to 50 times solar abundances are calculated . we also provide an analytic fit to the rosseland mean opacities over the grid in pressure , temperature , and metallicity . in addition to computing mean opacities at these local temperatures , we also calculate them with weighting functions up to 7000 k , to simulate the mean opacities for incident stellar intensities , rather than locally thermally emitted intensities . the chemical equilibrium calculations account for the settling of condensates in a gravitational field and are applicable to cloud - free giant planet and ultracool dwarf atmospheres , but not circumstellar disks . we provide our extensive opacity tables for public use .

1409.0026

a quantitative understanding of radiation transport in the cool molecule - dominated regions in planetary and ultracool dwarf atmospheres is essential to many aspects of understanding the temperature structure , thermal evolution , and formation of these objects . in freedman08 ( hereafter , f08 ) we presented a detailed discussion of atomic and molecular line opacities used by our group in modeling the atmospheres of brown dwarfs and giant planets [ e.g.][]marley02,fortney05,saumon08,fortney08a , marley10 . since mean opacities can also be widely used in many contexts , in f08 we also computed rosseland and planck mean opacities from 75 to 4000 k , 0.3 mbar to 300 bar , at metallicities of [ m / h ] = -0.3 , 0.0 , and + 0.3 . these tabulations have since found wide use in the communities working to understand giant planet formation [ e.g.,][]mordasini12a , bodenheimer13 , the temperature structure of giant planet atmospheres [ e.g.][]paxton13 , and planetary and ultracool dwarf thermal evolution [ e.g.,][]batygin11,valencia13,paxton13 . here we extend our previous work in a number of important aspects . in addition to updates in opacities of particular molecules ( described below ) , these new calculations are performed over a much larger phase space . mean opacity calculations at pressures from 1 @xmath0bar to 300 bars and 75 to 4000 k are presented , which is a much larger range in pressure , in particular at higher temperatures and lower pressures , compared to f08 . the overall temperature resolution of this compilation is also much finer . furthermore , we present calculations over a wide range of metallicities , from solar up to [ m / h]=+1.7 ( @xmath150@xmath2 solar ) , in several increments . these high metallicities may well - approximate the metal - rich atmospheres of giant planets , up to levels of the solar system s ice giant planets , uranus and neptune guillot07 . finally , for use in models of irradiated planetary atmospheres , we calculate mean opacities where the temperature in the weighting function is not the local temperature , but rather stellar blackbody temperatures from 3000 - 7000 k , to simulate mean `` incident flux '' or `` visible '' opacities to understand the absorption of incident stellar flux in planetary atmospheres . modern calculations of warm planetary atmospheres must draw from a variety of databases , including hitran rothman:2009 , hitemp rothman:2010 , exomol tennyson12 , and other sources . sharp07 and f08 provided excellent overviews of available opacities for giant planet and brown dwarf atmosphere modeling . much of this information remains current . our major changes since f08 are in ammonia , molecular hydrogen collision - induced absorption ( cia ) , methane , and carbon dioxide . for nh@xmath3 the updated reference is yurchenko:2011 , which is a first - principles calculation for temperatures up to 1500 k , which replaces our use of hitran . for h@xmath4 cia we use richard:2012 in place of previous work by borysow02 . in our recent work on brown dwarf model atmospheres , we have found these new opacity databases provide better fits to observed brown dwarf spectra saumon12,morley12 . two very recent opacity calculations for methane and carbon dioxide are now included . for methane we make use of the new first - principles line lists of yurchenko13 and yurchenko14 , which replaces our previous use of older sources brown:2005,strong:1993,wenger:1998 . we also now include the opacity contributions of co@xmath4 @xcite , which is a very minor player at solar metallicity , but rises in importance in metal - rich atmospheres as its abundance grows quadratically in metallicity lodders02 . all of our relevant molecular opacities , as well as the references , are collected in table 1 . as in our previous tabulations , here we do not account for the opacity of liquids or solids , which in the relatively high gravity atmospheres of planets and brown dwarfs may be confined to cloud decks . the opacity of these clouds decks is likely to be a strong function of poorly known parameters ( vigor of vertical mixing , particle size , particle size distribution ) and also other parameters that can not be readily incorporated into tables for general application over a range of object surface gravities and temperature structures . therefore , to maximize the utility of these calculations in the wide variety of potential applications , we continue to exclude the opacity of condensates . the condensation of cloud species , and their removal from the gas phase , is described below . the abundances of relevant atoms and molecules must be known at a given pressure and temperature in order to assess the wavelength dependent opacity . as in f08 we make use of the chemistry calculations of k. lodders and collaborators , using the base solar abundances of lodders03 . the chemistry calculations are described in a series of papers [ e.g.,][]fegley94,lodders99,lodders02,lodders02b , lodders06,lodders09 , which use the condor code . chemistry is calculated assuming that thermodynamic equilibrium is reached . this approximation is generally realized at high temperatures , where chemical reactions speed up dramatically . however , it becomes less correct at lower temperatures , where the timescale for vertical mixing in an atmosphere can be faster than relevant chemical conversion timescales . these effects have been explored in solar system giant planets , exoplanets , and brown dwarfs in many papers [ e.g.,][]prinn77,fegley96,saumon03,saumon06,hubeny07,visscher11 . this can generally lead to an incorrect estimation of the mixing ratios of carbon - bearing and nitrogen - bearing molecules , as the molecules co and n@xmath4 often have very long timescales for conversion to other molecules at low temperature . however , these effects depend strongly on the exact temperature structure of an atmosphere , as well the vertical mixing time . these two effects can not be generalized and incorporated with these tables , which are meant to be used over a wide phase space . perhaps most importantly , these mixing ratio differences are not a first - order effect . as described in f08 , the chemistry calculations are applicable to giant planets and ultracool dwarfs , which have significant self - gravity , unlike in protostellar disks . in disks , condensates remain well - mixed and in contact with the surrounding gas , which allows for further chemical reactions between condensates and gases . however , within planets and brown dwarfs , condensates form clouds that are typically vertically confined to a narrow region , often less than a gas scale height . above a given refractory cloud , at lower pressures , the atmosphere is then gradually depleted in the compounds ( atoms or molecules ) that have gone into forming the condensate . simply , the availability of these refractory compounds in the gas becomes greatly diminished . whether or not the gas phase remains in contact with condensates can have dramatic effects on chemical equilibrium calculations burrows99,lodders99,allard01,lodders02 with decreasing temperature . as discussed in f08 , all available evidence from brown dwarfs and the solar system s giant planets points towards this method of condensation , then depletion " of chemistry calculation as being most correct for these objects . our wavelength - dependent opacities are calculated in wavenumber space over a grid corresponding to wavelengths from 0.268 to 227 @xmath0 m . in figure [ example ] we show the wavelength dependent opacity at 400 k , 1400 k , and 2600 k , all at 1 bar . the plots are broken up to show the relative contributions of a few particular sources of opacity . in black is shown the dominant opacity source at all temperatures , which is that of neutral atoms and molecules . water vapor is present at all of these temperatures , and dominates in the infrared . in the optical at 400 k , there is relatively less opacity , but at warmer temperatures neutral atomic alkalis are present , which provide important opacities at 1400 k. at 2600 k , tio and vo gases are also important and generally overwhelm the alkali opacity . rayleigh scattering is generally important at the shortest wavelengths at all temperatures ( orange ) , as is the cia opacity of h@xmath4 molecules ( blue ) . the contributions of electrons ( thomson scattering , in green ) and negative hydrogen ions ( yellow ) are negligible at low temperatures but rise in importance at the highest temperatures as more free electrons are available . collision induced absorption ( cia ) is blue , negative hydrogen ions ( h@xmath5 , h@xmath4@xmath5 ) are shown in yellow , and opacity of all neutral atoms and molecules are shown in black.,title="fig:",width=259 ] collision induced absorption ( cia ) is blue , negative hydrogen ions ( h@xmath5 , h@xmath4@xmath5 ) are shown in yellow , and opacity of all neutral atoms and molecules are shown in black.,title="fig:",width=259 ] collision induced absorption ( cia ) is blue , negative hydrogen ions ( h@xmath5 , h@xmath4@xmath5 ) are shown in yellow , and opacity of all neutral atoms and molecules are shown in black.,title="fig:",width=259 ] the wavelength dependent opacities are tabulated at 1060 pressure - temperature _ p t _ points , on a nearly square grid . the high pressure and low temperature corner of phase space is not included ( @xmath1 100 k and 100 bar ) , as these conditions do not occur in atmospheres . for reference , the f08 calculations were at 324 points , and did not go to pressures below 0.3 mbar . in addition , high temperatures were only included at high pressures , while our revised grid is nearly square . the grid is sampled in temperatures from 754000 k and finely sampled from 100 - 350 k ( 10 k between points ) where water and ammonia condensation change the mixing ratio of these opacity sources quite dramatically . we compute the rosseland mean ( rm ) opacity , defined as : @xmath6 and the planck mean ( pm ) opacity , defined as : @xmath7 where @xmath8 is the wavelength - dependent opacity and @xmath9 is the planck function , at all of these 1060 _ p t _ points . as an illustration , in figure [ rmpm ] at 1 bar and 1000 k , we show in black the total wavelength - dependent opacity , with the weighting function for rm opacity ( top ) and pm opacity ( bottom ) . the peak of both weighting functions is at the same wavelength , but they have a different functional form and shape . in blue are the calculated rm and pm values . for reference the running integral of the mean opacities are shown in dotted red on an arbitrary linear axis , starting from long wavelengths . as expected , the dominant contributions to the rm opacities are in opacity minima , while for the pm opacities , these occur at opacity maxima . it is interesting to note that , as has been pointed out as least as far back as king56 [ see also][for a modern update]parmentier14a , that the ratio of the rm to the pm in an atmosphere is a measure of how important wavelength - dependnent ( non - gray ) effects are in determining the atmospheric temperature structure . figure [ rm ] shows the rm opacity calculation on the entire grid , presented as a colored contour plot . the general trend is one of the gradual loss of gaseous opacity with lower temperatures . g@xmath10 . [ rm ] ] over the _ t p _ range where the f08 and our present calculations overlap ( 324 points at @xmath11 mbar ) we can readily compare these calculations to the previous generation . this is shown for the rm opacities in figure [ f08compare1 ] , where pressure is color - coded . in general , the agreement is at the level of 10% or better . cia is important over the entire temperature range , and we find that @xmath180% of the generally small differences between the two calculations are due to this upgrade to the opacities from richard:2012 . larger differences ( though never more than a factor of 2 ) are seen below @xmath1 1000 k , where warm ammonia is the main nitrogen carrier . the first - princples yurchenko:2011 line list is a significant improvement over our previously - used hitran database , which lacked proper temperature dependence and was quite incomplete in the near infrared . as ammonia is lost into a cloud below @xmath1200 k , these differences between the f08 and new calculations are minimized . we can also compare our calculations to those of another group . in figure [ wich ] we compare our rm results at 1000 , 2000 , and 3000 k to those of ferguson05 . the chosen table , grain - free `` cunha06.nog.7.02.tron '' has a metallicity ( @xmath12,@xmath13 ) somewhat similar to our own solar composition . the ferguson05 calculations only have moderate density overlap with our own , since their calculations are most appropriate for low - density disks while ours are for denser planetary atmospheres . however , the agreement does appear reasonably good . opacity database . [ f08compare1 ] ] , @xmath13 calculation by ferguson05 , table `` cunha06.nog.7.02.tron '' ( dashed lines with open squares ) . the agreement appears reasonably good in the range of overlap , particularly at 2000 k. [ wich ] ] while the original f08 opacity calculations spanned a fairly narrow range of metallicities , from [ m / h]=-0.5 to + 0.5 , here we expand these calculations to include [ m / h ] = 0 , + 0.5 , + 0.7 , + 1.0 , + 1.5 , and + 1.7 , roughly corresponding to 1@xmath2 , 3@xmath2 , 5@xmath2 , 10@xmath2 , 30@xmath2 , and 50@xmath2 solar abundances . for reference , jupiter s atmosphere is approximately 3 - 5@xmath2 solar wong04 , saturn s atmosphere is enhanced in methane by a factor of @xmath1 10 flasar05 , and uranus and neptune in methane by a factor to @xmath150 guillot07 . we have calculated rm and pm opacities over the full pressure and temperature grid for these enhanced metallicity cases . the results for the rm opacities are shown in figure [ metals ] . the general trend of increased metallicity is of course increased opacity . over the temperature range from 250 k to 3000 k , the rm is a strong function of metallicity , rising not quite linearly . at very low temperatures , after the condensation of water , the effects of metallicity are much weaker . first water vapor and then ammonia are lost , essentially leaving only methane and h@xmath4 cia opacity as the main opacity sources at lower temperature . in particular below 150 k , the peak of the weighting function is at wavelengths beyond 20 @xmath0 m , where h@xmath4 cia opacity dominates over ch@xmath14 , leading to a minimal dependence of the opacity on metallicity . ( light blue ) to 50@xmath2 ( dark blue ) . from 200 to 3500 k , the local rm opacity increases not quite linearly with metallicity.,title="fig:",width=259 ] ( light blue ) to 50@xmath2 ( dark blue ) . from 200 to 3500 k , the local rm opacity increases not quite linearly with metallicity.,title="fig:",width=259 ] recently valencia13 used an older version of these tabulated mean opacities at metallicities from solar , 30@xmath2 solar ( [ m / h]=+1.5 ) , and 50@xmath2 solar ( [ m / h]=+1.7 ) to derive analytic fits to the rm opacities as a function of pressure , temperature , and metallicity . while interpolation in tables is certainly more accurate , there are instances where the speed of an analytic fit is more important than high accuracy . the behavior of opacities is quite complex and it is difficult to capture the full behavior with an analytic fit . impressively , valencia13 were able to provide a reasonable fit to the rm opacities with analytic equations that used 11 coefficients fit by least - squares minimization , by breaking the fit into regions above and below 800 k. their fit is shown in figure [ fit]_a_. their fit is generally good , but somewhat fails to reproduce the asymmetrical shape of the opacity maximum from 800 - 3000 k seen at pressures below @xmath1 1 bar . we find that we can generally reproduce this feature better by replacing the quadratic term in valencia13 ( their equation 2 ) with two terms , one an inverse tangent , and one an exponential with a pressure dependent term . the cost is relatively minor , with the inclusion of two addition coefficients , making our fit need 13 coefficients , instead of the 11 used in valencia13 . the analytic rm fit for the opacity @xmath15 ( in @xmath16 g@xmath10 ) is the sum of two components that have separate analytic fits , @xmath17 and @xmath18 , that are related as : @xmath19 where we define , @xmath20 and @xmath21 \label{highp}\end{aligned}\ ] ] with @xmath22 in k , @xmath23 in bar , and `` met '' as the metallicity , [ m / h ] . the coefficients @xmath24 through @xmath25 are given in table 2 . the fit error at solar metallicity is shown in figure [ errfit ] . differences are typically on the order of 50% . the fit reproduces the tables within a factor of 2 for 90% of the points at solar metallicity , and 80% at [ m / h]=+1.7 . for some users the flexibility of an analytic fit may be more important than accuracy . for a variety of pressures ( colors ) . on the left ( a ) , the analytic fit to the rm opacities described in valencia13 is over - plotted as dashed black lines . similarly , on the right ( b ) our analytic fit is over - plotted as dashed black lines showing an improvement in the region where @xmath26 . pressure increases from red to purple . [ fit ] ] b , with coefficients tabulated in table 2 . the fit reproduces the tables within a factor of 2 for 90% of the points at solar metallicity , and 80% at [ m / h]=+1.7 . [ errfit],width=259 ] over the past several years a variety of analytic work on the temperature structure of strongly irradiated planetary atmospheres has been published [ e.g.,][]hansen08,guillot10,heng12,robinson12,parmentier14a , parmentier14b . these works generally aim to understand the atmospheric temperature structure , and presence or absence of temperature inversions , with as few free parameters as possible . generally at a minimum one needs to know the thermal infrared opacity , and the relevant `` visible '' opacity for incoming stellar intensity . the nature of the visible opacity is in principle straightforward to physically understand . what one is interested in is how the stellar radiation field is attenuated by the wavelength - dependent opacity of the planetary atmosphere , as a function for depth . often this is prescribed as a gray atmospheric constant , meaning the visible opacity is chosen as a wavelength - independent constant [ e.g.][]heng12 or that the ratio of the visible to thermal opacity is a constant [ e.g.][]guillot10 . in principle , one could be interested in an opacity based either on the rm or pm , but we caution that either is an approximation to the real atmospheric bevahior , as the wavelength - dependent downwelling ( and upwelling ) stellar intensity can in general change strongly with depth . within the framework of an opacity table theses rm or pm visible opacities can be readily calculated across our _ p - t _ space by modifying the rm and pm formulas to use the temperature of a _ stellar _ @xmath27 , rather than the _ local _ gas temperature . hence the weighting function is no longer the intensity characteristic of the local temperature but that of the incoming stellar intensity . in practice here we use stellar blackbody intensities from 3000 to 7000 k , and the same tabulated wavelength - dependent opacities , including scattering . these visible opacities should be used with some caution . guillot10 ( his equation 9 ) postulate a use for the pm for the incident stellar intensity at the top of the atmosphere , however , the stellar intensity begins to dramatically deviate from a blackbody as the wavelength dependent flux is carved by absorption , with increasing depth . however , in a revision of the guillot10 work , parmentier14b ( see their 5.1.1 ) suggest that instead the stellar-@xmath27 weighted rm is much more useful to understand the depth to which incident stellar flux penentrates , since the rm most heavily weights at wavelengths where opacity is low . they base their suggestion of the utility of the weighted rm by comparing derived model temperature structures that utilize gray opacities , to those that use full frequency - dependent opacity calculations . for our purposes , we choose to calculate both the stellar - weighted rm and pm tables over our large _ p - t _ space , such that they can be of some use to the community , for current and future investigations . this change in the temperature of the weighting function can lead to dramatically different results for the rm and pm for a given composition . an example at 1 bar and 1000 k is shown in figure [ weight ] . in the top panel , as the weighting function moves from the near infrared ( 1000 k ) towards the optical ( 3000 k and higher ) the rm falls dramatically due to opacity minima between the pressure - broadened alkali lines , which dominates the rm calculation . in the bottom panel , for the pm , somewhat opposite behavior is found . as the weighting function overlaps with these same alkali lines they again dominate , leading to a dramatic increase in the pm opacity . some general behavior is shown in figure [ weight2 ] . this is a plot of the ratio of rm calculated at the stellar temperature , divided by the rm calculated at the local temperature , at 1 bar . a change in behavior is found around 1000 k. at higher temperatures , there is a much weaker dependence on the weighting function temperature , since the local temperature weighting function moves toward having a greater overlap with the stellar @xmath27s . the ratio approaches 1 and is exactly 1.0 for 3000 k , with a 3000 k @xmath27 , and at 4000 k , for a 4000 k @xmath27 , as expected . below 1000 k , even when the gas becomes cool enough that na and k condense and are lost from the gas phase , the optical opacity windows do not change dramatically , and water vapor dominates in the mid ir for the thermal opacity , such that the behavior does not have a strong dependence on the local gas temperature . we present our calculated opacities in tables 3 - 8 , which run from solar metallicity ( [ m / h]=0.0 ) to @xmath150@xmath2 solar ( [ m / h]=+1.7 ) , including intermediate increments of + 0.5 , + 0.7 , + 1.0 , and + 1.5 . the tables are over the same grid in temperature @xmath22 in k , and pressure @xmath23 in dyne @xmath28 . the ideal gas law is assumed , and using the appropriate chemical mixing ratios the density @xmath29 in g @xmath30 is also given . at the local temperature @xmath22 the rm and pm opacities are tabulated , in @xmath16 g@xmath10 . also tabulated are the rm and pm opacities tabulated at stellar incident blackbody weighting temperatures of 3000 , 4000 , 5000 , 6000 , and 7000 k ( see 4 ) . databases of the opacities at the relatively low temperatures relevant to giant planets and ultracool dwarfs have significantly advanced since the first model atmospheres of these objects nearly 20 years ago . in our tabulations we use state - of - the art calculations for all molecules and atoms . the generally good agreement between models of brown dwarf atmospheres and observations stephens09,saumon12,morley12 suggests that we are on the right track in understanding the chemical abundances and opacity of cool atmospheres . the general very close agreement at solar metallicity between our calculations here and f08 , suggests that future tabulations of purely gaseous opacities over this range of pressure , temperature , and metallicity , probably will not change dramatically in the future . future directions could include the use of a variety of c / o ratios , which change the mixing ratios of water vapor , carbon monoxide , and methane . with regard to the inclusion of cloud opacity in atmospheres , we note that cuzzi14 recently published a method to calculate the mean opacities of porous aggregates for use in dusty disks and cloudy atmospheres . with regard to main absorbers that could have an identifiable impact on the mean opacities , and where there is still likely considerable room for improvement , we suggest the alkali metals . our treatment of alkali metal opacity , following the work of bms , is still approximate . detailed calculations of alkali line shapes , perturbed by collisions with h@xmath4 and he are still in progress allard03,allard05 , but they do not yet reach to high enough pressures to be used over all of our phase space . also , for methane , detailed comparisons should be made between the databases of yurchenko14 , used here , and the recent calculations of rey14 . we hope that these calculations will find broad use in the community . a particularly interesting area of future study would be giant planet formation by core - nucleated accretion perri74,mizuno80,stevenson82b . the concurrent accretion of solids and gas pollack96 may allow for a wide range of metallicity enhancements , and subsequent opacity enhancements , that could change as a function of time and location within the accreting h / he envelope podolak88,mordasini06,fortney13 , and could plausibly lead to new , and previously unexplored behavior in formation models . we thank the anonymous referee for helpful comments , as well as v. parmentier , t. robinson , and t. guillot for enlightening discussions on analytic models of gray atmospheres . ll ch@xmath14 & @xcite;@xcite;@xcite ; hitran08 isotopes ; + co & hitemp10 ; @xcite + co@xmath4 & @xcite ; @xcite + crh & @xcite + feh & @xcite ; @xcite + h@xmath4o & @xcite ; @xcite ; hitran08 isotopes + h@xmath4s & @xcite;@xcite - private communication ; hitran08 isotopes + h@xmath4 & @xcite - cia + nh@xmath3 & @xcite ; @xcite + ph@xmath3 & @xcite ; geisa ; hitran08 + tio & @xcite;@xcite + vo & @xcite cccccc + @xmath31 & 10.602 & & @xmath32 & -14.051 & 82.241 + @xmath33 & 2.882 & & @xmath34 & 3.055 & -55.456 + @xmath35 & 6.09@xmath36 & & @xmath37 & 0.024 & 8.754 + @xmath38 & 2.954 & & @xmath39 & 1.877 & 0.7048 + @xmath40 & -2.526 & & @xmath41 & -0.445 & -0.0414 + @xmath42 & 0.843 & & @xmath25 & 0.8321 & 0.8321 + @xmath43 & -5.490 & & & & + ccccccccccccccc 75 & 1e+00 & 3.759e-10 & 1.019e-08 & 2.747e-05 & 8.173e-07 & 1.177e-01 & 1.814e-06 & 6.274e-02 & 3.371e-06 & 3.711e-02 & 5.497e-06 & 2.401e-02 & 8.088e-06 & 1.680e-02 + 75 & 3e+00 & 1.127e-09 & 3.037e-08 & 9.568e-06 & 9.905e-07 & 1.177e-01 & 2.178e-06 & 6.273e-02 & 4.027e-06 & 3.710e-02 & 6.542e-06 & 2.401e-02 & 9.597e-06 & 1.680e-02 + 75 & 1e+01 & 3.759e-09 & 1.007e-07 & 3.431e-06 & 1.200e-06 & 1.177e-01 & 2.618e-06 & 6.274e-02 & 4.812e-06 & 3.711e-02 & 7.787e-06 & 2.401e-02 & 1.139e-05 & 1.680e-02 + 75 & 3e+01 & 1.127e-08 & 3.015e-07 & 2.043e-06 & 1.426e-06 & 1.177e-01 & 3.088e-06 & 6.275e-02 & 5.648e-06 & 3.712e-02 & 9.106e-06 & 2.401e-02 & 1.328e-05 & 1.680e-02 + 75 & 1e+02 & 3.759e-08 & 1.003e-06 & 2.833e-06 & 1.751e-06 & 1.179e-01 & 3.761e-06 & 6.281e-02 & 6.839e-06 & 3.715e-02 & 1.097e-05 & 2.404e-02 & 1.595e-05 & 1.682e-02 + 75 & 3e+02 & 1.127e-07 & 3.006e-06 & 6.726e-06 & 2.181e-06 & 1.182e-01 & 4.642e-06 & 6.299e-02 & 8.381e-06 & 3.725e-02 & 1.337e-05 & 2.410e-02 & 1.935e-05 & 1.686e-02 + 75 & 1e+03 & 3.759e-07 & 1.001e-05 & 2.105e-05 & 2.907e-06 & 1.193e-01 & 6.098e-06 & 6.356e-02 & 1.089e-05 & 3.758e-02 & 1.723e-05 & 2.431e-02 & 2.476e-05 & 1.700e-02 + 75 & 3e+03 & 1.127e-06 & 3.002e-05 & 6.239e-05 & 3.950e-06 & 1.222e-01 & 8.136e-06 & 6.501e-02 & 1.432e-05 & 3.841e-02 & 2.242e-05 & 2.483e-02 & 3.194e-05 & 1.e-02 + 75 & 1e+04 & 3.759e-06 & 1.000e-04 & 2.068e-04 & 5.729e-06 & 1.285e-01 & 1.148e-05 & 6.824e-02 & 1.981e-05 & 4.026e-02 & 3.052e-05 & 2.599e-02 & 4.294e-05 & 1.815e-02 + 75 & 3e+04 & 1.127e-05 & 2.999e-04 & 6.190e-04 & 7.997e-06 & 1.328e-01 & 1.561e-05 & 7.038e-02 & 2.637e-05 & 4.148e-02 & 3.996e-05 & 2.676e-02 & 5.548e-05 & 1.867e-02 + 75 & 1e+05 & 3.759e-05 & 9.993e-04 & 2.061e-03 & 1.106e-05 & 1.303e-01 & 2.108e-05 & 6.900e-02 & 3.486e-05 & 4.065e-02 & 5.188e-05 & 2.622e-02 & 7.100e-05 & 1.830e-02 + 75 & 3e+05 & 1.127e-04 & 2.997e-03 & 6.182e-03 & 1.448e-05 & 1.300e-01 & 2.709e-05 & 6.884e-02 & 4.399e-05 & 4.057e-02 & 6.442e-05 & 2.618e-02 & 8.701e-05 & 1.828e-02 + 75 & 1e+06 & 3.759e-04 & 9.989e-03 & 2.060e-02 & 1.999e-05 & 1.329e-01 & 3.655e-05 & 7.056e-02 & 5.788e-05 & 4.166e-02 & 8.292e-05 & 2.691e-02 & 1.100e-04 & 1.879e-02 + 75 & 3e+06 & 1.127e-03 & 2.996e-02 & 6.182e-02 & 3.672e-05 & 1.411e-01 & 6.178e-05 & 7.550e-02 & 9.108e-05 & 4.476e-02 & 1.236e-04 & 2.898e-02 & 1.575e-04 & 2.026e-02 + 75 & 1e+07 & 3.759e-03 & 9.988e-02 & 2.060e-01 & 1.119e-04 & 1.701e-01 & 1.401e-04 & 9.277e-02 & 1.712e-04 & 5.563e-02 & 2.068e-04 & 3.626e-02 & 2.447e-04 & 2.542e-02 + 100 & 1e+00 & 2.821e-10 & 2.071e-08 & 3.332e-01 & 1.336e-06 & 1.432e-01 & 2.942e-06 & 7.613e-02 & 5.432e-06 & 4.501e-02 & 8.809e-06 & 2.910e-02 & 1.289e-05 & 2.032e-02 ccccccccccccccc 75 & 1e+00 & 3.784e-10 & 1.033e-08 & 2.777e-05 & 8.651e-07 & 3.708e-01 & 1.924e-06 & 1.974e-01 & 3.582e-06 & 1.164e-01 & 5.847e-06 & 7.490e-02 & 8.609e-06 & 5.186e-02 + 75 & 3e+00 & 1.135e-09 & 3.076e-08 & 9.997e-06 & 1.066e-06 & 3.708e-01 & 2.351e-06 & 1.974e-01 & 4.351e-06 & 1.164e-01 & 7.076e-06 & 7.491e-02 & 1.038e-05 & 5.186e-02 + 75 & 1e+01 & 3.784e-09 & 1.020e-07 & 3.910e-06 & 1.304e-06 & 3.709e-01 & 2.851e-06 & 1.974e-01 & 5.248e-06 & 1.164e-01 & 8.500e-06 & 7.491e-02 & 1.244e-05 & 5.187e-02 + 75 & 3e+01 & 1.135e-08 & 3.054e-07 & 2.549e-06 & 1.537e-06 & 3.710e-01 & 3.341e-06 & 1.975e-01 & 6.125e-06 & 1.165e-01 & 9.888e-06 & 7.493e-02 & 1.443e-05 & 5.188e-02 + 75 & 1e+02 & 3.784e-08 & 1.016e-06 & 3.394e-06 & 1.861e-06 & 3.713e-01 & 4.020e-06 & 1.977e-01 & 7.334e-06 & 1.166e-01 & 1.179e-05 & 7.500e-02 & 1.716e-05 & 5.193e-02 + 75 & 3e+02 & 1.135e-07 & 3.045e-06 & 7.410e-06 & 2.302e-06 & 3.724e-01 & 4.935e-06 & 1.982e-01 & 8.950e-06 & 1.169e-01 & 1.432e-05 & 7.520e-02 & 2.075e-05 & 5.206e-02 + 75 & 1e+03 & 3.784e-07 & 1.014e-05 & 2.207e-05 & 3.068e-06 & 3.760e-01 & 6.496e-06 & 2.000e-01 & 1.166e-05 & 1.179e-01 & 1.852e-05 & 7.585e-02 & 2.667e-05 & 5.251e-02 + 75 & 3e+03 & 1.135e-06 & 3.041e-05 & 6.425e-05 & 4.193e-06 & 3.850e-01 & 8.732e-06 & 2.046e-01 & 1.548e-05 & 1.206e-01 & 2.433e-05 & 7.751e-02 & 3.473e-05 & 5.363e-02 + 75 & 1e+04 & 3.784e-06 & 1.013e-04 & 2.108e-04 & 6.142e-06 & 4.049e-01 & 1.248e-05 & 2.147e-01 & 2.171e-05 & 1.264e-01 & 3.359e-05 & 8.116e-02 & 4.735e-05 & 5.612e-02 + 75 & 3e+04 & 1.135e-05 & 3.037e-04 & 6.276e-04 & 8.640e-06 & 4.182e-01 & 1.715e-05 & 2.214e-01 & 2.925e-05 & 1.301e-01 & 4.451e-05 & 8.353e-02 & 6.190e-05 & 5.772e-02 + 75 & 1e+05 & 3.784e-05 & 1.011e-03 & 2.085e-03 & 1.201e-05 & 4.099e-01 & 2.334e-05 & 2.166e-01 & 3.900e-05 & 1.273e-01 & 5.830e-05 & 8.168e-02 & 7.987e-05 & 5.644e-02 + 75 & 3e+05 & 1.135e-04 & 3.034e-03 & 6.252e-03 & 1.574e-05 & 4.069e-01 & 3.010e-05 & 2.151e-01 & 4.939e-05 & 1.264e-01 & 7.260e-05 & 8.111e-02 & 9.809e-05 & 5.605e-02 + 75 & 1e+06 & 3.784e-04 & 1.011e-02 & 2.083e-02 & 2.177e-05 & 4.097e-01 & 4.071e-05 & 2.168e-01 & 6.509e-05 & 1.274e-01 & 9.346e-05 & 8.182e-02 & 1.238e-04 & 5.655e-02 + 75 & 3e+06 & 1.135e-03 & 3.032e-02 & 6.250e-02 & 4.226e-05 & 4.179e-01 & 7.231e-05 & 2.217e-01 & 1.065e-04 & 1.305e-01 & 1.435e-04 & 8.388e-02 & 1.815e-04 & 5.801e-02 + 75 & 1e+07 & 3.784e-03 & 1.010e-01 & 2.083e-01 & 1.508e-04 & 4.466e-01 & 1.807e-04 & 2.388e-01 & 2.116e-04 & 1.413e-01 & 2.480e-04 & 9.108e-02 & 2.871e-04 & 6.312e-02 + 100 & 1e+00 & 2.844e-10 & 2.348e-08 & 9.863e-01 & 1.567e-06 & 4.472e-01 & 3.455e-06 & 2.375e-01 & 6.384e-06 & 1.401e-01 & 1.035e-05 & 9.014e-02 & 1.515e-05 & 6.240e-02 ccccccccccccccc 75 & 1e+00 & 3.806e-10 & 1.043e-08 & 2.816e-05 & 8.848e-07 & 5.859e-01 & 1.970e-06 & 3.119e-01 & 3.669e-06 & 1.839e-01 & 5.993e-06 & 1.181e-01 & 8.825e-06 & 8.166e-02 + 75 & 3e+00 & 1.142e-09 & 3.108e-08 & 1.036e-05 & 1.098e-06 & 5.860e-01 & 2.423e-06 & 3.119e-01 & 4.487e-06 & 1.839e-01 & 7.299e-06 & 1.181e-01 & 1.071e-05 & 8.167e-02 + 75 & 1e+01 & 3.806e-09 & 1.031e-07 & 4.317e-06 & 1.344e-06 & 5.860e-01 & 2.941e-06 & 3.119e-01 & 5.418e-06 & 1.839e-01 & 8.780e-06 & 1.181e-01 & 1.285e-05 & 8.167e-02 + 75 & 3e+01 & 1.142e-08 & 3.086e-07 & 2.979e-06 & 1.575e-06 & 5.862e-01 & 3.429e-06 & 3.120e-01 & 6.294e-06 & 1.840e-01 & 1.016e-05 & 1.182e-01 & 1.484e-05 & 8.169e-02 + 75 & 1e+02 & 3.806e-08 & 1.026e-06 & 3.875e-06 & 1.895e-06 & 5.854e-01 & 4.103e-06 & 3.116e-01 & 7.499e-06 & 1.837e-01 & 1.207e-05 & 1.180e-01 & 1.757e-05 & 8.157e-02 + 75 & 3e+02 & 1.142e-07 & 3.077e-06 & 7.990e-06 & 2.339e-06 & 5.870e-01 & 5.029e-06 & 3.124e-01 & 9.140e-06 & 1.842e-01 & 1.464e-05 & 1.183e-01 & 2.124e-05 & 8.178e-02 + 75 & 1e+03 & 3.806e-07 & 1.024e-05 & 2.293e-05 & 3.118e-06 & 5.927e-01 & 6.628e-06 & 3.153e-01 & 1.193e-05 & 1.858e-01 & 1.898e-05 & 1.193e-01 & 2.e-05 & 8.249e-02 + 75 & 3e+03 & 1.142e-06 & 3.072e-05 & 6.582e-05 & 4.275e-06 & 6.069e-01 & 8.944e-06 & 3.225e-01 & 1.590e-05 & 1.900e-01 & 2.504e-05 & 1.219e-01 & 3.578e-05 & 8.426e-02 + 75 & 1e+04 & 3.806e-06 & 1.023e-04 & 2.141e-04 & 6.290e-06 & 6.383e-01 & 1.286e-05 & 3.385e-01 & 2.245e-05 & 1.991e-01 & 3.480e-05 & 1.277e-01 & 4.911e-05 & 8.818e-02 + 75 & 3e+04 & 1.142e-05 & 3.068e-04 & 6.348e-04 & 8.873e-06 & 6.592e-01 & 1.774e-05 & 3.489e-01 & 3.039e-05 & 2.050e-01 & 4.635e-05 & 1.314e-01 & 6.453e-05 & 9.069e-02 + 75 & 1e+05 & 3.806e-05 & 1.022e-03 & 2.106e-03 & 1.234e-05 & 6.459e-01 & 2.420e-05 & 3.413e-01 & 4.064e-05 & 2.005e-01 & 6.089e-05 & 1.285e-01 & 8.349e-05 & 8.864e-02 + 75 & 3e+05 & 1.142e-04 & 3.065e-03 & 6.311e-03 & 1.617e-05 & 6.407e-01 & 3.122e-05 & 3.386e-01 & 5.150e-05 & 1.989e-01 & 7.587e-05 & 1.274e-01 & 1.025e-04 & 8.794e-02 + 75 & 1e+06 & 3.806e-04 & 1.021e-02 & 2.102e-02 & 2.239e-05 & 6.435e-01 & 4.228e-05 & 3.403e-01 & 6.793e-05 & 1.999e-01 & 9.770e-05 & 1.281e-01 & 1.294e-04 & 8.844e-02 + 75 & 3e+06 & 1.142e-03 & 3.062e-02 & 6.307e-02 & 4.444e-05 & 6.516e-01 & 7.671e-05 & 3.451e-01 & 1.131e-04 & 2.030e-01 & 1.522e-04 & 1.302e-01 & 1.919e-04 & 8.989e-02 + 75 & 1e+07 & 3.806e-03 & 1.020e-01 & 2.102e-01 & 1.692e-04 & 6.801e-01 & 1.996e-04 & 3.621e-01 & 2.300e-04 & 2.136e-01 & 2.662e-04 & 1.373e-01 & 3.056e-04 & 9.495e-02 + 100 & 1e+00 & 2.872e-10 & 2.684e-08 & 3.021e+00 & 1.766e-06 & 7.849e-01 & 3.894e-06 & 4.173e-01 & 7.194e-06 & 2.464e-01 & 1.166e-05 & 1.586e-01 & 1.705e-05 & 1.097e-01 ccccccccccccccc 75 & 1e+00 & 3.866e-10 & 1.071e-08 & 2.895e-05 & 9.110e-07 & 1.159e+00 & 2.031e-06 & 6.173e-01 & 3.787e-06 & 3.638e-01 & 6.188e-06 & 2.335e-01 & 9.117e-06 & 1.611e-01 + 75 & 3e+00 & 1.160e-09 & 3.191e-08 & 1.134e-05 & 1.140e-06 & 1.159e+00 & 2.518e-06 & 6.173e-01 & 4.669e-06 & 3.638e-01 & 7.600e-06 & 2.335e-01 & 1.116e-05 & 1.611e-01 + 75 & 1e+01 & 3.866e-09 & 1.058e-07 & 5.401e-06 & 1.391e-06 & 1.160e+00 & 3.049e-06 & 6.174e-01 & 5.625e-06 & 3.639e-01 & 9.122e-06 & 2.336e-01 & 1.336e-05 & 1.611e-01 + 75 & 3e+01 & 1.160e-08 & 3.168e-07 & 4.125e-06 & 1.614e-06 & 1.160e+00 & 3.523e-06 & 6.176e-01 & 6.478e-06 & 3.640e-01 & 1.048e-05 & 2.336e-01 & 1.531e-05 & 1.612e-01 + 75 & 1e+02 & 3.866e-08 & 1.054e-06 & 5.147e-06 & 1.925e-06 & 1.161e+00 & 4.184e-06 & 6.182e-01 & 7.667e-06 & 3.643e-01 & 1.236e-05 & 2.338e-01 & 1.802e-05 & 1.613e-01 + 75 & 3e+02 & 1.160e-07 & 3.159e-06 & 9.539e-06 & 2.369e-06 & 1.164e+00 & 5.118e-06 & 6.198e-01 & 9.333e-06 & 3.653e-01 & 1.499e-05 & 2.344e-01 & 2.177e-05 & 1.617e-01 + 75 & 1e+03 & 3.866e-07 & 1.052e-05 & 2.522e-05 & 3.163e-06 & 1.176e+00 & 6.765e-06 & 6.255e-01 & 1.223e-05 & 3.685e-01 & 1.950e-05 & 2.365e-01 & 2.816e-05 & 1.631e-01 + 75 & 3e+03 & 1.160e-06 & 3.155e-05 & 7.002e-05 & 4.358e-06 & 1.202e+00 & 9.187e-06 & 6.390e-01 & 1.642e-05 & 3.763e-01 & 2.593e-05 & 2.413e-01 & 3.712e-05 & 1.664e-01 + 75 & 1e+04 & 3.866e-06 & 1.051e-04 & 2.231e-04 & 6.458e-06 & 1.266e+00 & 1.332e-05 & 6.714e-01 & 2.341e-05 & 3.948e-01 & 3.642e-05 & 2.530e-01 & 5.149e-05 & 1.744e-01 + 75 & 3e+04 & 1.159e-05 & 3.150e-04 & 6.539e-04 & 9.143e-06 & 1.300e+00 & 1.849e-05 & 6.882e-01 & 3.191e-05 & 4.043e-01 & 4.885e-05 & 2.590e-01 & 6.813e-05 & 1.784e-01 + 75 & 1e+05 & 3.866e-05 & 1.049e-03 & 2.161e-03 & 1.272e-05 & 1.281e+00 & 2.529e-05 & 6.770e-01 & 4.282e-05 & 3.975e-01 & 6.442e-05 & 2.545e-01 & 8.849e-05 & 1.753e-01 + 75 & 3e+05 & 1.159e-04 & 3.150e-03 & 6.482e-03 & 1.667e-05 & 1.276e+00 & 3.268e-05 & 6.744e-01 & 5.439e-05 & 3.959e-01 & 8.047e-05 & 2.535e-01 & 1.089e-04 & 1.746e-01 + 75 & 1e+06 & 3.866e-04 & 1.048e-02 & 2.154e-02 & 2.307e-05 & 1.272e+00 & 4.427e-05 & 6.726e-01 & 7.178e-05 & 3.950e-01 & 1.036e-04 & 2.529e-01 & 1.374e-04 & 1.742e-01 + 75 & 3e+06 & 1.159e-03 & 3.142e-02 & 6.460e-02 & 4.745e-05 & 1.277e+00 & 8.320e-05 & 6.757e-01 & 1.232e-04 & 3.970e-01 & 1.654e-04 & 2.543e-01 & 2.078e-04 & 1.752e-01 + 75 & 1e+07 & 3.866e-03 & 1.047e-01 & 2.153e-01 & 1.999e-04 & 1.302e+00 & 2.313e-04 & 6.907e-01 & 2.602e-04 & 4.065e-01 & 2.957e-04 & 2.607e-01 & 3.349e-04 & 1.797e-01 + 100 & 1e+00 & 2.918e-10 & 2.714e-08 & 3.008e+00 & 1.835e-06 & 1.387e+00 & 4.054e-06 & 7.370e-01 & 7.499e-06 & 4.344e-01 & 1.216e-05 & 2.790e-01 & 1.779e-05 & 1.926e-01 ccccccccccccccc 75 & 1e+00 & 4.136e-10 & 1.178e-08 & 3.171e-05 & 9.155e-07 & 3.541e+00 & 2.045e-06 & 1.884e+00 & 3.819e-06 & 1.110e+00 & 6.249e-06 & 7.125e-01 & 9.213e-06 & 4.911e-01 + 75 & 3e+00 & 1.241e-09 & 3.512e-08 & 1.541e-05 & 1.149e-06 & 3.543e+00 & 2.545e-06 & 1.886e+00 & 4.727e-06 & 1.111e+00 & 7.705e-06 & 7.129e-01 & 1.132e-05 & 4.913e-01 + 75 & 1e+01 & 4.137e-09 & 1.165e-07 & 9.907e-06 & 1.381e-06 & 3.544e+00 & 3.039e-06 & 1.886e+00 & 5.622e-06 & 1.111e+00 & 9.136e-06 & 7.129e-01 & 1.339e-05 & 4.914e-01 + 75 & 3e+01 & 1.241e-08 & 3.488e-07 & 8.886e-06 & 1.575e-06 & 3.545e+00 & 3.458e-06 & 1.886e+00 & 6.383e-06 & 1.111e+00 & 1.035e-05 & 7.131e-01 & 1.515e-05 & 4.915e-01 + 75 & 1e+02 & 4.137e-08 & 1.160e-06 & 1.041e-05 & 1.858e-06 & 3.548e+00 & 4.067e-06 & 1.888e+00 & 7.490e-06 & 1.112e+00 & 1.212e-05 & 7.138e-01 & 1.771e-05 & 4.920e-01 + 75 & 3e+02 & 1.241e-07 & 3.479e-06 & 1.594e-05 & 2.280e-06 & 3.558e+00 & 4.968e-06 & 1.893e+00 & 9.115e-06 & 1.115e+00 & 1.470e-05 & 7.156e-01 & 2.141e-05 & 4.932e-01 + 75 & 1e+03 & 4.137e-07 & 1.158e-05 & 3.465e-05 & 3.057e-06 & 3.592e+00 & 6.606e-06 & 1.911e+00 & 1.203e-05 & 1.125e+00 & 1.928e-05 & 7.219e-01 & 2.794e-05 & 4.975e-01 + 75 & 3e+03 & 1.241e-06 & 3.474e-05 & 8.716e-05 & 4.248e-06 & 3.679e+00 & 9.071e-06 & 1.954e+00 & 1.636e-05 & 1.150e+00 & 2.599e-05 & 7.377e-01 & 3.e-05 & 5.082e-01 + 75 & 1e+04 & 4.135e-06 & 1.156e-04 & 2.591e-04 & 6.364e-06 & 3.854e+00 & 1.335e-05 & 2.043e+00 & 2.372e-05 & 1.201e+00 & 3.717e-05 & 7.695e-01 & 5.278e-05 & 5.299e-01 + 75 & 3e+04 & 1.240e-05 & 3.471e-04 & 7.321e-04 & 9.064e-06 & 3.997e+00 & 1.870e-05 & 2.115e+00 & 3.270e-05 & 1.242e+00 & 5.049e-05 & 7.953e-01 & 7.073e-05 & 5.473e-01 + 75 & 1e+05 & 4.136e-05 & 1.155e-03 & 2.379e-03 & 1.258e-05 & 3.910e+00 & 2.557e-05 & 2.066e+00 & 4.398e-05 & 1.213e+00 & 6.677e-05 & 7.762e-01 & 9.214e-05 & 5.341e-01 + 75 & 3e+05 & 1.241e-04 & 3.463e-03 & 7.089e-03 & 1.639e-05 & 3.873e+00 & 3.293e-05 & 2.046e+00 & 5.576e-05 & 1.201e+00 & 8.330e-05 & 7.685e-01 & 1.132e-04 & 5.288e-01 + 75 & 1e+06 & 4.136e-04 & 1.153e-02 & 2.361e-02 & 2.278e-05 & 3.889e+00 & 4.494e-05 & 2.055e+00 & 7.421e-05 & 1.206e+00 & 1.081e-04 & 7.720e-01 & 1.439e-04 & 5.312e-01 + 75 & 3e+06 & 1.240e-03 & 3.468e-02 & 7.111e-02 & 4.999e-05 & 3.928e+00 & 9.063e-05 & 2.075e+00 & 1.361e-04 & 1.218e+00 & 1.830e-04 & 7.799e-01 & 2.288e-04 & 5.367e-01 + 75 & 1e+07 & 4.137e-03 & 1.151e-01 & 2.355e-01 & 2.541e-04 & 3.910e+00 & 2.887e-04 & 2.068e+00 & 3.128e-04 & 1.214e+00 & 3.444e-04 & 7.777e-01 & 3.808e-04 & 5.353e-01 + 100 & 1e+00 & 3.121e-10 & 2.816e-08 & 2.803e+00 & 1.861e-06 & 3.889e+00 & 4.125e-06 & 2.063e+00 & 7.649e-06 & 1.214e+00 & 1.242e-05 & 7.784e-01 & 1.820e-05 & 5.363e-01 ccccccccccccccc 75 & 1e+00 & 4.383e-10 & 1.256e-08 & 3.413e-05 & 8.826e-07 & 5.454e+00 & 1.973e-06 & 2.903e+00 & 3.688e-06 & 1.710e+00 & 6.037e-06 & 1.097e+00 & 8.904e-06 & 7.561e-01 + 75 & 3e+00 & 1.314e-09 & 3.743e-08 & 1.868e-05 & 1.104e-06 & 5.454e+00 & 2.448e-06 & 2.903e+00 & 4.552e-06 & 1.710e+00 & 7.424e-06 & 1.097e+00 & 1.092e-05 & 7.561e-01 + 75 & 1e+01 & 4.383e-09 & 1.242e-07 & 1.351e-05 & 1.316e-06 & 5.455e+00 & 2.902e-06 & 2.903e+00 & 5.376e-06 & 1.710e+00 & 8.746e-06 & 1.097e+00 & 1.283e-05 & 7.562e-01 + 75 & 3e+01 & 1.314e-08 & 3.719e-07 & 1.269e-05 & 1.493e-06 & 5.457e+00 & 3.284e-06 & 2.904e+00 & 6.074e-06 & 1.711e+00 & 9.865e-06 & 1.097e+00 & 1.445e-05 & 7.564e-01 + 75 & 1e+02 & 4.383e-08 & 1.237e-06 & 1.463e-05 & 1.755e-06 & 5.462e+00 & 3.851e-06 & 2.907e+00 & 7.108e-06 & 1.712e+00 & 1.152e-05 & 1.098e+00 & 1.685e-05 & 7.571e-01 + 75 & 3e+02 & 1.314e-07 & 3.710e-06 & 2.103e-05 & 2.151e-06 & 5.478e+00 & 4.703e-06 & 2.915e+00 & 8.651e-06 & 1.717e+00 & 1.397e-05 & 1.101e+00 & 2.038e-05 & 7.590e-01 + 75 & 1e+03 & 4.383e-07 & 1.235e-05 & 4.208e-05 & 2.887e-06 & 5.530e+00 & 6.267e-06 & 2.941e+00 & 1.145e-05 & 1.732e+00 & 1.840e-05 & 1.111e+00 & 2.670e-05 & 7.656e-01 + 75 & 3e+03 & 1.314e-06 & 3.705e-05 & 1.005e-04 & 4.024e-06 & 5.663e+00 & 8.640e-06 & 3.009e+00 & 1.565e-05 & 1.771e+00 & 2.493e-05 & 1.135e+00 & 3.590e-05 & 7.821e-01 + 75 & 1e+04 & 4.383e-06 & 1.234e-04 & 2.871e-04 & 6.052e-06 & 5.956e+00 & 1.278e-05 & 3.157e+00 & 2.283e-05 & 1.856e+00 & 3.590e-05 & 1.189e+00 & 5.108e-05 & 8.186e-01 + 75 & 3e+04 & 1.314e-05 & 3.700e-04 & 7.894e-04 & 8.629e-06 & 6.150e+00 & 1.794e-05 & 3.254e+00 & 3.157e-05 & 1.911e+00 & 4.893e-05 & 1.223e+00 & 6.871e-05 & 8.420e-01 + 75 & 1e+05 & 4.383e-05 & 1.232e-03 & 2.542e-03 & 1.197e-05 & 6.023e+00 & 2.456e-05 & 3.182e+00 & 4.254e-05 & 1.868e+00 & 6.488e-05 & 1.195e+00 & 8.977e-05 & 8.224e-01 + 75 & 3e+05 & 1.314e-04 & 3.693e-03 & 7.551e-03 & 1.557e-05 & 5.968e+00 & 3.161e-05 & 3.153e+00 & 5.394e-05 & 1.850e+00 & 8.097e-05 & 1.184e+00 & 1.104e-04 & 8.146e-01 + 75 & 1e+06 & 4.383e-04 & 1.228e-02 & 2.508e-02 & 2.163e-05 & 5.970e+00 & 4.315e-05 & 3.154e+00 & 7.189e-05 & 1.851e+00 & 1.053e-04 & 1.184e+00 & 1.406e-04 & 8.150e-01 + 75 & 3e+06 & 1.314e-03 & 3.681e-02 & 7.517e-02 & 4.848e-05 & 5.976e+00 & 8.926e-05 & 3.158e+00 & 1.353e-04 & 1.853e+00 & 1.823e-04 & 1.186e+00 & 2.279e-04 & 8.161e-01 + 75 & 1e+07 & 4.383e-03 & 1.226e-01 & 2.505e-01 & 2.706e-04 & 6.000e+00 & 3.073e-04 & 3.171e+00 & 3.285e-04 & 1.862e+00 & 3.568e-04 & 1.192e+00 & 3.903e-04 & 8.202e-01 + 100 & 1e+00 & 3.305e-10 & 2.884e-08 & 2.655e+00 & 1.800e-06 & 5.900e+00 & 3.995e-06 & 3.129e+00 & 7.417e-06 & 1.841e+00 & 1.206e-05 & 1.180e+00 & 1.767e-05 & 8.127e-01

we show that the free - by - cyclic groups of the form act properly cocompactly on cat(0 ) square complexes . we also show using generalised baumslag - solitar groups that all known groups defined by a 2-generator 1-relator presentation are either sq - universal or are cyclic or isomorphic to . finally we consider free - by - cyclic groups which are not relatively hyperbolic with respect to any collection of subgroups .

1503.01989

the recent far reaching work of agol @xcite and wise @xcite proves that a word hyperbolic group @xmath2 acting properly and cocompactly on a cat(0 ) cube complex must be virtually special , implying that @xmath2 has a finite index subgroup which embeds in a right angled artin group ( raag ) . a host of very strong conclusions then apply to the group @xmath2 , of which the two that will concern us here are being linear ( which we take to mean over @xmath3 although it is even true over @xmath4 ) and ( if @xmath2 is not elementary ) being large , namely @xmath2 has a finite index subgroup surjecting to a non abelian free group . however , if @xmath2 is a finitely presented non hyperbolic group acting properly and cocompactly on a cat(0 ) cube complex , then the above consequences need no longer hold , indeed @xmath2 can even be simple @xcite . therefore suppose we have a class of finitely presented groups which is believed to be a well behaved class , but which contains both word hyperbolic and non word hyperbolic examples . we can ask : first , do all examples in this class have a nice geometric action , namely a proper cocompact action on a cat(0 ) cube complex , and second : do they all enjoy our strong group theoretic properties which are a consequence of being a virtually special group , namely being linear or being large . note that for non word hyperbolic groups , satisfying this geometric condition will not necessarily imply these group theoretic properties . in this paper the classes of groups we will be interested in are the following three : groups of the form @xmath5 for @xmath6 a free group of finite rank @xmath7 and @xmath8 an automorphism of @xmath6 ( we refer to these as `` free - by - cyclic groups '' ) ; the more general class of ascending hnn extensions @xmath9 of finite rank free groups , where rather than @xmath10 having to be an automorphism , as in the free - by - cyclic case , we allow @xmath10 to be any injective endomorphism of @xmath6 ; and finally the class of groups admitting a presentation with 2 generators and 1 relator , which we refer to as 2-generator 1-relator groups . this last class neither contains nor is contained in either of the other two classes but there is considerable overlap . in the free - by - cyclic case , it was recently shown in @xcite that such word hyperbolic groups do act properly and cocompactly on a cat(0 ) cube complex , and therefore are virtually special groups . however gersten in @xcite gave an example of a free - by - cyclic group which can not act properly and cocompactly on any cat(0 ) space , so this result can not hold in general in the non word hyperbolic case . moreover @xcite,@xcite shows that there are free - by - cyclic groups which are not automatic , whereas groups that act nicely on cat(0 ) cube complexes are automatic @xcite . in gersten s example the free group has rank 3 but in section 2 we consider free - by - cyclic groups of the form @xmath11 , none of which are word hyperbolic . therefore it is of interest to show directly that they act properly and cocompactly on cat(0 ) cube complexes , which we do in section 2 . this work is based on unpublished work of bridson and lustig . those authors give us the method of changing the natural topological model of the standard 2 complex ( shown in figure [ 2complex ] ) to get rid of a pocket of positive curvature , and in this way they go on to show that these groups act on 2-dimensional cat(0 ) complexes . we expand on this by showing that one can build these complexes from squares . this also strengthens a result of t. brady @xcite who showed that there is a 2-complex of non - positive curvature made from equilateral triangles with fundamental group @xmath11 . in section 3 we consider 2-generator 1-relator groups . it is conceivable , but very definitely open , that a word hyperbolic 2-generator 1-relator group always acts properly and cocompactly on a cat(0 ) cube complex ( for instance see @xcite conjecture 1.9 ) . however on moving to the non word hyperbolic case we see a different picture emerging because a group acting properly and cocompactly on any cat(0 ) space can not contain a baumslag - solitar group @xmath12 where @xmath13 . thus the examples of such nasty baumslag - solitar groups as @xmath14 mean that we need not always have largeness nor linearity , or even residual finiteness . in fact even restricting to residually finite 2-generator 1-relator groups will not imply linearity in general . this is because it was shown in @xcite that the group @xmath15 is not linear over any field if @xmath16 , and it was pointed out in @xcite that the group @xmath17 which is indeed a 2-generator 1-relator group @xmath18 that is known to be residually finite , contains this as an index 2 subgroup when @xmath19 . however , in looking for a `` large '' property that we hope is held by all 2-generator 1-relator groups except for the soluble groups @xmath20 and @xmath4 , including the non residually finite groups , we are led to the concept of a group @xmath2 being sq - universal : namely that every countable group embeds in a quotient of @xmath2 . it was conjectured by p.m.neumann in @xcite back in 1973 that a non cyclic 1-relator group is either sq - universal or isomorphic to @xmath20 . now it was shown in @xcite that a group having a 1-relator presentation with at least 3 generators is sq - universal , leaving the 2-generator 1-relator case . also @xcite from 1995 showed that all non elementary word hyperbolic groups are sq - universal and this was generalized to non elementary groups which are hyperbolic relative to any collection of proper subgroups in @xcite from 2007 . recently the concept of a group being acylindrically hyperbolic , which is more general than being hyperbolic with respect to a collection of proper subgroups and which implies sq - universality , was introduced in @xcite and studied in @xcite where one application was to 2-generator 1-relator groups . the authors divided these groups into three classes with the first consisting of groups that they could show were acylindrically hyperbolic . we prove in theorem [ sqcs2 ] that all the groups in their second case , which they show are not acylindrically hyperbolic , are indeed sq - universal unless equal to @xmath20 . in fact every group here is formed by taking an hnn extension with base equal to a quotient of some free - by - cyclic group @xmath5 for @xmath8 finite order , along with infinite cyclic edge groups . the proof proceeds by also identifying them as generalized baumslag - solitar groups , whereupon we show more generally in theorem [ sqcs2 ] that any generalized baumslag - solitar group either is sq - universal or is isomorphic to @xmath20 or @xmath4 . this leaves their third case , which is exactly the class of 2-generator 1-relator groups that are ascending hnn extensions of finite rank free groups . here we are not quite able to establish sq - universality of all of these groups not equal to @xmath20 or @xmath4 , though it is known to hold for the free - by - cyclic case , but we do show in corollary [ sq ] that the only possible exception would be a 2-generator 1-relator group equal to a strictly ascending hnn extension of a finite rank free group which either fails to be word hyperbolic and contains no baumslag - solitar subgroup , or does contain a baumslag - solitar subgroup ( but does not contain @xmath21 ) and where all finite index subgroups have first betti number equal to 1 . in both cases it is conjectured that no such examples exist , so we have established p.m.neumanns conjecture for all the known 2-generator 1-relator groups . we also obtain some general unconditional statements , such as corollary [ comm ] which says that if the relator is in the commutator subgroup of @xmath22 then @xmath2 is sq - universal or equal to @xmath21 . in the final section we show that free - by - cyclic groups formed using an automorphism @xmath8 of polynomial growth are not hyperbolic relative to any collection of proper subgroups ( thus sq - universality can not be established for all free - by - cyclic groups using only the result of @xcite ) , but are acylindrically hyperbolic unless @xmath8 has finite order as an outer automorphism . in our final class of ascending hnn extensions of finitely generated free groups , it was shown for the word hyperbolic case in @xcite that @xmath9 acts properly and cocompactly on a cat(0 ) cube complex if @xmath10 is an irreducible endomorphism , although that still leaves the case where @xmath10 is not irreducible but @xmath9 is word hyperbolic . as for a non word hyperbolic group of the form @xmath9 , there are some results on largeness for these groups in @xcite which are not quite exhaustive but make it likely that largeness holds throughout . however the example mentioned above of the injective endomorphism @xmath23 of the rank two free group @xmath24 for @xmath25 and @xmath26 both having modulus greater than 1 , shows that @xmath9 need not be linear over any field . moreover the possible existence in the non word hyperbolic case of baumslag - solitar subgroups @xmath20 where @xmath27 again means that such a group need not have a proper cocompact action on any cat(0 ) space . we finish by showing in theorem [ sap ] that the work in @xcite and the agol - wise machinery answers problem 17.108 in the recent edition of the kourovka notebook @xcite , namely that sapir s example of a strictly ascending hnn extension of @xmath22 is indeed linear . we would like to thank martin bridson and martin lustig for allowing us to reproduce the results from @xcite here . as mentioned before there are many nice consequences of word hyperbolic groups acting properly and cocompactly on cat(0 ) cube complexes . however , no automorphism of @xmath22 is hyperbolic since they all fix the conjugacy class of @xmath28^{\pm1}$ ] , therefore , agol s theorem does not apply and the complexes constructed are not known to be virtually special , though , due to the work of various authors the groups are virtually special . however , there are still advantages to having a group act on a cat(0 ) cube complex . for instance , abelian subgroups are quasi - isometrically embedded , and such groups are biautomatic @xcite,@xcite and have a deterministic solution to the word problem in quadratic time @xcite . groups which act on cat(0 ) square complexes have the further nice property that all of their finitely presented subgroups also act properly and cocompactly on cat(0 ) square complexes . this is proved using a tower argument ( see @xcite , p . 217 ) and the fact that a sub complex of a non - positively curved square complex is itself a non - positively curved square complex ( this may fail in higher dimensions ) . the construction also shows that for @xmath22-by-@xmath4 groups their geometric dimension is equal to their cat(0 ) dimension , namely 2 . we assume that the reader is familiar with the basics of cat(0 ) geometry for which the standard reference is @xcite . we say a metric space is _ non - positively curved _ if for each point there is a neighbourhood which is cat(0 ) . in the following we will study 2 dimensional piecewise euclidean ( @xmath29 ) complexes . these are complexes built from polygonal subsets of @xmath30 by gluing along edges , whereupon we put the natural path metric on the resulting complexes . for full details see @xcite . square complexes are special examples of @xmath29 complexes where all the cells are squares . the following theorems of gromov @xcite allow one to check whether a complex is non - positively curved just by looking at the links of vertices . @xcite a @xmath29 complex with finitely many isometry types of cells is non positively curved if and only if the link of each vertex is a cat(1 ) space . in the two dimensional case the link of any vertex is a graph and so this can be reduced to the following . @xcite a graph is cat(1 ) if it contains no circuits of length less than @xmath31 . we say that an action is _ proper _ if for each compact set @xmath32 the set @xmath33 is finite . as these groups will be the fundamental groups of non - positively curved spaces , they have an action on the universal cover . since the spaces are compact the action will be proper and cocompact and it will also be a free action since these groups are torsion free . the groups @xmath34 that we shall be concerned with are mapping tori of @xmath35 by a single automorphism @xmath36 . these groups have presentations of the form @xmath37 we start by considering the case of automorphisms which are of finite order . if @xmath38 has order @xmath39 in @xmath40 then @xmath34 is the fundamental group of a non - positively curved 2-complex . furthermore , this is finitely covered by @xmath41 where @xmath42 is a graph with fundamental group @xmath43 . every finite order automorphism @xmath44 of @xmath43 can be realised as an isometry of a finite graph @xmath42 ; see for instance @xcite theorem 2.1 . let @xmath45 /(x,0)\sim ( \phi(x),1)$ ] . @xmath46 is locally isometric to @xmath47 where @xmath48 is a contractible subset of a graph . this will be cat(0 ) and so @xmath46 is non - positively curved . if we take the cover corresponding to the obvious map to @xmath49 this will be @xmath50 . we now want to look at automorphisms of infinite order . we require the following lemmas to ensure that we account for the general case with our construction . @xmath34 is defined up to isomorphism by @xmath51\in \textnormal{out}(f_2)$ ] . if @xmath52 , for @xmath53 the inner automorphism conjugation by @xmath54 , then @xmath55 @xmath34 is defined up to isomorphism by the conjugacy class of @xmath51\in \textnormal{out}(f_2)$ ] . let @xmath56 then the following map defines an isomorphism . @xmath57 as such we will restrict to conjugacy in @xmath58 . in what follows we will need the following matrices : @xmath59 let @xmath60 be a matrix . then at least one of @xmath61 or @xmath62 is conjugate to a matrix with all non - negative coefficients . we will split into 2 cases . first we will deal with the case of matrices with no entry equal to 0 . conjugating and multiplying by @xmath63 , we may assume that @xmath64 . we may now replace @xmath54 by @xmath65 to make @xmath66 . we note that every matrix of this form in @xmath67 can not have one negative entry , as if this is the case then we see @xmath68 . we note that conjugation by @xmath69 changes the signs of @xmath70 and @xmath71 using this we can assume that @xmath72 , then we know that @xmath73 or @xmath74 . if we are in the second case we are done , so we assume we are in the first case . we now conjugate by @xmath75 : @xmath76 we know that @xmath77 are @xmath78 so we see that @xmath79 . now @xmath80 so once again we know that @xmath81 are non - negative so @xmath82 . if @xmath54 had an entry equal to 0 then either @xmath54 or @xmath83 is triangular . further conjugating by @xmath63 and multiplying by @xmath84 we can assume that @xmath85 and @xmath86 . we now have two cases ; namely @xmath87 . in the case where @xmath88 we can conjugate by @xmath69 to make @xmath89 . if @xmath90 then this matrix has order 2 and we do not worry about this case . the following lemmas are from @xcite . if @xmath54 is a non - diagonal matrix with all entries non - negative , then there is a subtraction of one row from another , which reduces the sum of the entries and produces a non - negative result . let @xmath91 with non - negative entries . then there is a unique sequence of row subtractions which keep @xmath54 non - negative and reduce it to @xmath92 or @xmath63 . we can think of these operations as multiplication by @xmath93 or @xmath94 which gives us the following result . the semigroup generated by @xmath95 and @xmath93 contains a conjugate of every infinite order matrix in @xmath96 . as such we will only need to realise the automorphisms corresponding to these in our groups . we will now construct non - positively curved complexes with @xmath34 as their fundamental groups , when @xmath44 is in the semigroup generated by @xmath97 we see that from the above this gives us all @xmath22-by-@xmath4 groups . we start with the obvious 2-complex shown in figure [ 2complex ] for the automorphism @xmath98 . this has a repeated corner which means it can not support a metric of non - positive curvature . to get rid of the repeated corner we cut our building blocks along the dotted line identifying the triangles with the repeated corner , resulting in our basic building blocks shown in figure [ buildingblocks ] . by the above lemmas we can without loss of generality assume that our automorphism has the form @xmath99 where @xmath100 or @xmath98 and @xmath10 is one of the following finite order automorphisms : @xmath101 we can assume that we only apply one of these and we do it at the end . this is because in @xmath102 the first and second give central elements whereas @xmath103 is equal to the composition @xmath104 , but @xmath105 . we can now glue these together to get an automorphism which is in the class defined previously , so up to isomorphism we have all groups @xmath34 . we do this in the following way : in figure 2 let @xmath106 so that we have the positive and the negative vertices @xmath107 at each end of @xmath108 . on performing the given gluing we have that @xmath107 are not identified , but if we further stick these two vertices together by identifying @xmath109 with @xmath110 respectively then the resulting 2-complex has fundamental group @xmath111 or @xmath112 . now suppose that our automorphism @xmath99 , where @xmath10 is one of the four special finite order automorphisms above . for each @xmath113 between 0 and @xmath114 we have a copy @xmath115 of the 2-complex associated to either @xmath98 or @xmath116 in figure 2 which contains the edge @xmath117 . we then glue @xmath115 to @xmath118 by the identity between the @xmath119 , which means that the vertex @xmath120 is identified with @xmath121 . finally we glue @xmath122 back to @xmath123 by identifying the @xmath124 with @xmath109 so that @xmath125 becomes equal to @xmath126 . we say a vertex is at _ time @xmath113 _ if it is the vertex where @xmath127 and @xmath117 meet . we start with the case of vertices of time not equal to 0 , thus these will be where our complexes @xmath128 and @xmath115 are glued together by the identity between the @xmath129 . the link of such a vertex is shown in figure [ links ] . we now want to assign angles such that there are no circuits of length less than @xmath31 . if we assign angles as in figure [ buildingblocks ] , then we get two types of link as shown in figure [ links ] . figure [ links ] i ) corresponds to the automorphisms at the @xmath113-th stage both being @xmath116 or both being @xmath98 . figure [ links ] ii ) corresponds to when one automorphism is @xmath98 and one is @xmath116 . we can see in either case that the link has no circuits of length less than @xmath31 . we now look at the case of the vertex at time 0 . this will have to take into account the map @xmath10 . we can consider the link as being split into 2 halves as shown in figure [ linkhalves ] . the finite order maps defined earlier give vertex identifications . there are 16 possible links we may get in this way , corresponding to which automorphisms meet and to one of the 4 finite order automorphisms . all of these give a link which is homeomorphic to the 1 skeleton of a tetrahedron with the set of angles depicted in figure [ links ] . with a more careful assignment of angles we can see that the complexes above can be made into square complexes . we split the automorphism @xmath44 into one of three types depending on its decomposition in the semigroup described earlier : 1 . @xmath130 or @xmath131 2 . @xmath132 or @xmath133 where @xmath134 and @xmath135 3 . all other automorphisms . we will assign angles to our building blocks based on which meet at time @xmath113 . we have depicted the two cases of our building blocks meeting at a vertex . in case 1 each time our building blocks meet , it will be of the type depicted in figure [ squares ] i ) . in this case we keep the angle assignment we had before and make the edges labelled @xmath136 or @xmath137 on the vertical sides of the two rectangles of length 2 and the other edges of length 1 . we then make the edge between the 2 triangles of length @xmath138 and subdivide the rectangles into squares of edge length 1 , thus replacing the two triangles with a new building block which is the square formed by gluing them together . the link of the original vertices in this complex are depicted in figure [ links2 ] i ) , where the edges corresponding to @xmath139 and @xmath140 have been suppressed as they have valence 2 . in case 2 each time our building blocks meet , it will be of the type depicted in figure [ squares ] ii ) . in this case we collapse the triangles to lines and then subdivide the resulting rectangles into 2 squares of side length 1 . the link of the original vertices in this complex are depicted in figure [ links2 ] ii ) , where the edges corresponding to @xmath139 and @xmath140 have been suppressed as they have valence 2 . in case 3 we will have a mix of both figures [ squares ] i ) and ii ) and we collapse all the triangles to lines as we did for case 2 . this introduces degenerate squares from building blocks meeting as in figure [ squares ] i ) , which could affect the topology of the overall complex if there were a cylinder of such squares . however here this will not happen as the only time a cylinder of degenerate squares could occur is if our automorphism is in case 1 . the links of vertices in these complexes are depicted in figure [ case3links ] , where the edges corresponding to @xmath139 and @xmath140 have been suppressed as they have valence 2 . the case of edges of 0 length are the degenerate squares where in fact the two edges become identified . in all the cases we see that the resulting complex will be a non - positively curved square complex . a countable group @xmath2 is said to be sq - universal ( standing for subgroup quotient ) if every countable group can be embedded in a quotient of @xmath2 . this immediately implies that @xmath2 contains a non abelian free group , and in turn is implied by @xmath2 being large ( having a finite index subgroup surjecting to a non abelian free group ) . however an infinite simple group @xmath141 containing @xmath22 would not be sq - universal , nor would a just infinite group such as @xmath142 for @xmath143 . as for examples of groups which are sq - universal , we have all non elementary word hyperbolic groups by @xcite . this means that word hyperbolic groups with property ( t ) provide lots of further examples of groups which are sq - universal but not large . moreover by @xcite a finitely generated group which is hyperbolic relative to any collection of proper subgroups is sq - universal ( or virtually cyclic ) . an important class of groups in this area is 1-relator groups . it was shown in @xcite in 1978 that any group with a presentation of deficiency at least 2 ( thus any group having an @xmath26-generator 1-relator presentation for @xmath143 ) is large , leaving 2-generator 1-relator groups . the question of when such a group @xmath2 contains @xmath22 has been known for some time : yes , unless @xmath2 is isomorphic to a baumslag - solitar group of the form @xmath144 ( where @xmath145 ) or is cyclic . largeness is a different matter ; for instance @xcite showed that the group @xmath12 is large if and only if @xmath25 and @xmath26 are not coprime . in @xcite we undertook extensive computation suggesting that non large groups with a 2-generator 1-relator presentation are few and far between , however there are more than just baumslag - solitar examples . moreover we did not see a clean criterion that presented itself for conjecture but in the case of sq - universality there is a statement that appeared in @xcite in 1973 : a non cyclic 1-relator group is sq - universal unless it is isomorphic to @xmath144 , thus if true this would be equivalent to containing @xmath22 . there have been a few results in this area since then ; for instance @xcite showed in 1974 that a group with an @xmath26-generator 1-relator presentation for @xmath143 is sq - universal ( which was then subsumed by the largeness result mentioned above ) . as for 2-generator 1-relator groups , edjvet s thesis @xcite proves sq - universality in some useful cases . we showed in @xcite corollary 7.5 that it is true if the group is lerf but that is a very strong condition to impose . recently the concept of a group being acylindrically hyperbolic was introduced in @xcite . it holds if our group is non elementary and is relatively hyperbolic with respect to a collection of proper subgroups . we will not need the definition here , just the fact also in @xcite that such a group is sq - universal . this was followed up in @xcite where the theory was applied to various situations , including 1-relator groups to obtain the following . here a subgroup @xmath146 of a group @xmath2 is @xmath147-normal in @xmath2 if @xmath146 is infinite and moreover @xmath148 is infinite for all @xmath149 . the relevance of this is that an @xmath147-normal subgroup of an acylindrically hyperbolic group must also be acylindrically hyperbolic , so for instance @xmath150 being @xmath147-normal in @xmath2 implies that @xmath2 is not acylindrically hyperbolic ( though it could certainly be sq - universal or even large ) . [ 1rel2gen ] ( @xcite proposition 4.20 ) let @xmath2 be a group with two generators and one defining relator . then at least one of the following holds : * @xmath2 is acylindrically hyperbolic ; * @xmath2 contains an infinite cyclic @xmath147-normal subgroup . more precisely , either @xmath2 is infinite cyclic or it is an hnn - extension of the form @xmath151 of a 2-generator 1-relator group @xmath152 with non - trivial center , so that @xmath153 in @xmath146 for some @xmath154 . in the latter case @xmath146 is ( finitely generated free)-by - cyclic and contains a finite index normal subgroup splitting as a direct product of a finitely generated free group with an infinite cyclic group . * @xmath2 is isomorphic to an ascending hnn extension of a finite rank free group . moreover , the possibilities ( i ) and ( ii ) are mutually exclusive . thus this establishes that groups in class ( i ) are sq - universal . we will show the same for groups in ( ii ) then discuss results for ( iii ) . when considering groups in case ( ii ) , we will use the class of generalized baumslag - solitar , or gbs , groups . these can be defined as those finitely generated groups which act on a tree with infinite cyclic vertex and edge stablisers . the two recent papers @xcite and @xcite cover a lot of ground in this area and we now mention the points we will be using , referring to them for more detail . we can describe a gbs group using the graph of groups theory , where a finite graph ( possibly with loops and/or multiple edges ) has labels consisting of a non zero integer at each end of each edge . this label tells us the index of this edge group in the adjacent vertex group , which determines the subgroup uniquely . in general many different finite labelled graphs can give rise to isomorphic gbs groups . one operation that can be performed without change of the underlying group is an elementary collapse . this is when one end of an edge @xmath155 next to a vertex @xmath156 is labelled @xmath157 and the edge is not a self loop . we can then contract this edge and multiply all other labels next to @xmath156 by the label at the other end of @xmath155 . by doing this repeatedly , we may assume that any edge with an end labelled by @xmath157 is a self loop . note that all gbs groups have deficiency 1 , that is they admit a presentation with one more generator than relator . in particular there always exists a surjective homomorphism from any gbs group to @xmath4 . [ sqcs2 ] if the group @xmath2 is as in case ( ii ) of the preceding proposition then @xmath2 is a generalized baumslag - solitar group . moreover any generalized baumslag - solitar group is either sq - universal or it is isomorphic to the baumslag - solitar group @xmath1 for some @xmath158 or is infinite cyclic . for the first part we can use theorem c of @xcite . this states that the non cyclic finitely generated groups of cohomological dimension 2 that have an infinite cyclic @xmath147-normal subgroup are exactly the generalized baumslag - solitar groups . now a 1-relator group has cohomological dimension 2 if the relator is not a proper power by @xcite , but a proper power gives rise to a group with torsion , whereas the groups in case ( ii ) are all torsion free . now let @xmath42 be the underlying graph of our graph of groups that results in the gbs group @xmath2 . as mentioned above we assume that the only edge labels equal to @xmath157 appear on self loops . it is well known that if @xmath42 contains more than one cycle ( here we include self loops as cycles ) then @xmath2 surjects to @xmath22 and so is sq - universal , because we introduce a stable letter for each cycle when forming @xmath2 , and all vertex subgroups can be quotiented out to leave only these stable letters which have no relations between them . it is also known that if @xmath42 is a tree then @xmath2 is virtually @xmath159 for @xmath160 which is large , hence so is @xmath2 . this can be seen by quoting proposition 4.1 of @xcite which states that a group is a gbs group with non trivial centre if and only if it is of the form @xmath5 with @xmath8 having finite order in @xmath161 . now here @xmath2 will certainly have a non trivial centre , namely an element which is a common power of all the generators of the vertex subgroups as these form a generating set for @xmath2 . moreover in the case of a tree the surjective homomorphism @xmath10 from @xmath2 to @xmath4 has the property that no non trivial element of a vertex ( or edge ) group lies in its kernel , as if so then the whole vertex group does , thus so do the neighbouring edge groups and so on across the whole tree . we now assume that @xmath42 has exactly one cycle @xmath162 . first assume this is not a self loop . we pick one edge @xmath155 lying in @xmath162 and remove the interior of @xmath155 to form a tree @xmath163 and a group @xmath146 coming from considering @xmath163 as the corresponding graph of groups . thus we have our homomorphism @xmath164 as above , with @xmath2 obtained from @xmath146 by taking generators @xmath165 of the vertex groups at each end @xmath166 of @xmath155 and then adding a stable letter @xmath167 which results in the presentation @xmath168 where @xmath169 are the labels at each end of @xmath155 , neither of which are 0 or @xmath157 . we now obtain a surjection from @xmath2 to a baumslag - solitar group using the following folklore lemma : let @xmath2 be an hnn extension of the group @xmath146 amalgamating the subgroups @xmath170 via the isomorphism @xmath171 . suppose we have a homomorphism @xmath10 from @xmath146 onto a quotient @xmath172 with @xmath173 isomorphic to @xmath174 such that @xmath44 descends to an isomorphism @xmath175 from @xmath173 to @xmath174 , meaning that @xmath175 is well defined and bijective with @xmath176 . ( this occurs if and only if @xmath177 for @xmath178 the kernels of the restriction of @xmath10 to @xmath170 respectively . ) then on forming the hnn extension @xmath93 of @xmath172 with stable letter @xmath147 amalgamating @xmath173 and @xmath174 via @xmath175 , we have that the original hnn extension @xmath2 has this new hnn extension @xmath93 as a quotient . we define a homomorphism from the free product @xmath179 onto @xmath93 sending @xmath167 to @xmath147 and @xmath180 to @xmath181 . we see that this factors through @xmath2 because any relation in @xmath2 of the form @xmath182 has the left hand side mapped by @xmath10 to @xmath183 and the right hand side to @xmath184 , and these two things are equal in @xmath93 by the hnn construction . consequently in our case we have @xmath2 surjects to @xmath185 , where @xmath186 which is not equal to zero as mentioned above because @xmath146 is formed from a tree , and similarly for @xmath187 . now a baumslag - solitar group @xmath188 is known to be sq - universal if neither of @xmath189 equal @xmath157 , by lemma 1.4.3 of @xcite if @xmath113 and @xmath190 are coprime , and by the well known trick of setting @xmath191 equal to the identity when @xmath192 divides @xmath113 and @xmath190 , to get a surjection to @xmath193 which is virtually free otherwise . as @xmath194 and @xmath195 are both greater than 1 , we have that @xmath2 surjects to an sq - universal group and so itself is sq - universal . we are now only left with the case where there is a single self loop @xmath94 in our graph @xmath42 , so that now @xmath196 and @xmath197 . the above proof also works here by removing @xmath94 this time , unless @xmath198 ( or @xmath199 in which case we replace @xmath200 with the isomorphic group @xmath201 ) . if @xmath94 is all of @xmath42 then we have that @xmath2 is just @xmath1 which is soluble , and so is a genuine exception to being sq - universal . otherwise we note that @xmath202 implies that @xmath203 is mapped by @xmath10 to a generator of @xmath4 . on taking any edge @xmath204 not equal to @xmath94 with one endpoint @xmath205 , let @xmath139 be the label of @xmath204 at this end and @xmath140 the label at the other end of @xmath204 next to the vertex group @xmath206 , say . we must have @xmath140 dividing @xmath139 because of the relation @xmath207 and the fact that @xmath208 . we now choose a particular edge @xmath204 and remove the loop @xmath94 and the vertex @xmath205 from @xmath42 , to form a possibly disconnected graph that is a union of trees @xmath209 , with @xmath210 the tree containing the vertex group @xmath206 . we then take a prime @xmath211 dividing @xmath140 and consider the quotient of @xmath2 formed by setting @xmath203 and @xmath212 equal to the identity . we have a surjective homomorphism with domain the group obtained as a graph of groups from the tree @xmath210 and image @xmath213 , consisting of the homomorphism to @xmath4 in the case of the tree @xmath210 and then composing with the map from @xmath4 to @xmath213 . we now extend this to a homomorphism from @xmath2 onto @xmath214 which is sq - universal as follows : send the stable letter @xmath167 to @xmath215 , so that the relation @xmath216 obtained from the loop @xmath94 now has both sides sent to the identity . moreover this holds for the relation @xmath217 obtained from the edge @xmath204 as @xmath211 divides @xmath140 . finally we send all vertex groups not in the component @xmath210 to the identity , with the @xmath218 relations obtained from the other edges @xmath219 that have endpoint @xmath205 all automatically satisfied . note : this result can be compared to @xcite theorem 6.7 in which the large gbs groups are determined , but there are cases for which the graph @xmath42 is a single cycle where the group @xmath2 is sq - universal but not large . we now come to case ( iii ) , that of @xmath2 being equal to the hnn extension @xmath220 , where @xmath221 is injective but need not be surjective ( if not then we call this a strictly ascending hnn extension of @xmath6 ) . here we can quote results of the first author in @xcite . theorem 5.4 of that paper states that @xmath2 will be sq - universal whenever @xmath10 is an automorphism ( unless @xmath222 or the klein bottle group when the rank @xmath7 is 0 or 1 ) . this is proved by showing that @xmath223 implies that @xmath2 is large , and then invoking olshanski s theorem on the sq - universality of word hyperbolic groups and the result in @xcite that not containing @xmath21 and being word hyperbolic are equivalent in the class of groups @xmath9 when @xmath10 is an automorphism . in the case where @xmath2 is a strictly ascending hnn extension of a finite rank free group , we have further results but they are not quite definitive . again we have that if @xmath223 then @xmath2 is large ( or is equal to @xmath21 or the klein bottle group ) by @xcite corollary 4.6 . however in the strictly ascending case there are examples where @xmath2 does not contain @xmath21 but does contain a baumslag - solitar subgroup , which must be of the form @xmath20 for @xmath27 so that @xmath2 fails to be word hyperbolic . in @xcite it is conjectured that a strictly ascending hnn extension of a finite rank free group is is word hyperbolic if it does not contain a baumslag - solitar subgroup and this conjecture seems to be widely believed , but a proof might well require the machinery of train track maps to be developed in full for injective endomorphisms of @xmath6 . moreover it is an open question whether a 1-relator group ( or indeed a group with a finite classifying space ) containing no baumslag - solitar subgroups is word hyperbolic , so we would be covered in our case if any of these ( or their intersection ) turned out to be true . as for when @xmath2 contains @xmath20 for @xmath27 , @xcite theorem 4.7 states that either @xmath2 is large , or @xmath2 is itself a baumslag - solitar group of the form @xmath144 , or @xmath224 but @xmath2 has virtual first betti number equal to 1 and it is conjectured that the last case does not occur . putting all this together , we have our result on the sq - universality of 2-generator 1-relator groups : [ sq ] if @xmath2 is a group given by a 2-generator 1-relator presentation that is not @xmath4 , @xmath21 or the klein bottle group then either @xmath2 is an sq - universal group , or @xmath2 is a strictly ascending hnn extension @xmath9 of a free group @xmath6 which is not word hyperbolic and such that : + ( i ) either @xmath2 contains no baumslag - solitar subgroup ( conjecturally this does not occur ) or + ( ii ) @xmath2 contains a baumslag - solitar group @xmath20 for @xmath27 but does not contain @xmath21 and the virtual first betti number of @xmath2 is 1 ( conjecturally this only occurs if @xmath225 for @xmath226 ) . we finish this section with a couple of unconditional results . [ comm ] if @xmath227 and @xmath228 is in the commutator subgroup of @xmath24 ( and without loss of generality cyclically reduced ) then @xmath2 is sq - universal , except when @xmath229 for @xmath228 a cyclic permutation of @xmath230 or its inverse . this proceeds by using the bns invariant @xmath231 of @xmath2 in @xcite and the proof is very similar to theorem d of that paper . the idea is that @xmath232 is an open subset of @xmath233 and if @xmath234 is not all of @xmath233 then we have a homomorphism @xmath235 that expresses @xmath2 as a non - ascending hnn extension . the magnus decomposition of this extension is such that @xmath2 is either in case ( i ) or case ( ii ) of proposition [ 1rel2gen ] , so that @xmath2 is sq - universal by that theorem for case ( i ) or by theorem [ sqcs2 ] for case ( ii ) . otherwise we have @xmath236 as @xmath233 is connected , which means that the kernel of @xmath237 is finitely generated and consequently the 1-relator group @xmath2 can be expressed as @xmath5 , which is sq - universal if @xmath160 . if @xmath238 then @xmath8 is the identity as @xmath2 surjects to @xmath21 , so @xmath239 and therefore admits only the above 2-generator 1-relator presentations by @xcite theorem 4.11 . finally an sq - universal group can be thought of as one with many infinite quotients whereas an infinite residually finite group can be thought of as having many finite quotients . we see that all 2-generator 1-relator groups therefore have many quotients of some kind : a group with a 2-generator 1-relator presentation is either sq - universal or residually finite . this follows from corollary [ sq ] because @xcite proved that a strictly ascending hnn extension of a finite rank free group is residually finite ( though as mentioned , not necessarily linear ) . for the three cases in the last section , we had that the groups in case ( i ) were all acylindrically hyperbolic whereas none in case ( ii ) were . however when considering groups in case ( iii ) for sq - universality , we did this independently of results on acylindrically hyperbolic groups . it can therefore be asked which mapping tori of finite rank free groups are acylindrically hyperbolic and indeed this is exactly problem 8.2 in section 8 of @xcite . moreover a solution just for the 1-relator groups in this class would then completely determine which 2-generator 1-relator groups are acylindrically hyperbolic , which is their problem 8.1 . it is clear that an ascending hnn extension @xmath5 of @xmath6 formed using an automorphism @xmath8 of finite order in @xmath161 will not be acylindrically hyperbolic because of the existence of an infinite order element in the centre . thus a possible answer to problem 8.2 is that all other ascending hnn extensions of @xmath6 are acylindrically hyperbolic with the exception of @xmath20 when @xmath238 . this would imply two mutually exclusive cases for these groups : either they are acylindrically hyperbolic or they are generalized baumslag - solitar groups , and it would also imply in answer to problem 8.1 that a 1-relator group is acylindrically hyperbolic if and only if it does not contain an infinite cyclic @xmath147-normal subgroup . as a partial answer to problem 8.2 we have [ coho ] if a finitely generated group @xmath2 of cohomological dimension 2 has a finite index subgroup @xmath146 splitting over @xmath4 then either @xmath2 is acylindrically hyperbolic or it is a generalized baumslag - solitar group . we can apply @xcite theorem c to @xmath146 as it also has cohomological dimension 2 . this implies that if @xmath146 splits over @xmath240 and @xmath241 is @xmath147-normal then @xmath146 is a generalised baumslag - solitar group and so is @xmath2 by @xcite corollary 3 ( ii ) as it is torsion free . otherwise we can apply corollaries 2.2 and 2.3 of @xcite which state that if @xmath146 is an amalgamated free product or hnn extension over an edge group which is not @xmath147-normal and not equal to a vertex group under any inclusion then @xmath146 is acylindrically hyperbolic and therefore @xmath2 is by @xcite lemma 3.8 . however if @xmath242 is equal to a vertex group then we have @xmath243 which is also a generalised baumslag - solitar group . all ascending hnn extensions of @xmath6 have geometric and thus cohomological dimension 2 , so by proposition [ coho ] we obtain acylindric hyperbolicity unless we have a generalised baumslag - solitar group . they are all also residually finite , but by @xcite corollary 7.7 a generalised baumslag - solitar group is not residually finite unless it is virtually @xmath159 or @xmath20 when @xmath238 . we now specialise to the case where the mapping torus is formed using an automorphism , so we are back in the class of free - by - cyclic groups @xmath244 , where we can say which such groups are acylindrically hyperbolic . of course this will be true if @xmath2 is word hyperbolic or hyperbolic with respect to a collection of proper subgroups . there are plenty of examples of word hyperbolic free - by - cyclic groups when @xmath245 . when @xmath246 there are none , but most are relatively hyperbolic with respect to the peripheral @xmath21 subgroup because they will be the fundamental group of a finite volume hyperbolic 1-punctured torus bundle . the exceptions are when the monodromy has finite order , giving the virtually @xmath247 case which can not be acylindrically hyperbolic , and parabolic monodromy where all groups will be commensurable with @xmath248 . here we consider the case where @xmath249\in \mbox{out}(f_k)$ ] is a polynomially growing automorphism , with recent results on this in @xcite which itself utilises the train track technology of bestvina , feighn and handel . the facts from @xcite section 5 that we need are : + if @xmath249 $ ] is polynomially growing then there is a positive power @xmath250 $ ] in upg@xmath251 , which is the subgroup of polynomially growing outer automorphisms whose abelianised action on @xmath252 has unipotent image . + if @xmath160 then every element of @xmath253 has in its class an automorphism @xmath8 of @xmath6 such that either : + ( i ) there exists a non trivial @xmath8-invariant splitting @xmath254 , so that @xmath8 restricts to an automorphism of @xmath255 and also of @xmath256 . + ( ii ) there exists a non trivial splitting @xmath257 , where @xmath255 is @xmath8-invariant and @xmath258 for @xmath228 an element of @xmath255 . + we also use the following two statements which follow immediately from corollary 4.22 and theorem 1.4 in @xcite : if @xmath2 is a finitely presented group that is torsion free and hyperbolic relative to a collection of proper subgroups @xmath259 ( the peripheral subgroups ) then + ( i ) any baumslag - solitar subgroup of @xmath2 is conjugate into some @xmath260 . + ( ii ) any @xmath260 is malnormal , so that if there is @xmath149 with @xmath261 non trivial then @xmath262 . moreover if there is @xmath149 with @xmath263 non trivial then @xmath264 . [ poly ] if @xmath249\in\mbox{out}(f_k)$ ] is a polynomially growing automorphism then @xmath244 is not hyperbolic relative to any collection of proper subgroups , but it is acylindrically hyperbolic unless @xmath249 $ ] has finite order in @xmath161 . by taking a power of @xmath8 , which corresponds to a finite index subgroup , we can assume that @xmath249 $ ] is in upg@xmath251 . thus let us repeatedly apply options ( i ) and ( ii ) until we have decomposed @xmath6 into a free product of cyclic groups . these provide splittings over @xmath4 of this finite index subgroup and so corollary [ ahyp ] immediately implies the second part of the statement . we can picture this process of repeatedly splitting @xmath2 over @xmath4 as a finite rooted tree , as in @xcite lemma 5.10 which involves describing it as a hierarchy . we have @xmath6 at the root vertex , with every other vertex being labelled by a proper non trivial free factor of @xmath6 . to explain this , we split @xmath5 using either case ( i ) or case ( ii ) , which in general changes the automorphism @xmath8 but only to something equal in @xmath161 which we can also call @xmath8 for now . if case ( i ) is used then we have an @xmath8-invariant decomposition @xmath254 , with stable letter @xmath167 inducing @xmath8 by conjugation , which means that @xmath265 has been split as an amalgamated free product @xmath266 over @xmath267 . we then draw two vertices labelled @xmath255 and @xmath256 as immediate descendents of the root vertex @xmath6 . if however case ( ii ) is used to split @xmath268 where @xmath269 has rank @xmath270 and @xmath271 for @xmath272 , then this corresponds to the hnn extension @xmath273 where @xmath274 , so now @xmath275 is the stable letter of this hnn extension conjugating the infinite cyclic groups @xmath276 and @xmath277 . in this case we have only one immediate descendent vertex @xmath269 of the root . having done this once , as the new vertex labels @xmath278 or @xmath269 will be @xmath8-invariant , we can take the appropriate restriction of @xmath8 and replace @xmath279 by the subgroup @xmath280 , @xmath281 or @xmath282 , which is also free by cyclic but where the free part has smaller rank . moreover this restriction will also be a polynomially growing outer automorphism whose abelianised action has unipotent image , so it is still in upg . thus we can then ( on changing the restriction of @xmath8 by an inner automorphism ) apply case ( i ) or case ( ii ) again , creating the next level of descendents and labelling them accordingly . this continues until we reach the leaf vertices ( those with no descendents ) , which are each labelled by an infinite cyclic free factor of @xmath6 , so no further splitting occurs . however each time we form the descendents of a vertex by applying case ( i ) or ( ii ) to the free factor of @xmath6 labelling this vertex , we are liable to change the given automorphism within the outer automorphism group of this free factor . now we also need to keep track of the actual automorphisms and how they are induced , so there are further labels on each vertex ( apart from the leaf vertices ) consisting of an automorphism of @xmath6 equal in @xmath161 to @xmath8 , along with a `` stable letter '' which is an element of @xmath5 inducing this automorphism under conjugation . this is defined inductively on the levels of the tree as follows : on splitting our root vertex @xmath6 , we have an automorphism equal to @xmath8 in @xmath161 respecting this splitting . now taking the composition of @xmath8 with an inner automorphism does not change @xmath5 , so we can assume that @xmath8 respects this splitting , whereupon we label the root vertex @xmath6 with this @xmath8 and the overall stable letter @xmath167 that induces @xmath8 by conjugation . this completes our labelling for the zeroth level . now suppose that @xmath283 are vertices on each successive level leading to the vertex @xmath284 , which is not a cyclic group as otherwise it is a leaf vertex and therefore requires no further labelling . therefore @xmath284 has successor(s ) @xmath285 in case ( ii ) ( as well as @xmath286 if case ( i ) is taken ) . in order to describe the labels for @xmath285 ( and @xmath286 ) if this too is not a leaf vertex , we suppose that @xmath284 is already labelled by the automorphism @xmath287 where @xmath288 is an element of @xmath289 and @xmath290 stands for the inner automorphism of @xmath6 consisting of conjugation by @xmath291 . moreover we assume that both @xmath284 and @xmath285 ( and @xmath292 are @xmath293-invariant . we further assume that also labelling @xmath284 is the element @xmath294 which we call the stable letter @xmath295 . of course if now @xmath285 ( or @xmath286 ) is cyclic then we do not need to label these any further , but otherwise we apply the splitting to @xmath285 under the automorphism obtained by restricting @xmath293 to @xmath285 , thus we obtain an automorphism of @xmath285 equivalent to @xmath296 in out@xmath297 which preserves @xmath298 ( and @xmath299 ) and which acts accordingly in case ( ii ) . thus there is an element @xmath300 such that we can define the automorphism @xmath301 of @xmath6 to be @xmath302 thus @xmath301 restricts to an automorphism of @xmath285 that preserves @xmath298 ( and @xmath299 ) . we then further label the vertex @xmath285 by this automorphism @xmath301 and the stable letter @xmath303 that will induce this map by conjugation . having completed the labelling , let us reverse the whole process and build up the free - by - cyclic group @xmath2 from these splittings . let us start at a leaf vertex @xmath304 of maximum depth . as the automorphism @xmath305 obtained from the vertex @xmath306 above preserves @xmath307 and @xmath308 is part of a free basis for @xmath6 , we must have @xmath309 as @xmath310=[\alpha]\in\mbox{upg}(f_k)$ ] and so all eigenvalues of the abelianised map equal 1 . thus @xmath311 giving us a copy of @xmath312 in @xmath2 , which must therefore lie in a subgroup @xmath162 which is a conjugate of one of the peripheral subgroups @xmath260 and so @xmath162 is also malnormal . if @xmath306 gave rise in case ( i ) to @xmath304 and @xmath313 , we would also have @xmath314 by maximality of depth . thus @xmath315 and the same stable letter @xmath316 commutes with @xmath317 , so we conclude by malnormality that @xmath318 . if however we had case ( ii ) then again there is @xmath317 with @xmath315 but now we have @xmath309 and @xmath319 . thus @xmath320 so again @xmath321 by malnormality . hence in either case we obtained from a leaf vertex a copy of @xmath322 lying in a conjugate @xmath162 of a peripheral subgroup and shown that @xmath323 is in @xmath162 too . we now contract edges starting at the leaf vertices , one edge at a time if the lower level vertex of this edge has one immediate descendent or a pair of vertices otherwise . we suppose that at each leaf vertex @xmath324 of the current contraction of the tree , we have @xmath325 lying in a conjugate @xmath162 of a peripheral subgroup ( although @xmath162 could vary over the current leaf vertices ) . first suppose that @xmath324 is the only current descendent of @xmath326 , so that @xmath324 was obtained by applying case ( ii ) to @xmath326 and the labelling indicates that @xmath327 , with @xmath328 sending @xmath275 to @xmath329 for @xmath330 a word in @xmath324 . thus @xmath331 , but @xmath332 for @xmath333 so @xmath334 is already in @xmath162 . consequently @xmath335 implies that @xmath275 and therefore @xmath336 is in @xmath162 as well . now suppose that there are other current descendents of @xmath326 in addition to @xmath324 . this means that after further contraction if necessary there will be the two leaf vertices @xmath324 and @xmath337 which are the descendents of @xmath338 , and we have by our assumption that @xmath325 lies in @xmath162 and similarly @xmath339 also lies in some conjugate @xmath340 of a peripheral subgroup . now @xmath341 as above , but the same argument also says that @xmath342 for @xmath343 . thus the same element @xmath334 is in both @xmath162 and @xmath340 , hence by malnormality they are equal and we have @xmath344 . we can now continue contracting until we are left with the root , concluding that all of @xmath345 lies in @xmath162 , thus @xmath2 is not relatively hyperbolic with respect to proper subgroups . this adds to results in the literature that provide a description of free - by - cyclic groups according to the type of hyperbolicity : for @xmath244 we have that : + @xmath2 is word hyperbolic if and only if no positive power of @xmath8 sends @xmath346 to a conjugate of itself . + @xmath2 is relatively hyperbolic if and only if @xmath249 $ ] is not of polynomial growth . ( the if direction requires one to accept the results of the unpublished manuscript @xcite where the peripheral subgroups are the mapping tori of the polynomially growing subgroups under @xmath249 $ ] of @xmath6 , whereas the only if direction comes from theorem [ poly ] . ) + @xmath2 is acylindrically hyperbolic but not relatively hyperbolic if and only if @xmath249 $ ] is polynomially growing and of infinite order in @xmath161 . ( the same comment as above applies here , with `` if '' and `` only if '' reversed . ) + @xmath2 is not acylindrically hyperbolic if and only if @xmath249\in\mbox{out}(f_k)$ ] has finite order . this leaves us with strictly ascending hnn extensions of finite rank free groups , which we have seen can be less well behaved and indeed need not be linear . however we finish with an example to show that the recent work of hagen and wise in @xcite on cubulation of ascending hnn extensions of finitely generated free groups solves a problem in the kourovka notebook @xcite on linearity of a particular group of this kind . problem 17.108 asks whether the group below is linear . the group is clearly a strictly ascending hnn extension of a finitely generated free group using the injective endomorphism @xmath348 of @xmath22 . corollary 6.20 of @xcite states that this group is virtually special , and hence linear over @xmath4 , if @xmath10 is irreducible and the hnn extension is word hyperbolic . here a reducible endomorphism @xmath10 of @xmath6 is one where there is a free product decomposition @xmath349 for @xmath350 and where each @xmath351 is non trivial ( though @xmath162 might be in which case @xmath352 ) such that @xmath353 is sent into a conjugate of @xmath354 with @xmath113 considered modulo @xmath355 . theorem a in @xcite considered the word hyperbolicity of strictly ascending hnn extensions of a finite rank free group @xmath356 and came up with a result when the endomorphism @xmath10 is an immersion , which is defined to mean that for all @xmath357 with @xmath358 , the word @xmath359 admits no cancellation ( so called because the standard selfmap of the @xmath360-petalled rose is an immersion ) . this implies that if @xmath361 is written as a reduced word on @xmath46 for @xmath362 then no cancellation can occur on the right hand side of @xmath363 . thus the word length of @xmath364 is the sum of that for the @xmath365 . this theorem states that if there is no periodic conjugacy class , meaning that there is no @xmath366 and @xmath367 such that @xmath368 is conjugate to @xmath369 in @xmath356 , then the strictly ascending hnn extension @xmath370 is word hyperbolic . we can apply this to our group as @xmath10 is indeed an immersion and the fact that each generator maps to a word of length two implies that the word length of @xmath368 is @xmath371 times that of @xmath228 . but we can assume by conjugating that @xmath228 is cyclically reduced , in which case @xmath368 is also cyclically reduced , as will be @xmath369 whose word length is @xmath190 times that of @xmath228 . now if @xmath368 and @xmath369 are both cyclically reduced and are conjugate in @xmath22 then they must have the same length and be cyclic permutations of each other , thus @xmath372 . next suppose the word length of @xmath228 is even and set @xmath373 for @xmath374 both half the length of @xmath228 . we have @xmath375 is a cyclic permutation of @xmath376 and so is equal to @xmath377 without cancellation in this expression , where @xmath378 ( but @xmath355 and @xmath147 are not necessarily of equal length ) . thus @xmath379 . as an immersion must be injective , we obtain @xmath380 and @xmath381 is a cyclic permutation of @xmath382 . thus we can continue to cut @xmath228 in half at each stage until it has odd length , whilst still preserving the fact that @xmath368 is a cyclic conjugate of @xmath376 . when we reach this point , we consider the number of appearances of @xmath383 and @xmath384 in @xmath228 , which can not be equal , thus nor can it be equal for @xmath376 or for any cyclic conjugate thereof . but as @xmath385 and @xmath386 both have equal exponent sums , this is true for any element in the image of @xmath228 which is a contradiction . now showing irreducibility of an endomorphism is straightforward in rank 2 because any proper non trivial free factor is just the cyclic group generated by a primitive element . thus to show @xmath10 is irreducible it is enough to rule out there being a primitive element @xmath275 such that @xmath387 or @xmath388 is conjugate to @xmath389 for some @xmath390 . but this has been eliminated for @xmath391 , and also for @xmath392 by considering @xmath388 and @xmath393 respectively . d.t.wise , _ from riches to raags : 3-manifolds , right - angled artin groups , and cubical geometry _ , cbms regional conference series in mathematics no . 117 , american mathematical society , providence , ri , 2012 .

we report the results of a study of the rest - frame ultraviolet ( uv ) spectrum of the cosmic eye ( j213512.73 ) , a luminous ( ) lyman break galaxy at magnified by a factor of via gravitational lensing by foreground mass concentrations at and 0.33 . the spectrum , recorded at high resolution and signal - to - noise ratio with the esi spectrograph on the keckii telescope , is rich in absorption features from the gas and massive stars in this galaxy . the interstellar absorption lines are resolved into two components of approximately equal strength and each spanning several hundred in velocity . one component has a net blueshift of relative to the stars and hii regions and presumably arises in a galaxy - scale outflow similar to those seen in most star - forming galaxies at . the other is more unusual in showing a mean _ redshift _ of relative to ; possible interpretations include a merging clump , or material ejected by a previous star formation episode and now falling back onto the galaxy , or more simply a chance alignment with a foreground galaxy . in the metal absorption lines , both components only partially cover the ob stars against which they are being viewed . however , there must also be more pervasive diffuse gas to account for the near - total covering fraction of the strong damped ly line , indicative of a column density(hi) . we tentatively associate this neutral gas with the redshifted component , and propose that it provides the dust ` foreground screen ' responsible for the low ratio of far - infrared to uv luminosities of the cosmic eye . the civ p cygni line in the stellar spectrum is consistent with continuous star formation with a salpeter initial mass function , stellar masses from 5 to 100 , and a metallicity . compared to other well - studied examples of strongly lensed galaxies , we find that the young stellar population of the cosmic eye is essentially indistinguishable from those of the cosmic horseshoe and ms1512-cb58 . on the other hand , the interstellar spectra of all three galaxies are markedly different , attesting to the real complexity of the interplay between starbursts and ambient interstellar matter in young galaxies observed during the epoch when cosmic star formation was at its peak . [ firstpage ] cosmology : observations galaxies : evolution galaxies : starburst galaxies : individual ( cosmic eye )

0910.0840

high redshift star - forming galaxies are often identified by a break in their ultraviolet continuum that is due to the lyman limit , partly from interstellar ( within the galaxy ) and primarily from intergalactic hi absorption below 912(steidel et al . 1996 ) . since the discovery of these ` lyman break galaxies ' ( or lbgs ) , samples of galaxies at @xmath64 have increased a thousand - fold . despite the variety of methods now employed to select high redshift galaxies , lbgs remain the most common and most extensively studied class of such objects . detailed studies of the physical properties of _ individual _ galaxies have in general been limited by their faintness ( @xmath14 , steidel et al . 1999 ; reddy et al . 2008 ) . in a few cases , however , gravitational lensing by foreground massive galaxies , groups , or clusters has afforded rare insights into the stellar populations and interstellar media of galaxies at @xmath64 ( e.g. pettini et al . 2000 , 2002 ; teplitz et al . 2000 ; lemoine - busserolle et al . 2003 ; smail et al . 2007 ; swinbank et al . 2007 ; cabanac , valls - gabaud , & lidman 2008 ; siana et al . 2008 , 2009 ; finkelstein et al . 2009 ; hainline et al . 2009 ; quider et al . 2009 ; swinbank et al . 2009 ; yuan & kewley 2009 and references therein ) . the order - of - magnitude boost in flux provided by strong lensing makes it possible to record the spectra of these galaxies with a combination of high resolution and high signal - to - noise ratio ( s / n ) which for normal , unlensed galaxies will have to wait until the next generation of 30+m optical / infrared telescopes . focusing on the rest - frame uv spectra in particular , our group has so far published observations of two of these strongly lensed high redshift galaxies : ms1512-cb58 and j1148 + 1930 ( cb58 and the ` cosmic horseshoe ' respectively ; pettini et al . 2002 ; quider et al . both are @xmath15 galaxies at redshifts @xmath16 magnified by a factor of @xmath17 ( seitz et al . 1998 ; dye et al . 2008 ) . a considerable amount of data that are inaccessible to typical low - resolution studies is contained in these rest - uv spectra , including information on interstellar gas composition and kinematics , on the initial mass function ( imf ) of starbursts at high @xmath18 , and clues to the geometry of the gas and stars . one of the motivations for this work is to assess the range of properties possessed by high redshift star - forming galaxies . by comparing the uv spectra of cb58 and the cosmic horseshoe , quider et al . ( 2009 ) showed that the young stellar populations of these galaxies are very similar , as are the metallicities and kinematics of their interstellar media . on the other hand , the two galaxies exhibit clear differences in the covering fractions of their stars by the interstellar gas , differences which are reflected in the strikingly different morphologies of their lyman @xmath9 lines ( a damped absorption profile in cb58 and a double - peaked emission profile in the horseshoe ) . studies at other wavelengths ( teplitz et al . 2000 ; hainline et al . 2009 ) have also shown that cb58 and the cosmic horseshoe have comparable star formation rates , @xmath19100 yr@xmath20 , at the upper end of the range of values measured in star - forming galaxies at @xmath63 , and similar dynamical masses , @xmath21 , typical of luminous galaxies at this epoch ( pettini et al . 2001 ; erb et al . 2006b , c ; reddy et al . 2006 ) . clearly , several examples of strongly lensed galaxies need to be studied at such levels of detail for a full characterization of galaxy properties at these redshifts . fortunately , many new cases have been discovered recently , mostly from dedicated searches in imaging data from the sloan digital sky survey ( estrada et al . 2007 ; belokurov et al . 2007 , 2009 ; ofek et al . 2008 ; shin et al . 2008 ; kubo et al . 2009 ; lin et al . 2009 ; wen et al . other surveys have capitalised on the superior spatial resolution of _ hubble space telescope ( hst ) _ to identify high redshift galaxies strongly lensed into multiple images or arcs ; the @xmath22 galaxy which is the focus of the present paper , dubbed the ` cosmic eye ' , was indeed discovered by smail et al . ( 2007 ) from an _ hst _ snapshot program targeting high luminosity x - ray clusters . in this paper we present high resolution observations of the rest - frame uv spectrum of the cosmic eye . the paper is organized as follows . in section [ sec : eyesummary ] we summarize the known properties of the cosmic eye from previous studies at a variety of wavelengths , while section [ sec : observations ] has details of our observations and data reduction . we analyze the interstellar spectrum of the cosmic eye in section [ sec : ism ] , and the composite spectrum of its young stellar population in section [ sec : stars ] . section [ sec : intervening ] focuses on the numerous intervening absorption systems that potentially confound our interpretation of the galaxy s intrinsic features . in section [ sec : discussion ] we discuss the implications of our findings ; finally , we summarize our main conclusions in section [ sec : conclusions ] . throughout the paper , we adopt a cosmology with @xmath23 , @xmath24 , and @xmath25 km s@xmath20 mpc@xmath20 . entrance slit of esi used for the observations reported here . ] among the bright , strongly lensed , high redshift galaxies , the cosmic eye is the most well - studied to date after cb58 . here we summarize available information which is relevant to our analysis . [ fig : fullspectrum ] the cosmic eye , also known as lbg j213512.73@xmath0010143 , was so named because it consists of two bright arcs extending @xmath26 which nearly fully surround a compact galaxy ( figure [ fig : eyeimage ] ) . from low resolution spectroscopy , the central galaxy was classified as a massive early - type spiral , or s0 , galaxy at @xmath4 , and from rest - frame optical spectroscopy the source redshift was estimated to be @xmath27 ( smail et al . 2007 ) . the detailed lensing model by dye et al . ( 2007 ) concluded that the lensing of the cosmic eye is attributable to both the galaxy at @xmath28 and a foreground cluster , macs j2135.2 - 0102 , at @xmath29 . their models require two source components at similar redshifts : a main component which creates the eye s ringed structure , and a secondary component , offset @xmath30 to the west , responsible for producing faint extensions at either end of the northern arc . the total magnification of the system is estimated to be @xmath3 ; when this is taken into account , the observed magnitude @xmath31 corresponds to an intrinsic luminosity @xmath32@xmath33 compared to the luminosity function of @xmath34 galaxies estimated by steidel et al . ( 1999 ) and more recently reddy et al . ( 2008 ) . the geometry and kinematics of the cosmic eye have begun to be explored . most recently , siana et al . ( 2009 ) used the galaxy s far - infrared luminosity , which they found to be @xmath35 times lower than expected from its rest - frame uv luminosity , to argue that a reddening curve similar to those determined from observations of stars in the large and small magellanic clouds ( lmc / smc ) is more appropriate for the dust in the cosmic eye than the calzetti et al . ( 2000 ) attenuation curve normally adopted for starbursts . the difference between the two reddening laws is one of geometry : the calzetti et al . ( 2000 ) curve is thought to arise from a configuration where stars and dust are mixed together , whereas the lmc / smc ( or . for that matter , milky way ) extinction curves are more appropriate to the reddening produced by a ` screen of dust ' which is in the foreground of the sources of uv light . integral field spectroscopy of h@xmath36 and [ oiii]@xmath37 nebular emission lines with the oh - suppressing infrared integral field spectrograph ( osiris ) on the keck ii telescope ( stark et al . 2008 ) showed a regular rotation pattern @xmath38 on which are superimposed random motions with a comparable velocity dispersion of @xmath39 . coppin et al . ( 2007 ) detected co emission from the cosmic eye and found the signal to peak at the location of the secondary component of the source , suggesting that this component may be the remaining reservoir of gas available to fuel star formation . the star formation rate of the galaxy has been estimated from its luminosity in the uv stellar continuum , h@xmath36 emission line , and infrared dust emission to be @xmath40 , @xmath41 , and @xmath42 yr@xmath20 , respectively ( smail et al . 2007 ; stark et al . 2008 ; coppin et al . 2007 ; siana et al . the metallicity determined from the ratio of strong emission lines ( the @xmath43 method of pagel et al . 1979 ) , is @xmath44 ( stark et al . 2008 ) . at the current rate of star formation , it would have taken the galaxy @xmath41myr to build its stellar mass of @xmath45(coppin et al . 2007 ) . to this growing body of data we now add a high resolution study of the rest - frame uv spectrum of the cosmic eye . at uv wavelengths the spectrum is dominated by the combined emission from early - type stars and absorption from interstellar gas . such data can shed new light on the kinematics , chemical composition and stellar populations of this galaxy , complementing the information gleaned from other wavelengths . [ fig : stellarlines ] the procedures we followed for our observations and data reduction were very similar to those described by quider et al . ( 2009 ) in their study of the cosmic horseshoe , and we refer the interested reader to that paper for a comprehensive account . briefly , we used the echellette spectrograph and imager ( esi ; sheinis et al . 2000 ) on the keckii telescope to record the rest - frame uv spectrum of the cosmic eye on the nights of 2007 september 13 and 14 ut . with its high efficiency , wide wavelength coverage ( from @xmath46 to 10000 ) , and moderately high resolution ( @xmath47 ) , esi is the instrument of choice for this work . on the first night , the @xmath48 wide slit of esi was positioned on the northern arc of the cosmic eye at position angle @xmath49 ( see figure [ fig : eyeimage ] ) ; on the second night we recorded the light of the southern arc at @xmath50 . the total exposure time _ for each arc _ was 16000s , broken up into eight 2000s long integrations on the esi detector . the seeing was sub - arcsecond on both nights . inspection of the two - dimensional ( 2d ) spectra shows that only the light from the main ring structure was recorded ; the fainter extensions to the east and the west of the northern ring which , as explained above ( section [ sec : eyesummary ] ) , are due to the secondary component of the source in the lensing model of dye et al . ( 2007 ) , were only partially covered by the slit placements ( see figure [ fig : eyeimage ] ) and are in any case much fainter than the main ring structures . the data were processed with standard iraf tasks following the steps outlined in quider et al . ( 2009 ) ; compared to that work , subtraction of the background was considerably more straightforward because in the present case light from the lensing galaxy did not fall within the esi slit ( see figure [ fig : eyeimage ] ) . nevertheless , once the data were fully reduced as described below , we found residual flux in the core of the damped ly@xmath9 absorption line ( see section [ sec : lya ] ) , amounting to @xmath51 of the continuum level . we can not determine whether the effect is real , indicating that the damped ly@xmath9 line only covers 95% of the ob stars producing the continuum near 1216 , or is an artifact of errors in the determination of the background level to be subtracted . nor do we have any other ` markers ' in the spectrum as definitive as the damped ly@xmath9 line to help us assess whether this background correction , whatever its origin , is constant along the spectrum or varies with wavelength . in the circumstances , we limited ourselves to applying to the final one - dimensional ( 1d ) spectrum a uniform background correction of 5% of the continuum flux . at the s / n ratio of the present data , we could not find any significant differences between the extracted 1d spectra of the northern and southern arcs ; their similarity is consistent with the lensing model of dye et al . ( 2007 ) which shows them to be images of the same source , mostly contained within the caustic ( see figure 4 of dye et al . therefore , we averaged the spectra of the two arcs and rebinned the resulting spectrum to 0.5 bins to maximise the s / n ratio . the final composite 1d spectrum ( reproduced in figure [ fig : fullspectrum ] ) has an average @xmath52 per 0.5 bin between 5200 and 7500 ( 12751840 in the rest - frame of the source ) , and a resolution @xmath53 , sampled with @xmath54 wavelength bins . [ fig : ismvelocity3 ] an accurate determination of the systemic redshift of the lensed galaxy in the cosmic eye is critical to the rest of our analysis . our rest - frame uv spectrum includes several absorption features from the photospheres of hot stars which can be used for this purpose . the cleanest among these are oiv@xmath55 , sv@xmath56 , and a close blend of cii and niii lines centred at @xmath57 ( all vacuum wavelengths ) ; see figure [ fig : stellarlines ] . table [ tab : systemic_z ] lists the redshifts deduced from each of these stellar lines , as well as the values of redshift we measured from the h@xmath36 and [ oiii]@xmath58 emission lines in the osiris spectrum of stark et al . as can be seen from table [ tab : systemic_z ] , the mean redshift of the uv stellar lines , @xmath59 , differs by @xmath60 from the mean @xmath61 of the three nebular lines . reported by smail et al . ( 2007 ) , but the latter was not on a vacuum heliocentric scale ( m. swinbank , private communication ) . ] this small difference may reflect a small offset between the wavelength calibrations of esi and osiris , or may be simply due to noise in the spectra . it could also be real but , as we have no other reason to suspect that the massive stars in this galaxy and the nebulae they ionize should be at systematically different redshifts , we decided to average together all the values of redshift in table [ tab : systemic_z ] to deduce a mean systemic redshift for the galaxy @xmath62 ( @xmath63 ) . the standard deviation corresponds to a velocity uncertainty @xmath64 , or just over one half of a wavelength bin in the final 1d esi spectrum . [ cols="<,<,<,<,^,^,^,<,^,^,^ , < " , ] @xmath65 vacuum wavelength and @xmath66-values are from morton ( 2003 ) with updates by jenkins & tripp ( 2006 ) . + @xmath67 velocity range for measurements of blueshifted absorption component . + @xmath68 redshift of the centroid of blueshifted absorption component . + @xmath69 rest - frame equivalent width of blueshifted absorption component . + @xmath70 @xmath63 random error on the rest - frame equivalent width of blueshifted absorption component . + @xmath71 velocity range for measurements of redshifted absorption component . + @xmath72 redshift of the centroid of redshifted absorption component . + @xmath73 rest - frame equivalent width of redshifted absorption component . + @xmath74 @xmath63 random error on the rest - frame equivalent width of redshifted absorption component . + @xmath75 these values refer to the blend of blueshifted and redshifted absorption components . + we closely inspected the spectrum of the cosmic eye for interstellar absorption lines and identified 20 transitions from eight elements in different ionization states , from oi to civ . the absorption lines are listed in table [ table : is_lines ] and all but two ( for clarity purposes ) are reproduced in the montage in figure [ fig : ismvelocity3 ] . are siii@xmath76 , which is actually visible in the plot for the nearby oi@xmath77 where it is indicated by a dotted line , and niii@xmath78 which is of similar strength to the other two niii lines reproduced in the middle panel of figure [ fig : ismvelocity3 ] . ] the most striking aspect of the cosmic eye s interstellar absorption lines is that they consist of two distinct components : a ( mostly ) blueshifted component centred at @xmath79relative to @xmath80 , and a redshifted component centred at @xmath81(these velocities refer to the wavelength bins with the highest apparent optical depth in the line profiles ) . as we shall see below ( section [ sec : intervening ] ) , the sightline to the cosmic eye intersects several lower redshift absorbers that produce a multitude of intervening absorption lines unrelated to the lensed galaxy at @xmath82 . these features do complicate the interpretation of the interstellar absorption due to the cosmic eye itself ; in figure [ fig : ismvelocity3 ] we have indicated with downward pointing arrows instances where the absorption lines due to the interstellar medium in the cosmic eye are contaminated by intervening absorbers at lower redshifts . the two - component structure of the interstellar lines in the cosmic eye is most clearly seen in figure [ fig : ismvelocity3 ] in unblended transitions recorded at high s / n , such as siii@xmath83 and alii@xmath84 . from these absorption lines it appears that the blueshifted component extends over a velocity range @xmath85 , from @xmath86 to @xmath87 relative to the systemic redshift @xmath80 , while the redshifted component is somewhat narrower with @xmath88(from @xmath87 to @xmath89 relative to @xmath8 ) . furthermore , it can be seen from figure [ fig : ismvelocity3 ] that the redshifted component is absent in the absorption lines of the most highly ionized gas , civ and siiv , and is relatively weak in the intermediate ionization stage aliii . whenever possible , we have listed measurements of absorption redshift and equivalent width separately for the two components in table [ table : is_lines ] . note , however , that the values of equivalent width given in the table may include blends with intervening absorption features , as explained above ; in this respect , the purely random errors quoted in columns ( 7 ) and ( 11 ) underestimate the true uncertainties ( random and systematic ) affecting these measurements . the kinematic pattern revealed by figure [ fig : ismvelocity3 ] is unexpected for star - forming galaxies at all redshifts , where the interstellar medium seen in absorption against the starburst is primarily moving with _ negative _ velocities , commonly interpreted as large - scale outflows . for example , in the stack of more than 800 low resolution spectra of galaxies at @xmath90 constructed by shapley et al . ( 2003 ) , the interstellar lines have a net blueshift of @xmath91 , and an analogous result is found in stacked spectra of galaxies at @xmath6 ( law et al . 2007 ) , and @xmath92 , 5 and 6 ( vanzella et al . 2009 ) . similarly , in her esi study of the interstellar nai d lines in 18 local ultraluminous infrared galaxies , martin ( 2005 ) found signatures of outflows in 15 cases and of inflow in only one case . our earlier high - resolution work on other strongly lensed galaxies at @xmath93 ( pettini et al . 2002 ; quider et al . 2009 ) did show a tail of absorption extending to positive velocities , but what we see in the cosmic eye is qualitatively different , with the redshifted component being of comparable strength to the blueshifted absorption . when observed at the low resolutions of most spectra of high-@xmath18 galaxies ( @xmath94 ) , the stronger interstellar lines in the cosmic eye would have a net redshift of @xmath95 km s@xmath20 , which is highly unusual ( e.g. steidel et al . in preparation ) . we discuss possible interpretations in section [ sec : discuss_kinematics ] . the profiles of the interstellar lines in the cosmic eye hold clues to the geometry of gas , dust , and stars in this galaxy . an inspection of figure [ fig : ismvelocity3 ] reveals that the cores of the strongest lines which are not contaminated by intervening absorption at lower redshifts ( that is , lines not marked by a downward pointing arrow ) do not reach zero flux in either of the two kinematic components . we consider the implications of this residual intensity in the line cores for each component in turn . first we consider the redshifted component . siii@xmath96 , alii@xmath97 and feii@xmath98 all exhibit flat cores at a residual intensity of @xmath99 ( where @xmath100 is the measured intensity in the line at wavelength @xmath101 and @xmath102 is intensity in the continuum ) . the most straightforward interpretation is that we are seeing the superposition within the spectrum of saturated absorption lines and escaping continuum photons . such a scenario would apply if the absorbing material only covers @xmath103% of the continuum source . the analysis of the blueshifted component is less clear - cut . it appears to be covering even less of the stellar light than the redshifted component , with the saturated cores of siii@xmath96 and oi@xmath77 levelling at a residual intensity @xmath104 , corresponding to an apparent optical depth @xmath105 [ with the usual definition of the optical depth @xmath106 given by @xmath107 . alii@xmath97 and feii@xmath98 may not be fully saturated in this component , since their minimum residual intensities ( in the blue component ) are higher : @xmath108 , and @xmath109 respectively . on the other hand , we do not have a satisfactory explanation for the higher apparent optical depth of cii@xmath110 , unless this transition is blended with other unrecognized absorption ( in addition to the fine structure line cii@xmath111 which would fall @xmath112 to the red in the top left - hand panel of figure [ fig : ismvelocity3 ] ) . in any case , a lower covering factor of the blueshifted gas compared to the redshifted absorption is required in order to make sense of the apparent contradiction between the relative strengths of the red and blue components in absorption lines of differing @xmath66-values . for example , the two components are observed to have comparable values of @xmath113 in siii@xmath114 but different values in siii@xmath96 , with the blue component being weaker than the red one ( see lower two panels in the left - hand column of figure [ fig : ismvelocity3 ] ) . such an arrangement would at first sight appear to be unphysical . these two siii transitions have widely different @xmath66-values ( see table [ table : is_lines ] ) ; since they both arise from the same ground state of si@xmath115 , their _ intrinsic _ optical depths must be in the same ratio as their @xmath116 products , or @xmath117 . this expectation can only be reconciled with the measured values of _ apparent _ optical depth , @xmath113 , if we assume that the covering factor of the blue component is lower than that of the red one , and that both components are in fact saturated in siii@xmath96 . once that is taken into account , and the zero level adjusted accordingly , the intrinsic ( rather than apparent ) optical depth is actually higher in the blue component than in the red one . the combination of : ( a ) partial coverage by differing factors for the blue and red components , and possibly even varying as a function of velocity within each component ( see martin & bouch 2009 ) ; ( b ) contamination by unrelated absorption features due to lower redshift systems ; and ( c ) limited s / n ratio , makes it very difficult in our view to deduce reliable ion column densities from the analysis of the line profiles . accordingly , we have refrained from attempting to do so . the ly@xmath9 line in the cosmic eye shows a strong damped absorption profile , from which we deduce a neutral hydrogen column density @xmath118@xmath12 ( see figure [ fig : dla ] ) . this is a high value of @xmath10(hi ) , near the upper end of the column density distribution of qso - dlas ( noterdaeme et al . 2009 ) and more typical of the dlas seen in the spectra of the afterglows of high redshift gamma - ray bursts ( pontzen et al . the corresponding line width is @xmath119 or @xmath120 ( jenkins 1971 ) . on kinematic grounds it is therefore impossible , with the s / n ratio of the present data , to assign the dla absorption to one or other of the two main components visible in the metal absorption lines , which are separated by only @xmath121(section [ sec : kinematics ] ) . nevertheless , we suspect that most of the dla absorption may be due to the redshifted component , which has the larger covering fraction of the two and consists primarily of neutral gas . the @xmath122% covering factor we surmised for the outflowing , blueshifted , component may lead to the escape of a substantial fraction of the ly@xmath9 photons produced by the hii regions surrounding the ob stars , as found by quider et al . ( 2009 ) in the case of the cosmic horseshoe . and yet no ly@xmath9 emission line is seen here , not even as residual emission when the damped profile is divided out , as is the case in cb58 ( pettini et al . 2000 , 2002 ) and in the lensed @xmath123 lbg forj0332@xmath03557 observed by cabanac et al . it is possible that any ly@xmath9 photons that escape from the star - forming regions in the cosmic eye are absorbed by the foreground neutral gas giving rise to the redshifted component of the metal absorption lines . it is interesting that the damped ly@xmath9 line has a larger covering factor than any other absorption feature in the spectrum . recall that at the outset we subtracted off 5% of the continuum flux from the entire spectrum ( section [ sec : observations ] ) , in order to bring the core of the damped ly@xmath9 line to zero . at this small level , we can not be sure whether this residual flux in the core of the line is due to inaccurate background estimation , or is the true level of unobscured continuum flux . in any case , the correction we have applied to our data translates to a lower limit of 95% to the covering factor of the ob star continuum by foreground hi gas . the fact that this is larger than the values we deduced in section [ sec : partial_cover ] for the metal lines is not surprising , given that for solar abundances the optical depth in the ly@xmath9 line will exceed that of even the strongest metal lines from hi regions by factors of @xmath124 . in other words , low column densities of gas , undetectable even in the strongest metal absorption lines , could be adding to the apparent optical depth in ly@xmath9 to give a higher overall covering fraction . bringing together the points discussed in this section , our high resolution esi observations have revealed that the interstellar spectrum of the cosmic eye is unusual in showing two absorption components , of approximately equal strengths , one blueshifted by @xmath5 and the other redshifted by @xmath7 relative to the systemic redshift @xmath80 defined by the ob stars and their hii regions . the redshifted component is mostly of low ionization . the interpretation of the interstellar spectrum is complicated by the fact that the metal lines appear to only partially cover the starburst , with differing covering factors for blueshifted and redshifted absorption . there is a very large column of foreground neutral gas , @xmath10(hi)@xmath125@xmath12 , giving rise to a strong damped ly@xmath9 line which covers at least 95% of the stellar continuum . line . overlaid on the spectrum are three theoretical damped profiles , centred at @xmath80 , produced by column densities of neutral hydrogen @xmath118@xmath12 . ] the stars which provide the rest - frame uv continuum light in a high redshift galaxy leave their imprint on the galaxy s spectrum . the stellar wind features are a clear signature from the most massive o stars which have the strongest stellar winds ( kudritzki & puls 2000 ) . rix et al . ( 2004 ) have shown that the p cygni profile of the stellar wind features is sensitive to the metallicity of the stars , as well as to the slope and upper mass cut - off of the imf . while stellar and interstellar features are usually blended at the low resolution of typical high redshift galaxy spectra ( crowther et al . 2006 ) , the high resolution of our spectrum allows us to easily separate them and focus on the stellar component . we used model spectra computed with the spectral synthesis code _ starburst99 _ to exploit the metallicity and imf sensitivities of the p cygni profiles evident in the spectrum of the cosmic eye . the two model spectra considered here were generated using libraries of empirical uv spectra from stars either within the galaxy ( @xmath126 ) , or within the large and small magellanic clouds ( @xmath127 ; leitherer et al . 1999 , 2001 ) . both models are for 100myr old continuous star formation with a salpeter imf with upper mass limit @xmath128 , which are reasonable parameters for the current star formation episode ( coppin et al . 2007 ; pettini et al . 2000 ; quider et al . 2009 ) . the strongest stellar wind feature present in the cosmic eye is the civ@xmath129 p cygni profile shown in figure [ fig : civ ] . also plotted in the top panel are the two _ starburst99 _ model spectra for comparison . clearly there is a good agreement between the models and the data . the main deviations from the models are due to _ interstellar _ siii@xmath96 and civ@xmath129 ( see section [ sec : ism ] ) , which are evident in the blue absorption trough , and narrower intervening alii@xmath97 and civ@xmath129 absorption ( see section [ sec : intervening ] ) which can be seen in the p cygni emission and absorption respectively ( the intervening absorptions are highlighted with upward pointing arrows in figure [ fig : civ ] ) . of the two model spectra shown , the one computed with libraries of magellanic cloud stars is the better match , while milky way metallicities overpredict the strength of the absorption . this conclusion is also consistent with the p cygni profile of nv@xmath130 , although the analysis of the latter is complicated by blending with the red wing of the damped ly@xmath9 line and some intervening absorption . thus , we infer that @xmath131 . we also investigated the effects on the emergent p cygni profile of varying the slope and upper mass cut - off of the imf . as found in previous studies , ( e.g. pettini et al . 2000 ) , there are no indications of a departure from the ` standard ' salpeter imf for stars with masses @xmath132 . the lower panel of figure [ fig : civ ] compares the p cygni region of the cosmic eye with that of the cosmic horseshoe , also recorded with esi ( reproduced from quider et al . the p cygni profiles of these two galaxies are very similar , and both are similar to that seen in cb58 ( quider et al . the most striking difference between the cosmic horseshoe and the cosmic eye is in their interstellar absorption profiles : the horseshoe s interstellar absorption extends to higher outflow velocities , while in the cosmic eye there is strong absorption at positive velocities in siii@xmath96 ( and other ions dominant in neutral gas ) , as discussed above ( section [ sec : kinematics ] ) . the emission component of the civ p cygni profile , which can be clearly seen in the cosmic horseshoe and cb58 , seems to be absent from the cosmic eye . smail et al . ( 2007 ) interpreted this as evidence that the massive stellar population of the galaxy is deficient in o - type stars ; with our higher spectral resolution we can now see that the lack of emission is partly due to the presence of foreground absorption by an intervening triplet of alii@xmath97 lines at @xmath133 ( see table [ tab : intervening ] ) . finally , nebular siii@xmath134 emission , which is evident in the horseshoe and cb58 , is weak in the cosmic eye . several blends of stellar photospheric absorption lines in the integrated spectra of star - forming galaxies have been shown to be sensitive to the metallicity of early - type stars ( leitherer et al . 2001 ; rix et al . two of these regions are a blend of siiii@xmath135 , ciii@xmath136 , and fev@xmath137 absorption lines spanning the wavelength interval 14151435 ( the `` 1425 '' index of rix et al . 2004 ) , and a blend of several feiii absorption lines from b stars between 1935 and 2020 ( the `` 1978 '' index ) . previous work has shown the success of using the 1425 index to estimate the stellar metallicity ( rix et al . 2004 ; halliday et al . 2008 ; quider et al . 2009 ) , but these same studies are divided on the effectiveness of the 1978 index . in the case of the cosmic eye , we can not use the 1425 index because it is heavily contaminated by intervening absorption lines at lower redshifts ( section [ sec : intervening ] ) . while these features are significantly narrower than the broad blend of stellar absorption , they nevertheless compromise the placement of the continuum in this region of the spectrum , rendering the measurement of the 1425 index too uncertain to be useful . the 1978 index region appears to be ` clean ' , but we found it to be a generally poor match to the model spectra by rix et al . ( 2004 ) , in particular being stronger than any of the metallicities considered by those authors between 1985 and 2015 . quider et al . ( 2009 ) also found the 1978 index to be problematic for the cosmic horseshoe ( although in that case the index failed between 1960 and 1980 ) , so we reiterate their cautionary note that the 1978 index needs to be tested on more galaxies to establish its usefulness as an abundance indicator . to sum up this section , from a consideration of the civ stellar wind profile , the population of early - type stars in the lensed galaxy of the cosmic eye was found to be very similar to those of the cosmic horseshoe and cb58 . in all three cases , _ starburst99 _ models with metallicity @xmath138 , continuous star formation , and salpeter imf in the mass range @xmath139 provide satisfactory fits to the uv spectra . the 1978 index is not well - matched by the spectrum of the cosmic eye at any of the model metallicities , indicating that further testing of this index is necessary before it can be reliably used as a metallicity measure . llcc l @xmath140 ( ) & identification & @xmath141 & @xmath142 ( ) & comments + + 5277.15 & siii 1526.7070 & 2.4567 & 1.31 & blended with siiv 1393.7602 in system 4 and stellar absorption features + 5351.27 & civ 1548.204 & 2.4564 & 1.80 & + 5360.11 & civ 1550.781 & 2.4564 & 1.26 & blended with niii 1317 in cosmic eye + 5559.42 & feii 1608.4511 & 2.4563 & 0.44 & + 5774.69 & alii 1670.7886 & 2.4563 & 1.88 & + 6409.68 & aliii 1854.7184 & 2.4559 & 0.69 & + 6438.52 & aliii 1862.7910 & 2.4564 & 0.48 & + 8102.21 & feii 2344.2139 & 2.4563 & 0.85 & + 8235.86 & feii 2382.7652 & 2.4564 & 1.37 & + + 5100.64 & siiv 1393.7602 & 2.6596 & 1.08 & + & si iv 1402.7729 & & & blended with siii 1260 in cosmic eye + 5665.89 & civ 1548.204 & 2.6597 & 1.07 & + & civ 1550.781 & & & blended with siiv 1393 in cosmic eye and civ 1548 in system 3 + + 5106.67 & siiv 1393.7602 & 2.6640 & 2.61 & blended with sii 1253 in cosmic eye + & siiv 1402.7729 & & & blended with siii 1260 in cosmic eye + 5672.94 & civ 1548.204 & 2.6642 & 5.42 & blended with siiv 1393 in cosmic eye and civ 1550 in system 2 + 5682.19 & civ 1550.781 & 2.6641 & 3.64 & blended with siiv 1393 in cosmic eye + 6794.61 & aliii 1854.7184 & 2.6634 & 1.06 & blended with alii 1670 in cosmic eye + + 5053.73 & cii 1334.5323 & 2.7869 & 0.67 & + 5277.15 & siiv 1393.7602 & 2.7864 & 1.31 & blended with siii 1526 in system 1 and stellar absorption features + & siiv 1402.7729 & & & blended with oi 1302 in cosmic eye + 5780.88 & siii 1526.7070 & 2.7865 & 0.69 & + 5862.35 & civ 1548.204 & 2 .7865 & 1.93 & + 5871.72 & civ 1550.781 & 2.7863 & 2.12 & blended with civ 1548 in system 6 + 6090.54 & feii 1608.4511 & 2.7866 & 0.78 & + 6326.42 & alii 1670.7886 & 2.7865 & 0.57 & blended with civ p cygni emission in cosmic eye + 7023.16 & aliii 1854.7184 & 2.7866 & 0.60 & + 7053.51 & aliii 1862.7910 & 2.7865 & 0.34 & + + 5057 : & cii 1334.5323 & 2.79 : & & blended + 5281.97 & siiv 1393.7602 & 2.7896 & 0.61 & blended with stellar absorption features + & siiv 1402.7729 & & & blended with siii 1304 in cosmic eye + 5784.63 & siii 1526.7070 & 2.7890 & 0.83 & + 5866.87 & civ 1548.204 & 2.7895 & 1.45 & + 5876.32 & civ 1550.781 & 2.7893 & 1.00 & + 6330.05 & alii 1670.7886 & 2.7887 & 0.41 & blended with civ p cygni emission in cosmic eye + 7026.91 & aliii 1854.7184 & 2.7887 & 0.24 & + 7058.23 & aliii 1862.7910 & 2.7891 & 0.30 & + + 5061.13 & cii 1334.5323 & 2.7924 & 1.91 & + 5789.76 & siii 1526.7070 & 2.7923 & 0.84 & + 5871.72 & civ 1548.204 & 2.7926 & 2.12 & blended with civ 1550 in system 4 + 5881.45 & civ 1550.781 & 2.7926 & 0.86 & + 6336.16 & alii 1670.7886 & 2.7923 & 1.06 & blended with civ p cygni emission in cosmic eye + + 5899.50 & civ 1548.204 & 2.8105 & 1.49 & + 5909.52 & civ 1550.781 & 2.8107 & 1.25 & + + 6275.06 & civ 1548.204 & 3.0531 & 0.60 & blended with c iv p cygni absorption in cosmic eye + 6284.55 & civ 1550.781 & 3.0525 & 0.58 & blended with civ p cygni absorption in cosmic eye + [ tab : intervening ] possibly because of its high redshift , and the high resolution and s / n ratio of our data , the spectrum of the cosmic eye shows numerous narrow absorption lines at redshifts @xmath143 due to intervening absorbers along the line of sight . limiting ourselves to the wavelength interval longwards of the ly@xmath9 line ( so as to avoid the ly@xmath9 forest ) , we identified 41 such absorption lines associated with seven definite and one possible systems at redshifts from @xmath144 to @xmath145 . every system has a civ@xmath146 doublet associated with it and six systems have additional low or high ionization species present . table [ tab : intervening ] lists the identified intervening absorption lines with their redshifts and observed equivalent widths . although for clarity we have labelled each resolved set of absorption lines as an ` absorption system ' , some of the absorbers are separated by redshift differences which correspond to relative velocities of only a few hundred km s@xmath20 . if , for ease of comparison with earlier absorption line statistics from lower resolution work , we group together into one ` absorption system ' lines which fall within a velocity interval @xmath147 km s@xmath20 , we have four such ( definite ) systems within a redshift interval @xmath148 between @xmath149 and @xmath150 , considering that we could have detected civ doublets between @xmath151and @xmath152 ( approximately the redshifted wavelengths of ly@xmath9 and civ in the cosmic eye ) . all four systems thus defined have @xmath153 , where @xmath154 is the rest - frame equivalent width of the stronger member of the civ doublet . thus , in the cosmic eye the number of such systems per unit redshift is @xmath155 , or @xmath156 times higher than the mean @xmath157 for absorbers of this strength and in the same redshift interval in qso spectra ( steidel 1990 ) . it is hard to conclude , without a more detailed analysis which is beyond the scope of this paper , whether this is just a statistical fluctuation , or whether the excess number of strong civ absorbers in front of the cosmic eye is at least partly due to the extension of the image on the plane of the sky ( two arcs , each @xmath26 long ) . as can be seen in figure [ fig : ismvelocity3 ] , the intervening absorption lines are scattered throughout the spectrum of the cosmic eye , inconveniently falling within several features of interest . column ( 5 ) of table [ tab : intervening ] details the blending of the intervening absorbers with interstellar lines in the cosmic eye as well as each other . there is no doubt that the presence of so many intervening absorption lines complicates the interpretation of the cosmic eye spectrum . this is a clear case where a high resolution spectrum reveals a more complex situation than was inferred from lower resolution observations . thus , the cosmic eye serves as a cautionary tale for the interpretation of high redshift galaxy spectra which , without the boost provided by gravitational lensing , generally can only be recorded at low resolution with available instrumentation . we also searched for mgii@xmath158 absorption near @xmath4 , the redshift of the lensing galaxy of the cosmic eye and of at least two other galaxies in the field ( smail et al . however , these lines would fall in the ly@xmath9 forest and , if present , are difficult to disentangle from the intergalactic medium absorption at the s / n ratio of our data . one of the advantages offered by strongly lensed lbgs is that they can be observed in numerous wavebands allowing us to put together a more extensive picture of their properties than is normally possible in the absence of lensing . in this respect , the cosmic eye is among the better studied galaxies at @xmath34 ( see section [ sec : eyesummary ] ) ; in this section we discuss our findings from the analysis of its rest - frame uv spectrum in the light of previous work carried out at other wavelengths . when siana et al . ( 2009 ) compared the far - infrared ( at wavelengths @xmath159120@xmath160 m , @xmath161 ) and uv ( 1600 , @xmath162 ) luminosities of the cosmic eye , they found the ratio @xmath163 to be @xmath35 times lower than the value predicted with the relationship by meurer , heckman , & calzetti ( 1999 ) , which relates @xmath163 to the slope @xmath36 of the uv stellar continuum in local starburst galaxies ( assumed to be a power law of the form @xmath164 ) . an analogous discrepancy , albeit by a smaller factor , was found by siana et al . ( 2008 ) for ms1512-cb58 . these two studies attributed such departures to a steeper extinction curve for the dust in the cosmic eye and cb58 than that applicable to dust in local starbursts ( calzetti et al . 2000 ) : a faster rise in the extinction @xmath165 with decreasing @xmath101 would have the net effect of increasing @xmath36 ( i.e. making the uv spectral slope redder ) for a lower overall degree of dust extinction , as measured by the @xmath163 ratio . siana et al . ( 2009 ) showed that a uv extinction curve similar to those measured in the large and small magellanic clouds could account for the observed values of @xmath163 and @xmath36 in the cosmic eye and cb58 . the difference between the calzetti and lmc / smc extinction curves is thought to be one of geometry , depending on whether the bulk of the dust is mixed with , or lies in front of , the ob stars whose light is being attenuated . the implication that in cb58 and the cosmic eye most of the dust may be located in a ` foreground screen ' is consistent with the finding that in both cases the uv spectra of the galaxies exhibit a damped ly@xmath9 line with near 100% covering factor ( pettini et al . 2002 for cb58 , and section [ sec : lya ] of the present paper for the cosmic eye ) . the ubiquitous presence of galaxy - scale outflows in star - forming galaxies at @xmath93 may lead to a geometrical configuration of dust and stars more analogous to that described by the lmc / smc curves in many lbgs , as considered by pettini et al . ( 1998a ) . another factor may be age : in their recent study , reddy et al . ( 2010 ) find that it is the galaxies that are younger than @xmath41myr that exhibit a tendency to lie below the meurer et al . ( 1999 ) relation . looking ahead , with a larger sample of high resolution spectra of strongly lensed lbgs it may be possible to disentangle the effects of geometry and age on the emergent uv flux of star - forming galaxies . an immediately obvious test is to measure the @xmath163 ratio in lbgs where the ly@xmath9 line is predominantly in emission , such as the cosmic horseshoe where our earlier work also showed that the insterstellar medium only covers @xmath166 of the uv stellar continuum ( quider et al . 2009 ) . in the physical picture put forward by shapley et al . ( 2003 ) and more recently kornei et al . ( 2009 ) , the three parameters ( age , ly@xmath9 luminosity , and covering factor ) are related , in that galaxies with strong ly@xmath9 emission represent a later evolutionary stage , in which supernova - induced outflows have reduced the dust covering fraction . the interstellar absorption lines provide information on the kinematics of the interstellar gas along our line of sight to the ob stars of the cosmic eye . this only allows us to make a one dimensional assessment of the gas motions in the galaxy , given that the uv light is dominated by a single source with a half - light radius of @xmath167kpc in the lensing model of dye et al . ( 2007see section [ sec : eyesummary ] ) . however , using laser - guided adaptive optics , stark et al . ( 2008 ) were able to probe the two - dimensional distribution of velocities of nebular emission lines ( [ oiii ] and h@xmath36 ) across the face of the galaxy on unprecedented small scales of @xmath41pc . they found a regular pattern ( see figure 2 of stark et al . 2008 ) from which they constructed a rotation curve extending to @xmath168kpc from the dynamical centre with an amplitude of @xmath169 , where @xmath170 is the inclination angle . superposed on this regular pattern are chaotic motions with a dispersion @xmath171 . such a low ratio of ordered to random motions , @xmath172 , is commonly found in star - forming galaxies at @xmath63 ( e.g. law et al . 2009 ; frster schreiber et al . 2009).kpc ) of the galaxy ; larger motions may well be found if one were able to probe baryons over the full virial radius of the galaxy . ] to these kinematic data we now add measurements of outflowing gas seen in absorption . as discussed in section [ sec : kinematics ] , the blueshifted component of the absorption lines has maximum apparent optical depth at a velocity @xmath173 , relative to @xmath80 , and extends out to @xmath174(see figure [ fig : ismvelocity3 ] ) . in their recent analysis of outflows in low redshift starburst galaxies , martin & bouch ( 2009 ) associate the velocity of maximum apparent optical depth with the speed of a shell of swept - up , interstellar gas at the time of blowout from the disk of the galaxy , at a few pressure scale heights . in their picture , gas at higher negative velocities is located further away from the disk ( and is therefore accelerating ) , and exhibits lower apparent optical depths as a result of geometrical dilution . if this scenario also applies to the cosmic eye , we would conclude that rotation , velocity dispersion , and outflow speed of swept - up interstellar matter at blowout are all of comparable magnitude . however , it is still far from clear how all these motions fit into one coherent picture . it is interesting that the outflow speeds in the cosmic eye are lower than the values measured in most other lbgs . the value @xmath173 is only about half the mean blueshift of the strongest interstellar lines in the composite spectrum of 811 lbgs constructed by shapley et al . ( 2003 ) , and lower than @xmath175and @xmath176measured in the cosmic horseshoe ( quider et al . 2009 ) and ms1512-cb58 ( pettini et al . 2002 ) respectively . the maximum blueshift at which absorption is detected , @xmath174is also lower than @xmath177 in both the horseshoe and cb58 . however , this limit may simply be an observational one , determined by the decrease in covering factor with distance from the central starburst , and there may be gas at higher negative velocities which is too diluted to produce detectable absorption ( martin & bouch 2009 ) . while there are indications at both low ( @xmath178 , martin 2005 ; rupke et al . 2005 ) and intermediate ( @xmath179 , weiner et al . 2009 ) redshifts that ` superwind ' speeds scale with galactic mass and star formation rate , there is sufficient scatter in these relations to accommodate the differences we have uncovered between the three strongly lensed high-@xmath18 galaxies , even though they are thought to have comparable masses and star formation rates . finally we note that , alongside rotation , random motions , and outflows , our uv spectrum has shown the presence of high column density , neutral gas apparently moving _ towards _ the ob stars in the cosmic eye , further complicating the kinematic picture of this galaxy . with the available data , we can not arrive at definite conclusions concerning the nature and location of this gas . one possibility is that it may be a chance superposition of another galaxy ( or damped ly@xmath9 system ) along the line of sight . alternatively , it may be dynamically related to the cosmic eye . for example , it may be gas ejected from the galaxy by a previous episode of star formation and now falling back onto it , or we may be viewing a merger along a favourable sightline , although in both cases the 350 velocity offset from @xmath8 seems very high , given that the cosmic eye does not appear to be a very massive galaxy [ e.g. @xmath180 ( coppin et al 2007 ) , and @xmath181 km s@xmath20 ( stark et al . 2008 ) ] . the lensing model of dye et al . ( 2007 ) does include a second , separate clump of uv light in the source plane , which could indicate a merger , or a foreground galaxy . however , coppin et al . ( 2007 ) reported a redshift @xmath182 for the co emission which they found to be peaked at the position of this second source ; the small velocity difference @xmath183from the systemic redshft @xmath62 ( section [ sec : z_sys ] ) makes it unlikely that the absorbing gas we have found with @xmath184 is associated with the co emitting clump . additional kinematic data on strongly lensed galaxies spanning a range of physical properties are clearly needed in order to explore how outflows ( and in some cases presumably inflows ) tie into the other motions within a galaxy , as well as the galaxy s intrinsic characteristics . the civ p cygni profile encodes information about the metallicity of the o stars , as well as the relative numbers of massive stars in the cosmic eye . here we consider each in turn . up to now , the degree of metal enrichment attained by high redshift star - forming galaxies has been determined primarily from the analysis of the most prominent nebular lines emitted from their hii regions ( e.g. pettini et al . 2001 ; shapley et al . 2004 ; erb et al . 2006a ; maiolino et al . strongly lensed galaxies give us the opportunity to conduct a complementary metallicity measurement based on uv spectral features due to massive stars . young stars and hii regions are expected to have intrinsically the same chemical composition , since the former have only recently formed out of the gas which they ionize into the latter ; thus a comparison between stellar and nebular abundances is essentially a consistency check which may reveal systematic offsets between different metallicity indicators . in the case of the cosmic eye , stark et al . ( 2008 ) deduced a metallicity @xmath185 assuming that the upper - branch solution of the r23 method first introduced by pagel et al . ( 1979 ) applies . on the other hand , we showed in section [ sec : stars ] that at solar metallicities the civ p cygni profile would be stronger than observed , and that a metallicity @xmath186 gives a better match to the data . discrepancies by factors of @xmath187 between different metallicity indicators in high redshifts galaxies are probably to be expected ( pettini 2006 ) . on the other hand , in our earlier study of the cosmic horseshoe ( quider et al . 2009 ) , we did find the metallicity deduced from the r23 index to be the odd one out ( in the sense of being a factor of @xmath54 higher ) among several stellar and nebular measures . at lower redshifts too , methods relying on the r23 index result in metallicities towards the upper end of the range spanned by different indicators ( see , for example , figure 2 of kewley & ellison 2008 ) . in light of these considerations , the apparent difference between nebular and stellar metallicity in the cosmic eye is not surprising . as mentioned in section [ sec : stars ] , the civ p cygni profiles of the cosmic eye , cosmic horseshoe , and ms1512-cb58 are all remarkably similar . while initially this may seem to be an odd coincidence , considering that these three galaxies were randomly selected by gravitational lensing , on further reflection this result is perhaps to be expected . the key issue here is continuous star formation . when considering the composite spectrum of an entire galaxy , @xmath34 lbg is equivalent to the integrated output of @xmath188 o7 stars ( pettini et al . 1998b ) . ] it is exceedingly unlikely that we would pick out a special time in its stellar evolution . while in individual regions star formation is likely to proceed in bursts , a succession of such bursts on a galaxy - wide scale will approximate a continuous process of star formation . under such circumstances , the contrast of the civ ( and other ) wind lines over the underlying ob photospheric continuum will stabilize after @xmath189myr from the onset of star formation , corresponding to the lifetime of the lowest mass stars contributing to the light at 1550 . since this time interval is comparable to the dynamical timescale of most lbgs ( e.g. erb et al . 2006c ) , only a fairly exceptional star - formation history would produce significant changes in the integrated uv stellar spectrum . as discussed above , the stellar metallicity is also an important factor in determining the strengths of the p cygni lines , but on this basis too we would not expect marked differences between the cosmic eye , the cosmic horseshoe and ms1512-cb58 , given that the three galaxies have similar metallicity , @xmath190 . such a uniform degree of metal enrichment is in turn not as surprising as it may seem , since the three objects are of comparable luminosities , reflecting the fact that most strongly lensed galaxies targeted for detailed analyses so far are intrinsically luminous , with @xmath191 . a test of these ideas would be provided by observations of highly magnified galaxies of lower intrinsic luminosity ( and presumably metallicity ) than the few examples studied up to now . strong gravitational lensing of a @xmath82 star - forming [ @xmath192 yr@xmath20 for a chabrier ( 2003 ) imf ] galaxy magnifies it by a factor of @xmath3 and distorts its image into two @xmath193 long arcs which have been collectively named the ` cosmic eye ' by their discoverers , smail et al . the high magnification has allowed us to use the esi spectrograph on the keckii telescope to record the galaxy s rest - frame uv spectrum with high resolution and s / n ratio . by analyzing these data together with existing observations of the cosmic eye at other wavelengths , we have reached the following main conclusions . \(i ) the interstellar absorption lines exhibit two components , of approximately equal strength , which are respectively blueshifted by @xmath5 and redshifted by @xmath7 relative to the stars and hii regions . while these values apply to the gas with the highest apparent optical depths , both components include absorption spanning several hundred . we associate the blueshifted component with a galaxy - wide outflow similar to , but possibly weaker than , those seen in most star - forming galaxies at @xmath63 . the redshifted absorption is very unusual , and may represent gas ejected by a previous episode of star formation and now falling back onto the galaxy , or a merger viewed along a favourable line of sight . alternatively , it may just be a chance superposition of another galaxy along the line of sight . \(ii ) both components of the metal absorption lines show indications that they do not fully cover the ob stars against which they are being viewed ; we estimate covering fractions of @xmath122% and @xmath103% for the blueshifted and redshifted component respectively . there must also be more pervasive diffuse gas , because the strong damped ly@xmath9 line , corresponding to a column density @xmath10(hi)@xmath194@xmath12 , covers at least 95% of the uv stellar continuum . we tentatively associate this high column density of hi with the redshifted component of the metal lines , where absorption from ionized species is weak or missing altogether , and propose that it provides the ` foreground screen ' of dust responsible for the lower - than - expected far - infrared luminosity of the cosmic eye found by siana et al . ( 2009 ) with the _ spitzer space telescope_. \(iii ) the internal kinematics of the galaxy lensed into the cosmic eye are very complex , our data now adding outflow , and possibly even inflow , to the rotation and velocity dispersion already known from the integral field spectroscopy with adaptive optics by stark et al . ( 2008 ) . ordered rotation , chaotic motions , and outflow all seem to be of comparable magnitude , with @xmath19570 . however , we do not have a model yet of how these different motions fit together into one coherent kinematic picture . \(iv ) turning to the stellar spectrum , we find that the civ p cygni profile is well fit by a _ starburst99 _ stellar population model spectrum having continous star formation with a salpeter imf , stellar masses from 5 to 100 , and a lmc / smc metallicity of @xmath127 . the p cygni profiles of the cosmic eye , the cosmic horseshoe , and ms1512-cb58 , three high redshift star - forming galaxies studied at high spectral resolution , are all nearly identical . this is not unexpected , however , when we consider that in each case we see the integrated light of several hundred thousand o - type stars , and that these three galaxies have similar metallicities and dynamical timescales over which star formation is taking place . \(v ) the metallicity @xmath190 deduced for the o stars in the cosmic eye is lower by a factor of @xmath187 than the value derived from the analysis of the strong nebular lines from its hii regions using the r23 index . we consider this apparent discrepancy to reflect systematic offsets between different abundance estimators , rather than intrinsic inhomogeneities in the chemical composition of stars and ionized gas . ( vi ) the interpretation of both interstellar and stellar features in the uv spectrum of the cosmic eye is complicated by the presence of numerous intervening absorption lines associated with eight absorption systems at redshifts @xmath1443.0528 . these narrow features are not resolved in existing low - resolution data , highlighting the caution that should be exercised in interpreting the spectra that are typically available for unlensed lbgs . in closing , the new data presented here further emphasize the complexity of the physical conditions which prevailed in actively star - forming galaxies at redshifts @xmath63 . it is remarkable that the three strongly - lensed galaxies targeted by our esi observations , while showing very similar young stellar populations , are all different in the detailed properties of their interstellar media . whether such variety is simply the result of different geometries and viewing angles , or has its roots in more fundamental physical reasons remains to be established . fortunately , with the increasing attention being given to detailed studies of gravitationally lensed galaxies , we can look forward with optimism to a more comprehensive empirical picture of galaxy formation coming together in the years ahead . we are grateful to the staff at the w. m. keck observatory for their competent assistance with the observations . valuable suggestions by an anonymous referee improved the final version of this paper . amq s research is funded by a scholarship from the marshall foundation , and a national science foundation graduate research fellowship . aes acknowledges support from the david and lucile packard foundation and the alfred p. sloan foundation , and ccs from nsf grant ast-0606912 and the john d. and catherine t. macarthur foundation . dps s research is support by the uk stfc through the award of a postdoctoral fellowship . finally , we wish to extend thanks to those of hawaiian ancestry on whose mountain we are privileged to be guests . pettini , m. , steidel , c. c. , adelberger , k. l. , kellogg , m. , dickinson , m. , & giavalisco , m. 1998b , in woodward , c. e. , shull , j. m. , thronson , h. a. , eds . , asp conf . . vol . 148 , origins , astron . , san francisco , p. 67

using high - resolution , two - dimensional hydrodynamic simulations , we investigate nonlinear gravitational responses of gas to , and the resulting drag force on , a very massive perturber moving at velocity through a uniform gaseous medium of adiabatic sound speed . we model the perturber as a plummer potential with softening radius , and run various models with differing and by imposing cylindrical symmetry with respect to the line of perturber motion . for supersonic cases , a massive perturber quickly develops nonlinear flows that produce a detached bow shock and a vortex ring , which is unlike in the linear cases where mach cones are bounded by low - amplitude mach waves . the flows behind the shock are initially non - steady , displaying quasi - periodic , overstable oscillations of the vortex ring and the shock . the vortex ring is eventually shed downstream and the flows evolve toward a quasi - steady state where the density wake near the perturber is in near hydrostatic equilibrium . we find that the detached shock distance and the nonlinear drag force depend solely on such that and for , where is the linear drag force of ostriker ( 1999 ) . the reduction of compared with is caused by front - back symmetry in the nonlinear density wakes . in subsonic cases , the flows without involving a shock do not readily reach a steady state . nevertheless , the subsonic density wake near a perturber is close to being hydrostatic , resulting in the drag force similar to the linear case . our results suggest that dynamical friction of a very massive object as in a merger of black holes near a galaxy center will take considerably longer than the linear prediction .

0908.1391

a massive object in orbital motion suffers from orbital decay due to a negative torque caused by gravitational interaction with its own gravitationally induced wake created in the background medium . this process , commonly referred to as dynamical friction ( df ) , occurs in not only collisionless environments ( e.g. , @xcite ) but also collisional gaseous backgrounds ( e.g. , @xcite ) . the df in a gaseous medium is of great importance in understanding the formation and evolution of planets , binary stars , supermassive black holes , etc . for instance , gravitational interaction between a protoplanet and its environmental disk causes the former to migrate toward a central star , naturally explaining the presence of `` hot jupiters '' found from radial velocity surveys ( * ? ? ? * and references therein ) . the migration also helps a planet grow faster in mass by providing an expanded gas - feeding zone at an enhanced accretion rate , which may overcome the failure of _ in situ _ core - accretion scenario in building a giant planet within a typical disk lifetime ( see , e.g. , * ? ? ? in the case of nuclear black holes in merging galaxies , they are expected to first experience the df to form a binary and then coalesce into a supermassive black hole by emitting gravitational waves . friction of nuclear black holes against the collisionless _ stellar _ background appears to be inefficient due to scattering and depletion of stars near the black holes , which is known as the `` final - parsec problem '' ( see * ? ? ? * and references therein ) . however , recent numerical @xmath13-body / sph simulations show that the gravitational drag from the _ gaseous _ background is sufficient to form a black - hole binary in a relatively short time ( @xmath14 myrs ) ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? thanks to a seminal paper of @xcite , df in a gaseous medium is well understood as long as density wakes have small amplitudes . earlier theoretical work by @xcite , @xcite , and @xcite considered density wakes in a steady state and found that the drag force vanishes for a subsonic perturber , while it becomes remarkably similar to the collisionless counterpart for supersonic cases . using a time - dependent linear perturbation theory , on the other hand , @xcite found that the gravitational drag force on a point - mass perturber with mass @xmath0 moving at velocity @xmath1 on a straight - line trajectory through a uniform gaseous medium with density @xmath15 and sound speed @xmath2 is given by @xmath16 where @xmath17 is the mach number , @xmath18 is the time elapsed after the introduction of the perturber , and @xmath19 is the minimum radius introduced to avoid the singularity in the force evaluation . equation ( [ eq : f_pt_linear ] ) shows that the gaseous df force becomes identical to the @xcite ( @xcite ) formula for the collisionless drag for @xmath20 , and is , albeit small , non - zero even for subsonic perturbers because the time dependency breaks the symmetry in the density wakes ( see also @xcite ) . equation ( [ eq : f_pt_linear ] ) has been applied to various astrophysical situations including orbital decay of compact objects in accretion disks ( e.g. , @xcite ) and heating of an intracluster medium by supersonically moving galaxies in clusters ( e.g. , @xcite ) . while the result of @xcite is valid in a strict sense only for a linear - trajectory perturber in a uniform medium , it has proven to be applicable to more general cases . for example , @xcite considered a circular - orbit perturber with orbital radius @xmath21 in a uniform gaseous medium and found that equation ( [ eq : f_pt_linear ] ) is a reasonable approximation to the gaseous drag force on it , provided @xmath22 . @xcite numerically found that the orbital decay of a plummer sphere with a softening radius @xmath3 in a radially - stratified medium is consistent with the prediction of equation ( [ eq : f_pt_linear ] ) , if @xmath23 . also , @xcite showed that equation ( [ eq : f_pt_linear ] ) remains valid even for a perturber with relativistic speed if the relativistic correction factors are included . while @xcite found that the orbital decay of black hole binaries took longer than the prediction of equation ( [ eq : f_pt_linear ] ) for a single perturber , the discrepancy between the numerical and analytical results can be reconciled , at least partly , by taking allowance for fact that an object in a binary experiences not only a negative torque due to its own wake but also a positive torque from the companion wake . for an equal - mass binary , @xcite found that the positive torque is on average about 40% of the negative torque . since the results of @xcite are based on the assumption that density wakes remain in the linear regime , the validity of equation ( [ eq : f_pt_linear ] ) for very massive perturbers has yet to be seen . the strength of gravitational perturbations due to a body with mass @xmath0 can be measured by the dimensionless parameter @xmath24 which roughly corresponds to the perturbed density at a distance @xmath3 from the perturber relative to the background density ( e.g. , @xcite ) , and is equal to the bondi radius @xmath25 relative to @xmath3 . for systems with @xmath26 , the density wakes are clearly in the nonlinear regime and the linear perturbation analyses are likely to fail . identifying @xmath3 with the gravitational softening radius of a perturber ( or , equivalently , its size ) , @xmath27 is in the ranges of @xmath28 for galaxies embedded in typical intracluster media , @xmath29 for protoplanets in protostellar disks , and @xmath30 for supermassive black holes near galaxy centers , suggesting that the wakes of massive compact objects can readily be nonlinear . indeed , @xcite reported that the orbital decay time of supermassive black hole binaries with @xmath31 depends on @xmath0 much less sensitively than the results of the linear theory , which may be caused primarily by the nonlinear effects . in this paper , we investigate nonlinear df of a very massive perturber in a gaseous medium using numerical hydrodynamic simulations . in order to isolate the effects of the perturber mass and its velocity on the df force , we consider a perturber following a straight - line trajectory in a uniform gaseous medium , similarly to in @xcite . we model the perturber as a plummer sphere that does not possess any solid surface and merely provides gravitational potential perturbations to the background medium that would otherwise remain static and uniform ; to make contact with the results of the linear theory , we ignore the accretion of gas onto the perturber in the current work . our primary objectives are to find the changes in distributions of density wakes with @xmath27 and @xmath32 , and to quantify the resulting gravitational drag forces in comparison with the linear cases . nonlinear responses of a background to a massive perturber moving at a supersonic speed have been extensively studied in the context of the bondi - hoyle - lyttleton ( bhl ) accretion ( @xcite ; see also review of @xcite and references therein ) . unlike our models where mass accretion to a perturber is prohibited , however , the bhl accretion problem considered a perturber containing a defined surface through which gas is either accreted or reflected . it was @xcite who first solved for bhl accretion flows numerically , finding that the collisional nature of gas supports a bow shock in front of a supersonic perturber . later studies found that the bhl accretion flows exhibit unstable behaviors such as flip - flop motions of the accretion shocks and vortex shedding when the condition of axisymmetry is relaxed ( e.g.,@xcite ) . while these numerical works on the bhl accretion explored temporal evolution and distribution of density wakes as well as nonlinear features in some great detail , they mainly concentrated on the gravitational focusing and resulting accretion rate of gas onto the perturber . although some authors ( e.g. , @xcite ) presented values for aerodynamic and gravitational drag forces , only a limited range of @xmath27 was covered . in this work , we run a number of numerical simulations by varying @xmath27 and @xmath32 systematically in order to quantify the dependences of the gravitational drag force on these parameters . a brief comparison between our results with those from the bhl accretion studies will be presented . this paper is organized as follow : in [ sec : code ] , we describe numerical methods we employ for nonlinear simulations . in [ sec : linear ] , as a code test we revisit the cases with a spatially - extended , linear perturber with @xmath33 . we compare the resulting distributions of density and velocity wakes with those of analytical results and provide a way to handle the effect on the df force of a softening radius which is necessary for numerical simulations . evolution and quasi - steady distributions of fully nonlinear density wakes and the associated drag forces are presented in [ sec : nonl ] . finally , in [ sec : diss ] we summarize our findings and discuss their astrophysical implications . in this paper , we consider an unmagnetized , inviscid , non - radiating , non - self - gravitating gaseous medium and study its gravitational responses to a massive perturber moving along a straight - line trajectory using numerical simulations . the background gaseous medium is initially static and uniform with density @xmath15 and adiabatic sound speed @xmath2 . we adopt an adiabatic equation of state with an index @xmath34 throughout simulations . the simulations are carried out on a two - dimensional @xmath35 plane in cylindrical symmetry , where @xmath36 and @xmath37 denote the distances from and along the axis of symmetry , respectively . we assume that the perturber exerts only gravity to the surrounding medium and does not possess any surface , so that neither accretion nor reflection of the gas is allowed . the perturber is modeled as a plummer sphere with mass @xmath0 that is moving at a constant speed @xmath1 along the @xmath38 axis toward the positive @xmath37-direction , with the gravitational potential @xmath39 ^ 2)^{1/2 } } , \;\;\;{\rm for } \;t\geq0,\ ] ] where @xmath3 is the softening radius . we take @xmath3 , @xmath2 , and @xmath40 as the units of length , velocity , and time , respectively , in our simulations . then , our models are completely parametrized by @xmath27 and @xmath32 . we run a total of 58 models with @xmath27 varying from 0.01 to 600 and @xmath32 in the range of 0.5 to 4.0 . we solve the basic equations of ideal hydrodynamics using flash3 @xcite , an eulerian hydrodynamics code that implements a direct piecewise - parabolic method solver of @xcite for high - order spatial reconstruction . although the flash3 code is capable of both uniform grid and adaptive mesh refinement calculations , we adopt the uniform grid method since the accurate evaluation of the drag force requires the whole computational domain to be well resolved . as we will show below , we find that it is necessary to have at least 5 zones per @xmath3 to obtain converged results for the drag forces . our largest grid models have 3,072@xmath4112,288 zones in @xmath42 ; we make sure that our computational domain is taken to be large enough to contain the whole density wake in a given model . the simulations are typically carried out until @xmath43 when most of the wakes are well resolved and reach a quasi - steady state . time - dependent linear perturbation theories for the df drag force in a gaseous medium usually study the responses of gas to a low - mass , point - mass perturber corresponding to @xmath44 in the plummer potential ( e.g. , @xcite ) , which requires to introduce the cut - off radius @xmath19 in the linear force formula ( eq . [ [ eq : f_pt_linear ] ] ) . in numerical simulations , on the other hand , one needs to assign a non - zero value to the softening radius , which in turn makes it unnecessary to use the cut - off radius in the force evaluation . since our goal is to compare the nonlinear drag force on a massive perturber with the linear prediction , we have to first find a proper relationship between @xmath3 and @xmath19 that makes the numerical and analytical results consistent with each other when @xmath45 . motivated by this consideration , in this section we briefly present the results of numerical simulations for a low - mass perturber with @xmath33 , and compare the resulting distributions of density and velocity wakes and the drag forces with those from the linear theories . this will also allow us to check the accuracy of our numerical experiments . regarding the linear wakes with which numerical results will be compared , it is worth mentioning that there are several analytical methods for finding solutions for the perturbed density and velocity fields . @xcite utilized fourier transform for the space variables and laplace transform for the time variable , finding expressions both for the density and velocity wakes . instead , @xcite used a retarded green s function technique and found an expression only for the density wake that is identical to the result of @xcite . we found that while the analytical formula for the density wake agrees well with our numerical results for a low - mass perturber , the simulated velocity field differs from the expression given by @xcite . in appendix [ sec : ft ] , we revisit the time - dependent linear theory using fourier transforms both for the space and time variables . as we will show below , our expressions ( [ eq : vz ] ) and ( [ eq : vr ] ) for the perturbed velocities are in good agreement with the numerical results , confirming that equation ( 47 ) of @xcite contains a typographical mistake . figure [ fig : linear ] shows as color - scale images the snapshots on the @xmath37@xmath36 plane of the perturbed density @xmath46 ( _ top _ ) , the parallel velocity @xmath47 ( _ middle _ ) , and the perpendicular velocity @xmath48 ( _ bottom _ ) to the line of motion for a model with @xmath33 and @xmath49 . note that the coordinates are normalized by @xmath50 . the perturber initially introduced at @xmath51 has moved to @xmath52 at time @xmath18 . the characteristic features of a supersonic wake consisting of a sonic sphere with radius @xmath50 centered at the initial perturber location and a mach cone bounded by mach waves located at @xmath53 for @xmath54 are apparent in the top panel . also plotted as black solid contours are the results of the linear perturbation theory ( eqs . [ [ eq : alpha ] ] , [ [ eq : vz ] ] , and [ [ eq : vr ] ] ) , which are overall in good agreement with the numerical results . a careful inspection of figure [ fig : linear ] , however , reveals that the numerical results deviate slightly from the analytical ones especially near the sonic radius and the mach waves . this is due to the fact that the perturber in our numerical models is modeled as an extended plummer sphere rather than a point mass . one can semi - analytically construct the density wake @xmath55 of an extended perturber by convolving the density wake @xmath56 due to the corresponding point mass with the extended mass distribution @xmath57 of the perturber ( e.g. , @xcite ) . for a plummer sphere we use , the convolution theorem gives @xmath58 where @xmath59 . figure [ fig : linprof ] plots as solid lines the profiles of @xmath55 for @xmath33 and @xmath49 along the cuts at @xmath60 and @xmath61 marked as dotted lines in the top panel of figure [ fig : linear ] , which are in excellent agreement with the simulation outcomes ( _ open circles _ ) . compared with the point - mass results ( _ dashed lines _ ) , the extended mass distribution tends to smear out the discontinuities at the boundary of the sonic sphere and the mach cone . this makes sense since a perturbed density at one location is a superposition of sonic perturbations with various strengths launched by all the mass elements comprising the extended body . allowing for the extended mass distribution , the gravitational drag force exerted on the perturber in the negative @xmath37-direction can be obtained by directly evaluating the integral @xmath62 where @xmath63 is the wake distribution . although equation ( [ eq : f_integ ] ) requires the @xmath64-integration to be performed over the entire plummer sphere , we empirically found that the drag force on the region with distance from the perturber larger than @xmath65 has a negligible contribution to the total . thus , in practice , we limit the integration to the region within @xmath65 that contains about 98.5% of the total perturber mass . figure [ fig : fmlin ] plots as open squares the numerical drag forces on a low - mass perturber with @xmath33 but differing @xmath32 . the solid line corresponds to equation ( [ eq : f_pt_linear ] ) with @xmath66 which gives the best fit to the supersonic results of our adiabatic simulations . equation ( [ eq : rmin ] ) is our prescription for the cut - off radius when we compare the numerical drag forces with the analytic results . note that @xcite suggested @xmath67 based on their isothermal simulations . the difference between the two prescriptions may be due in part to using different equations of state and in part to low resolution ( 1 cell per @xmath3 ) in @xcite . we begin by describing the temporal evolution of our fiducial model with @xmath68 and @xmath49 ; the evolution of other supersonic models are qualitatively similar . figure [ fig : evol ] illustrates the density and velocity structures of this model in a comoving frame with the perturber located at @xmath69 . only the region with @xmath70 and @xmath71 is shown . unlike the linear cases with @xmath72 where the density wakes have too small amplitudes to launch shock waves , the perturber with @xmath68 emits strong perturbations that quickly develop into a bow shock . the upstream gas moving toward the perturber along the symmetry axis is first accelerated by the gravity of the perturber , is shocked to subsonic speed , and then piles up near the center of the perturber . this creates a steep pressure gradient in between the perturber and the shock , tending to push the shock away from the perturber . on the other hand , the gas flowing above ( not far away from ) the symmetry axis is deflected toward the perturber even before entering the shock and decreases its speed after the shock . this gas thus has a longer time to be exposed to the gravity of the perturber as it moves toward the symmetry axis . the gravitational potential is so deep that the material arriving at the rear side of the perturber can be pulled back toward the perturber , creating a stagnation point just as in the bhl accretion flows ( e.g. , @xcite ) . this produces a strong counterstream that moves into the upstream direction along the symmetry axis , as well as a primary vortex in the @xmath73@xmath36 plane ( fig . [ fig : evol]_a_)@xmath36 plane is a vortex ring in three dimensions , and the counterstream is a part of the vortex ring near the symmetry axis . ] . the counterstream combined with the pressure gradient in the front side pushes the shock front away from the perturber . the advance of the shock front in the upstream direction allows more time for the shocked gas to be affected by the gravity , strengthening the counterstream ( as well as the primary vortex ) and thus increasing the detached distance @xmath6 of the shock measured along the symmetry axis . at the same time , the center of the vortex with a lower density than the surrounding buoyantly rises toward the high-@xmath36 regions . this decreases the ram pressure of the counterstream exerted on the shock front and the shock advance slows down . figure [ fig : dtime ] plots the time evolution of @xmath6 for some selected models , while figure [ fig : vort ] traces the trajectory of the center of the primary vortex in our standard model ( @xmath68 and @xmath49 ) on the @xmath73@xmath36 plane . for our standard model , @xmath6 keeps increasing to @xmath74 at @xmath75 . at this time , the vortex is located near at @xmath76 ( figs . [ fig : evol]b and [ fig : vort ] ) . the shock front soon overshoots a potential equilibrium position where the thermal pressure of the postshock gas supports it against the perturber gravity , and begins oscillating around the equilibrium position . the shock oscillation changes the velocity field in the postshock region , causing the vortex to move in the counterclockwise direction around its mean position @xmath77 following the background flow . kelvin s circulation theorem implies that the vortex becomes stronger as it moves toward the symmetry axis , amplifying the strength of the counterstream near the perturber . the velocity of the counterstream becomes largest when the vortex arrives closest to the perturber on its oscillation path ( fig . [ fig : evol]_d _ ) . the shock front that was moving toward the perturber is pushed by the strong counterstream , and reverses its motion . the vortex rises to high @xmath36 as the shock moves away from the perturber , weakening the counterstream , and the oscillation cycle repeats quasi - periodically . figures [ fig : dtime ] and [ fig : vort ] show that the amplitudes of the detached shock oscillations as well as the vortex movements grow secularly with time as the shock oscillation continues . this can be understood as follows . when the shock front is displaced from the equilibrium position away from the perturber ( e.g. , fig . [ fig : evol]_c _ ) , the shocked subsonic gas can acquire an extended time to be gravitationally influenced , similarly to the early situation when the shock advances away from the perturber . with the stagnation point moving away from the perturber , more mass and momentum are added to the counterstream , amplifying the vortex oscillation . in addition , when the strong counterstream collides directly with the shocked subsonic flow near the symmetry axis , the material at the interface is injected upward in the lateral direction and then carried in the negative @xmath73-direction by the background flow ( see , e.g. , the region at @xmath78 and @xmath79 in fig . [ fig : evol]_e _ ) . this produces small vortices near the interface that move downstream and merge with the primary vortex , again strengthening the latter . consequently , the primary vortex is able to move closer to the symmetry axis and amplify the counterstream in the next cycle , making the shock oscillation overstable . as the center of the primary vortex moves away from the symmetry axis during its last phase of the overstable oscillation ( @xmath80 ) , it becomes less gravitationally bound and can thus be more easily influenced by the background flow . for all the _ supersonic _ models we run in this paper , we find that whenever the vortex rises above a half of the accretion radius , also known as the hoyle - lyttleton radius , defined by @xmath81 , it is swept away by the background flow in the negative @xmath73-direction ( fig . [ fig : vort ] ) . when the primary vortex is located outside @xmath82 , the counterstream associated with the vortex is weak and occurs in the far rear side of the perturber , unable to pass through the center of the perturber . the counterstream is instead resisted by the flow moving downstream right across the perturber and pushed up in the lateral direction ( fig . [ fig : evol]_g _ ) . this develops another vortex with an opposite sense of rotation to the primary one , capable of pushing the latter away from the perturber ( fig . [ fig : evol]_h _ ) . with the primary vortex carried away downstream , the flow near the perturber reaches a quasi - steady state in which shocked subsonic gas moves almost parallel to the symmetry axis . since the associated kinetic energy is much smaller than the thermal and gravitational energies of the gas , the density distribution around the perturber becomes nearly hydrostatic , as will be shown below . figure [ fig : subs ] displays density and velocity structures of a nonlinear subsonic model with @xmath68 and @xmath83 in a comoving frame with the perturber . snapshots at @xmath84 , 150 , and 600 are shown . sound waves launched from the perturber at @xmath85 propagate radially outward into the surrounding medium , forming a spherical causal region within which the medium is affected by the sonic perturbations . since the perturber is spatially extended , however , the boundary of the casual region is not as sharp as in the case of a point mass , although the most dominant perturbations still come from the perturber center . unlike supersonic cases , this model always involves subsonic flows and never produces a shock . nevertheless , the overall flow pattern and late - time density structure near the perturber of this model is very similar to those in the postshock subsonic regions of supersonic models . first of all , the strong gravitational pull forms a counterstream and an associated vortex ring near the symmetry axis ( fig . [ fig : subs]_a _ ) . the counterstream moving in the upstream direction interacts with the incident flow ( fig . [ fig : subs]_b _ ) . the gas at the interface is pushed up toward the high-@xmath36 regions and then carried downstream , creating small vortices with low density ( fig . [ fig : subs]_c _ ) . the primary vortex slowly rises in the @xmath36-direction due to buoyancy , and merge with the small vortices . since this model does not contain a shock that would confine the region of influence and since the causal region keeps expanding at a sonic speed , the flows in the high-@xmath36 regions are almost parallel to the symmetry axis . as a result , the primary vortex keeps rising as there is no momentum input in the background flow capable of pushing it back toward the symmetry axis . at the end of the run ( @xmath43 ) , the primary vortex in this model arrives at @xmath86 , corresponding to @xmath87 . it is uncertain whether the primary vortex will be carried downstream when it goes beyond @xmath88 in a manner similar to supersonic cases . at any event , the density distribution close to the perturber is well described by the condition of hydrostatic equilibrium at late time ( see [ sec : wake ] below ) . figure [ fig : compadens ] illustrates changes in the quasi - steady density wakes with varying @xmath27 on the @xmath73@xmath36 plane for supersonic models with fixed @xmath49 at @xmath43 . the left panels show large - scale views of the wakes at @xmath89 and @xmath90 , while the region near the perturber with @xmath91 and @xmath92 is enlarged in the right panels . in each panel , the perturber is located at @xmath93 , and the black line connecting the points @xmath94 and @xmath95 marks the boundary of the mach cone characteristics of the linear density wake for @xmath49 . when @xmath72 , the sonic perturbations are too weak to produce a shock , and the high - density ridge of the wake follows the mach cone fairly well , although it is broadened due to the extended mass distribution of the perturber . as @xmath27 increases to unity , sonic perturbations even outside the mach cone attain substantial amplitudes enough to induce a bow shock that is attached to the center of the perturber ( within the resolution limit ) . since the density wake is effectively shifted toward the perturber compared with the linear case and still located preferentially at the rear side of the perturber , the resulting drag force will be larger than the linear counterpart ( see [ sec : force ] ) . as @xmath27 increases further , the shocked material gathered around the perturber begins to build up a strong pressure barrier which the incident flow can not easily penetrate . this naturally makes the shock detached . figure [ fig : compadens]_b , c _ shows that the postshock density distribution around the perturber with @xmath96 or 20 is almost spherically symmetric , indicating that the kinetic energy of the gas there is much smaller than the thermal and gravitational potential energies . to check if this is indeed the case , we plot in figure [ fig : hse ] the density profiles along the symmetry axis and the @xmath36-axis from the center of a massive perturber with @xmath68 ; both supersonic and subsonic models with @xmath49 and @xmath97 at @xmath43 are presented . also shown as dotted line is the density distribution under the assumption of hydrostatic equilibrium @xmath98 \right\}^{1/(\gamma-1)},\ ] ] where @xmath99 denotes the distance from the perturber center and @xmath100 and @xmath101 are the density and the adiabatic sound speed at @xmath102 , respectively . for both supersonic and subsonic models , the density distributions at @xmath103 along the @xmath104- , @xmath105- , and @xmath36-cuts are virtually identical to each other , and are in remarkable agreement with the predictions of equation ( [ eq : hse ] ) with @xmath106 and @xmath107 for @xmath49 , and @xmath108 and @xmath107 for @xmath83 . the sharp drop offs of the density in the supersonic model near @xmath109 and @xmath110 along the @xmath104- and @xmath36-directions , respectively , are of course due to the bow shock . in the subsonic model , the presence of the primary vortex makes the local density decreased at @xmath111 . note that the quantity @xmath112 measures the specific entropy and thus is conserved in an adiabatic flow without involving a shock , as is the case in the subsonic model . for the supersonic model , however , @xmath112 is increased from unity to 1.34 because of a shock jump . this corresponds to the shock mach number of @xmath113 , which is larger than @xmath49 because the flow is accelerated by the perturber even before experiencing the shock . in the regions with @xmath114 , the density is overall larger along the @xmath105- than @xmath104-directions , providing non - vanishing drag forces . nonetheless , the presence of hydrostatic cores in the density wakes of massive supersonic perturbers makes the drag forces smaller than the linear results . while the density wake _ near _ a massive perturber with @xmath115 is quite different from the linear counterpart , the distant part of the wake is more or less the same . figure [ fig : profiles ] plots the exemplary profiles of the normalized perturbed density , @xmath116 , along the cuts with @xmath38 and @xmath117 for models shown in figure [ fig : compadens ] . note that @xmath116 is nearly independent of @xmath27 in regions far away from the perturber ( e.g. , @xmath118 region in fig . [ fig : profiles]_a _ and @xmath119 region in fig . [ fig : profiles]_b _ ) ; in these regions , the gravitational potential perturbations are weaker by more than two orders of magnitudes than at the perturber location and thus locally in the linear regime . even with low amplitudes of local perturbations , however , the region with @xmath120 along the symmetry axis behind the perturber has the nonlinear density wake that deviates considerably from the linear case . this is because the gas flowing in this region was already affected by strong gravitational potential in the upstream region , and has a diverging velocity field that reduces the perturbed density . small fluctuations of nonlinear density wakes apparent in figures [ fig : compadens ] and [ fig : profiles ] near the @xmath38 axis is thought of as arising from sonic perturbations induced by the primary vortex and its oscillations discussed in [ sec : general ] . we have shown in the previous subsection that a sufficiently - massive supersonic perturber generates a density wake that is characterized by a bow shock standing ahead of the perturber and a surrounding hydrostatic envelope . figure [ fig : dtime ] shows that the quasi - steady value of the detached shock distance @xmath6 is larger for models with larger @xmath27 or smaller @xmath121 . to quantify the dependences of @xmath6 upon @xmath27 and @xmath32 , we introduce the nonlinearity parameter @xmath122 and plot in figure [ fig : deta ] the normalized shock distance against @xmath123 . the various symbols give the mean values of @xmath6 temporally averaged over @xmath124 that ignores the initial wake - development phase . the standard deviations of @xmath6 are also indicated by errorbars . the numerical results are remarkably well described by the two simple power laws : @xmath125 for @xmath126 and @xmath127 for @xmath11 . the behavior of @xmath6 with @xmath128 can be qualitatively understood as follows . for a very massive perturber , the kinetic energy of the incident flow along the symmetry axis is almost entirely converted to the thermal energy of the postshock flow that supports a hydrostatic envelope against the gravitational potential of the perturber . when the shock is strong ( @xmath20 ) , the postshock thermal energy proportional to @xmath129 balances the gravitational potential energy @xmath130 at the shock location , resulting in @xmath131 . in the limit of @xmath132 , the flow does not in principle produce a shock , corresponding to @xmath133 . in practice , the gravitational acceleration is able to turn an incident nearly - transonic gas into a weakly supersonic flow , but the shock that may form is located very far away from the perturber anyway . in view of the detached shock distance , @xmath128 may be a better indicator of nonlinearity than @xmath27 ; for fixed @xmath27 , the shape of a density wake due to a supersonic perturber becomes similar to the linear counterpart as @xmath32 increases . it is well known from hydrodynamic experiments and corresponding theories that a supersonic flow over a solid object with a blunt nose develops a detached bow shock when the nose angle is larger than the maximum angle allowed by the postshock flow ( e.g. , @xcite , and references therein ) . the shocked gas becomes subsonic and slowly adjusts its velocity as it flows downstream to meet the boundary conditions at the surface of the object . in our simulations , the incoming flow toward a massive perturber recognizes the hydrostatic envelope as a spherical obstacle . since the nose angle of a spherical body with respect to the incident flow is 90 degrees , the shock must be detached . even though near - hydrostatic envelopes that form in our simulations are not entirely impenetrable , we want to measure their effective sizes as perturbing obstacles . this can be achieved by comparing our numerical results for the detached shock distances with those from non - gravitating hydrodynamic theories ( or lab experiments ) . assuming that a bow shock ahead of a spherical body with radius @xmath134 has a spherical shape near the symmetry axis , @xcite showed that the standoff distance of the shock is approximately given by @xmath135^{1/(2k ) } -1,\ ] ] where @xmath136 \left [ 2\mu + 1 + \frac{1}{{\mathcal{m}}^2}\right],\ ] ] with @xmath137 . equation ( [ eq : guy ] ) has proven to explain the experimental data quite well ( ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; see also @xcite ) . figure [ fig : dm_rk ] plots as a solid line @xmath138 from equation ( [ eq : guy ] ) with @xmath34 as a function of @xmath32 . also plotted as various symbols are our numerical results for the ratio of @xmath6 to the bhl radius , @xmath139 for models with @xmath11 , averaged over @xmath124 . again , errorbars indicate the standard deviations . a rough agreement between the two results suggests that the bhl radius can be a useful measure of the effective size of the hydrostatic sphere . for a given wake distribution @xmath63 at time @xmath18 of a perturber , it is straightforward to calculate the df force on it by performing integration in equation ( [ eq : f_integ ] ) . figure [ fig : ftime ] plots temporal changes of the df forces normalized by @xmath140 for models with differing @xmath27 , but with fixed @xmath49 , over the course of the wake evolution . the dotted line corresponding to the linear df force ( eq . [ [ eq : f_pt_linear ] ] ) , with @xmath19 given in equation ( [ eq : rmin ] ) , closely follows the numerical results for @xmath33 , showing that the linear drag force increases logarithmically with time . the drag forces for nonlinear cases with high @xmath27 also have a similar logarithmic time dependence , although they fluctuate for a while in response to the oscillations of primary vortices as well as detached bow shocks before a quasi - steady state is attained ; the fluctuation amplitudes are typically @xmath141 , with a smaller value corresponding to larger @xmath27 and @xmath32 . note that for @xmath49 , the normalized drag force decreases with increasing @xmath27 , indicating that the nonlinear effect makes the df force smaller than the linear estimate . this is because a higher value of @xmath27 implies a correspondingly larger detached shock distance , and a hydrostatic envelope with front - back symmetry near the perturber contributes a negligible amount to the net df force . the dimensionless drag forces at @xmath43 when the wakes are in a quasi - steady state are given in figure [ fig : fm ] for various models with different @xmath27 and @xmath32 . for all the supersonic models , the df force is a decreasing function of @xmath27 . the reduction of the normalized df force is larger for models with @xmath142 than highly supersonic models . the mach number corresponding to the maximum drag force shifts from unity to @xmath143 as the wake becomes highly nonlinear . for subsonic models , on the other hand , the nonlinear drag forces for @xmath68 and 50 show some fluctuations ( represented by errorbars ) associated with slowly - evolving vortices present in the wakes , but their respective time - averaged values are very close to the drag forces in the linear regime . in fact , the similarity between the linear and nonlinear drag forces on subsonic perturbers is expected since the linear density wakes also possess front - back symmetry in the vicinity of the perturber @xcite . since the subsonic df forces are dominated by the far field where perturbations are weak regardless of the perturber mass , the linear results should be valid even for very massive perturbers . the gravitational drag force certainly depends on both @xmath27 and @xmath32 , but the discussions given above suggest that it may be through the nonlinearity parameter @xmath123 . to check this , we plot in figure [ fig : fa ] the ratio of the nonlinear df force @xmath7 to the linear prediction @xmath144 as a function of @xmath123 . again , various symbols give temporal averages of @xmath145 over @xmath124 , and their standard deviations are indicated by errorbars . for @xmath146 with which a bow shock that barely forms is attached to a perturber , @xmath147 . when @xmath123 is increased to @xmath1482 , the shock becomes detached , but its standoff distance is not so large . in this case , most of the material in the wake is still located behind of , but closer to the perturber in comparison with the linear wake ( see , e.g. , fig . [ fig : compadens]_b _ ) , resulting in the df force slightly larger ( by less than 20% ) than the linear counterpart . in highly nonlinear cases with @xmath11 , however , the presence of a large hydrostatic envelope makes the drag force reduced considerably . for @xmath11 , the numerical results are well fitted by @xmath149 figure [ fig : fa ] also plots the gravitational drag force from the hydrodynamic simulations of @xcite for bhl accretion flows , which are consistent with our numerical results . we defer to [ sec : diss ] a more detailed discussion of our results in connection with the bhl flows . to ascertain that the reduction of the df force in highly nonlinear supersonic cases is really caused by the presence of spherically - symmetric hydrostatic envelopes near the perturbers , we calculate the drag force by imposing a cut - off radius @xmath19 such that only the regions in the wake with @xmath150 participate in the force evaluation ( eq . [ [ eq : f_integ ] ] ) . figure [ fig : frmin ] plots the resulting dependence of @xmath7 upon @xmath19 for the model with @xmath68 and @xmath49 at @xmath43 . the vertical dashed line marks the location of the bow shock along the symmetry axis , while the dotted line indicates a slope of @xmath151 . note that the drag force is independent of @xmath19 for @xmath152 , clearly demonstrating that the hydrostatic sphere surrounding the perturber has a negligible contribution to the net drag force . figure [ fig : frmin ] shows @xmath153 for large @xmath19 , analogous to the linear cases ( see eq . [ [ eq : f_pt_linear ] ] ) . from the study of bhl accretion flows to large gravitating bodies , @xcite similarly found that the drag force declines logarithmically as the size of the accretor increases . finally , we remark on the effect of numerical resolution on the df force . figure [ fig : res ] plots the time evolution of the detached shock distance as well as the drag force on a perturber with @xmath96 and @xmath49 from models with different resolution : @xmath154 , 0.2 , 0.4 , and 0.8 , where @xmath155 is the grid spacing . in models with @xmath156 and 0.8 , the shock has a larger standoff distance without noticeable oscillations and the df force is correspondingly smaller than in models with higher resolution . compared with the @xmath157 model , the model with @xmath158 presents the shock oscillations with higher amplitudes and arrives at quasi - steady equilibrium earlier . this resolution dependence of the shock oscillations was also noticed by @xcite for adiabatic bhl flows onto an absorbing perturber . the resolution study shown in figure [ fig : res ] suggests that our numerical results for the df forces are reliable as long as 5 or more grids per @xmath3 are taken . df of bodies orbiting in a gaseous medium is of great importance in various astronomical systems ranging from protoplanetary disks to galaxy clusters . in previous analytic studies of df , it has been assumed that the mass of a moving object is small enough for the induced density wake to have low amplitudes and be thus in the linear regime . however , there are many astronomical situations such as in a merger of black holes near a galaxy center and migration of protoplanets , where a perturber is so massive that the induced wakes are well in the nonlinear regime . in this paper , we use numerical hydrodynamic simulations to explore nonlinear gravitational responses of the gas to , and the resulting drag force on , a perturber with mass @xmath0 moving straight at velocity @xmath1 in an initially static , uniform background with density @xmath15 and adiabatic sound speed @xmath2 . the perturber is represented by a plummer sphere with softening radius @xmath3 . unlike in the bhl problems , the perturber in our models does not possess a defined surface through which gas is accreted or reflected . assuming an axial symmetry , we solve the basic equations on the ( @xmath36 , @xmath37 ) plane , where @xmath36 and @xmath37 denote the direction perpendicular and parallel to the motion of the perturber , respectively . our numerical models are completely characterized by two dimensionless parameters : @xmath4 and @xmath5 . to study df in various situations , we run as many as 58 models that differ in @xmath27 and @xmath32 . our standard models have 5 zones per @xmath3 , but we also run models with different resolutions to ensure that the density and velocity wakes are fully resolved . for supersonic models ( @xmath159 ) , we find that a massive perturber with @xmath26 produces a bow shock ahead of it through which the incident supersonic flow becomes subsonic . in the beginning , the postshock flow in the nonlinear cases develops transient features such as a primary vortex ring and an associated counterstream near the symmetry axis , which causes the shock to oscillate around its equilibrium position . the shock oscillation added by the counterstream is overstable , amplifying the amplitudes of its oscillation and the vortex movement in the ( @xmath36 , @xmath37)-plane ( see [ sec : general ] ) . the vortex ring is eventually shed downstream from the perturber , leaving the nearly - stationary bow shock and a quasi - hydrostatic envelope that surrounds the perturber . on the other hand , subsonic models with @xmath26 without involving a shock retain a primary vortex and many other small - scale structures until the end of runs . nevertheless , strong gravity makes the density distribution near the perturber still very close to that under hydrostatic equilibrium . by comparing the numerical results from various supersonic models with differing @xmath27 and @xmath32 , we find that the simulation outcomes such as the detached shock distance and the gravitational df force are very well quantified by a single parameter @xmath128 defined in equation ( [ eq : amod ] ) . when @xmath146 , the system is in the linear regime where a mach cone is bounded by low - amplitude mach waves , and the drag force is just the same as the analytic linear value @xmath12 of @xcite . when @xmath123 is moderate at @xmath126 , the mach waves turn into a bow shock that is weakly detached , with the standoff distance @xmath160 from the perturber . in this case , the density wake is slightly shifted toward the perturber compared with the linear counterpart , causing the drag force to be larger than @xmath12 by a factor of less than 1.2 . in the highly nonlinear regime with @xmath11 , however , the detached shock distance behaves as @xmath161 , and the nonlinear drag force @xmath7 is given by @xmath162 . the reduction of the drag force compared with the linear value is because the density wake close to the perturber is in near - hydrostatic equilibrium and thus contributes a negligible amount to the net df force . since front - back symmetry notable for nonlinear wakes exists also in the linear wakes with @xmath163 , the nonlinear drag force on a massive subsonic perturber is similar to the linear prediction . as mentioned in [ sec : intro ] , our model simulations differ from those of the bhl accretion flows in terms of the boundary conditions . in our models , a perturber simply provides a gravitational potential for the background gas and does not hold any surface , while models for the bhl accretion considered a defined surface through which the gas is accreted or reflected . the different boundary conditions might lead to different evolution and structure of wakes . for instance , the bhl accretion flows around a perturber present nonlinear features including a detached bow shock , vortex oscillations , and counterstreams in the forward / backward directions , just as in our models , if the perturber is not perfectly absorbing @xcite . when the perturber has a totally absorbing surface , on the other hand , the wakes are relatively quiescent , flowing nearly spherically into the perturber that absorbs the angular momentum carried by the accreting gas @xcite . in the case of an extremely small perturber compared with the accretion radius , even an absorbing body produces violent features since it is not able to accept the whole accreting gas @xcite . despite these differences , the detached distances of bow shocks and the resulting df forces appear to be not so sensitive to the boundary conditions adopted . figure [ fig : detabhl ] plots the detached shock distances inferred from the results of two - dimensional axisymmetric simulations @xcite and the tabulated values from three - dimensional nonaxisymmetric runs @xcite for the bhl accretion flows , by taking @xmath3 equal to the radius of the perturber . the bhl results under the reflecting boundary conditions are larger than our results by a factor of only @xmath164 , while those under the absorbing boundary conditions are smaller by a factor of @xmath165 . since the work on the bhl accretion mostly focused on mass accretion rates , only a few of them evaluated the gravitational drag forces . in figure [ fig : fa ] , we plot using star symbols the drag forces from the adiabatic runs with @xmath34 in @xcite , which are the only data we acquired from the literature that allow a reliable comparison with our results . note that the drag force is larger on a purely - absorbing perturber for which the detached shock distance is smaller . the df forces from @xcite are roughly consistent with , and follow a similar trend to , our results . this is probably because even though density wakes are nowhere close to being hydrostatic in bhl accretion flows , they somehow maintain front - back symmetry that makes the net df force smaller . while we have considered in this paper only an adiabatic gas with index @xmath34 , we note that the df force may depends sensitively on @xmath166 . simulations of the bhl accretion flows reported that as @xmath166 decreases , the detached shock distance decreases and the density wake becomes shaped increasingly into a cone - like structure similar to the linear wake @xcite . this is presumably because bow shocks in lower-@xmath166 models should be stronger in order to compensate for the diminished pressure in supporting the postshock gas against the perturber s gravity . consequently , drag forces are larger for the gas with smaller @xmath166 @xcite . @xcite carried out numerical simulations of a df - induced merger of supermassive binary black holes due to an adiabatic gas with @xmath34 . their figure 9 presents the effect of the black hole mass on the evolution of the binary separation . from this figure , we infer that the orbital decay time scales with the black hole mass as @xmath167 in their simulations , which is somewhat shallower than the linear expectation of @xmath168 ( e.g. , @xcite ) . on the other hand , our results predict @xmath169 if the perturber is sufficiently massive , suggesting that the delayed orbital decay of supermassive black holes in @xcite is partly due to the nonlinear effect . of course , there are many other factors that may change @xmath170 . while we consider quite an ideal situation in which a perturber is moving straight in a static , uniform medium , the gas in @xcite is distributed in a rotating , radially - stratified , self - gravitating disk and the black holes follow curvilinear , possibly eccentric , orbits . in addition , the mach number of the black holes in their models is likely to vary during the decay toward the orbital center , which may also modify the decay time . since these compact objects have a very small size , it is also an issue whether numerical models resolve detached bow shocks , which is crucial in evaluating the drag force accurately . we will discuss these in subsequent work . we acknowledge a helpful report from an anonymous referee and eve c. ostriker for useful suggestions . the software used in this work was in part developed by the doe - supported asc / alliance center for astrophysical thermonuclear flashes at the university of chicago . this work was supported by kicos through the grant k20702020016 - 07e0200 - 01610 provided by most . simulations were performed by using the supercomputing resource of the korea institute of science and technology information through the grant ksc-2009-s02 - 0008 . using the fourier transform for the space variables and the laplace transform for the time variable , @xcite derived an analytic solution for the velocity field in the linear wake , but unfortunately the resulting expression ( their equation [ 47 ] ) contains an typographical mistake . here we rederive the expression using fourier transforms for both space and time variables . we consider a point - mass perturber with mass @xmath0 on a straight - line trajectory through a uniform medium with density @xmath15 and adiabatic sound speed @xmath2 . the basic equations of hydrodynamics for an inviscid , nonmagnetic , non - self - gravitating gas are @xmath171 and @xmath172 where @xmath173 is the gravitational potential of the perturber with density @xmath174 , satisfying the poisson equation @xmath175 . assuming that the density and velocity fields induced by the perturber have small amplitudes , we linearize equations ( [ eq : con ] ) and ( [ eq : mom ] ) to obtain @xmath176 and @xmath177 where @xmath178 is the dimensionless perturbed density . taking fourier transformation of equations ( [ eq : con_p ] ) and ( [ eq : mom_p ] ) , we have @xmath179 and @xmath180 where symbols with and without hat denote a fourier pair defined by @xmath181 and @xmath182 with @xmath183 representing any physical quantities . for a perturber introduced at @xmath85 and moving at a constant speed @xmath1 along the positive @xmath37-direction , @xmath184 , where @xmath185 is the kronecker delta and @xmath186 is a heaviside step function . this results in @xmath187.\ ] ] the second term in equation ( [ eq : ftd_ext ] ) is originated from the step function , which vanishes if a steady state is assumed ( e.g. , * ? ? ? substituting equation ( [ eq : ftd_ext ] ) into equation ( [ eq : ftd ] ) , and taking an inverse fourier transform of the resulting expression for @xmath188 , one can obtain the solution of the perturbed density for finite - time perturbations . the detailed procedure is similar to that of the fourier - laplace transform method of @xcite , so that we simply write the result @xmath189 where @xmath190 and @xmath191 , 1 , and 0 in region i ( @xmath192 , @xmath159 ) , region ii ( @xmath193 ) , and any other region , respectively . note that equation ( [ eq : alpha ] ) is identical to equation ( 10 ) of @xcite based on the retarded green s function technique . combining equations ( [ eq : ftd ] ) , ( [ eq : ftv ] ) , and ( [ eq : ftd_ext ] ) together in favor of @xmath194 and taking its inverse fourier transform , we have @xmath195 \frac{\omega{\mbox{\boldmath{$k$}}}}{k^2 } e^{i({\mbox{\boldmath{$k$}}}\cdot{\mbox{\boldmath{$x$}}}-\omega t)}\,d^3{\mbox{\boldmath{$k$}}}\,d\omega.\ ] ] the second term in the integrand of equation ( [ eq : iftv ] ) has three simple poles on the real @xmath196 axis at @xmath197 and @xmath198 . to ensure the proper analyticity , we must displace them by an infinitesimal amount into the upper or lower half @xmath196-plane . since our problem requires @xmath199 everywhere for @xmath200 , the proper choice of the pole displacements should be @xmath201 into the upper half @xmath196-plane and @xmath202 to the lower half @xmath196-plane . for @xmath200 , then , a contour of integration consisting of the real @xmath196 axis and an infinitely - large semicircle in the upper half @xmath196-plane encloses only @xmath201 , and the calculus of residues guarantees that after the integration over @xmath196 , the first and second terms in the integrand of equation ( [ eq : iftv ] ) become identical with each other with opposite sign , resulting in no velocity perturbations before the introduction of the perturber . for @xmath203 , the contour that goes along an infinitely - large semicircle in the lower half @xmath196-plane instead contains the poles at @xmath202 . the remaining steps in the integration with respect to @xmath196 is quite straightforward to yield @xmath204 to carry out the integration in equation ( [ eq : iftv_a ] ) , it is convenient to use spherical coordinates in the @xmath205-space as @xmath206 which is oriented relative to the space coordinates @xmath207 satisfying @xmath208 . we then obtain @xmath209 e^{ikr\sin\theta\cos\varphi}\nonumber\\ \times ( \sin^2\theta\cos(\varphi+\zeta){{\mbox{\boldmath{$\mathrm{e_x}$}}}}+\sin^2\theta\sin(\varphi+\zeta){{\mbox{\boldmath{$\mathrm{e_y}$}}}}+\sin\theta\cos\theta{{\mbox{\boldmath{$\mathrm{e_z}$ } } } } ) \,\,dk\,d\theta\,d\varphi,\end{aligned}\ ] ] where @xmath210 , @xmath211 , and @xmath212 are the unit vectors in the @xmath213- , @xmath214- , and @xmath37-directions , respectively . without loss of generality , we take @xmath215 , in which case @xmath210 and @xmath211 correspond to the unit vectors in the radial ( @xmath216 ) and azimuthal ( @xmath217 ) directions , respectively . it then follows that the azimuthal velocity vanishes ( i.e. , @xmath218 ) since this component in the integrand is an odd function of @xmath219 . for the velocity component in the direction of motion , we substitute @xmath220 to the second term inside the square bracket of equation ( [ eq : v_int3 ] ) to obtain @xmath221 -\cos(ks\cos\theta ) \right\}.\ ] ] using [ g6.671.8 ] ] refers to the number of formula in the integral table of @xcite . ] , inserting the substitutions @xmath222 and @xmath223 for the first and second terms of equation ( [ eq : vz_int2 ] ) , respectively , and applying [ g2.551.3 ] for the @xmath128-integrations , we obtain @xmath224,\ ] ] where @xmath2252 , 1 , 0 and @xmath226 , 0 , 1 for region i ( @xmath192 , @xmath159 ) , region ii ( @xmath193 ) , and any other region , respectively . for the velocity component in the radial direction , the use of [ g6.671.1 ] similarly leads to @xmath227.\ ] ] the results of our time - dependent linear analyses for density and velocity fields are confirmed by direct numerical simulations for a low - mass perturber with @xmath33 presented in [ sec : linear ] . in terms of our notation , @xmath228 , @xmath229 , @xmath230 , @xmath231 , and @xmath232 . ] , equation ( 47 ) of @xcite can be rewritten as @xmath233.\ ] ] comparison of equation ( [ eq : just ] ) with equations ( [ eq : vz ] ) and ( [ eq : vr ] ) shows that equation ( 47 ) of @xcite takes @xmath226 everywhere , which is clearly incorrect . alibert , y. , mordasini , c. , benz , w. , & winisdoerffer , c. 2005 , , 434 , 343 armitage , p. j. , & rice , w. k. m. 2005 , astro - ph/0507492 barausse , e. 2007 , , 382 , 826 bondi , h. , & hoyle , f. 1944 , , 104 , 273 bondi , h. 1952 , , 112 , 195 butler , r. p. , wright , j. t. , marcy , g. w. et al . 2006 , , 646 , 505 chandrasekhar , s. 1943 , , 97 , 255 colella , p. , & woodward , p. r. 1984 , journal of computational physics , 54 , 174 conroy , c. , & ostriker , j. p. 2008 , , 681 , 151 cuadra , j. , armitage , p. j. , alexander , r. d. , & begelman , m. c. 2009 , , 393 , 1423 dokuchaev , v. p. 1964 , soviet astron . , 8 , 23 dotti , m. , colpi , m. , & haardt , f. 2006 , , 367 , 103 dotti , m. , colpi , m. , haardt , f. , & mayer , l. 2007 , , 379 , 956 edgar , r. 2004 , new astronomy reviews , 48 , 843 el - zant , a. a. , kim , w .- t . , & kamionkowski , m. 2004 , , 354 , 169 escala , a. , larson , r. b. , coppi , p. s. , & mardones , d. 2004 , , 607 , 765 escala , a. , larson , r. b. , coppi , p. s. , & mardones , d. 2005 , , 630 , 152 faltenbacher , a. , kravtsov , a. v. , nagai , d. , & gottloeber , s. 2005 , , 358 , 139 foglizzo , t. , & ruffert , m. 1997 , , 320 , 342 foglizzo , t. , & ruffert , m. 1999 , , 347 , 901 foglizzo , t. , galletti , p. , & ruffert , m. 2005 , , 435 , 397 fryxell , b. a. , taam , r. e. , & mcmillan , s. l. w. 1987 , , 315 , 536 fryxell , b. a. , & taam , r. e. , 1988 , , 335 , 862 fryxell , b , . olson , k. , ricker , p. , timmes , f. x. , zingale , m. , lamb , d. q. , macneice , p. , rosner , r. , truran , j. w. , & tufo , h. 2000 , , 131 , 273 furlanetto , s. r. , & loeb , a. 2002 , , 565 , 854 gradshteyn , i. s. , & ryzhik , i. m. 1994 , table of integrals , series , and products ( new york : academic press ) guy , t. b. 1974 , aiaa journal , 12 , 380 heberle , j. w. , wood , g. p. , & gooderum , p. b. 1950 , nasa tn 2000 hoyle , f. , & lyttleton , r. a. 1939 , proceedings of the cambridge philosophical society , 34 , 405 hunt , r. 1971 , , 154 , 141 hunt , r. 1979 , , 188 , 83 just , a. , & kegel , w. h. 1990 , , 232 , 447 karas , v. , ubr , l. 2001 , , 376 , 686 kim , h. , & kim , w .- t . 2007 , , 665 , 432 kim , h. , kim , w .- t . , & snchez - salcedo , f. j. 2008 , , 679 , l33 kim , w .- t . , el - zant , a. a. , & kamionkowski , m. 2005 , , 632 , 157 kim , w .- t . 2007 , , 667 , l5 koide , h. , matsuda , t. , & shima , e. 1991 , , 252 , 473 liepmann , h. , & roshko , a. 1957 , elements of gasdynamics , galcit aeronautical series ( new york : wiley , 1957 ) matsuda , t. , inoue , m. , & sawada , k. 1987 , , 226 , 785 matsuda , t. , sekino , n. , shima , e. , & sawada , k. 1989 , , 236 , 817 mayer , l. , kazantzidis , s. , madau , p. , colpi , m. , quinn , t. , & wadsley , j. 2007 , science , 316 , 1874 milosavljevi , m. , & merritt , d. 2003 , in aip conf . proc . 686 , the astrophysics of gravitational wave sources , ed . j. m. centrella ( new york : aip ) , 201 narayan , r. 2000 , , 536 , 663 ostriker , e. c. 1999 , , 513 , 252 rephaeli , y. , & salpeter , e. e. 1980 , , 240 , 20 ruderman , m. a. , & spiegel , e. a. 1971 , , 165 , 1 ruffert , m. 1994 , , 106 , 505 ruffert , m. 1995 , , 113 , 133 ruffert , m. 1996 , , 311 , 817 ruffert , m. 1999 , , 346 , 861 snchez - salcedo , f. j. , & brandenburg , a. 1999 , , 522 , l35 snchez - salcedo , f. j. , & brandenburg , a. 2001 , , 322 , 67 schreier , s. compressible flow ( john wiley & sons ) , 185 schwartz , r. n. , & eckerman , j. 1956 , journal of applied physics , 27 , 169 shankar , a. , kley , w. , & burkert , a. 1993 , , 274 , 955 shima , e. , matsuda , t. , takeda , h. , & sawada , k. 1985 , , 217 , 367 taam , r. e. , & fryxell , b. a. 1988 , , 327 , l73 van dyke , m. d. , & milton , d. 1958 , journal of the aeronautical sciences , 25 , 485

non - linear voter models assume that the opinion of an agent depends on the opinions of its neighbors in a non - linear manner . this allows for voting rules different from majority voting . while the linear voter model is known to reach consensus , non - linear voter models can result in the coexistence of opposite opinions . our aim is to derive approximations to correctly predict the time dependent dynamics , or at least the asymptotic outcome , of such local interactions . emphasis is on a probabilistic approach to decompose the opinion distribution in a second - order neighborhood into lower - order probability distributions . this is compared with an analytic pair approximation for the expected value of the global fraction of opinions and a mean - field approximation . our reference case are averaged stochastic simulations of a one - dimensional cellular automaton . we find that the probabilistic second - order approach captures the dynamics of the reference case very well for different non - linearities , i.e for both majority and minority voting rules , which only partly holds for the first - order pair approximation and not at all for the mean - field approximation . we further discuss the interesting phenomenon of a correlated coexistence , characterized by the formation of large domains of opinions that dominate for some time , but slowly change . _ keywords : _ opinion dynamics ; voter model ; pair approximation ; higher - order probability distribution , cellular automata

1604.07778

the concept of _ entropy _ plays a key role in describing the transition between disorder and order . specifically , the emergence of order in a random phase is indicated by a significant reduction of entropy . calculating this reduction requires to know the probability of each possible system configuration that is compatible with the given system constraints - a challenging problem both methodologically and computationally . if we consider a system of @xmath0 elements each of which can be in one of two states for example up and down spins in a physical system , or agents with opposite opinions in a social system , or agents with the two strategies _ cooperate _ or _ defect _ in an economic system the number of possible configurations is @xmath1 , which can be quite large . in fact , statistical physics was founded in the 19th century to provide an efficient solution based on the concept of statistical ensembles and state sums . in this paper , we address the problem by proposing a stochastic approach that allows to decompose such probabilities for systems characterized by neighbor - neighbor interactions . our candidate model to describe this interaction is the so - called _ voter model _ which is discussed in more detail in sect . [ sec : voter ] . in this model , agents are in one of two discrete states , @xmath2 , denoted as `` opinions '' . they change their opinion in response to the opinions in their neighborhood . in order to define such a neighborhood , we have chosen a one - dimensional cellular automaton ( ca ) , in which consecutively numbered cells @xmath3 represent agents ( see figure [ fig:1dca ] ) . we assume that the ca forms a ring to close the system . each agent @xmath4 then has two neighbors @xmath5 , @xmath6 i.e. their opinions @xmath7 form a _ the second - order neighborhood that also takes the neighbors of @xmath5 , @xmath6 into account , then results in a _ quintuplet _ of opinions @xmath9 . ) and quintuplet ( @xmath10 ) definition of the neighborhood of agent @xmath4,scaledwidth=50.0% ] our major assumption of the voter model is that changes in the opinion of agent @xmath4 are only caused by the first - order neighborhood . specifically , the agent responds to the _ local frequency _ of opinions in the triplet that also includes its own opinion . however , because the dynamics of agent @xmath4 depends on its neighbors , it is also coupled to their dynamics , i.e. to the second - order neighborhood of agent @xmath4 , and so forth . this specifically denotes the problem that we are going to discuss in this paper . in a stochastic approach , we are faced with a system of @xmath0 coupled dynamic equations for the probabilities @xmath11 to find any agent @xmath4 with opinion @xmath7 at time @xmath12 . this coupling exists only through the two neighbors of each agent , i.e. changes in a far distant cell only propagate slowly through the ca by means of neighbor - neighbor interactions . so , precisely , how large should be the neighborhood taken into account for the dynamics of the ca ? or , how far reaching are correlations in the changes of opinions ? to answer this question , in this paper we propose three different analytic approximations for the dynamics which take different neighborhood sizes into account : zero - order no correlations between neighboring agents , first - order correlations between an agent and one of its nearest neighbors ( so called pair approximation ) , second - order correlations between an agent and its second - nearest neighbors . we test the validity of these dynamic approximations by comparing them to stochastic simulations of the ca , averaged over a number of runs . our emphasis is of course on the second - order approximation , which extends previous investigations . but we want to understand under which circumstances this approximation fares better than the simpler ones . therefore , we have chosen different variants of the voter model , also known as _ nonlinear voter models_. the non - linearity is with respect to the response to the local frequency of opinions . we further compare deterministic and stochastic dynamics , to show whether our analytic approximations can correctly predict the dynamics . our variable of interest is the expected fraction of opinion 1 , denoted as @xmath13 , which is an aggregated variable for which we derive the dynamics based on a stochastic approach . our paper combines , and extends , previous investigations in different directions . one line of research refers to the approximation of higher - order probability distributions by means of lower - order distributions @xcite which has been also applied to the voter model @xcite . this implies a loss of ( microscopic ) information which can be quantified by means of information - theoretic measures @xcite . the question whether the coarse - grained dynamics is still markovian can be answered by analyzing emergent macroscopic memory effects using information - theoretic measures @xcite . such measures have also been applied to voter models @xcite . in most cases , the markov chain analysis becomes quite cumbersome and therefore is restricted to one - dimensional ca @xcite . another line of research considers different forms of neighborhood approximations specifically for binary state - dynamics ( see @xcite for a good overview ) , which also has been applied to the voter model , already @xcite . we note that our pair approximation approach follows @xcite , but applies it here to a one - dimensional ca , which results in different expressions for @xmath13 and the correlations @xmath14 . compared to two - dimensional ca @xcite or even complex networks @xcite , one could find one - dimensional ca too simple . but this judgment is in fact not justified . already one - dimensional ca have proven to exhibit a really complex dynamics , with a chance to derive analytic expressions . extensions of the simple voter model , for example the sznajd model @xcite or the @xmath15-voter model @xcite , could be thoroughly analyzed for one - dimensional ca . the emphasis of our investigations is on the validity of the analytic approximations for _ nonlinear _ voter models . this non - linearity can be introduced in different ways . in ref . @xcite , the authors discuss it on the level of individual agents that respond to neighboring influences in a _ heterogeneous _ manner . specifically , @xcite assumes a heterogeneous inertia for agents to change their opinions , @xcite assumes a heterogeneous neighborhood size to influence agents , while @xcite assumes a heterogeneous weight for the influence of agents . compared to these approaches , we assume a _ homogeneous _ , but nonlinear response of agents on the local frequency of opinions . specifically , we consider the _ majority _ rule , where the tendency of agents to change their opinion _ increases _ with the frequency of the opposite opinion . the _ minority _ rule , on the other hand , assumes exactly the opposite , i.e. a _ decreasing _ tendency . this can be simply varied by one parameter @xmath16 , which however is assumed to be the same for all agents . eventually , we would like to point out that we refrain from interpreting our model in a social context . although agents are called `` voters '' and their states are called `` opinions '' , the simplicity of the underlying assumptions does not justify to sell the model as a reflection of a social system . we see it rather as a very generic setup to better understand the impact of local feedback on emerging systemic properties , such as _ consensus _ ( i.e. a `` ferromagnetic '' phase ) or _ coexistence _ ( i.e. a `` paramagnetic '' phase ) . but we acknowledge that , despite this basic limitations , the voter model has been applied in various context , e.g. to model investors behavior in financial markets @xcite , emerging communication networks @xcite , or invasion of species @xcite . a good overview of spin - type models in sociophysics is given in @xcite . in this paper , we consider a one - dimensional cellular automaton ( ca ) consisting of @xmath0 cells , each of which is identified by the index @xmath17 ( see figure [ fig:1dca ] ) . each cell shall be characterized by a discrete value @xmath18 , hence the total distribution of states is given by the vector @xmath19 . assuming a torus space , each cell @xmath4 has a clearly defined neigborhood of first and second nearest neighbors , @xmath20 and @xmath21 . the probability to find cell @xmath4 in state @xmath7 at time @xmath12 ( where time shall be measured in discrete steps ) is @xmath22 . consequently , the conditional probabilities @xmath23 and @xmath24 describe the probability to find cell @xmath4 in state @xmath7 given that it has the first and second nearest neighbors @xmath25 , @xmath26 under markov assumptions the chapman - kolmogorov equation holds for the probability @xmath11 to find cell @xmath4 in state @xmath7 at time @xmath27 : @xmath28 \;p(\hat{\theta}_{i},t)\ ] ] the propagator @xmath29 $ ] denotes the transition probability to go from any given state @xmath30 at time @xmath12 to the assumed state @xmath7 in the next time step , @xmath27 . here the summation is over all possible realizations @xmath31 of @xmath30 , i.e. the @xmath32 possible states @xmath33 . at this point , we make our _ 1st assumption _ , namely that any change of @xmath7 depends on the _ nearest neighbors _ , given by @xmath20 . that means a _ triplet _ @xmath8 decides about the value of @xmath7 in the next time step . if we define the probability of a triplet configuration as @xmath34 then the probability @xmath22 results as the marginal distribution of the triplet probability : @xmath35 here the summation is over all possible realizations @xmath36 of the nearest neighborhood @xmath37 , i.e. @xmath38 different possibilities . @xmath39 denotes the conditional probability to find @xmath7 as the focal cell given a triplet @xmath40 , and the summation is over all possible realizations @xmath41 of @xmath40 , i.e. @xmath42 diffferent possibilities . based on the assumption that the nearest neighborhood matters , we can rewrite eqn . ( [ eq : chap0 ] ) as @xmath43 \;p(\hat{\theta}_{i}|\underline{\theta}'_{i},t)\ ] ] where @xmath44 denotes the conditional probability to find the focal cell in state @xmath7 given the neighborhood @xmath45 . using bayes rule , we can express this probability as : @xmath46 with this , we can eventually rewrite the chapman - kolmogorov eqn . ( [ eq : chap0 ] ) for the single cell @xmath4 in terms of the triplet probabilities , @xmath47 : @xmath48 \;p(t_{i},t)\ ] ] the propagator @xmath49 $ ] describes the transition probabilities to go from any possible triplet @xmath40 to a state @xmath7 during the next time step . this equation leaves us with the further specification of the triplet probability , @xmath47 . while , according to our 1st assumption , the occurence of @xmath7 is just determined by the nearest neighbors @xmath20 , the occurrence of either @xmath50 or @xmath51 also depends on their nearest neighbors , i.e. the second nearest neighbors of @xmath4 , @xmath21 . that means a _ quintuplet _ @xmath9 decides about the value of the triplet @xmath8 in the next time step . if we consider a quintuplet configuration @xmath10 , then we can define the triplet probability as the marginal probability : @xmath52 here the summation is over all possible realizations @xmath53 of the nearest neighborhood @xmath54 , i.e. @xmath38 different possibilities . @xmath55 denotes the conditional probability to find the focal triplet @xmath40 given a quintuplet @xmath10 , and the summation is over all possible realizations @xmath56 of @xmath10 , i.e. @xmath57 diffferent possibilities . for the dynamics for the probability to find a triplet in state @xmath40 at time @xmath27 , we can write a chapman - kolmogorov equation quite similar to eqn . ( [ eq : chap ] ) @xmath58 \;p(q_{i},t)\ ] ] the propagator @xmath59 $ ] describes the transition probabilities to go from any possible quintuplet @xmath10 to a triplet @xmath40 during the next time step . in order to specify the quintuplet probability @xmath60 , we may consider a neighborhood of seven , etc . however , following the above procedure repeatedly is neither convenient nor practicable since at the end we have to consider @xmath1 possible configurations . therefore , in the next section , we present a more convenient method to determine the quintuplet probability distribution . instead of defining the quintuplet distribution in terms of higher - order probability distributions , following the procedure of the previous section , we now make our _ 2nd assumption _ by expressing these probabilities in terms of _ lower - order distributions _ , i.e. triplet distributions , this way arriving at a closed form description of the problem . in the following we refer to the theory of approximating discrete probability distributions already developed in the late 1950 @xcite , which was also applied to vm @xcite . the idea is to approximate an @xmath61th order probability distribution , @xmath62 by products containing the probabilities of given subsets , e.g. @xmath63 . it is known that such a product approximation contains at most @xmath61 terms , since every new term has to contain at least one variable @xmath64 not contained in previous terms . for example , the factorization @xmath65 already satisfies this condition , although it may be not a good approximation , since it holds only for ideal systems . we note that the approximation procedure is not unique , i.e. there are various ways of constructing a product approximation . for example , taking @xmath66 , this can be approximated by @xmath67 as well as by @xmath68 or by @xmath69 . which one of the above approximations is the most suitable has to be determined based on the information available about the triplet distribution @xmath66 . in some cases , there are measurements of this distribution , or additional information about the dependency between the @xmath64 as we will use below . with reference to our quintuplet approximation using the triplets defined , we propose the following procedure : we first note that there are different triplets contained in the quintuplet configuration , for example : @xmath8 , @xmath70 , @xmath71 . these are valid triplets because they consider the correct neighborhood relations , whereas for example @xmath72 would be inappropriate . out of the valid triplets , we may choose for example the first one , @xmath73 and complete the product approximation as follows : @xmath74 in order to express the conditional probabilities using the triplet distribution , we apply bayes rule : @xmath75 which results in the final quintuplet approximation : @xmath76 we note that this product approximation is the only possible one , given that we have an ordered set of variables @xmath7 according to their neighborhood relations . using the approximation explained above , we have reduced the higher - order description to the level of triplet distributions , @xmath77 , etc . , which results in a closed form dynamics . we now have to specify the initial conditions for the distributions . while we are able to make any suitable assumption about the initial triplet distribution , we assume here that the occupation of the different cells is initially statistically independent , i.e. @xmath78 for further investigations , we have set @xmath79 . in order to solve the above set of equations , we have two different possibilities : we can solve the dynamics ( a ) on the level of the distribution @xmath11 , eqn . ( [ eq : chap ] ) , or ( b ) on the level of the triplet distribution @xmath47 , eqn . ( [ eq : chap2 ] ) . variant ( a ) requires more computational effort , since we have to replace any @xmath47 as a marginal distribution of triplets , whereas in variant ( b ) , given the reduction of the quintuplet distributions , we already have a closed form dynamics for the triplet distributions , which is computationally more efficient . variant ( b ) , however , requires to know the transition probabilities on the level of the _ triplets _ , whereas these are usually specified on the level of single cell changes ( see also the following section ) . hence , we have to determine the propagators @xmath80 $ ] in terms of @xmath81 $ ] . since our 1st assumption specifies that the change of every cell in the triplet only depends on its nearest neighbors , we are able to factorize these transition probabilities : @xmath82= p\left[(\theta_{i-1},t+1 ) \gets ( t_{i-1},t)\right ] \;\times \nonumber \\ & \ ; \times \ ; p\left[(\theta_{i},t+1 ) \gets ( t_{i},t)\right ] \ ; p\left[(\theta_{i+1},t+1 ) \gets ( t_{i+1},t)\right]\end{aligned}\ ] ] this completes our closed form description . to calculate the time - dependent probability distribution of triplets , we have to insert eqs . ( [ eq:5app - fin ] ) , ( [ eq : factor ] ) into eqn . ( [ eq : chap2 ] ) and from the result calculate @xmath11 as the marginal distribution , eqn . ( [ eq : marginal ] ) . the only remaining task is now to specify the transition probabilities for single cells , @xmath81 $ ] at which point the vm comes into play . the _ voter model _ , in its basic form , is applied to a population of @xmath0 agents , where each agent is characterized by a discrete value , its _ `` opinion '' _ , @xmath83 . the _ opinion dynamics _ at the level of the agents works as follows : two agents @xmath4 and @xmath84 are _ randomly _ chosen from the population , and agent @xmath84 adopts the opinion of agent @xmath4 , i.e. , @xmath85 . after @xmath0 such update events , time is increased by 1 ( random sequential update ) . the rather simplified rule limits the applications of the voter model to any real voting process or opinion dynamics . in a well mixed population , the probability for agent @xmath84 to adopt a given opinion @xmath2 is simply proportional to the global frequency @xmath86 of agents with the respective opinion : @xmath87 this dynamics is known to always converge to complete _ consensus _ , @xmath88 or @xmath89 . the only interesting question is then how long it will take to reach this state dependent on the system size @xmath0 and the initial condition @xmath90 . this picture becomes more complex if instead of a well - mixed population a defined neighborhood for each agent is assumed . this can be a social network where agents have _ links _ to other agents , or simply a lattice where the neighborhood is given by the geometry that defines the nearest and second nearest neighbors of an agent . then , instead of the global frequency , the probability of an agent to adopt a given opinion depends on the _ local frequency _ , @xmath91 , of this opinion in the neighborhood of that agent : @xmath92 \;;\quad f_{i}^{(1{{\kern0.1em-\kern0.1em}}\sigma ) } = 1 - f_{i}^{\sigma}\ ] ] here @xmath93 means the kronecker delta , which is 1 only for @xmath94 and zero otherwise . in this paper , we restrict ourselves to the one - dimensional regular lattice , where each agent is represented by a cell @xmath4 that has precisely two neighbors . different from the basic voter rule , in the calculation of the local frequency we have taken the opinion @xmath7 of the focal cell @xmath4 into account . this adds some inertia to the dynamics as agent @xmath4 counts towards the local minority / majority . it also avoids stalemate situations where the two neighbors have different opinions . the linear voter model would assume that the transition rate @xmath95 of a focal cell @xmath4 to change its opinion from @xmath96 to @xmath97 is directly proportional to the local frequency @xmath98 , i.e. the _ opposite _ opinion in the neighborhood . the expression of eq . ( [ eq:2 ] ) is a specification of the more general transition probability @xmath81 $ ] used in eq . ( [ eq : factor ] ) . it takes into account that for frequency dependent processes the change of @xmath7 depends not directly on the local distribution @xmath40 , but only on the _ local frequency _ @xmath98 . we are more interested in the _ non - linear _ case , where the local frequency still plays a role , however the response to the opposite opinion can be different . the linear voter model can then be generalized into a _ majority _ rule where the transition toward the opposite opinion _ monotonously increases _ with the respective local frequency . different from this , the _ minority _ rule would assume that the transition toward the opposite opinion _ monotonously decreases _ with the respective local frequency , i.e. the minority opinion is favored . eventually , we could also have _ mixed rules _ with a _ non - monotonous _ frequency dependence , which become possible only if the local frequency depends at least on the opinion of three agents , as it is the case in our model . instead of an analytical expression , we specify the transition rate @xmath99 by using free parameters @xmath100 , @xmath101 , @xmath102 , @xmath103 which cover all of the above cases in a general manner . with these , the transition rates shall be defined as follows : @xmath104 the general case of four independent transition rates in eqn . ( [ trans2 ] ) can be reduced to two transition rates @xmath100 , @xmath101 by assuming a symmetry between the two states @xmath105 and 1 , i.e. @xmath106 , @xmath107 . the condition @xmath108 then denotes the above mentioned _ majority rule _ , because the transition rate _ increases _ with an increasing fraction of the _ opposite _ opinion @xmath109 in the neighborhood . this process is also known as _ positive freqency dependent invasion_. in ecology , it means that individuals of abundant species have a better chance to survive . opposite to that , for the so - called _ negative freqency dependent invasion _ process the probability that cell @xmath4 shall switch to the opposite opinion @xmath109 _ decreases _ with the fraction number of individuals of subpopulation @xmath110 in the neighborhood . this implies @xmath111 in an ecological context negative frequency dependent invasion means that individuals of a rare subpopulation have a better chance to survive . the rate @xmath100 in eqn . ( [ trans2 ] ) applies for the case where cell @xmath4 is only surrounded by opinions of the same kind . in a deterministic model , there would be no force to change the current state . in a stochastic ca however all possible processes should have a certain non - zero probability to occur , therefore a rather small value @xmath112 is used to avoid absorbing states in the dynamics . we will then only vary the remaining rate @xmath113 , which is the only free parameter in the model . table summarizes the different transition rates given the possible local configurations @xmath40 . it should be read as follows : given that the local configuration is e.g. @xmath114 at time @xmath12 , the probability to find @xmath115 is @xmath16 . if @xmath116 this transition rate increases with @xmath117 , hence it denotes a _ majority voting _ rule , otherwise it denotes a _ minority voting _ rule . .transition probabilities for @xmath7 given different local configurations . @xmath116 denotes _ majority voting _ rules , @xmath118 denotes _ minority voting _ rules . [ cols="^,^,^,^,^",options="header " , ] now that the transition rates for the nonlinear voter model are specified , we have different ways to proceed : ( 1 ) we can run stochastic computer simulations of the one - dimensional ca , to get some intuition about the dynamics and the role of the parameter @xmath16 . this will be described in the following sect . ( 2 ) we can also derive an approximate macroscopic dynamics for the global frequency @xmath119 , which will be done in sections [ 3.1 ] , [ 4.1 ] using two different approximations . ( 3 ) eventually , we can use the closed - form description already derived in sect . [ sec : dyn ] , to numerically calculate @xmath119 . the stochastic simulations will denote our _ reference case _ , used to compare the two different approximate macroscopic dynamics and the numerical calculations . for a first insight into the dynamics , we have conducted stochastic simulations of the ca described above . in this section , we will only refer to particular runs , to show some snapshots of the dynamics , while in sect . [ sec : num ] also averaged simulation results of the global variables are discussed . for our simulations , we have used a one - dimensional ca of @xmath120 cells with periodic boundary conditions , where all cells are simultaneously updated . the discrete time scale is defined by _ generations_. we have checked that the main results do not change if @xmath0 is increased to 6400 . the initial configuration of the ca refers to a homogeneous distribution ( within discrete limits ) of both opinions , i.e. initially each cell is randomly assigned one of the possible states , @xmath33 , with a probability that is equal to the initial global frequency @xmath121 . at each time step , the transition rates are calculated for each cell according to eqn . ( [ trans2 ] ) and the values are compared with a random number ` rnd ` drawn from the intervall @xmath122 . if ` rnd ` is less than the calculated transition rate , the respective transition process is carried out , otherwise the cell remains in its current state . since each transition only depends on the current local configuration , memory effects are not considered here . it follows from the above description that the case @xmath123 and @xmath124 refers to a _ deterministic _ positive majority voting rule , simply because the state of cell @xmath4 _ never _ changes unless the two nearest neighbor cells have adopted the same opinion . but then , it will _ always _ change , such that all three cells have the same opinion . similarly , a _ deterministic _ minority voting rule is described by @xmath125 , and @xmath126 . , @xmath127 , ( top right ) @xmath128 , @xmath129 , ( bottom left ) @xmath130 , @xmath131 , ( bottom right ) @xmath130 , @xmath132 . @xmath120 , initial condition : @xmath133 , ( black ) indicates @xmath134 , ( white ) indicates @xmath135 . , title="fig:",width=264 ] , @xmath127 , ( top right ) @xmath128 , @xmath129 , ( bottom left ) @xmath130 , @xmath131 , ( bottom right ) @xmath130 , @xmath132 . @xmath120 , initial condition : @xmath133 , ( black ) indicates @xmath134 , ( white ) indicates @xmath135 . , title="fig:",width=264 ] , @xmath127 , ( top right ) @xmath128 , @xmath129 , ( bottom left ) @xmath130 , @xmath131 , ( bottom right ) @xmath130 , @xmath132 . @xmath120 , initial condition : @xmath133 , ( black ) indicates @xmath134 , ( white ) indicates @xmath135 . , title="fig:",width=264 ] , @xmath127 , ( top right ) @xmath128 , @xmath129 , ( bottom left ) @xmath130 , @xmath131 , ( bottom right ) @xmath130 , @xmath132 . @xmath120 , initial condition : @xmath133 , ( black ) indicates @xmath134 , ( white ) indicates @xmath135 . , title="fig:",width=264 ] the results of the computer simulations are illustrated in figure [ fig : ca1dp ] . we first note the remarkable difference in the outcome between majority and minority voting . in the latter , cells change their state frequently , however no pattern evolves because the cell always adopts an opinion different from the two neighbors , if these have the same opinion . thus , we see a rather random and unstructured pattern , in which both opinions _ coexist _ with a long - term fraction @xmath136 . this overall behavior is also not changed in case of a stochastic dynamics . in contrast , for the majority voting rule we observe the formation of local clusters of the same opinion . in the deterministic case , the resulting pattern becomes stationary very fast , such that the clusters remain rather small . in the stochastic case , however , we observe the interesting phenomenon that very large clusters of cells with the same opinion emerge . these domains do not grow up to system size because of the small perturbation @xmath130 , that enables cells to adopt a different opinion even in a homogeneous neighborhood . the non - linearity ( @xmath131 is different from the linear case @xmath137 ) then amplifies such random deviations . this phenomenon has been discussed , for the case of a two - dimensional ca , in @xcite . as a consequence , we observe a _ correlated coexistence _ of both opinions characterized by the facts that ( i ) each opinion at times is the majority and ( ii ) forms larger clusters , and ( iii ) the dynamics is always in non - equilibrium . this is clearly visible in the long - term behavior , shown in figure [ fig : ca1dp ] ( bottom left ) . in sect . [ sec : num ] , we further investigate these different regimes numerically , after we have derived some appropriate approximations of the macroscopic dynamics . a formal description for the dynamics of the ca starts with the probability @xmath11 used in eqn . ( [ eq : chap0 ] ) . by means of @xmath138 we can obtain the key variable of the macroscopic dynamics @xmath139 , which is the expected global frequency of each opinion in the population . note that , different from eq . ( [ eq:1 ] ) where @xmath86 is used , @xmath139 describes the ensemble average over very many simulations . this is of some importance when interpreting the results . for the _ linear voter model _ , it is known that the system converges always to consensus , i.e. @xmath89 or @xmath88 . if we use the initial condition @xmath140 , then in 50% of the cases we observe @xmath89 , and in 50% @xmath88 . however , when averaging over all of these outcomes , we find @xmath141 . assuming a master equation for the dynamics of @xmath11 with the transition rates specified in eqn . ( [ trans2 ] ) , we can derive a rate equation for @xmath142 as discussed in detail in @xcite : @xmath143 \end{aligned}\ ] ] here , the summation is over all possible @xmath144 opinion patterns @xmath145 for the neighborhood of cell @xmath4 . these are binary strings @xmath146 that indicate the particular values of the nearest neighbors @xmath5 , @xmath6 . with @xmath147 neighbors , these would be @xmath148 , @xmath149 , @xmath150 , @xmath151 . together with the possible state for cell @xmath4 , i.e. @xmath134 or @xmath135 , the respective local frequencies @xmath91 can be derived according to eqn . ( [ sum ] ) , such that all transition rates in eq . ( [ eq : master_macrof ] ) are specified . the solution of eqn . ( [ eq : master_macrof ] ) would however require the computation of the averaged global frequencies @xmath152 and @xmath153 for all possible opinion patterns @xmath145 over time , which would be a tremendous effort . therefore , in @xcite two analytic approximations have been discussed to solve this problem . here , we only summarize the results . in the first approximation , the so - called mean - field limit , the state of each cell does not depend on the opinion of its neighbors , but is only influenced via a mean field . in this case the opinion distribution factorizes : @xmath154 and we find with @xmath155 for the macroscopic dynamics , eqn . ( [ eq : master_macrof ] ) in the mean - field limit : @xmath156 \end{aligned}\ ] ] for the calculation of the @xmath157 we have to look at each possible opinion pattern @xmath145 . the mean - field approach assumes that the occurence of each 1 or 0 in the pattern can be described by the global frequencies @xmath158 and @xmath159 , respectively ( for simplicity , the abbreviation @xmath160 will be used in the following ) . taking the example @xmath161 , i.e. @xmath162 for cell @xmath4 and @xmath163 , @xmath164 for its neighbors , would result in @xmath165 . inserting further the transition rates , eqn . ( [ trans2 ] ) , we find this way the equation for the mean - field dynamics as : @xmath166+(1 - 3\alpha)\ , x(1-x)\ , ( 2x-1 ) \end{aligned}\ ] ] the fixed points of the mean - field dynamics can be calculated from eqn . ( [ eq : mean_field ] ) by means of @xmath167 . in the limit @xmath125 , we find : @xmath168 the three stationary solutions denote either consensus toward one of the opinions , or coexistence of both opinions with an equal share . in order to verify the stability of the solutions , we have further investigated the jacobian @xmath169 of eqn . ( [ eq : mean_field ] ) . the results can be concluded as follows : below a critical reinforcement , @xmath170 , @xmath171 and @xmath172 are stable attractors and @xmath136 is the separatrix . above the critical reinforcement , @xmath173 , @xmath136 becomes the stable attractor , while @xmath171 and @xmath172 are unstable . we will come back on this after discussing the second approximation . the mean - field approximation assumes that the local occurrence of opinions is determined by the _ frequencies rather than by local interactions . a better approximation should take local correlations between neighboring cells into account . our second approximation , the so - called _ pair approximation _ , is based on the assumption that the state of each cell @xmath4 is only correlated to the states of each of its nearest neighbors , separately . i.e. , the two neighbors are only correlated through the focal cell and not to each other . therefore , the neighborhood @xmath174 is decomposed in _ pairs _ of nearest neigbor cells , @xmath175 , @xmath176 which are called _ doublets_. the expected value of the global frequency of doublets is denoted as @xmath177 where @xmath110 refers to the focal cell and @xmath178 to one of its neighbors . we can then approximate the global frequency of a specific opinion pattern @xmath179 in eqn . ( [ eq : master_macrof ] ) as : @xmath180 where the @xmath181 are the correlations : @xmath182 that depend on the doublet frequency @xmath177 and the global frequency of opinions @xmath183 neglecting higher order correlations . @xmath181 can be interpreted as the conditional probability that a randomly chosen nearest neighbor of a cell in state @xmath178 is in state @xmath110 . with the relations : @xmath184 and using again @xmath185 , these correlations can be expressed in terms of only @xmath186 and @xmath187 as follows : @xmath188 we then find for the macroscopic dynamics , eqn . ( [ eq : master_macrof ] ) , in pair approximation : @xmath189 \end{aligned}\ ] ] with respect to eqn . ( [ eq : cond_prob ] ) , eqn . ( [ macro - pair ] ) now depends on two variables , @xmath187 and @xmath186 . in order to derive a closed description , we need an additional equation for @xmath190 , that can be obtained from eqn . ( [ eq : c - prob ] ) : @xmath191 eqn . ( [ eq : local1 ] ) requires additionally the time derivative of the global doublet frequency @xmath192 . we note that the three coupled equations for @xmath187 , @xmath186 and @xmath193 can be easily solved numerically . in the appendix , we have derived explicit expressions for these equations for the one - dimensional ca discussed here , using the transition rates of eqn . ( [ trans2 ] ) . in this section , we aim at a comparison of the different approaches developed so far . we recall that our _ reference case _ is the stochastic simulation of the ca , where some sample runs were already presented in sect . [ 2.3 ] . to allow a comparison with the other approaches that predict the expected behavior , we will average the ca simulations over 50 independent runs . the initial condition is given by the global fraction of opinion 1 , @xmath121 , which is realized as a uniform random distribution of the opinions as described already in sect . thus , the initial value for the pair correlation is @xmath194 . to decide when the ca simulations have reached stationarity , we verify that changes of @xmath13 are less than @xmath195 . the dynamics of the reference case ( a ) shall be compared to the two other dynamic approximations , ( b ) quintuplet approximation ( see sect . [ sec : dyn ] ) and ( c ) pair approximation ( see sect . [ 4.1 ] and appendix ) . to facilitate the understanding , we summarize here the computational procedure . for the _ quintuplet approximation _ our starting point is eq . ( [ eq : chap2 ] ) which contains two terms that need a further approximation . the quintuplet probability @xmath60 can be decomposed in terms of triplet probabilities by means of eq . ( [ eq:5app - fin ] ) . the propagator @xmath196 $ ] can be decomposed in terms of propagators @xmath197 $ ] by means of eq . ( [ eq : factor ] ) . the latter are given as the transition rates of the vm in table [ tab : trans ] . let us illustrate the computation by one example . a quintuplet has five states and hence @xmath198 configurations denoted as @xmath56 in eq . ( [ eq : chap2 ] ) . each of these possible configurations will contribute to the triplet distribution according to eq . ( [ eq : chap2 ] ) . let us for the quintuplet take the sample configuration @xmath199 and for the triplet @xmath200 . the right - hand side of eq . ( [ eq : chap2 ] ) then reads : @xmath201\ p[(0,t+1)\gets ( 110,t)]\ p[(0,t+1)\gets ( 101,t ) ] \times \nonumber \\ \times \frac{p(011,t)\ p(1 1 0,t ) \ p(1 0 1,t)}{p(1 1,t)\ p(1 0,t)}\end{aligned}\ ] ] in order to determine @xmath202 according to eq . ( [ eq : chap2 ] ) , we have to sum up over the 32 different quintuplet configurations . and we have to follow this procedure 8 times , because there are @xmath203 possible triplet configurations . at the end , we arrive at a new triplet distribution , which serves as the starting point for the next generation , and so forth . regarding the computational effort , it is comparably low because of the straight forward calculation . because in the vm the transition rates do not depend on the lattice position @xmath4 but rather on the local frequencies @xmath204 , it would be even possible to reduce the higher order approximation to a framework of frequencies of motives @xmath200 , @xmath205 , etc . but to test the resulting improvement in computation speed is not the aim of this paper . for the _ pair approximation _ , we repeat again that we have to solve a system of three coupled equations for @xmath187 , @xmath186 and @xmath193.the appendix provides explicit expressions for the involved quantities , where the transition rates are obtained from table [ tab : trans ] . as a first test of the validity of our approximations , we focus on the correct prediction of the stationary solution , @xmath206 . we have two parameters to vary : ( i ) the initial condition @xmath121 which is changed from 0.1 to 0.9 in steps of 0.1 , ( ii ) the parameter @xmath16 that decides about majority and minority voting . for majority voting . ( left ) deterministic case , @xmath124 , @xmath125 , ( right ) stochastic case , @xmath131 , @xmath130 . curves from different approaches : ( a ) ca simulation ( reference case , green ) , ( b ) quintuplet approximation , ( c ) pair approximation , ( d ) mean - field approximation , title="fig:",width=264 ] for majority voting . ( left ) deterministic case , @xmath124 , @xmath125 , ( right ) stochastic case , @xmath131 , @xmath130 . curves from different approaches : ( a ) ca simulation ( reference case , green ) , ( b ) quintuplet approximation , ( c ) pair approximation , ( d ) mean - field approximation , title="fig:",width=264 ] for the majority voting , the results are shown in fig . [ fig : pos1dstat ] for the deterministic ( @xmath124 ) and the stochastic ( @xmath131 ) case . there are two remarkable observations . first , there is a noticeable difference between the deterministic and the stochastic outcome . in the latter , for the reference case , we always observe @xmath207 , regardless of the initial condition . this is understandable also from fig . [ fig : ca1dp ] ( lower left ) , where we observe the _ correlated coexistence _ of the two opinions . the small disturbance @xmath130 acts as a repulsion from consensus and averaging over the large opinion domains in the long - term limit leads to the balance between the two opinions . in the deterministic case , however , the dynamics freezes really fast , within the first 20 generations , which is also shown in fig . [ fig : ca1dp ] ( upper left ) . hence , not many deviations from @xmath121 can happen . but @xmath206 is slightly below ( for @xmath208 ) or above ( for @xmath209 ) this value , which indicates the trend towards consensus . the second remarkable observation is the difference between the mean - field approximation ( d ) and the other approximations ( b ) , ( c ) in predicting the stationary value . regardless of the details of the dynamics , the mean - field approximation always predicts _ consensus _ for the opinion that was initially the majority , i.e. @xmath210 for @xmath208 and vice versa . the quintuplet and the pair approximation , however , correctly predict the expected stationary solution , as the match with the reference case ( a , green curve ) indicates . hence , we can conclude that the mean - field approximation is not suitable to describe the dynamics of the one - dimensional ca for majority voting , which is not very surprising . but we are more interested to find out to what extend the pair and the quintuplet approximation , which _ both _ predict the expected _ stationary _ solution correctly for both the deterministic and the stochastic case , are also able to describe the _ dynamics toward the stationary state _ correctly . this is discussed in the following section . we note that the _ minority voting _ rule , not further discussed here , always leads to symmetric coexistence of both opinions , @xmath211 , for each initial condition . this was already indicated in fig . [ fig : ca1dp ] ( right column ) , . for minority voting , the mean - field approximation indeed predicts the correct stationary value , @xmath141 , but it miserably fails for majority voting . again , we are interested in the dynamics of the _ expected _ fraction of opinion 1 , @xmath13 . we concentrate only on the stochastic dynamics , since the deterministic dynamics freezes very fast . in fig . [ fig : pos1df ] , we compare the validity of our two different approximations , the pair and the quintuplet approximation , with the reference case , the stochastic ca simulations , for both the majority and the minority voting rule . for the case of _ majority voting _ , all the three curves for @xmath13 actually start at @xmath212 , but then decrease rapidly during the first few generations ( which can not be noticed on the long time scale ) , before they steadily evolve toward the quasistationary state @xmath141 . this kind of overshooting in the very initial phase can be also observed for other extreme initial conditions , e.g. @xmath213 . in this case @xmath13 will first increase before settling at @xmath141 . we see that the two approximations reach this state much faster than the simulations . while this gap is noticable in particular in the beginning , the quintuplet approximation fares better than the pair approximation . the evolution of the correlation term @xmath14 starts at @xmath214 and reaches values near one very quickly . this indicates the formation of large domains of the same opinion , i.e. domains with different opinions are well separated , but can coexist . however , we should notice that the stochastic simulations do not reach the value of 1 , because there is never a real _ consensus_. again , in the beginning , there is some discrepancy between the approximations and the simulations , but the quintuplet approximation is slightly more accurate . and the correlation @xmath14 . ( upper row ) majority voting , parameters : @xmath131 , @xmath130 , @xmath215 , @xmath216 . ( lower row ) minority voting , parameters : @xmath132 , @xmath130 , @xmath215 , @xmath217 . curves from different approaches : ( a ) ca simulation ( reference case , green ) ; ( b ) quintuplet approximation ; ( c ) pair approximation . , title="fig:",width=264 ] and the correlation @xmath14 . ( upper row ) majority voting , parameters : @xmath131 , @xmath130 , @xmath215 , @xmath216 . ( lower row ) minority voting , parameters : @xmath132 , @xmath130 , @xmath215 , @xmath217 . curves from different approaches : ( a ) ca simulation ( reference case , green ) ; ( b ) quintuplet approximation ; ( c ) pair approximation . , title="fig:",width=264 ] and the correlation @xmath14 . ( upper row ) majority voting , parameters : @xmath131 , @xmath130 , @xmath215 , @xmath216 . ( lower row ) minority voting , parameters : @xmath132 , @xmath130 , @xmath215 , @xmath217 . curves from different approaches : ( a ) ca simulation ( reference case , green ) ; ( b ) quintuplet approximation ; ( c ) pair approximation . , title="fig:",width=264 ] and the correlation @xmath14 . ( upper row ) majority voting , parameters : @xmath131 , @xmath130 , @xmath215 , @xmath216 . ( lower row ) minority voting , parameters : @xmath132 , @xmath130 , @xmath215 , @xmath217 . curves from different approaches : ( a ) ca simulation ( reference case , green ) ; ( b ) quintuplet approximation ; ( c ) pair approximation . , title="fig:",width=264 ] for the case of _ minority voting _ , we notice that the quintuplet approximation fares considerably better than the pair approximation . the latter fails to predict the dynamics for both @xmath13 and @xmath218 , whereas the quintuplet approximation is almost exact , except for a short time interval after the beginning , where it predicts a faster relaxation for @xmath13 . values of @xmath141 and @xmath219 clearly indicate a random pattern as also shown in the snapshots of fig . [ fig : ca1dp ] . with our investigations of the one - dimensional voter model , we want to achieve two goals : ( i ) a better understanding of the _ non - linearity _ in the voter dynamics that was mostly studied as a linear model , only , ( ii ) a probabilistic description , and possible approximations , of the dynamics for the fraction of opinion 1 , @xmath119 . the non - linearity can be easily expressed by means of a free parameter @xmath16 . the critical value @xmath220 distinguishes between two different rules , _ majority _ voting ( @xmath221 ) and _ minority _ voting ( @xmath222 ) , whereas @xmath223 refers to the border case of the _ linear _ voter model . it is known from well - mixed populations that majority voting should result in _ consensus _ , i.e. the asymptotic dominance of only one opinion , whereas minority voting should result in _ coexistence _ , i.e. the occurrence of both opinions in different fractions . our main focus was the role of local correlations in determining this outcome . for this we have used one - dimensional cellular automata ( ca ) in which each cell @xmath4 , characterized by its opinion @xmath7 , has a defined neighborhood of two cells with possibly different opinions , denoted as _ triplet_. the transition probability of a cell to change its opinion is then determined by the local _ frequency of opinions _ in its neighborhood ( including the focal cell ) and the _ non - linear response _ to this information , expressed by means of @xmath16 . the values of @xmath16 define a certain probability to switch to the opposite opinion , i.e. we can use them to switch between a deterministic ( @xmath224 ) or a stochastic ( @xmath225 ) dynamics . in the latter case , we have further assumed a very small probability @xmath226 to perturb a state of complete consensus , which allows a non - stationary dynamics of the ca as shown in figure [ fig : ca1dp ] ( bottom right ) . while the minority rule only results in _ random coexistence _ of the two different opinions , the majority rule generates more interesting results . in particular , we observe a _ correlated coexistence _ characterized by the formation of large domains of the same opinion which change continuously . i.e. , we have a non - equilibrium dynamics in which each opinion , for a certain time , can form large clusters of the majority opinion . the question then is how to describe this dynamics mathematically . in this paper , we follow a probabilistic approach , i.e. each cell has a certain probability of a given opinion @xmath227 , which also depends on the probabilities of the nearest neighbors , second - nearest neighbors , and so forth . in order to close the dynamics , we have proposed three different approximations at different levels of the description . the first level is the aggregated description in terms of the global fraction of opinion 1 , @xmath119 , for which we derive a dynamics for the _ expected value _ , @xmath13 . on this level , we discuss two approximations . the simplest one is the _ mean - field approximation _ , in which _ no correlations _ between neighboring states are considered . so , we call this the zero - order approximation . it gives us a prediction for @xmath13 derived from the well - mixed case . in contrast , the _ pair approximation _ considers a correlation between a cell and its neighbor , i.e. the triplet consisting of a cell and its two neighbors is decomposed in two cell - neighbor pairs . correlations between neighbors are not considered , so we call this the 1st - order approximation . hence , we have a prediction for @xmath13 coupled to the dynamics of the pair correlations @xmath14 . the third approximation does not refer to the aggregated level , but to the stochastic dynamics of a triplet , i.e. a cell with its two neighbors , that is determined by the larger neighborhood of a quintuplet , i.e. considers also the second nearest neighbors of the cell . therefore , we call this the 2nd - order approximation . using certain assumptions , we are able to provide a _ closed form dynamics _ for this larger neighborhood in terms of a probabilistic equation . to compare the validity of these mathematical approximations , we use as a _ reference case _ stochastic computer simulations of the one - dimensional ca , which are averaged over a larger number of runs . we have discussed the majority and the minority voting , as well as the deterministic and the stochastic dynamics , separately . in conclusion , we can summarize that the zero - order approximation only predicts the stationary outcome of the minority voting correctly , but fails for the majority voting rule , both with respect to the dynamics and the stationary outcome . the first - order approximation performs comparably well in comparison to the second - order approximation only for the majority voting . the stationary outcome is correctly predicted both for the deterministic and the stochastic case , also the dynamics is covered fairly good . however , the first - order approximation fails to predict the dynamics of the minority voting . this case is only well covered by the second - order approximation that gives not only a correct description of @xmath13 , but also of the pair correlations @xmath218 . commenting specifically on the correlations , we recall again that _ both _ the first and second - order approximations lead to comparably good asymptotic results only for the majority voting . but they clearly predict a faster formation of domains , i.e. a convergence to their stationary value , as compared to the ca simulations . this limits their usability to fully understand the emergence of long - range correlations . in the case of minority voting , our first - order approximation fails , while the second - order approximation for @xmath218 could be even seen as accurate in its computational prediction . this is not so surprising if we recall that minority voting rules , different from majority voting , do _ not _ result in long - range correlations . hence , we can conclude that the quintuplet approximation that covers also the second - order neighborhood , is accurate enough to describe the dynamics of the ca on the macroscopic level . while it is of course understandable , that an approximation that considers more information is usually more accurate , we should also relate this conclusion to the computational effort . here , it turns out that the 2nd - order approximation , although stochastic , i.e. needs to be averaged over a number of runs , performs very fast because of the closed form dynamics . of course , the 1st - order approximation , the closed form of which is given in the appendix , is computationally even simpler . but there is a trade - off with the accuracy of the prediction . still , as long only majority voting is considered , the 1st - order approximation should be preferred , both for simplicity , accuracy and computational effort . our last remark is about the _ coexistence _ of the two opinions , which is the more interesting scenario compared to _ consensus _ , i.e. the existence of only one opinion . here , we are not so much interested in the trivial case of _ random coexistence _ without any structure formation , which is characterized by @xmath141 and @xmath228 . we focus more on the case of _ correlated coexistence _ which has an interesting complexity because of the formation of local structures , nicely shown in fig . [ fig : ca1dp ] ( lower left ) . on the level of our approximations , this state is characterized by @xmath141 and @xmath229 . i.e. , both opinions form at times large domains , indicated by the high pair correlation , but none of the two opinions entirely dominates the dynamics , as the expected value for its fraction is about 0.5 . such insights can be generalized to other cases of frequency dependent processes which e.g. play a role in population ecology ( invasion or extinction of species ) . for most of these applications a two - dimensional ca is more appropriate as it was discussed in @xcite . the one - dimensional ca investigated here , on the other hand , allows a mathematical approximation of the stochastic dynamics in terms of the 2nd - order neighborhood , which gives a much higher predictive power . this formalism can be also used for other frequency dependent processes in one - dimensional ca . the authors acknowledge fruitful discussions with h. mhlenbein on an early version of this paper . 21 natexlab#1#1url # 1`#1`urlprefixselectlanguage # 1 agapie , a. ; hns , r. ; mhlenbein , h. ( 2004 ) . markov chain analysis for one - dimensional asynchronous cellular automata . _ methodology and computing in applied probability _ * 6(2 ) * , 181201 . banisch , s. ( 2014 ) . from microscopic heterogeneity to macroscopic complexity in the contrarian voter model . _ advances in complex systems _ * 17(05 ) * , 1450025 . banisch , s. ; arajo , t. ; lou , j. ( 2010 ) . opinion dynamics and communication networks . _ advances in complex systems _ * 13(01 ) * , 95111 . behera , l. ; schweitzer , f. ( 2003 ) . _ international journal of modern physics c _ * 14(10 ) * , 13311354 . brown , d. t. ( 1959 ) . a note on approximations to discrete probability distributions . _ information and control _ * 2(4 ) * , 386392 . castellano , c. ; fortunato , s. ; loreto , v. ( 2008 ) . statistical physics of social dynamics . _ review of modern physics _ . castellano , c. ; muoz , m. a. ; pastor - satorras , r. ( 2009 ) . nonlinear q - voter model . _ physical review e _ * 80(4 ) * , 041129 . gleeson , j. p. ( 2013 ) . binary - state dynamics on complex networks : pair approximation and beyond . _ physical review x _ * 3(2 ) * , 021004 . grnerup , o. ; jacobi , m. n. ( 2008 ) . a method for inferring hierarchical dynamics in stochastic processes . _ advances in complex systems _ * 11(01 ) * , 116 . keitt , t. h. ; lewis , m. a. ; holt , r. d. ( 2001 ) . allee effects , invasion pinning , and species borders . _ the american naturalist _ * 157(2 ) * , 203216 . krause , s. m. ; bttcher , p. ; bornholdt , s. ( 2012 ) . mean - field - like behavior of the generalized voter - model - class kinetic ising model . _ physical review e _ * 85(3 ) * , 031126 . mhlenbein , h. ; hns , r. ( 2002 ) . stochastic analysis of cellular automata with applications to the voter model . _ advances in complex systems _ * 5(2 ) * , 301337 . l . ; wang , j. ( 2015 ) . entropy and recurrence measures of a financial dynamic system by an interacting voter system . _ entropy _ * 17(5 ) * , 2590 . pfante , o. ; bertschinger , n. ; olbrich , e. ; ay , n. ; jost , j. ( 2014 ) . comparison between different methods of level identification . _ advances in complex systems _ * 17(02 ) * , 1450007 . przybya , p. ; sznajd - weron , k. ; tabiszewski , m. ( 2011 ) . exit probability in a one - dimensional nonlinear q - voter model . _ physical review e _ * 84(3 ) * , 031117 . schweitzer , f. ; behera , l. ( 2009 ) . nonlinear voter models : the transition from invasion to coexistence . _ the european physical journal b _ * 67(3 ) * , 301 318 . stark , h .- u . ; tessone , c. j. ; schweitzer , f. ( 2008 ) . slower is faster : fostering consensus formation by heterogeneous inertia . _ advances in complex systems _ * 11(04 ) * , 551563 . stauffer , d. ( 2002 ) . better be third than second in a search for a majority opinion . _ advances in complex systems _ * 5(1 ) * , 97100 . suchecki , k. ; eguluz , v. m. ; san miguel , m. ( 2005 ) . conservation laws for the voter model in complex networks . _ europhysics letters _ * 69 * , 228234 . suchecki , k. ; eguluz , v. m. ; san miguel , m. ( 2005 ) . voter model dynamics in complex networks : role of dimensionality , disorder , and degree distribution . _ physical review e _ * 72 * , 036132 . xiong , f. ; liu , y. ; zhu , j. ( 2013 ) . competition of dynamic self - confidence and inhomogeneous individual influence in voter models . _ entropy _ * 15(12 ) * , 52925304 . in @xcite , we describe a method to calculate @xmath187 , @xmath186 and @xmath192 for a two - dimensional non - linear vm . here , we apply this approach to the one - dimensional ca , i.e. we consider a cell and its two nearest neighbors in pair approximation , using the transition rates of eqn . ( [ trans2 ] ) . for simplicity , we use again the notation @xmath230 . with eqn . ( [ eq : cond_prob ] ) , we find for eqn . ( [ macro - pair ] ) @xmath231+\frac{(1 - 3\alpha)x(2x-1)(1-c_{1|1})^2}{1-x } \end{aligned}\ ] ] we note that @xmath232 if all neighboring states are uncorrelated , in this case eqn . ( [ eq : master1d ] ) reduces to the mean - field eqn . ( [ eq : mean_field ] ) . in order to calculate the time derivative of the doublet frequency @xmath233 we have to consider how it is affected by changes of @xmath110 in a specific opinion pattern @xmath234 , where the @xmath235 refer to the given states of the @xmath236 neighbors of a cell . in a frequency dependent process it is assumed that the transition does not depend on the exact distribution of the @xmath235 , but only on the frequency of a particular state @xmath110 in the neighborhood . let @xmath237 describe a neighborhood where the center cell in state @xmath110 is surrounded by @xmath15 cells of the same state @xmath110 . for any given @xmath238 , there are @xmath239 different opinion patterns with the same value of @xmath15 . the global frequency to find a neighborhood @xmath237 is denoted as @xmath240 with the expectation value @xmath241 . regarding the possible transitions , we are only interested in changes of the doublet ( 1,1 ) , i.e. transitions @xmath242 or @xmath243 . the transition rates shall be denoted as @xmath244 and @xmath245 respectively , which of course depend on the local neighborhood @xmath237 . the dynamics of the expected doublet frequency can then be described by the rate equation : @xmath246 \end{aligned}\ ] ] in order to specify the transition rates of the doublets @xmath247 , with @xmath248 and @xmath249 , we note that there are only 6 distinct configurations of the neighborhood . let us take the example @xmath250 . a transition of the center cell @xmath251 would lead to the extinction of 2 doublets @xmath252 . on the other hand , the transition rate of the center cell is @xmath253 as known from eqn . ( [ trans2 ] ) . this would result in @xmath254 . for a lattice of size @xmath0 the number of doublets is @xmath0 , so also the number of neighborhoods @xmath255 is @xmath0 . therefore , if we apply the transition rates of the single cells , eqn . ( [ trans2 ] ) to the transition of the doublets , their rates remain unchanged . similarly , if we take the example @xmath256 , a transition of the center cell @xmath257 would occur at the rate @xmath258 and would create 2 new doublets . we verify that @xmath259 . this way we can also determine the other possible transition rates : @xmath260 note that two of the transition rates are zero , because the respective doublets ( 1,1 ) or ( 0,1 ) do not exist in the assumed neighborhood . eventually , we express @xmath241 in eqn . ( [ d_frequency2 ] ) by @xmath261 and apply the pair approximation , eqn . ( [ eq : pairapprox ] ) , to @xmath179 , to obtain the dynamic equation for @xmath192 : @xmath262 finally , we obtain for the change of the correlation @xmath186 using equations , , and : @xmath263 \\ & + \frac { ( 1-c_{1|1})^2\bigl ( -2\alpha - c_{1|1}+3\alpha c_{1|1}-2(1 - 3\alpha)(1-c_{1|1})x\bigr)}{1-x } \end{aligned}\ ] ] thus equations and are the final closed form for the dynamics of the one - dimensional ca .

we prove optimal systolic inequalities on finsler mobius bands relating the systole and the height of the mobius band to its holmes - thompson volume . we also establish an optimal systolic inequality for finsler klein bottles of revolution , which we conjecture to hold true for arbitrary finsler metrics . extremal metric families both on the mobius band and the klein bottle are also presented .

1503.01282

optimal systolic inequalities were studied since the mid - twentieth century after c. loewner proved in an unpublished work the following result , _ cf . _ @xcite . every riemannian two - torus @xmath0 satisfies @xmath1 with equality if and only if @xmath0 is a flat hexagonal torus . recall that the _ systole _ of a nonsimply connected riemannian surface @xmath2 , denoted by @xmath3 , represents the length of the shortest noncontractible loop of @xmath2 . this inequality leads us to introduce the _ systolic area _ of @xmath2 defined as @xmath4 where @xmath5 runs over all the riemannian metrics on @xmath2 ( hence the subscript @xmath6 for riemannian ) . thus , @xmath7 . following this direction , p. pu @xcite showed that @xmath8 , where the infimum is attained exactly by the riemannian metrics with constant ( positive ) curvature on the projective plane @xmath9 . in the eighties , c. bavard @xcite proved that @xmath10 , where the infimum on the klein bottle @xmath11 is not attained by a smooth riemannian metric . see also @xcite , @xcite and @xcite for other proofs and variations on this inequality . these are the only nonsimply connected closed surfaces with a known systolic area . the existence of extremal metrics in higher dimension is wide open . the original proofs of the optimal riemannian systolic inequalities on @xmath0 , @xmath9 and @xmath11 rely on the conformal representation theorem ( a consequence of the uniformization theorem on riemann surfaces ) and proceed as follows . by the uniformization theorem , every riemannian metric @xmath5 on a closed surface is conformally equivalent to a riemannian metric @xmath12 of constant curvature . taking the average of @xmath5 over the isometry group of @xmath12 gives rise to a new metric @xmath13 with the same area as @xmath5 . by the cauchy - schwarz inequality , the systole of @xmath13 is at most the systole of @xmath5 . thus , the new metric @xmath13 has a lower ratio @xmath14 than the original metric @xmath5 . now , if the isometry group of @xmath12 is transitive , which is the case for @xmath0 and @xmath9 , the metric @xmath13 has constant curvature . hence the result for the projective plane . then , it is not difficult to find the extremal metric among flat torus . the case of the klein bottle requires an extra argument since the isometry group of @xmath12 is not transitive , _ cf . _ section [ sec : klein ] . + indeed , let @xmath15 be a riemannian surface . then by the uniformization theorem , @xmath5 is conformally equivalent to a riemannian metric @xmath12 of constant curvature on @xmath2 , _ i.e. _ , there exists a function @xmath16 such that @xmath17 . we deduce by the cauchy - schwarz inequality that @xmath18 where @xmath19 is the integration of @xmath20 over the isometry group of @xmath21 . now , if the latter is transitive ( this is the case of @xmath0 and @xmath9 ) , we conclude that @xmath19 is constant and we can easily then deduce the extremal riemannian metrics on @xmath2 . in this article , we consider finsler systolic inequalities . loosely speaking , a finsler metric @xmath22 is defined as a riemannian metric except that its restriction to a tangent plane is no longer a euclidean norm but a minkowski norm , _ cf . _ section [ preliminaries ] . from a dynamical point of view , the function @xmath23 can be considered as a lagrangian which induces a lagrangian flow on the tangent bundle @xmath24 of @xmath2 . thus , finsler manifolds can be considered as degree @xmath25 homogeneous lagrangian systems . the trajectories of the lagrangian correspond to the geodesics of the finsler metric . there exist several definitions of volume for finsler manifolds which coincide in the riemannian case . we will consider the holmes - thompson volume @xmath26 , _ cf . _ section [ preliminaries ] . as previously , we can define the systolic area @xmath27 , with the subscript @xmath22 for finsler , by taking the infimum in over all finsler metrics on @xmath2 . contrary to the riemannian case , there is no uniformization theorem for finsler surfaces . as a result , the classical riemannian tools to prove optimal systolic inequalities on surfaces , which are based on the conformal length method described above , do not carry over to the finsler case . new methods are thus required to deal with finsler metrics . the first optimal finsler systolic inequality has been obtained by s. ivanov @xcite who extended pu s systolic inequality to finsler projective planes . [ iv02 ] let @xmath9 be a finsler projective plane . then @xmath28 furthermore , equality holds if all the geodesics are closed of the same length . in particular , the systolic area of the projective plane is the same in the riemannian and finsler settings , that is , @xmath29 note that theorem [ iv02 ] provides an alternate proof of pu s inequality in the riemannian case which does not rely on the uniformization theorem . using a different method based on @xcite and @xcite , a finsler version of loewner s inequality has been obtained by the first author @xcite . [ sa10 ] let @xmath0 be a finsler two - torus . then @xmath30 equality holds if @xmath0 is homothetic to the quotient of @xmath31 , endowed with a parallelogram norm @xmath32 , by a lattice whose unit disk of @xmath32 is a fundamental domain . observe that @xmath33 contrary to the riemannian case . an optimal finsler systolic inequality holds for non - reversible finsler metrics on @xmath0 , _ cf . _ @xcite . note also that there is no systolic inequality for non - reversible finsler two - tori if one considers the busemann volume instead of the holmes - thompson volume , _ cf . _ @xcite . no systolic inequality holds for manifolds with boundary either . however , p. pu @xcite and c. blatter @xcite obtained optimal riemannian systolic inequalities in each conformal class of the mobius band and described the extremal metrics , _ cf . _ section [ sec : candidates ] . later , these inequalities were used by c. bavard @xcite and t. sakai @xcite in their proofs of the systolic inequality on the klein bottle in the riemannian case . the proof of the optimal conformal riemannian systolic inequalities on the mobius band relies on the uniformization theorem and the conformal length method ( as in the original proofs of the riemannian systolic inequalities on @xmath0 , @xmath9 and @xmath11 ) . among manifolds with boundary , one can not expect to get a positive systolic constant . we only know that in the riemannian case , the mobius band @xmath34 does have an extremal metric in each conformal class , _ cf . _ @xcite . pu @xcite and c. blatter @xcite obtained conformal systolic inequalities on @xmath34 , _ cf . _ section [ sec : candidates ] . these inequalities were used later by c. bavard @xcite and t. sakai @xcite in their proofs of the systolic inequality on the klein bottle in the riemannian case . as in the original proofs of the riemannian systolic inequalities on @xmath0 , @xmath9 and @xmath11 , the uniformization theorem was also essential in the proof of the conformal systolic inequalities on @xmath34 . in this article , we first prove a finsler generalization of the optimal systolic inequality on @xmath0 extending loewner s inequality , _ cf . _ @xcite , and derive further optimal geometric inequalities on finsler cylinders , _ cf . _ section [ sec : torus ] . these results allow us to establish an optimal inequality on every finsler mobius band @xmath34 relating its systole @xmath35 , its height @xmath36 and its ( holmes - thompson ) volume @xmath37 at least when @xmath34 is wide enough , _ cf . _ section [ sec : wide ] . here , the height @xmath36 represents the minimal length of arcs with endpoints on the boundary @xmath38 , which are not homotopic to an arc in @xmath38 , _ cf . _ definition [ heightdefinition ] . more precisely , we prove the following . [ finslermobius ] let @xmath34 be a finsler mobius band . let @xmath39 . then @xmath40 \\ \frac{1}{\pi}\frac{\lambda+1}{\lambda } & \text{otherwise}. \end{cases}\ ] ] moreover , the above inequalities are optimal for every value of @xmath41 . we describe extremal and almost extremal metric families in details in section [ sec : candidates ] , example [ ex : extremal - wide ] and example [ ex : extremal - narrow ] . + the optimal finsler systolic inequality on the klein bottle is still unknown . however , based on the inequality on finsler mobius bands , we obtain a partial result for finsler klein bottles with nontrivial symmetries . we refer to definition [ def : sym ] for a description of the symmetries considered in the statement of the following theorem . [ finslerkleinbottle ] let @xmath42 be a finsler klein bottle with a soul , soul - switching or rotational symmetry . then @xmath43 moreover , the inequality is optimal . we also present some extremal metric family in example [ ex : extremal2 ] . + finally , we present as a conjecture that the inequality should hold for every finsler klein bottle with or without symmetries . that is , @xmath44 should be equal to @xmath45 ( as @xmath46 and @xmath47 ) . in this section , we introduce general definitions regarding finsler manifolds . a ( reversible ) _ finsler metric _ @xmath48 on the tangent bundle @xmath24 of a smooth n - dimensional manifold @xmath2 is a continuous function satisfying the following conditions ( for simplicity , let @xmath49 ) : 1 . smoothness : @xmath22 is smooth outside the zero section ; [ 1 ] 2 . homogeneity : @xmath50 for every @xmath51 and @xmath52 ; [ 2 ] 3 . quadratic convexity : for every @xmath53 , the function @xmath54 has positive definite second derivatives on @xmath55 , _ i.e. _ , if @xmath56 , @xmath57 , @xmath58 , the symmetric bilinear form @xmath59 is an inner product . [ 3 ] the pair @xmath60 is called a _ finsler manifold_. if @xmath22 is only positive homogeneous instead of homogeneous , that is , only holds for @xmath61 , we say that the finsler metric is non - reversible . for simplicity , we will only consider reversible finsler metrics . conditions , and imply that @xmath22 is strictly positive outside the zero section and that for every @xmath53 and @xmath62 , we have @xmath63 with equality if and only if @xmath64 or @xmath65 for some @xmath66 , _ cf . _ ( * ? ? * theorem 1.2.2 ) . hence , @xmath22 induces a strictly convex norm @xmath67 on each tangent space @xmath68 with @xmath69 . more specifically , it gives rise to a minkowski norm @xmath70 on each tangent symmetric space @xmath71 . working with quadratically convex norms and not merely ( strictly ) convex norms provides nice dynamical features such as a geodesic flow and a legendre transform , _ cf . _ @xcite . as in the riemannian case , notions of length , distance , and geodesics extend to finsler geometry . let @xmath72\rightarrow m$ ] be a piecewise smooth curve . the length of @xmath73 is defined as @xmath74 by condition , @xmath75 does not depend on the parametrization of @xmath73 . moreover , the functional @xmath76 gives rise to a distance function @xmath77 defined as @xmath78 , where the infimum is taken over all piecewise smooth curves @xmath73 joining @xmath79 to @xmath80 . a geodesic is a curve which locally minimizes length . it is a critical point of the energy functional @xmath81 ( here the quadratic convexity condition is necessary ) . for @xmath53 , we denote by @xmath82 the unit ball of the minkowski norm @xmath70 , that is , @xmath83 . furthermore , the norm @xmath70 is euclidean if and only if @xmath82 is an ellipsoid . the dual of @xmath82 is defined as @xmath84 where @xmath85 is the dual norm of @xmath67 . note that @xmath86 identifies with the polar body @xmath87 of @xmath82 , where @xmath88 is a given scalar product on @xmath89 . in the riemannian case , there exists a unique notion of volume , up to normalization , which agrees both with the n - dimensional hausdorff measure determined by the riemannian metric and with the projection of the liouville measure from the unit tangent bundle , _ cf . _ however , in the finsler case , there is no notion of volume that satisfies both properties , _ cf . _ this leads to two distinct notions of finsler volume presented below . denote by @xmath90 the euclidean volume of the euclidean unit ball in @xmath91 . let @xmath92 represent a given volume form on @xmath2 and @xmath93 be the restriction of this volume form to each tangent space @xmath68 . similarly , let @xmath94 be the restriction of the volume form dual to @xmath92 to each cotangent space @xmath95 . busemann volume _ _ @xcite , is defined as @xmath96 the busemann volume is sometimes called the busemann - hausdorff volume as it agrees with the @xmath97-dimensional hausdorff measure of @xmath2 ( at least when the finsler metric @xmath22 is reversible ) . another volume frequently used in finsler geometry is the _ holmes - thompson volume _ _ @xcite , defined as @xmath98 it is equal to the liouville volume of its unit co - disc bundle divided by @xmath99 , _ cf . _ @xcite . note that the integrals in and do not depend on the chosen volume form . sine the volume is a local notion , it is possible to extend this definition even when @xmath2 is nonorientable , that is , when volume forms do not exist on @xmath2 . in @xcite , c. durn proved the following volume comparison inequality for finsler manifolds : @xmath100 with equality if and only if @xmath2 is riemannian . hence , every systolic inequality for the holmes - thompson volume remains true for the busemann volume . however , the inequality may fail for non - reversible finsler metrics . in this section we establish a finsler version of the minkowski second theorem for the two - torus . more precisely , l. keen proved the following . [ ke67 ] let @xmath0 be a riemannian two - torus . there exist two closed curves of lengths @xmath101 and @xmath102 generating the first integral homology group of @xmath0 such that @xmath103 equality holds if and only if @xmath0 is homothetic to the flat torus obtained as the quotient of @xmath31 by a hexagonal lattice . the proof of proposition [ ke67 ] relies on the uniformization theorem and the cauchy - schwarz inequality . a finsler version of proposition [ ke67 ] is given by the following result . [ ke67finsler ] let @xmath0 be a finsler two - torus . there exist two closed curves of lengths @xmath101 and @xmath102 generating the first integral homology group of @xmath0 such that @xmath104 equality holds if @xmath0 is homothetic to the quotient of @xmath31 , endowed with a parallelogram norm @xmath32 , by a lattice generated by two vectors of lengths @xmath101 and @xmath102 , parallel to the sides of the unit ball of @xmath32 . since there is no uniformization theorem for finsler metrics , the proof of this proposition differs from the proof of proposition [ ke67 ] . let @xmath105 be a systolic loop of @xmath0 and @xmath106 be the shortest closed curve of @xmath0 homologically independent with @xmath105 . denote by @xmath101 and @xmath102 the lengths of @xmath105 and @xmath106 . the loops @xmath105 and @xmath106 are simple and intersect each other at a single point . cutting @xmath0 open along @xmath105 and @xmath106 gives rise to a surface @xmath107 isometric to a fundamental domain of @xmath0 . let @xmath108 be a positive number greater than @xmath109 . denote by @xmath56 and @xmath110 the smallest integers such that @xmath111 and @xmath112 . then , glue @xmath113 copies of @xmath107 in such a way that the resulting shape is isometric to the fundamental domain of a finsler torus of volume @xmath113 times the volume of @xmath0 and of systole equal to @xmath114 . by construction , this new finsler torus is a degree @xmath113 cover of @xmath0 . then , by theorem [ sa10 ] , we have @xmath115 hence , @xmath116 by choosing @xmath108 large enough , the integers @xmath56 and @xmath110 can be made arbitrarily large , which leads to the desired inequality . now , if @xmath0 is the quotient of @xmath31 , endowed with a parallelogram norm , by a lattice generated by two vectors of lengths @xmath101 and @xmath102 which are parallel to the sides of the unit ball of the parallelogram norm , then @xmath117 . briefly speaking , the idea of the proof of proposition [ ke67finsler ] is to use finite covers to get a quasi - isosystolic two - torus ( _ i.e. _ , whose first homology group is generated by two loops of lengths nearly the systole ) and to apply the systolic inequality of theorem [ sa10 ] to this two - torus . this argument also applies in the riemannian case and gives an alternative proof of proposition [ ke67 ] without the use of the uniformization theorem . we can apply proposition [ ke67finsler ] to prove a systolic inequality on finsler cylinders . first , we give the following definition [ heightdefinition ] let @xmath2 be a compact finsler surface with boundary . the _ height _ @xmath118 of @xmath2 is the minimal length of arcs with endpoints on the boundary @xmath119 , which are not homotopic to an arc in @xmath119 . more formally , @xmath120\rightarrow m \mbox { with } \gamma(0 ) , \gamma(1 ) \in \partial m \text { and } [ \gamma]\neq 0\in \pi_1(m,\partial m)\}.\ ] ] a _ height arc _ of @xmath2 is a length - minimizing arc of @xmath73 of @xmath2 with endpoints in @xmath119 inducing a nontrivial class in @xmath121 . by definition , the length of a height arc of @xmath2 is equal to @xmath118 . [ finslercylindre ] let @xmath122 be a finsler cylinder . then , @xmath123 let @xmath124 be a positive even integer . we glue @xmath124 copies of @xmath122 by identifiying the identical boundary components pairwise . the resulting space is a torus @xmath0 . every loop of @xmath0 non freely homotopic to a multiple of a boundary component of @xmath122 is of length at least @xmath125 . therefore , for symmetry reasons , if @xmath124 satisfies @xmath126 , the systole of the torus @xmath0 is equal to the systole of the cylinder @xmath122 . applying proposition [ ke67finsler ] to this torus , we derive @xmath127 hence the result . we will make use of proposition [ finslercylindre ] in the proof of theorem [ finslermobius ] for wide finsler mobius bands , _ cf . _ section [ sec : wide ] . in this section , we first review the extremal riemannian metrics for systolic inequalities on the mobius band and the klein bottle presented in @xcite . by analogy with the riemannian metrics , we construct finsler metrics which are natural to consider when studying optimal finsler systolic inequalities . consider the standard sphere @xmath128 . denote by @xmath57 and @xmath129 the longitude and the latitude on @xmath128 . let @xmath130 . the @xmath101-tubular neighborhood of the equator @xmath131 is a spherical band @xmath132 which can be represented as @xmath133 the quotient of @xmath132 by the antipodal map is a riemannian mobius band with curvature @xmath134 denoted by @xmath135 . the conformal modulus space of the mobius band is parametrized by @xmath135 with @xmath130 . more precisely , every conformal class on the mobius band agrees with the conformal structure induced by some @xmath135 with @xmath136 . furthermore , the conformal classes of the @xmath135 s are pairwise distinct . the spherical mobius bands @xmath135 are involved in several extremal conformal systolic inequalities for riemannian metrics . more precisely , we define the _ orientable systole _ of a riemannian mobius band @xmath34 as the shortest length of a noncontractible orientable loop in @xmath34 . it will be denoted by @xmath137 . similarly , we define the _ nonorientable systole _ of @xmath34 and denote it by @xmath138 . observe that @xmath139 . moreover , we define @xmath140 as the minimal length of the arcs joining @xmath141 to @xmath142 in the @xmath143-coordinates of @xmath135 , which are homotopic with fixed endpoints to the projections of the meridians in @xmath132 . for instance , @xmath144 , @xmath145 and @xmath146 . note that the definition of @xmath147 relies on conformal data , namely the longitude - latitude coordinates to define the endpoints of the arcs involved in the length minimization . in @xcite , c. blatter obtained optimal lower bounds for the functionals @xmath148 in each conformal class of the mobius band . more precisely , for every riemannian metric conformally equivalent to @xmath135 , we have the sharp lower bound @xmath149 we also have the sharp inequality @xmath150 \\ \sigma_2({\mathbb{m}}_{\alpha(a)}\cup c_{a,\alpha(a ) } ) & \text{if } a \in [ b,\frac{\pi}{2 } ) \end{cases}\ ] ] where @xmath102 is the unique solution in @xmath151 of the equation @xmath152 and @xmath153 is the mobius band obtained by attaching a flat cylinder @xmath154 to the spherical mobius band @xmath155 along their boundary . here , the angle @xmath156 $ ] is implicitly given by a nonlinear equation depending on the conformal type @xmath101 and the flat cylinder @xmath157 is defined as the product @xmath158 $ ] , where @xmath159 is a boundary component of @xmath160 . alternately , @xmath157 is the mercator projection of a connected component of @xmath161 to the vertical cylinder generated by @xmath162 . + finally , we have the third sharp inequality @xmath163\\ \sigma_3({\mathbb{m}}_{\frac{\pi}{3}}\cup c_{a,\frac{\pi}{3 } } ) & \text{if } a\in[\frac{\pi}{3},\frac{\pi}{2 } ) \end{cases}\ ] ] with the help of , c. bavard @xcite established the optimal isosystolic inequality on the klein bottle . later , t. sakai @xcite used the inequalities and to give an alternative proof of bavard s isosystolic inequality for the klein bottle . the extremal riemannian metric on the klein bottle is obtained by gluing two copies of the spherical mobius band @xmath164 along their boundary . the closed geodesics in @xmath132 project down to systolic loops in @xmath135 which differ only by rotations and can be described as follows . let @xmath165 be the equator @xmath131 of @xmath166 parametrized by arclength . every great circle @xmath73 in @xmath132 different from @xmath165 intersects @xmath165 at a unique point @xmath167 with @xmath168 . the great circles @xmath165 and @xmath73 form at @xmath167 an oriented angle @xmath169 $ ] with @xmath170 . such great circle @xmath73 is denoted by @xmath171 . from the sinus formula in spherical trigonometry , the great circle @xmath171 exactly lies between the circles of latitude @xmath172 . thus , @xmath173 $ ] . for @xmath174 , let @xmath175 $ ] . by clairaut s relation , _ cf . _ @xcite , the positive angle @xmath176 between the circle of latitude @xmath129 and the great circle @xmath171 satisfies @xmath177 . in particular , the unit tangent vectors to the systolic loops of @xmath135 at a point of latitude @xmath129 generate a symmetric cone of half angle @xmath178 in the tangent plane , _ cf . _ figure [ extremalriemannian ] . these unit tangent vectors of @xmath135 are referred to as _ systolic directions_. the unit tangent vectors to the meridians are called _ meridian directions_. despite the risk of confusion , we will also call great circles of @xmath135 the projections of the great circles of @xmath132 to @xmath135 . in @xmath135.,width=113 ] the spherical mobius band @xmath135 , which is extremal for some conformal systolic inequalities among riemannian metrics , _ cf . _ , , and , is not extremal among finsler metrics . indeed , by slightly perturbing the quadratically convex norm in each tangent plane away from the systolic and meridian directions of the spherical metric , _ cf . _ figure [ extremalriemannian ] , we can decrease the area of the mobius band without changing the systole and the height . this shows that any unit tangent vector to an extremal ( quadratically convex ) finsler mobius band is tangent either to a systolic loop or a height arc , _ cf . _ definition [ heightdefinition ] . in other words , the unit tangent vectors induced by the systolic loops and the height arcs of an extremal ( quadratically convex ) finsler mobius band fill in its unit tangent bundle . with this observation in mind , it is natural to consider the following ( non - quadratically convex ) finsler metrics as potential extremal metrics . the idea is to adjust the shapes of the unit balls in the tangent bundle of the mobius band so that the systolic and meridian directions fill in the unit tangent bundle . more precisely , define a finsler metric @xmath179 on @xmath132 whose restriction to each tangent plane @xmath180 is a norm @xmath181 of unit ball @xmath182 given by the convex hull of the systolic directions of @xmath135 , _ cf . _ figure [ sabourauextremal ] . in longitude and latitude coordinates , the ball @xmath182 at @xmath183 can be represented as @xmath184 hence , the finsler metric @xmath179 can be represented in local coordinates as @xmath185 \\ \sqrt{du^2+dv^2 } & \text{if } \arctan\left(\frac{dv}{du}\right ) \in [ \theta(v),\frac{\pi}{2 } ] \end{cases}\ ] ] this metric passes to the quotient by the antipodal map to a finsler metric still denoted by @xmath179 . denote by @xmath186 the finsler mobius band so obtained . at a point of latitude @xmath129 in @xmath132.,width=113 ] since the definition of @xmath147 relies on conformal data which do not extend to the finsler case , it is more natural to consider the height @xmath36 , _ cf . _ definition [ heightdefinition ] , in the finsler case . some geometric features of the finsler mobius bands @xmath186 are summarized in the following two propositions . [ height ] let @xmath136 . then , @xmath187 , @xmath188 and @xmath189 . in particular , if @xmath190 then @xmath191 . let us start with a useful observation . denote by @xmath192 the interior of the domain of @xmath193 formed by the unit tangent vectors of the great circles of @xmath135 . the finsler metric @xmath179 coincides with the round riemannian metric of @xmath135 on @xmath192 . therefore , the subset @xmath192 is stable under the geodesic flow of @xmath179 ( which is well - defined on @xmath192 ) . furthermore , the length of a great circle with respect to @xmath179 is equal to @xmath194 . let us show that @xmath189 . consider a height arc @xmath73 of @xmath186 . the arc @xmath73 can be parametrized with respect to the latitude . otherwise , we could remove a subarc of @xmath73 joining two points at the same latitude and still make up an arc in the same relative homotopy class as @xmath73 with the remaining pieces using the rotational symmetry of @xmath186 . this would contradict the length - minimizing property of @xmath73 . hence , @xmath195 now , let us show that the systolic curves of @xmath186 agree with the great circles of @xmath135 in the nonorientable case and with the boundary of @xmath135 in the orientable case . consider an orientable or nonorientable noncontractible loop @xmath73 of minimal length in @xmath186 . if @xmath73 lies in the boundary of @xmath186 then the loop @xmath73 is orientable of length @xmath196 . thus , we can assume that @xmath73 passes through an interior point @xmath56 of @xmath186 . if a tangent vector of @xmath73 lies in @xmath192 then the geodesic arc @xmath73 coincides with a great circle of @xmath135 in the nonorientable case and with a great circle run twice in the orientable case . ( recall that @xmath192 is stable by the geodesic flow of @xmath179 . ) in the former case , the curve @xmath73 is of length @xmath194 , while in the latter , it is of length @xmath197 . thus , we can assume that the tangent vectors of @xmath73 do not lie in @xmath192 . consider the closed lift @xmath198 of @xmath73 in @xmath132 . let @xmath199 be the two extreme great circles of @xmath132 passing through the lifts of @xmath56 and tangent to the boundary of @xmath132 . that is , @xmath199 are the great circles of @xmath132 making an angle of @xmath200 with the curves of constant latitude @xmath201 in @xmath132 passing through the lifts of @xmath56 . since the tangent vectors of @xmath73 do not lie in @xmath192 , the curve @xmath198 does not intersect @xmath199 in the interior of @xmath132 , except at the lifts of @xmath56 . therefore , there exists a subarc of @xmath198 ( actually two subarcs of @xmath198 ) joining the two boundary components of @xmath132 in the region delimited by the great circles @xmath199 and the boundary of @xmath132 , see the gray region of figure [ propositionheight ] . thus , @xmath202 with equality if @xmath73 agrees with @xmath199 . we conclude that @xmath187 and @xmath203 . hence , @xmath204 \\ { { \rm sys}}_{+}({\mathbb{m}}_{f_a } ) & \mbox{if } a \in [ \frac{\pi}{3},\frac{\pi}{2 } ) \end{cases}\ ] ] we claim that the systolic curves are only the great circles . observe first that the set of unit vectors in @xmath193 that are pointing in the directions of great circles is stable under the geodesic flow , _ cf . _ @xcite , chapter 1.f . moreover , it is easy to see that the length of a great circle is equal to @xmath194 . now , let @xmath73 be a geodesic loop in @xmath186 that lifts to an arc joinning the point @xmath205 to the point @xmath206 in @xmath132 , where @xmath207 $ ] , pointing in a meridian direction . let @xmath208 be a geodesic loop in @xmath186 that lifts to the extreme great circle @xmath209 in @xmath132 ( that is , the angle between the meridian @xmath210 and @xmath209 is equal to @xmath211 , @xmath212 $ ] ) . by a calibration argument we have @xmath213 , with equality if and only if @xmath73 coincides with @xmath209 . in fact , assume that the geodesics loops @xmath73 and @xmath208 are parametrized by the latitude @xmath129 , then @xmath214 hence , @xmath187 . moreover , it is not hard to show that @xmath203 . hence , @xmath2150,\frac{\pi}{3 } ] \\ { { \rm sys}}_{+}({\mathbb{m}}_{f_a } ) & \text{if } a \in [ \frac{\pi}{3},\frac{\pi}{2}[. \end{cases}\ ] ] finally , let @xmath73 be a unit speed ( non closed ) geodesic arc pointing in a meridian direction , and of length @xmath216 . then @xmath73 joins two points of @xmath217 and induces a nontrivial class in @xmath218 s. we claim that its length is in fact equal to @xmath194 . indeed , we have @xmath219 passing through a lift @xmath220 of @xmath56 in @xmath132.,width=188 ] the finsler mobius bands @xmath186 are not pairwise isometric since they have distinct orientable systoles . [ volume ] let @xmath136 . then , @xmath221 the unit ball @xmath222 coincides with the polar body of @xmath82 described in figure [ dualsabourauextremal ] . the area of @xmath222 is equal to @xmath223 . by definition of the holmes - thompson volume , _ cf . _ , we have @xmath224 or polar body of @xmath82.,width=113 ] as a consequence of propositions [ height ] and [ volume ] , we observe the following couple of points . 1 . the orientable and nonorientable systoles of @xmath179 agree with those of its dual @xmath225 . hence , @xmath226 . moreover , computations similar to those in propositions [ height ] and [ volume ] show that @xmath227 and @xmath228 . this means that for both @xmath179 and its dual @xmath225 , we have @xmath229 , when @xmath230 . 2 . the finsler mobius bands @xmath231 with @xmath232 $ ] attain the equality case in when @xmath233 . in this section , we give a proof of theorem [ finslermobius ] for wide finsler mobius bands , that is , when @xmath234 . more precisely , we prove the following result . [ prop : wide ] let @xmath34 be a finsler mobius band with @xmath235 . then @xmath236 we present examples showing this result is optimal at the end of this section , _ cf . _ example [ ex : extremal - wide ] . consider @xmath237 slightly perturbing the distance function @xmath238 if necessary , we can assume that this distance function is a morse function on @xmath34 for which @xmath239 is a regular value . in this case , @xmath240 is a surface with boundary . if @xmath34 has some big bumps " , the surface @xmath240 may possibly have some holes . more precisely , the surface @xmath240 may not be a topological cylinder as some of its boundary components may bound topological disks in @xmath34 . let @xmath241 be the union of @xmath240 with the topological disks of @xmath34 bounded by the boundary components of @xmath240 . under this construction , @xmath242 is a cylinder one of whose boundary components agrees with @xmath243 . clearly , the height of @xmath242 is equal to @xmath239 . furthermore , since the inclusion @xmath244 induces a @xmath245-isomorphism , we have @xmath246 . applying proposition [ finslercylindre ] to the cylinder @xmath242 yields @xmath247 now , consider the finsler mobius band @xmath248 . [ case2h ] the height and systole of @xmath249 satisfy @xmath250 let @xmath251 be a height arc of @xmath249 , _ cf . _ definition [ heightdefinition ] . by construction , @xmath252 and the points of @xmath253 are at distance @xmath239 from @xmath243 . therefore , the two endpoints of @xmath251 can be connected to @xmath243 by two arcs @xmath254 and @xmath255 of @xmath242 , each of length @xmath239 . moreover , the arc @xmath256 with endpoints in @xmath243 induces a nontrivial class in @xmath257 . therefore , since @xmath258 , we obtain @xmath259 now , let @xmath73 be a height arc of @xmath34 . by definition , we have @xmath260 . the part @xmath261 of @xmath73 in @xmath242 is made of two arcs , each of length at least @xmath239 . moreover , the arc @xmath262 with endpoints in @xmath253 induces a nontrivial class in @xmath263 . hence , @xmath264 since the inclusion @xmath265 induces a @xmath245-isomorphism , we obtain @xmath266 consider the projective plane @xmath9 defined as the quotient @xmath267 , where the boundary @xmath253 is collapsed to a point . strictly speaking , the finsler metric on @xmath9 has a singularity at the point to which @xmath253 collapses , but we can smooth it out . the following result allows us to derive the systole of @xmath9 . [ metricmobius ] let @xmath9 be the projective plane defined as the quotient @xmath268 of a finsler mobius band @xmath34 . then , @xmath269 where @xmath9 is endowed with the quotient metric . let @xmath73 be a noncontractible loop in @xmath9 . the curve @xmath73 lifts either to a noncontractible loop in @xmath34 or to a noncontractible arc in @xmath34 joining two points of the boundary @xmath243 . in the former case , the length of @xmath73 is at least @xmath35 , while in the latter , it is at least @xmath36 . on the other hand , we can easy construct noncontractible loops in @xmath9 of length @xmath35 or @xmath36 . from lemma [ metricmobius ] and lemma [ case2h ] , the systole of @xmath9 is equal to @xmath35 . applying theorem [ iv02 ] to @xmath9 , we obtain @xmath270 this inequality combined with yields @xmath271 hence the result . we conclude this section by describing extremal and almost extremal finsler metrics when @xmath234 . [ ex : extremal - wide ] let @xmath272 . * the horizontal translation @xmath273 of vector @xmath274 is an isometry of the plane @xmath31 endowed with the sup - norm . the quotient of the strip @xmath275 $ ] by the isometry group @xmath276 generated by @xmath273 is a cylinder @xmath122 . the finsler mobius band @xmath34 obtained by gluing a boundary component of @xmath122 to @xmath231 along @xmath277 with @xmath190 , _ cf . _ section [ sec : candidates ] , satisfies @xmath278 , @xmath279 and @xmath280 . see figure [ fig : extmobius].(a ) . * endow the plane @xmath31 with the sup - norm . the quotient of the strip @xmath281 $ ] by the group generated by the map @xmath282 is a finsler mobius band @xmath283 with @xmath284 , @xmath285 and @xmath286 . let @xmath122 be the finsler cylinder defined in ( e.1 ) . attach @xmath122 to @xmath287 via a cylindrical part of arbitrarily small area , _ cf . _ figure [ fig : extmobius].(b ) , so that the resulting space is a finsler mobius band @xmath34 with , @xmath288 and @xmath289 , where @xmath290 are arbitrarily close to @xmath134 . this finsler mobius band is almost extremal for the inequality when @xmath234 . 0.4 .,title="fig : " ] 0.4 .,title="fig : " ] in this section , we give a proof of theorem [ finslermobius ] for narrow finsler mobius bands , that is , when @xmath291 . more precisely , we prove the following result . [ prop : narrow ] let @xmath34 be a finsler mobius band with @xmath292 . then @xmath293 this inequality is optimal . extremal finsler metrics can be constructed as follows . [ ex : extremal - narrow ] let @xmath294 . endow the plane @xmath31 with the sup - norm . the quotient of the strip @xmath295 $ ] by the group generated by the map @xmath282 is a finsler mobius band @xmath34 with @xmath296 , @xmath279 and @xmath280 . before proceeding to the proof of this proposition , we need to introduce a few definitions and notions . [ def : k ] the orientable double cover of a finsler mobius band @xmath34 is a cylinder denoted by @xmath122 . the points of @xmath122 which are at the same distance from each boundary component of @xmath122 form a simple closed curve @xmath297 invariant under deck transformations . the _ soul _ of @xmath34 , denoted by @xmath298 , is defined as the projection of @xmath297 to @xmath34 . note that @xmath298 is a nonorientable simple closed curve of @xmath34 whose homotopy class generates @xmath299 . let @xmath42 be the finsler klein bottle obtained by attaching to @xmath34 another copy @xmath300 of @xmath34 along their boundary . ( the finsler metric on @xmath42 may have a singularity line along @xmath38 . ) the isometry of @xmath42 switching the souls of @xmath34 and @xmath300 , and leaving @xmath38 pointwise fixed is called the _ soul - switching symmetry _ of @xmath42 . let @xmath301 . the _ systole - height inequality _ on the mobius band is said to be satisfied for @xmath302 if for every finsler mobius band @xmath34 with @xmath303 and @xmath304 , we have @xmath305 by scale invariance , if the systole - height inequality is satisfied for @xmath302 , then it is also satisfied for @xmath306 with @xmath307 . we first prove the following preliminary result . [ lem:1/2 ] let @xmath308 $ ] . suppose that the systole - height inequality on the mobius band is satisfied for @xmath302 with @xmath309 . then , it is also satisfied for @xmath302 with @xmath310 . let @xmath34 be a finsler mobius band with @xmath303 and @xmath304 , where @xmath311 . consider the klein bottle @xmath42 made of two copies of @xmath34 defined in definition [ def : k ] and cut it open along the soul @xmath312 of @xmath300 . the resulting surface is a finsler mobius band denoted by @xmath313 whose boundary component double covers @xmath312 in @xmath42 . let @xmath105 be a noncontractible loop of @xmath313 . decompose @xmath105 into two parts @xmath314 and @xmath315 with @xmath316 . the parts @xmath101 and @xmath317 form two collections of arcs with endpoints lying in @xmath318 . by construction , the image @xmath319 of @xmath317 by the soul - switching symmetry lies in @xmath34 . furthermore , the union @xmath320 forms a closed curve lying in @xmath34 and homotopic to @xmath105 in @xmath313 ( and so noncontractible in @xmath34 ) . since @xmath321 has the same length as @xmath105 , we conclude that @xmath322 . actually , since the inclusion @xmath323 is a strong deformation retract , we derive the relation @xmath324 . but we will not make use of this equality in the sequel . by construction , the distance between the soul @xmath298 and @xmath38 ( and between @xmath312 and @xmath325 ) is at least @xmath326 . this implies that @xmath327 . actually , we can show that @xmath328 ( but we will not make use of this relation afterwards ) . indeed , let @xmath105 be a height arc of @xmath34 . by definition , @xmath329 . denote by @xmath330 its image in @xmath300 by the soul - switching symmetry of @xmath42 . the trace of the union @xmath331 to @xmath313 defines an arc with endpoints in @xmath332 inducing a nontrivial class in @xmath333 . the length of this arc is twice the length of @xmath105 . therefore , @xmath334 . hence , the equality @xmath328 . in conclusion , the mobius band @xmath313 satisfies @xmath335 and @xmath336 . since @xmath337 , the systole - height inequality is satisfied for @xmath338 by the lemma assumption . therefore , @xmath339 and the result follows . we establish a second preliminary result . [ lem:(x+y)/2 ] let @xmath340 such that @xmath341 . suppose that the systole - height inequality on the mobius band is satisfied for @xmath302 with @xmath342 or @xmath343 . then , it is also satisfied for @xmath302 with @xmath344 . let @xmath34 be a finsler mobius band with @xmath303 and @xmath304 , where @xmath344 . consider the klein bottle @xmath42 made of two isometric mobius bands @xmath34 and @xmath300 with souls @xmath298 and @xmath312 defined in definition [ def : k ] . consider also @xmath345 and @xmath346 note that if we drop the multiplicative constants @xmath347 and @xmath343 in the definitions of @xmath348 and @xmath349 , we obtain @xmath34 and @xmath300 . the subset @xmath348 is a finsler mobius band contained in @xmath34 . similarly , the subset @xmath349 is a finsler mobius band containing @xmath300 . observe also that the mobius bands @xmath348 and @xmath349 cover @xmath42 and that their interiors are disjoint . every point @xmath350 satisfies the equality @xmath351 by symmetry , the distance between @xmath352 and @xmath353 is equal to @xmath36 . it follows by the triangle inequality that @xmath354 as a result of the relations and , we obtain @xmath355 now , let @xmath105 be an arc of @xmath356 with endpoints @xmath357 inducing a nontrivial class in @xmath358 . as @xmath105 intersects @xmath359 , we deduce from that @xmath360 therefore , @xmath361 in another direction , we can also bound from below the systole of @xmath348 and @xmath349 as follows . for the systole of @xmath348 , since the inclusion @xmath362 induces a @xmath245-isomorphism , we derive @xmath363 note that the first inequality may be strict as the inclusion @xmath362 is strict . for the systole of @xmath349 , we argue as in lemma [ lem:1/2 ] . let @xmath105 be a noncontractible loop of @xmath349 . decompose @xmath105 into two parts @xmath314 and @xmath315 with @xmath364 . the union @xmath365 , where @xmath366 is the image of @xmath101 by the soul - switching symmetry of @xmath42 , forms a closed curve of length @xmath367 lying in @xmath300 and homotopic to @xmath105 in @xmath349 . hence , @xmath368 the systole - height inequality on the mobius band is satisfied for @xmath369 from the lemma assumption since @xmath370 . from the bounds , and , this inequality applies to @xmath356 and yields @xmath371 finally , recall that the mobius bands @xmath348 and @xmath349 cover @xmath42 and that their interiors are disjoint . by adding up for @xmath372 , we conclude that @xmath373 hence the result . at first glance , it seems more natural to assume that @xmath374 and @xmath375 in the definition of the systole - height inequality , _ cf . _ definition [ def : k ] . observe that the proof of lemma [ lem:1/2 ] carries over with this alternative notion . however , we have not been able to directly prove a result similar to lemma [ lem:(x+y)/2 ] with this more restrictive notion . the reason is that the inequality , namely @xmath376 , may be strict as the inclusion @xmath362 is strict . to get around this subtle difficulty , we relaxed the original definition and formulated the systole - height inequality in terms of lower bounds for the systole and the height of the mobius band . we can now proceed to the proof of proposition [ prop : narrow ] . by lemma [ lem:1/2 ] , for every nonnegative integer @xmath124 , the systole - height inequality on the mobius band is satisfied for @xmath302 with @xmath377 . combined with lemma [ lem:(x+y)/2 ] , this implies that the systole - height inequality is satisfied for every @xmath302 where @xmath378 is a dyadic rational of @xmath379 . since the height , the systole and the volume are continuous over finsler metrics , the result follows from the density of the dyadic rationals in @xmath380 $ ] . ( the inequality @xmath376 may be strict as the inclusion @xmath381 is strict . this is the reason why we formulated the @xmath302-systole - height inequality in terms of lower bounds for the systole and the height . ) similarly , the subset @xmath349 is a finsler mobius band containing in @xmath34 with the same soul @xmath298 as @xmath34 . furthermore , @xmath382 ( in this case , we can show that @xmath383 as in lemma [ lem:1/2 ] ) and @xmath384 . we can now proceed to the proof of proposition [ prop : narrow ] . in particular , for every nonnegative integer @xmath124 , the systole - height inequality on the mobius band is satisfied for @xmath302 with @xmath377 . in this section , we show that the systolic area of finsler klein bottles with soul , soul - switching or rotational symmetries is equal to @xmath45 . [ def : sym ] recall that every riemannian klein bottle is conformally equivalent to the quotient of @xmath31 by the isometry group @xmath385 generated by the glide reflection @xmath282 and the vertical translation @xmath386 with @xmath387 . the flat klein bottle @xmath388 decomposes into two mobius bands whose souls correspond to the projections of the lines @xmath389 and @xmath390 . the boundary of the two mobius bands agrees with the projection of the line @xmath391 ( or @xmath392 ) . a finsler metric on @xmath42 has a _ soul symmetry _ if its lift to @xmath31 is invariant by the map @xmath393 . similarly , a finsler metric on @xmath42 has a _ soul - switching symmetry _ if its lift to @xmath31 is invariant by the map @xmath394 . finally , a finsler metric on @xmath42 has a _ rotational symmetry _ if its lift to @xmath31 is invariant by the map @xmath395 for every @xmath396 $ ] . these definitions are consistent with the notions introduced in [ sec : narrow ] . in 1986 , c. bavard established an optimal isosystolic inequality for riemannian klein bottles , _ cf . _ @xcite . alternative proofs can be found in @xcite . all the proofs are based on the uniformization theorem . in fact , as mentioned in the introduction , the problem boils down to consider riemannian klein bottles invariant under soul ( and rotational ) symmetry . these klein bottles are made of two isometric copies of riemannian mobius bands . thus , in the end , the systolic inequality on riemannian klein bottles follows from optimal systolic inequalities on riemannian mobius bands , _ cf . _ @xcite . + there is no known optimal isosystolic inequality on the klein bottle for finsler metrics . however , we obtain the following partial result similar to the riemannian case . note that in the riemannian case , the hypothesis is automatically satisfied by an average argument . [ finslerkleinbottle2 ] let @xmath42 be a finsler klein bottle with a soul , soul - switching or rotational symmetry . then @xmath397 moreover , the inequality is optimal . let @xmath388 be a finsler klein bottle ( with or without symmetry ) . denote by @xmath34 the finsler mobius band obtained by cutting @xmath42 open along the soul given by the projection of the line @xmath398 to @xmath42 . the proof of the inequality in theorem [ finslerkleinbottle2 ] follows by combining the next three lemmas . [ kcase1 ] if @xmath399 then the inequality holds true . the inclusion @xmath400 induces a @xmath245-isomorphism . hence , @xmath401 . now , we only have two cases to consider . first , if @xmath402 , then we deduce from theorem [ finslermobius ] , first case , that @xmath403 second , if @xmath404 , then we deduce from theorem [ finslermobius ] , second case , that @xmath405 since both @xmath406 and @xmath407 are greater or equal to @xmath134 . the next three lemmas show that the assumption of lemma [ kcase1 ] is satisfied when the finsler metric on @xmath42 has soul , soul - switching or rotational symmetries . [ kcase2 ] if @xmath42 is a finsler klein bottle with a soul symmetry then @xmath399 . observe that the soul symmetry of @xmath42 leaves both @xmath34 and @xmath243 invariant . given an arc @xmath105 of @xmath34 , we denote by @xmath408 the arc of @xmath34 symmetric to @xmath105 by the soul symmetry . let @xmath73 be a height arc of @xmath34 , _ cf . _ definition [ heightdefinition ] . this arc decomposes into two subarcs @xmath105 and @xmath106 connecting @xmath243 to the soul of @xmath34 with @xmath409 . the arc @xmath410 with endpoints in @xmath243 induces a nontrivial class in @xmath411 of length at most the length of @xmath412 . by definition of @xmath36 , we conclude that @xmath410 is as long as @xmath73 and so is length - minimizing in its relative homotopy class . in particular , it is geodesic . thus , the arc @xmath73 , which has the subarc @xmath105 in common with @xmath413 , agrees with @xmath410 . in particular , it is invariant by the soul symmetry . as a result , the arc @xmath73 induces a noncontractible loop on @xmath42 after identification of the points of @xmath243 under the soul symmetry . hence , @xmath414 . [ kcase4 ] if @xmath42 is a finsler klein bottle with a soul - switching symmetry then @xmath399 . since the finsler klein bottle @xmath42 has a soul - switching symmetry , we can assume that it is composed of two isometric finsler mobius bands @xmath348 and @xmath349 . by symmetry , we have @xmath415 . given an arc @xmath105 of @xmath348 , we denote by @xmath408 the arc of @xmath349 symmetric to @xmath105 by the soul - switching symmetry . let @xmath73 be a height arc of @xmath348 . this arc decomposes into two subarcs @xmath416 and @xmath417 connecting @xmath418 to the soul of @xmath348 . the arc @xmath419 , _ i.e. _ , the union of the arc @xmath73 and its symmetric image @xmath420 , has endpoints in @xmath243 and induces a nontrivial class in @xmath411 of length equal to @xmath421 . moreover , this arc @xmath422 induces a noncontractible loop on @xmath42 after identification of the points of @xmath243 under the soul - switching symmetry . we conclude that @xmath399 . [ kcase3 ] if @xmath42 is a finsler klein bottle with rotational symmetry then @xmath399 . observe that the rotational symmetries of @xmath42 leave both @xmath34 and @xmath243 invariant . given an arc @xmath105 of @xmath34 , we denote by @xmath423 the arc of @xmath34 symmetric to @xmath105 by the rotational symmetry of angle @xmath424 . let @xmath73 be a length - minimizing arc of @xmath34 parametrized ( proportionally to its length ) by @xmath380 $ ] with endpoints in @xmath38 inducing a nontrivial class in @xmath411 . note that @xmath73 is a geodesic arc of length @xmath36 . by the first variation formula for finsler metrics , _ cf . @xcite , the geodesic arc @xmath73 is perpendicular to @xmath243 . it follows that the endpoints @xmath425 and @xmath426 of @xmath73 in @xmath38 are distinct . since the finsler metric is invariant under rotational symmetry , there exists @xmath427 such that @xmath428 . both symmetric arcs @xmath73 and @xmath429 are perpendicular to @xmath243 . in particular , their tangent vectors at @xmath428 coincide up to sign . therefore , the geodesic arcs @xmath73 and @xmath430 agree up to reparametrization . more precisely , @xmath431 for every @xmath432 $ ] . thus , @xmath433 . hence , @xmath434 , that is , @xmath435 . therefore , the arc @xmath73 projects to a closed curve in @xmath42 . it follows that the length of @xmath73 is at least @xmath436 . [ ex : extremal2 ] the following two examples show that the inequality is optimal . 1 . the quotient of @xmath31 , endowed with the sup - norm , by the isometry group @xmath385 generated by the glide reflection with parameter @xmath437 , _ cf . _ definition [ def : sym ] , is a finsler klein bottle with soul , soul - switching and rotational symmetries , of area @xmath197 and systole @xmath194 . 2 . the finsler mobius band @xmath438 , _ cf . _ section [ sec : candidates ] , whose opposite boundary points are pairwise identified defines a finsler klein bottle with soul , soul - switching and rotational symmetries , of area @xmath197 and systole @xmath194 . we believe that theorem [ finslerkleinbottle2 ] holds true for every finsler klein bottle ( not necessarily invariant under soul , soul - switching or rotational symmetries ) . more precisely , we state the following conjecture . let @xmath42 be a finsler klein bottle . then @xmath439 moreover , the inequality is optimal . if the conjecture is true , the finsler systolic areas of @xmath9 , @xmath0 and @xmath42 would be the same . in this section , we present a non - optimal systolic inequality on finsler klein bottles . every symmetric convex body @xmath441 admits a unique ellipsoid @xmath442 of maximal volume among the ellipsoids contained in @xmath122 . this ellipsoid , called john s ellipsoid , continuously varies with @xmath122 for the hausdorff topology . furthermore , it satisfies the double inclusion , _ cf . _ * corollary 11.2 ) for instance , @xmath443 given a klein bottle @xmath42 with a finsler metric @xmath22 , we define a continuous riemannian metric @xmath5 on @xmath42 by replacing the minkowski norm @xmath70 on each tangent space @xmath444 by the inner product induced by the john ellipsoid @xmath445 , where @xmath82 is the unit ball of @xmath70 . the double inclusion satisfied by @xmath445 implies that @xmath446 . hence , @xmath447 from the optimal riemannian systolic inequality on the klein bottle @xcite , we obtain @xmath448 lvarez - paiva j .- c . ; thompson a .- c . : volumes on normed and finsler spaces , a sampler of riemann - finsler geometry ( d. bao , r. bryant , s .- s . chern , and z. shen , eds . ) , math . 50 , cambridge univ . press , cambridge , ( 2004 ) , 148 .

we discuss how to analytically obtain an essentially infinite matrix product state ( mps ) representation of the ground state of the xy model . on the one hand this allows to illustrate how the ornstein - zernike form of the correlation function emerges in the exact case using standard mps language . on the other hand we study the consequences of truncating the bond dimension of the exact mps , which is also part of many tensor network algorithms , and analyze how the truncated mps transfer matrix is representing the dominant part of the exact quantum transfer matrix . in the gapped phase we observe that the correlation length obtained from a truncated mps approaches the exact value following a power law in effective bond dimension . in the gapless phase we find a good match between a state obtained numerically from standard mps techniques with finite bond dimension , and a state obtained by effective finite imaginary time evolution in our framework . this provides a direct hint for a geometric interpretation of finite entanglement scaling at the critical point in this case . finally , by analyzing the spectra of transfer matrices , we support the interpretation put forward by [ v. zauner _ at . al_. , new j. phys . * 17 * , 053002 ( 2015 ) ] that the mps transfer matrix emerges from the quantum transfer matrix though the application of wilson s numerical renormalisation group along the imaginary - time direction .

1411.2607

over the recent decades , matrix product states ( mps ) @xcite and related numerical techniques have become the standard framework for simulating low energy states of local hamiltonians in 1d . mps with finite bond dimension provide an exact representation of the ground state only for a certain well - designed class of so - called parent hamiltonians ( such as the celebrated aklt model @xcite ) . nevertheless , for generic local gapped hamiltonians , mps of finite bond dimension approximate local quantities in the ground state essentially to arbitrary precision @xcite . for the long - range behavior of the system , however , this is not necessarily the case as correlations of mps with finite bond dimension by construction must decay purely exponentially at sufficiently long distances @xcite . the question of how well mps are able to reproduce correlations at long distances becomes particularly interesting in view of recent observations linking the minima of dispersion relations of elementary excitations of local , translationally invariant hamiltonian with the rate of decay of momentum - filtered correlations in its ground state @xcite . moreover , a novel interpretation of the transfer matrix obtained in the mps algorithm was also proposed in ref . . namely , that it reproduces the quantum transfer matrix in the euclidean path - integral representation of the quantum state through a renormalisation group procedure equivalent to the seminal wilson s numerical renormalisation group @xcite . in that picture the physical spin is interpreted as an impurity and the mps transfer matrix contains only the subset of degrees of freedom , out of exponentially many for the quantum transfer matrix , which are relevant for description of its correlations . this interpretation still requires corroboration by explicit calculations , which we partially address in this article see also ref . in that context . a closely related topic is that of the finite entanglement scaling at the critical point @xcite a numerically established fact that in an mps approximation of the ground state of a critical system , there emerges a long - distance correlation length as an artifact of the finite bond dimension of the mps . the state exhibits scaling behavior as a function of growing bond dimension , and it is even possible to extract the conformal information about the critical point by analyzing it @xcite . nevertheless , it remains an interesting topic to get a better understanding of how finite entanglement scaling arises . in this article we study a particular example where such questions can be addressed analytically , albeit in the framework of mps . to that end , in sec . ii , we show how to construct an exact mps representation of the ground state of the xy model a prototypical spin model in one dimension with in principle exponentially diverging bond dimension . as a corollary , we use this representation to illustrate how the ornstein - zernike form of the correlation function naturally emerges in this ( exact ) case using the standard language for mps . subsequently , in sec . iii , we show how to obtain an mps representation with finite bond dimension from the one discussed above . we examine how this truncation which is also a part of many numerical algorithms affects the state , focusing mostly on the long distance correlations and on the spectrum of the transfer matrix . in particular , in sec . iiia , we analyze the error in reproducing the correlation length in the gapped system . in sec . iiib we examine the relation between our construction and the finite entanglement scaling at the critical point , and find close similarities , allowing a geometric interpretation . finally , in secs . iiic and iiid , we take a comprehensive view on the spectrum of the transfer matrix , as well as the form factors for the correlation functions , providing direct evidence in support of the impurity picture , as proposed in ref . the @xmath0 xy model on a chain of @xmath1 spins is defined by the hamiltonian @xmath2},\ ] ] where @xmath3 are standard pauli operators acting on site @xmath4 , and periodic boundary conditions are assumed . in order to construct an mps representation of the ground state @xmath5 , we exploit the 1d - quantum to 2d - classical mapping ( commonly used e.g. in the context of quantum monte carlo @xcite , corner transfer matrix dmrg @xcite , or bethe ansatz at finite temperature @xcite , to name just a few ) , and above all the original observation by suzuki @xcite that @xmath6 commutes and more importantly shares the ground state with an operator @xmath7 given by @xmath8 } ; \ v_2 = \exp{\left [ k_2 \sum_{n=1}^n \sigma^x_n \sigma^x_{n+1}\right ] } , \nonumber\end{aligned}\ ] ] which appears naturally as the transfer matrix in the solution of the classical 2d ising model @xcite . we follow the notation of @xcite for convenience but in our case the parameters @xmath9 and @xmath10 , which we assume are non - negative , a priori do not have any specific physical interpretation . for the sake of clarity we briefly reiterate the main steps of diagonalizing @xmath7 . the subsequent use of a jordan - wigner transformation @xmath11 , @xmath12 [ with fermionic annihilation operators @xmath13 , a fourier transform @xmath14 , and a bogolyubov transformation @xmath15 allows to rewrite @xmath7 as @xcite : @xmath16=\exp \left[- h_v \right],\ ] ] where the single particle energies @xmath17 are given by @xmath18 , and the bogolyubov angles @xmath19 are determined as @xmath20 the hamiltonian of the xy model can be diagonalized following exactly the same steps as those for @xmath7 , but with the bogolyubov angles given by @xmath21 therefore , they can be simultaneously diagonalized if @xmath22 in which case the bogolyubov angles in eqs . and match and @xmath6 and @xmath23 commute . subsequently we check that the ground state of one is also the ground state of the other . the mapping above covers the part of the phase diagram of the xy model where @xmath24 and @xmath25 as shown in fig . [ fig : pd ] . and @xmath26 . red dashed lines indicate critical lines . ] it is worth pointing out , that one could match the bogolyubov angles also by considering complex @xmath27 , which , as follows from eq . , would cover the incommensurate region of @xmath28 . however , while the hamiltonians @xmath6 and @xmath29 do commute , they only share the ground state if an additional condition of @xmath30 is satisfied . for that reason , a more general approach is required to extend the mapping to the incommensurate region with oscillating correlation function and , in this article , we limit ourself only to the commensurate case of @xmath25 and real @xmath9 and @xmath10 . at the risk of stating the obvious , @xmath7 appears naturally also as a result of the second order suzuki - trotter expansion of ( the exponent of ) the quantum ising model hamiltonian @xmath31}$ ] , in which case @xmath32 and @xmath33 , where @xmath34 is a discrete trotter step . equation precisely quantifies the trotter errors for this specific , but commonly employed case , by showing that one approaches the ground state of the xy model as a result of them . exponentials of operators of the form in eq . can be efficiently decomposed in terms of matrix product operators ( mpos ) with bond dimension @xmath35 @xcite : @xmath36 } x^{s_1 } \otimes \dotsm \otimes x^{s_n},\ ] ] up to normalization @xcite , where @xmath37 we refer to @xcite for details of the derivation . following @xcite , we employ the convenient notation where @xmath38 , and @xmath39 . by applying @xmath7 to some initial state @xmath40 @xmath41 times , the ground state @xmath42 of the xy model is obtained in the limit @xmath43 ( pending normalization ) . this is equivalent to performing imaginary time evolution with the hamiltonian @xmath23 in eq . , where the effective imaginary time of evolution is proportional to @xmath41 . this procedure is depicted in the top half of fig . [ fig : network]a , where each row represents a single operator @xmath7 in mpo form with local tensors @xmath44 . alternatively , one can look at this picture in the vertical direction , interpreting each column @xmath45 as an exact mps representation of @xmath42 with bond dimension @xmath46 , that is , @xmath47 } \vert i_1\dotsc i_n \rangle = \vert \psi(a)\rangle$ ] . due to the symmetry between @xmath48 and @xmath49 , which is apparent from eq . , we can obtain @xmath45 simply by inverting the steps leading to the mpo decomposition of @xmath7 . the only additional complication comes from the boundaries of @xmath50 , representing the physical ( spin ) degrees of freedom and the initial state @xmath51 , respectively . after some algebra we obtain @xmath52 } ; \ u_2 = \exp{\left [ k_1 \sum_{l=1}^{l-1 } \tau^x_{l } \tau^x_{l+1}\right]}. \nonumber\end{aligned}\ ] ] here @xmath53 labels the auxiliary degrees of freedom along the vertical direction and @xmath54 are pauli operators acting on these . @xmath55 is a localized operator acting on auxiliary site @xmath56 , with @xmath57 and @xmath58 . for a graphical representation see fig . [ fig : network]b . notice that @xmath55 commutes with @xmath59 , but not with @xmath60 . as an initial ( top ) state we use for convenience a product state @xmath61 with @xmath62 fully polarized in the @xmath63 direction . this state is an eigenstate of the parity operator @xmath64 with eigenvalue @xmath65 . as @xmath66 commutes with @xmath7 , the final state @xmath67 has the same parity as @xmath51 . in particular , this means that in the ferromagnetic phase we consider the symmetric superposition of the two symmetry broken ground states . from now on , for the rest of this article , we will assume that the system is in the thermodynamic limit @xmath68 . into a two - dimensional tensor network . rows represent operator @xmath7 in mpo form with local tensors @xmath69 . half - columns constitute the mps decomposition of @xmath5 with mps matrices @xmath45 , while a full ( infinite ) column represents the ( quantum ) transfer matrix @xmath70 at zero temperature . ( b ) mps matrix @xmath71 as given by eq . . ] we have cast the ground state of the xy model in an exact mps form , where each mps matrix @xmath45 has bond dimension @xmath46 and the limit @xmath72 is assumed . however , before we address the question of finding an efficient mps approximation with low bond dimension , it is illuminating to discuss the asymptotic behavior of the correlation functions in the exact case . following standard notation @xcite we define the mps transfer matrix ( see fig . [ fig : network]a ) as @xmath73 which , up to the open boundary conditions and exchanging @xmath74 and @xmath10 , has exactly the same form as @xmath7 in eq . . we discuss this in more detail in the appendix . the subscript _ f _ denotes that this is the full transfer matrix of the exact mps representation of the ground state , as opposed to the transfer matrix of the truncated mps , which will be introduced in the next section . we also define the ( spin ) operator transfer matrix @xmath75 , which obviously simplifies to the one above for @xmath76 . for @xmath77 , which will be of most interest to us in this section , we find it convenient to express it as @xmath78 , where @xmath79}$ ] and @xmath80 . this way the effect of @xmath81 in the spin transfer matrix is encoded in an operator which is localized in the virtual direction at sites with @xmath82 , see fig . [ fig : network]a . note however that some care is needed here as @xmath83 is not hermitian and does not commute with @xmath70 . the static correlation function is then calculated as @xmath84 . for illustrative purposes , we further consider only the connected correlation function @xmath85 where the second equality is valid in the thermodynamic limit when @xmath86 projects onto the ground space of @xmath70 which is hermitian by construction . here , we have defined the form factors as @xmath87 using the localized operator @xmath83 for simplicity . @xmath88 are the eigenvectors of the transfer matrix @xmath89 to eigenvalues @xmath90 and @xmath91 is the dominant one . below we will only consider the case where @xmath91 is unique . otherwise , e.g. in the ferromagnetic phase , the definition of form factor has to be generalized to include properly normalized sum over all dominant eigenvectors . @xmath89 can be diagonalized by mapping onto a free - fermionic system as @xmath92 } , \label{eq : diagonaltf}\ ] ] where @xmath93 are fermionic annihilation operators . for simplicity , in this section , we approximate @xmath70 by using periodic boundary conditions , which does not affect the results . it is then diagonalized following the same steps as for @xmath7 in eqs . - . in the limit @xmath94 the spectrum of the transfer matrix consists of continuous bands as the states @xmath95 are obtained from the vacuum @xmath96 by exciting free - fermionic quasiparticles and @xmath97 follows from summing up the corresponding single particle energies @xmath98 , where @xmath99 now , in order to obtain the leading asymptotic of the correlation function , it is sufficient to know the dispersion around the minimum of the lowest relevant band i.e. for which the form factors @xmath100 are nonzero and the scaling of those form factors . in the case of @xmath101 the only nonzero form factor contributions come from @xmath102 , that is , where two quasiparticles with momenta @xmath103 and @xmath104 are excited . notice that form factors corresponding to the lowest single particle band @xmath105 vanish , since both @xmath106 and @xmath83 conserve parity . below , we consider two cases : \i ) critical point for @xmath107 , in which case @xmath108 . expanding around the minimum of @xmath109 at @xmath110 we obtain @xmath111 with the coefficients @xmath112 and @xmath113 . now , for large @xmath114 the correlation function behaves asymptotically as @xmath115 in the limit of @xmath43 we treat @xmath116 as continuous variables with @xmath117 and we have @xmath118 where we can extend the limits of integration to @xmath119 for large @xmath120 . rescaling the variables in the integral leads to the algebraic dependence on @xmath114 , @xmath121 where we recover the classic result by barouch and mccoy @xcite . \ii ) paramagnetic phase for @xmath122 . the dispersion relation exhibits qualitatively different behavior around its minimum when compared to the critical case discussed above . expanding around @xmath110 we obtain @xmath123 with the gap @xmath124 and the coefficients @xmath125 and @xmath126 . now , for large @xmath114 and in the continuous limit of @xmath127 the correlation function behaves asymptotically as @xmath128 where we can extend the limits of integration to @xmath129 for large @xmath130 . naturally , we recognize the correlation length as @xmath131 , which is the slowest possible decay resulting from @xmath132 . performing integrals or just rescaling variables to extract @xmath114 from the integrals yields the leading algebraic dependence on @xmath114 , @xmath133 again in agreement with @xcite . it is worth pointing out that this expansion is valid for @xmath134 , what can be seen from the size of @xmath130 . for smaller @xmath114 , when the exponential does not suppress other terms , the correlation function would behave similarly as at the critical point . while this remark is not so important for zz correlation here , other correlation functions might have other exponents of the algebraic part at the critical point and away from it @xcite . to summarize , notice that on the one hand the gap of the transfer matrix sets the correlation length while on the other hand , the full ( low energy part of the ) continuous band contributes _ equally _ ( in the sense of form factors being proportional to @xmath135 ) to form the algebraic part as a consequence of the nonflat dispersion relation . the value of the exponent in this algebraic behavior is determined by a combination of the dispersions of the leading eigenvalues of the transfer matrix , and of the corresponding form factors , as well as symmetries . for instance in the case of @xmath136 discussed above , the conserved parity symmetry modifies this exponent due to the double integrals appearing in eqs . and . this is also accompanied by halving the correlation length in eq . comparing to the slowest possible decay suggested by the transfer matrix . we also note that , at the critical point , this picture might be more complicated as contributions from many bands are relevant in some cases ; we refer to the appendix for additional details . such an asymptotic behavior of the correlation function is often labeled in literature as the ornstein - zernike form @xcite , especially in context of systems significantly away from the critical points . for 1d quantum systems , it is typically expected that the exponent in the algebraic behavior is equal to @xmath137 , namely @xmath138 . notice that this form of asymptotic behavior would indeed emerge in our treatment in the most simple case when single - particle form factors are nonzero ( they might be zero for example as a result of some symmetries of local observables ) and in the leading order independent on @xmath139 : @xmath140 , and the dispersion relation around the minimum is smooth and quadratic @xmath141 . this happens e.g. for the xx correlation function in the paramagnetic phase . we discuss this further in the appendix , where we numerically obtain the behavior of the form factors for other correlation functions in various phases . the results above were obtained analytically in the limit of exponentially diverging bond dimension @xmath142 , where @xmath143 with @xmath127 . however , above all , mps serve as a class of variational states that lie at the heart of many numerical techniques , where only modest bond dimensions are feasible . it is therefore important to understand what information about the quasi - exact state is retained after truncating to an efficient mps approximation with finite bond dimension @xmath142 , i.e. @xmath144 , with @xmath145 matrices @xmath146 . we follow the standard mps truncation procedures for infinite , translationally invariant systems @xcite . it is based on finding the schmidt decomposition of the state along a single cut and retaining only the dominant schmidt values . while such a procedure would be optimal for truncation at a single bond , in the infinite system it is performed at all sites simultaneously @xcite . this is then equivalent to finding the reduced density operator of the mps @xmath147 on a half - infinite chain and finding its diagonal basis in which we truncate by keeping only the dominant eigenvalues . for this particular case , where the transfer matrix @xmath70 is hermitian and its left and right dominant eigenvector are both given by @xmath148 , the physical density operator of the half infinite chain shares the spectrum with the reduced density matrix @xmath149 of @xmath148 with support on the @xmath41 site auxiliary system with @xmath150 ( see fig . [ fig : network ] ) . we briefly outline the main steps of the procedure below and refer to the appendix for details . we map the full transfer matrix onto a system of free fermions and , following ref . , describe the transfer matrix using _ the transformation matrix _ , which describes the transformation of fermionic operators under the similarity transformation given by @xmath70 . since part of the operations has to be performed numerically , we keep @xmath41 large but finite . bringing the transformation matrix into the canonical form allows to find the bogolyubov transformation which diagonalizes the transfer matrix and @xmath151 is the vacuum state in that basis . we describe the reduced density matrices in a standard way @xcite by using the two - point correlation matrix . subsequently , it can then be expressed as @xmath152 , where @xmath153 are fermionic annihilation operators and @xmath154 is the entanglement spectrum arranged in ascending order . @xmath153 and @xmath155 are obtained by finding the bogolyubov transformation which brings the correlation matrix into its canonical form . efficiently truncating to an mps with bond dimension @xmath156 is now obtained by keeping only the first @xmath157 ( most relevant ) _ fermionic modes _ of @xmath149 . this amounts to the projection @xmath158 where @xmath159 . we point out that this procedure is _ not _ fully equivalent to just keeping the @xmath142 largest singular values of @xmath149 as it additionally retains the free - fermionic structure of the problem , as the reduced density matrix of the truncated state and consequently its schmidt values have the form @xmath160 this means that some care is needed when we compare the bond dimensions ( and states ) obtained with our procedure with those from conventional numerical mps methods . for instance , @xmath161 is going to contain some very small schmidt values together with the dominant one , especially for larger @xmath157 . as a trade - off , the structure we retain allows for a clean description and interpretation of the spectra , which would be nearly impossible using the standard numerical approach . finally , as described in the appendix , we obtain the transformation matrix describing the transfer matrix generated by @xmath162 , i.e. @xmath163 . by finding the bogolyubov transformation which brings it into the canonical form we diagonalize the truncated transfer matrix as @xmath164,\ ] ] where the spectrum is determined by single particle energies @xmath165 arranged in ascending order . for the remainder of this article we will mostly focus on this spectrum . firstly , we discuss the non - critical case focusing on the dominant eigenvalues of the transfer matrix @xmath166 . to that end , we simulate the xy model for a particular set of parameters @xmath167 and @xmath168 in the paramagnetic phase . we show the resulting single particle energies @xmath169 for several different values of @xmath157 in fig [ fig : numgap]a . notably , we observe that the low energy part of the spectrum collapses onto a single curve when the index @xmath170 is rescaled by @xmath157 . moreover , the lowest part of the spectrum shows quadratic behavior in @xmath171 , which corroborates a similar observation made in ref . for the case of conventional mps calculations and part of the transfer matrix spectrum most relevant for the xx correlation function , see fig . 9 therein . in other words the results are consistent with the scaling of the form @xmath172 , at least for the first few @xmath171 s . this universal quadratic behavior implies that the gap of the truncated transfer matrix @xmath173 should be shifted from the value of the true gap @xmath174 , and approach it as a power law in @xmath157 with exponent equal 2 . that is indeed observed in fig [ fig : numgap]b where we show the relative error in the gap ( i.e. the inverse of the correlation length ) for increasing @xmath157 . we fit @xmath175 with @xmath176 close to 2 . notice , that for this particular set of parameters , even for @xmath177 the correlation length obtained from the free - fermionic mps still underestimates the exact value by @xmath178 , where the exact gap of the full transfer matrix is given as @xmath179 those observations also hold for other values of the magnetic field , where @xmath180 remains close to 2 and @xmath181 grows slowly when approaching the critical point . and @xmath182 . ( a ) single particle energies @xmath169 of eq . for different @xmath157 . for rescaled index @xmath183 , the data points collapse onto a single curve . the solid line is a quadratic fit to the first 4 points with @xmath177 . ( b ) relative error of the correlation length @xmath184 as a function of @xmath157 . it shows power - law behavior @xmath185 = -2.0518 \log \chi + 1.6944 $ ] ( green line ) . ( c ) comparison with idmrg calculations . the black line represents the exact gap @xmath174 and stars represent the gap @xmath173 of the finite @xmath157 approximation of the transfer matrix . the blue line shows ( minus log of ) the transfer matrix gap obtained from an idmrg ground state approximation , plotted as a function of @xmath186 . finally , circles represent data obtained from truncating idmrg with @xmath187 down to a smaller bond dimension while only selecting schmidt states preserving the free - fermionic structure in eq . . ] it is apparent that a finite value of @xmath157 results both in an underestimation of the correlation length and the breakdown of the asymptotic algebraic dependence of the correlation function on @xmath114 above some length scale dictated by @xmath157 . correspondingly , by increasing @xmath157 , the mps is able to better reconstruct the low energy part of the continuous band which is responsible for the asymptotic algebraic part of the exact correlation function . the neat behavior of the gap observed above should be contrasted with the results from the conventional mps calculations using idmrg @xcite , which we plot in fig . [ fig : numgap]c . first of all , we obtain a very good match between free - fermionic results and idmrg , provided the correct @xmath188 schmidt - states are selected from an mps obtained from idmrg with initially larger bond dimension ( labeled as idmrg * @xcite ) corroborating our procedure . without preserving the structure of eq . , but rather just keeping the largest schmidt values during truncation , standard idmrg is able to approach the exact gap @xmath174 faster with increasing @xmath142 , but in a rather irregular , step - wise way . remarkably , we can use the observations made for free - fermions above to estimate this behavior . we employ the fact that in the paramagnetic phase considered here the single particle schmidt spectrum in eq . has a simple form @xmath189 @xcite . we define @xmath190 as the index of the largest schmidt value _ not _ reproduced by eq . for given @xmath157 , and use the form above to calculate the value of @xmath191 . we can expect that the error of the gap obtained with idmrg with bond dimension @xmath192 should be lower - bounded by the error of @xmath193 from our fermionic procedure , as the second one represents the state containing additional information coming from some smaller schmidt values as well ( numerics confirms this ) . @xmath192 grows quickly with @xmath157 , and locally , for @xmath194 we numerically see the scaling @xmath195 , @xmath196 . by substituting this into eq . with @xmath197 we expect that the error of the gap obtained with idmrg with bond dimension @xmath142 should be vanishing slower than @xmath198 , at least for @xmath142 of the order of hundred . the fit ( not plotted ) to the data in fig . [ fig : numgap]c and @xmath199 shows that the error is shrinking on average as @xmath200 , remarkably close to our prediction . those results illustrate that one has to take some care when extracting the correlation length directly from the transfer matrix , as the error can be vanishing slowly ( and increasingly so ) with growing bond dimension . and @xmath201 . ( a ) smallest single particle energies @xmath169 for different @xmath157 quickly approach the exact smallest values from @xmath202 . ( b ) schmidt values obtained with idmrg with @xmath203 and corresponding schmidt values obtained from @xmath204 and truncating to @xmath205 . ( c ) comparison of the transfer matrix spectrum obtained for @xmath204 truncated to @xmath206 , and the spectrum from truncating idmrg results with @xmath203 down to @xmath207 @xcite . ( d ) comparison of single particle energies in eq . obtained from truncating from @xmath204 and idmrg with @xmath203 , see text for details . ] secondly , we consider the gapless case , where we simulate the xy model for a particular set of critical parameters @xmath107 and @xmath208 . contrary to the gapped system , the energy gap of @xmath7 is vanishing here and the actual ground state can not be reached for any finite value of @xmath41 . the dominant part of the spectrum of the full transfer matrix @xmath70 behaves as @xmath209 , where the quasi - momenta take necessarily discrete values @xmath210 , which are universal for systems with open boundary conditions . this is shown in fig . [ fig : num_cri]a for @xmath204 , together with the dominant @xmath169 from eq . obtained from truncation for several different @xmath157 . we observe that the @xmath169 reproduce the discrete low energy structure for finite @xmath41 increasing well with growing @xmath157 . in other words , even for relatively small @xmath157 the gap of the transfer matrix is effectively determined by @xmath41 , which in turn is proportional to the imaginary time evolution of the initial state with hamiltonian @xmath23 , see fig . [ fig : network ] . on the other hand , conventional mps calculations , where we use idmrg @xcite , try to approximate the critical ground state by the best possible state with finite correlation length @xcite . a priori there is no reason to expect a good match between these two approximations ( idmrg and free - fermionic mps with finite @xmath41 ) , we however obtain a surprisingly good agreement nonetheless . to that end we analyze the state obtained with idmrg and bond dimension @xmath203 and compare it with data for @xmath204 . the value of @xmath41 is obtained here by matching the ratio of the first two schmidt values , corresponding to @xmath211 in eq . , to the one obtained from idmrg . in fig . [ fig : num_cri]b we plot the schmidt values from idmrg and the corresponding values for @xmath206 and @xmath212 . most importantly , even for idmrg at the critical point , we are able to identify groups of schmidt values corresponding to the underlying free - fermionic structure given by eq . . that such a structure is visible for the non - critical system is to be expected , however for the critical case there is a priori no reason for a numerical algorithm not to break it . even more , the spectra obtained with idmrg and with finite @xmath41 coincide rather well ( i.e. corresponding values of @xmath155 in eq . are matching ) . we further corroborate this by comparing the spectra of the transfer matrix ( @xmath213 eigenvalues ) for several values of @xmath157 and the corresponding idmrg * @xcite . as can be seen in fig . [ fig : num_cri]c for @xmath206 , picking the correct schmidt values results in a clear structure of the transfer matrix spectrum which is consistent with the one given by eq . for finite @xmath41 . the structure of the transfer matrix in fig . [ fig : num_cri]c allows us to compute single particle energies corresponding to the ones in eq . directly from idmrg * and compare them with the ones obtained with free - fermions for finite @xmath41 . the results are shown in fig . [ fig : num_cri]d where the match for @xmath214 is remarkably good . it is worth stressing here one more time , that all the points labeled idmrg * in fig . [ fig : num_cri ] are acquired from the same initial state obtained from idmrg with @xmath203 , which was subsequently truncated down by picking the correct schmidt values . the obtained results allow us to conclude that the state obtained with idmrg contains the structure which is fully consistent with a free - fermionic theory on a strip of finite width with open boundary conditions for similar observation in finite system where the exact ground state can be reached due to the finite size effects , see @xcite . this provides a strong hint that so called finite entanglement scaling @xcite scaling observed while simulating the ( conformally invariant ) critical theory using mps with finite @xmath142 can be interpreted in a geometric way . this can not be seen that easily when looking directly at idmrg and the ratios of the dominant eigenvalues of the transfer matrix ( cf . @xcite ) , since the ratios are both highly susceptible to further truncations and even then , for a given state and @xmath215 they are still far from expected values of @xmath216 on a strip , see fig . [ fig : num_cri]a . and @xmath201 . ( a ) dominant schmidt values at snapshots of the imaginary time ( @xmath217 ) evolution with the tdvp algorithm with @xmath218 and corresponding schmidt values from our free fermionic procedure with finite @xmath41 . ( b ) for a large window of intermediate time the values of @xmath41 are proportional to @xmath217 ( plotted in a log - log scale for convenience ) . ] however , while we were able to find a value of @xmath41 in a free - fermionic mps which is a good match to a particular mps obtained from idmrg with given bond dimension @xmath219 , the above analysis does not provide any hint why , for given bond dimension @xmath219 , idmrg should yield an mps approximation corresponding to an effective imaginary time evolution of the system up to some finite imaginary time proportional to @xmath41 . or equivalently , how to choose @xmath219 for an idmrg calculation to reproduce results from a finite @xmath41 free - fermionic calculation . in order to shred more light on this , we performed similar analysis for the tdvp algorithm @xcite which is based on the imaginary time evolution . likewise , we find a reasonable agreement between the state obtained with tdvp and our free - fermionic construction for some finite value of @xmath41 , provided that tdvp was initialized with the spin polarized state with even parity ( all spins pointing in + z direction ) . then , the tdvp algorithm does not manifestly break the parity symmetry and along the evolution @xmath220 ( this is also the case for the state obtained with idmrg above ) . for tdvp initialized with the random state no good match with our construction could be found . in fig . [ fig : tdvpvsff]a we show the dominant schmidt values taken from a _ single _ run of the tdvp algorithm with @xmath218 , where we looked at the snapshots at different values of the imaginary time @xmath217 . we compare them with the corresponding schmidt values from our free fermionic procedure with finite @xmath41 , where the values of @xmath41 were obtained by matching the ratios of the two largest schmidt values . for intermediate values of time the match is very good . we also observe that it is getting considerably worse for large enough values of @xmath217 ( around @xmath221 in our case ) , where the energy of the tdvp state is effectively saturating at @xmath222 above the exact ground state energy . not surprisingly , for a wide range of intermediate times , the values of @xmath41 are proportional to @xmath217 , see [ fig : tdvpvsff]b . this suggest that instead of the standard approach based on analyzing the converged states for various values of d , it would be more natural to extract the conformal information about the ground state just from the snapshots of single run of the tdvp algorithm for fixed value of @xmath219 . finally , while the previous two sections were focusing on the dominant part of the transfer matrix , we take a more comprehensive view on the spectrum here , valid both for the critical and gapped case alike . to that end , we define discrete momenta @xmath223 corresponding to the spectrum of the truncated transfer matrix @xmath169 through the relation @xmath224 above , @xmath225 is the dispersion relation of the full transfer matrix , given by eq . , where the momenta take continuous ( in the limit @xmath43 ) values @xmath226 . we show the resulting @xmath223 , both for the gapped and critical cases , in fig . [ fig:5 ] . most importantly , as can be seen in that plot , the value of momenta which are effectively selected during the truncation procedure satisfy the relation @xmath227 for all but the first few @xmath171 s . [ t ] corresponding to the single particle energies @xmath169 of the truncated transfer matrix for several @xmath157 . data for ( a ) the gapped system with @xmath167 and @xmath168 , ( b ) critical system with @xmath107 , @xmath208 and @xmath204 . in both cases we find a clear linear dependence @xmath228 , valid for all but few smallest m s ( i.e. smallest @xmath169 ) . ] in ref . , it was proposed that one can understand the transfer matrix obtained in the mps algorithm as resulting from a renormalisation group procedure applied to the ( full ) quantum transfer matrix in the imaginary time direction . more precisely , the physical spin at @xmath229 in the virtual direction in fig . [ fig : network ] plays a distinguished role in this tensor network as physical operators are applied there during the calculation of expectation values . it can then be interpreted as an impurity in the two - dimensional tensor network and the degrees of freedom relevant for the description of impurity and at the same time the concise description of the state in the mps algorithm emerge as a result of application of wilson s numerical renormalisation group ( nrg ) @xcite to the quantum transfer matrix along the virtual ( imaginary time ) direction . qualitatively , this procedure boils down to dividing the continuous momenta @xmath230 into windows which are logarithmically shrinking toward @xmath110 ( minimum of @xmath225 ) , and representing each window by a single effective mode , as pictorially presented in fig . [ fig:6 ] . the mode corresponding to the largest momentum represents the action of a few sites localized close to the impurity in the virtual direction ( small @xmath231 in fig . [ fig : network ] ) , while smaller momenta modes describe the relevant degrees of freedom which can not be sharply localized around the impurity and are supported on sites extending to larger values of @xmath231 . for gapped systems , this procedure can be terminated at the infrared cut - off related to the correlation length and a good approximation is obtained with a finite number of modes , resulting in a finite bond dimension . therefore , fig . [ fig:5 ] and eq . provide a direct evidence that such an interpretation of the origin of the transfer matrix obtained in the mps algorithm is indeed correct . this picture is further validated in ref . , where a general tensor - network ansatz based on this idea is constructed . , is represented by single effective mode . this is also visible in fig . [ fig:5 ] and eq . , where @xmath223 are obtained from the transfer matrix , as a result of standard mps truncation . ] the values of @xmath232 in eq . depend both on the bond dimension @xmath157 and the correlation length in the system , i.e. @xmath233 in the gapped system and @xmath234 in the critical one . qualitatively , one could expect that @xmath235 and @xmath236 . we get rid of the proportionality constant appearing in eq . by considering the ratio @xmath237 and thus expect that @xmath238 quantitatively , we check this by extracting @xmath239 ( the slope of the linear dependence in fig . [ fig:5 ] ) for various values of @xmath240 and @xmath157 . while we see that eq . requires some corrections , it is capturing the _ leading _ tendency quite well . one of the reasons behind the correction could be , for example , that the smallest values of @xmath241 are visibly ( in fig . [ fig:5 ] ) affected by the infrared cut - off and eq . does not describe them well . we obtain further evidence in support of the interpretation presented in the previous section by looking at the form factors calculated for the truncated state . we focus on the zz correlation function and define the form factors @xmath242 where @xmath243 is the operator transfer matrix for @xmath81 , @xmath244 is the dominant eigenstate of the truncated transfer matrix @xmath245 and @xmath246 are eigenstates corresponding to two excited quasiparticles . the form factors corresponding to single particle excitations are zero due to fermionic parity symmetry . we plot the unique , non - zero form factors @xmath247 for a specific point with @xmath167 and @xmath182 in the paramagnetic phase , and bond dimension @xmath177 in fig . [ fig:7 ] . we point out , that the form factors most relevant for the long range correlations ( i.e. corresponding to smallest single particle energies ) have the smallest values in that plot . for the zz correlation function for the system with @xmath248 , @xmath168 in the truncated state with @xmath177 . only unique , non - zero form factors are plotted . for comparison , red circles show the results of a simple renormalisation procedure , where @xmath249 ( see eq . ) is obtained by integration of the form factors before truncation over the momenta window surrounding @xmath250 and @xmath251 , depicted pictorially in fig . [ fig:6 ] . ] additionally , we observe that the form factors are zero when the indices @xmath252 are simultaneously odd or even , which can be attributed to the symmetry when reflecting bra and ket parts of the transfer matrix and we do not show them in the plot . both the details of the calculations and the above symmetry in the exact case where it is also present are discussed in the appendix . we argue that the values of the form factors can be understood as emerging through the renormalization group procedure described in sec . iiic . according to this each mode is effectively representing the window of momenta of the original not truncated transfer matrix @xmath132 , as depicted in fig . [ fig:6 ] . at the same time , the dominant , two - particle contribution to the correlation function is truncated as @xmath253 . consequently , we could expect that the form factor @xmath247 is obtained as a sum of all the form factors @xmath254 in that window ( notice that @xmath255 was defined using only @xmath83 , in the expression for the operator transfer matrix @xmath256 , resulting in the additional factor @xmath257 appearing above , and also that @xmath258 ) . in order to test this hypothesis we introduce , @xmath259 where @xmath260 represents the window around the momentum @xmath223 in fig . [ fig:6 ] . as the most crude approximation , this means that the form factor would be proportional to the ( logarithmically shrinking toward @xmath110 ) size of the corresponding momentum window . the results of the above procedure are shown in fig . [ fig:7 ] reproducing the actual form factors with the relative error of maximally a couple of percent , which could be probably brought down even more by picking the exact size of the windows @xmath260 in a more sophisticated way here we used @xmath261 taking into account that roughly half of the form factors are zero . still , even for such a simple procedure the agreement is exceptionally good as the form factors in fig . [ fig:7 ] span almost 4 orders of magnitude . in this article , we constructed an exact mps representation for the ground state of the xy model and showed how the ornstein - zernike form of the correlation function is naturally emerging in this picture . subsequently we truncated this state ( which has an exponentially - diverging mps bond dimension ) , obtaining its approximation with relatively small bond dimension a procedure which is commonly employed in the representation of quantum states using tensor networks . by analyzing this truncation , we can conclude , that an mps with finite bond dimension can be understood as a particular renormalization group procedure applied to the exact transfer matrix , whose dominant part is increasingly well reproduced with increasing bond dimension . the free - fermionic nature of the problems allows for a concise description of the state , and , for instance , its transfer matrix . the obtained spectrum of the mps transfer matrix behaves as would be expected from the rg scheme based on a description in terms of an effective impurity proposed in ref . while the current discussion was limited to exactly solvable system , it motivates further studies of the intrinsic structure of the mps matrices . such an analysis is indeed done in @xcite , where a general tensor network algorithm based on the impurity picture , discussed in sec . iii.c , is designed . likewise , the mapping employed in this article in eq . is limited to the commensurate phase of the xy model , where the transfer matrix is hermitian and most of our procedures were based on that fact . it would be interesting to extend the mapping into the incommensurate phase and non - hermitian transfer matrices , especially in context of testing ref . @xcite which connects such oscillations with the position of minima of the dispersion relation of the hamiltonian . discussions with vid stojevic , viktor eisler are gratefully acknowledged . we acknowledge support by ncn grant 2013/09/b / st3/01603 ( m.m.r . ) , eu grants siqs and querg and the austrian fwf sfb grants foqus and vicom ( v.z . and f.v . ) , and research foundation flanders ( j.h . ) . 99 m. fannes , b. nachtergaele , and r. werner , comm . phys . * 144 * , 443 ( 1992 ) . f. verstraete , v. murg , and j.i . cirac , adv . phys . * 57 * , 143 ( 2008 ) . u. schollwck , annals of physics * 326 * , 96 ( 2011 ) . i. affleck , t. kennedy , e.h . lieb , h. tasaki , phys . lett . * 59 * , 799 ( 1987 ) . f. verstraete , j.i . cirac , phys . b * 73 * , 094423 ( 2006 ) . v. zauner at . al . , new j. phys . * 17 * , 053002 ( 2015 ) . j. haegeman , v. zauner , n. schuch and f. verstraete , arxiv:1410.5443 ( 2014 ) . wilson , rev . phys . * 47 * , 773 ( 1975 ) . l. tagliacozzo , t. de oliveira , s. iblisdir , and j. latorre , phys . rev . b * 78 * , 024410 ( 2008 ) . f. pollmann , s. mukerjee , a.turner , and j. moore , phys . 102 * , 255701 ( 2009 ) . v. stojevic , j. haegeman , i. p. mcculloch , l. tagliacozzo and f. verstraete , phys . b * 91 * , 035120 ( 2015 ) . w. m. c. foulkes , l. mitas , r. j. needs , and g. rajagopal , rev . phys . * 73 * , 33 ( 2001 ) . t. nishino and k. okunishi , j. phys . . jpn . * 65 * 891 ( 1996 ) ; _ ibid . _ * 66 * 3040 ( 1997 ) ; a. klmper and r.z . bariev , nucl . b , * 458 * 623 ( 1996 ) ; a. klmper , z. phys . b * 91 * 507 ( 1993 ) ; j. sirker , j. stat . mech . theor . exp . , p12012 ( 2012 ) ; m. suzuki , prog . . phys . * 46 * , 1337 ( 1971 ) ; _ ibid . _ * 56 * , 1454 ( 1976 ) . schultz , d.c . mattis , and e.h . lieb , rev . phys . * 36 * , 856 ( 1964 ) . v. murg , j.i . cirac , b. pirvu , and f. verstraete , new j. phys * 12 * , 025012 ( 2010 ) . e. barouch , and b.m . mccoy , phys . rev . a * 3 * , 786 ( 1971 ) . ornstein , f. zernike , proc . amsterdam * 17 * , 795 ( 1914 ) . t. kennedy , comm . phys . * 137 * , 599 ( 1991 ) . r. ors , and g. vidal , phys . b * 78 * , 155117 ( 2008 ) ; g. vidal , phys . * 98 * , 070201 ( 2007 ) . abraham , stud . app . math . * 1 * , 71 ( 1971 ) . i. peschel , j. phys . a : math . gen . * 36 * , l205 ( 2003 ) . i. peschel , and v. eisler , j. phys . a : math . theor . * 42 * 504003 ( 2009 ) . s.r . white , phys . lett . * 69 * , 2863 ( 1992 ) ; phys . b * 48 * , 10345 ( 1993 ) . mcculloch , arxiv:0804.2509 ( 2008 ) . b. pirvu , g. vidal , f. verstraete , and l. tagliacozzo , phys . b * 86 * , 075117 ( 2012 ) . a.m. luchli , arxiv:1303.0741 ( 2013 ) . j. haegeman , j.i . cirac , t.j . osborne , i. piorn , h. verschelde , f. verstraete , phys . lett . * 107 * , 070601 ( 2011 ) . j. haegeman , c. lubich , i. oseledets , b. vandereycken , and f. verstraete , arxiv:1408.5056 ( 2014 ) . m. bal , m.m . rams , v. zauner , j. haegeman , f. verstrate , arxiv:1509.01522 ( 2015 ) . . we obtain an mps for some larger bond dimension @xmath219 and subsequently truncate to @xmath262 by keeping only schmidt values corresponding to eq . . they can be identified by finding groups of schmidt values which have constant ratios ( with reasonable precision ) , resulting from the structure of eq . . in this section we extend the discussion from sec . iib by considering other correlation functions . we numerically obtain the form factors , extract their relevant scaling and show how the asymptotic scaling of the correlation functions is emerging as a result . we compare these observations with the analytical results obtained in ref . @xcite by direct calculation of the correlation functions in the xy model an approach which is computationally significantly less complicated . consequently , our analysis is intended just as an illustration of the underlaying mechanism . as discussed is sec . iib , in our framework , the connected correlation function is calculated as : @xmath263 where we consider local operators @xmath264 , and the index @xmath265 enumerates all the eigenstates of the hermitian transfer matrix @xmath70 . the form factors are defined as @xmath266 where @xmath267 is the number of degenerated and orthogonal , dominant eigenvectors of @xmath70 , which we label as @xmath268 . the relevant , localized part of the operator transfer matrix , which we decompose as @xmath269 , is @xmath270 } , \nonumber \\ \mathcal{q}_x & = & \sqrt{1-e^{-4 k_1 } } ( \tau_{-1 } + \tau_{1 } ) /2 , \label { eq : qxyz } \\ \mathcal{q}_y & = & i \sqrt{e^{4 k_1}-1 } ( \tau_{-1 } -\tau_{1 } ) /2 , \nonumber\end{aligned}\ ] ] where @xmath80 and following the convention in the main text where the position in the virtual direction @xmath271 is consistent with fig . [ fig : network ] . the operators @xmath272 and @xmath273 , when mapped onto a fermionic system as in the appendix b , contain a string operator which extends over half of the chain . therefore , we limit ourselves here to numerical calculations of the expectation values for some large but finite value of @xmath41 . the transfer matrix @xmath70 is diagonalized as discussed in the next section , where its diagonal form is given by eq . with the single particle energies given by eq . , ( almost ) uniformly distributed momenta @xmath226 ( for non - zero @xmath274 ) and the effective @xmath275 . the transfer matrix is invariant with respect to the transformation @xmath276 . likewise , @xmath272 and @xmath83 are even with respect to that transformation and @xmath273 is odd . this suggest that the form factors related to the excited states with non - matching symmetry would be zero , which we indeed observe . for that reason one also expects that the mixed form factor @xmath277 . similarly , @xmath278 as @xmath83 is conserving fermionic parity and @xmath272 and @xmath273 are changing it . this is consistent with the fact that in the ground state of xy model @xmath279 . _ paramagnetic phase. _ we simulate the system for @xmath282 . from the numerics , we observe that the non - zero form factors , which correspond to single - particle band , behave as @xmath283 and @xmath284 around @xmath110 , and the dispersion relation @xmath285 . we show the form factors in fig . [ fig : ff_paramagnetic ] . the asymptotic form of the correlation functions is then determined by the single - particle band , as @xmath286 in agreement with @xcite . _ ferromagnetic phase. _ we simulate the system for @xmath287 . there is a single mode with @xmath288 and the dominant eigenstate of the transfer matrix is degenerated with @xmath289 . the rest of the spectrum is still given by eq . with @xmath290 around the minimum at @xmath110 . we observe that the form factors corresponding to single - particle band are zero , @xmath291 , and likewise @xmath292 . for xx and yy correlations , the first nonzero form factors come from 3-quasiparticles excitations including @xmath288 , i.e. @xmath293 . from the specific example the data are consistent with the quadratic scaling of @xmath294 and quartic scaling of @xmath295 . likewise for zz correlation , we see quadratic scaling @xmath296 . by performing double integrals like in the main text , this would lead to @xmath297 , @xmath298 and @xmath299 in agreement with @xcite . notice that none of the correlation functions is falling off exponentially with the correlation length suggested by the transfer matrix @xmath300 , but actually twice as fast . interestingly , it is possible to recover the actual correlation length by adding the string operator between the two end points of the xx or yy correlation function , namely by considering @xmath301 . _ critical point for g=1. _ we simulate the system for @xmath302 . the zz correlation function , discussed in the main text , follows the scheme presented above , as the specific form of the @xmath83 makes all form factors , expect for the ones corresponding to the two - quasiparticle excitations , equal zero . for the xx and yy correlations this is no longer the case , as the gap is equal zero ( @xmath303 ) and the contributions from many - quasiparticles - bands might be relevant for the algebraic scaling . we see that this is indeed the case . the data for growing @xmath41 suggest the scaling @xmath304 , and @xmath305 . due to additional factors of @xmath306 the contribution from the single - particle band is vanishing in the limit of @xmath43 . in order to recover the actual asymptotic behavior @xmath307 and @xmath308 @xcite , the contribution coming from all multi - particle bands would have to be taken into account and the simple single - particle picture presented in this section does no longer apply . _ transfer matrix. _ we define the transfer matrix as @xmath309 , where , for the xy model , matrices @xmath45 are given by eq . , resulting in @xmath310 } ; \ w_2 = \exp{\left [ k_1 \sum_{l=1}^{2l-1 } \tau^x_l \tau^x_{l+1}\right]}. \nonumber\end{aligned}\ ] ] we have reindexed the auxiliary spins along the vertical direction for convenience , so that sites with @xmath311 correspond to @xmath45 and sites with @xmath312 to @xmath313 . operators of this form were diagonalized by abraham @xcite using the formalism of transformation matrices . here we reiterate the main steps of the derivation . firstly , @xmath70 is mapped onto a free - fermionic model by means of a jordan - wigner transformation @xmath314 , @xmath315 , where @xmath316 are fermionic annihilation operators . it is convenient to introduce majorana fermions @xmath317 and @xmath318 , with @xmath319 , where we will use superscript @xmath320 to indicate majorana fermions . now , @xmath321 } , \\ & w_2 = \exp{\left [ k_1 \sum_{l=1}^{2l-1 } i c^m_{2l } c^m_{2l+1}\right]}. \end{aligned}\ ] ] we define a ( row ) vector @xmath322 and for an operator @xmath323 , which is an exponential of a free - fermionic hamiltonian , we consider the similarity transformation @xmath324,\ ] ] which defines a @xmath325 transformation matrix @xmath326 $ ] . above , it is understood that @xmath327 . it is convenient to introduce a @xmath328 matrix @xmath329 the transformation matrices for @xmath330 and @xmath331 are block diagonal @xmath332 = \bigoplus_{n=1}^{2l } u(\overline k_2 ) , \\ & r[w_2 ] = 1 \oplus \left ( \bigoplus_{n=1}^{2l-1 } u(2 \overline k_1 ) \right ) \oplus 1 . \end{aligned}\ ] ] the transformation matrix for @xmath70 is found simply by multiplying transformation matrices for @xmath330 and @xmath331 @xmath333 = r[w_1^{\frac12 } ] r[w_2 ] r[w_1^{\frac12}].\ ] ] subsequently , the transfer matrix is diagonalized by finding @xmath334 which brings @xmath335 $ ] into canonical form @xmath336 = u_\mathcal{t } r_a^m u_\mathcal{t}^t , \\ & r_a = \bigoplus_{n=1}^{2l } u(\epsilon^{\mathcal{t}_f}_n).\end{aligned}\ ] ] this gives single particle energies @xmath337 and @xmath338 defines a new fermionic basis @xmath339 , for which ( up to normalization ) @xmath340 } = \exp { \left[-\sum_{n=1}^{2l}\epsilon_n^{\mathcal{t}_f } a_n^\dagger a_n\right]}.\ ] ] the dominant eigenvector of @xmath70 is annihilated by all annihilation operators @xmath341 . _ reduced density matrix. _ the reduced density matrix @xmath149 of @xmath342 with support on sites @xmath343 is diagonalized following standard techniques @xcite by considering a @xmath344 covariance matrix for a half - chain @xmath345 } = { \left[(\psi_\emptyset| c^m_m c^m_n | \psi_\emptyset)\right]}_{m , n = 1,2 , \dotsc,2l } = \mathbb{1 } + i b^{m},\ ] ] where @xmath346 is skew - symmetric . the reduced density matrix is diagonalized by bringing @xmath347 into canonical form @xmath348 where @xmath155 defines the entanglement spectrum . @xmath349 defines a new set of majorana fermions @xmath350 , for which @xmath351 } = \frac{1}{z } \exp { \left[- 2 \sum_{n=1}^{l } \delta_n f^\dagger_n f_n\right]},\ ] ] where @xmath352 and @xmath353 are the normalization factors . similarly , we obtain @xmath354 by considering the reduced density matrix supported on sites @xmath355 . _ truncation. _ finally , we describe how to obtain the truncated transfer matrix @xmath356 , where @xmath145 are given by eq . and we keep the @xmath157 most relevant fermionic modes , i.e. we discard @xmath357-modes for @xmath358 . we work directly with the transfer matrix and find @xmath359 where states @xmath360 , for which @xmath361 , are obtained from the diagonalization of the reduced density matrix . we use the formalism of the transformation matrix , extending it to the case of non - invertible projections , cf . eq . . first , we obtain the transformation matrix for @xmath70 in the @xmath357-fermionic base ( it is convenient to work with dirac fermions here ) as @xmath362^f,\ ] ] where for convenience we reorder @xmath363 with @xmath364 describing the relevant modes @xmath365 , @xmath366 are annihilation operators corresponding to the truncated modes , and finally @xmath367 denote the corresponding creation operators . @xmath368^f$ ] can be directly obtained from @xmath368 $ ] in eq . by a suitable basis rotation from @xmath369 to @xmath370 . here , the relevant sub - matrices of @xmath368^f$ ] are @xmath371^f\right]}_{m , n = 1,\dotsc 4 \chi } \\ r_{ab } & = { \left[r[\mathcal{t}^f]^f\right]}_{m = 1,\dotsc 4 \chi , n = 4\chi + 1 , \dotsc , 2 l + 2 \chi } \\ r_{ba } & = { \left[r[\mathcal{t}^f]^f\right]}_{m = 4\chi + 1 , \dotsc 2 l + 2 \chi , m = 1,\dotsc 4 \chi } \\ r_{bb } & = { \left[r[\mathcal{t}^f]^f\right]}_{m , n = 4\chi + 1 , \dotsc 2 l+2 \chi } \end{aligned}\ ] ] namely , @xmath372 describes transformation of @xmath364 into @xmath364 under the similarity transformation given by @xmath132 , @xmath373 corresponds to the transformation @xmath364 into @xmath374 , etc . the transformation matrix corresponding to @xmath245 , @xmath375,\ ] ] is found as @xmath376 = r_{aa } - r_{ab } r_{bb}^{-1 } r_{ba}.\ ] ] now , bringing @xmath377 $ ] into canonical form similar to @xmath378 $ ] yields the spectrum @xmath169 , where @xmath379}.\ ] ] where the projection @xmath381 . notice that @xmath382 and @xmath383 . rewriting eq . we obtain @xmath384 eliminating @xmath385 from the above equation leads to @xmath386 now it is enough to notice that @xmath387 and since @xmath388 works nontrivially only on modes @xmath389 with @xmath390 , and @xmath245 on modes with @xmath391 we obtain eq . with @xmath392 $ ] given by eq . . _ form factors in the truncated state. _ we focus on the form factors for the zz correlation function . the ( full ) operator transfer matrix is @xmath393 w_2 w_1^{\frac12},\ ] ] and after mapping onto free - fermionic system we describe it using the transformation matrix @xmath394 $ ] in an analogous way to the transfer matrix @xmath70 . we work directly with this transfer matrix and calculate its form after truncation by using eqs . and finding @xmath395 $ ] in the process . now , in order to find the form factors in eq . we rewrite @xmath396 , \label{eq : fffromfermions}\ ] ] where @xmath397 is projector onto the dominant eigenstate of @xmath166 . we can think about @xmath398 as a reduced density operator and using wick s theorem calculate the above expression ( up to normalization ) as a pfaffian of the two - point correlation matrix @xmath399 , where @xmath400 are all pairs of @xmath401 operators appearing in eq . . they , in turn , can be easily calculated by finding the canonical form of @xmath395 $ ] , calculating the two point correlations in that base and subsequently rotating them into @xmath401 fermions .

is a high - mass x - ray binary with an orbital period of 3.9 days . the mass of the neutron star is as low as , suggesting that it was likely to be formed through an electron - capture supernova rather an iron - core collapse supernova . from the present system configurations , we argue that the orbital period at the supernova was days . since the mass transfer process between the neutron star s progenitor and the companion star before the supernova should have increased the orbital period to tens of days , a mechanism with efficient orbit angular momentum loss and relatively small mass loss is required to account for its current orbital period . we have calculated the evolution of the progenitor binary systems from zero - age main - sequence to the pre - supernova stage with different initial parameters and various mass and angular momentum loss mechanisms . our results show that the outflow from the outer langrangian point or a circumbinary disk formed during the mass transfer phase may be qualified for this purpose . we point out that these mechanisms may be popular in binary evolution and significantly affect the formation of compact star binaries .

1311.1649

high - mass x - ray binaries ( hmxbs ) usually contain an accreting neutron star ( ns ) and an early - type ( o or b ) companion with mass exceeding @xmath2 . the ns is fed by the stellar wind or beginning roche - lobe overflow ( rlof ) from the companion . since the nss in hmxbs have experienced very little accretion because of their young ages , their masses should be very close to those at birth . eclipsing x - ray binary systems where the x - ray source is a pulsar can be ideal systems for a dynamical determination of the ns masses . the measured masses of nss in hmxbs range from @xmath3 for @xcite to @xmath4 for @xcite . recently @xcite presented an improved method for determining the masses of nss in eclipsing x - ray pulsars and applied it to six systems . they found the ns masses of @xmath5 for and @xmath6 for . in this paper , we focus on with a ns mass near the minimum mass limit expected for a ns produced in a supernova ( sn ) @xcite . was first detected during a rocket flight @xcite . the discovery of x - ray eclipses with the _ uhuru _ satellite established the binary nature of . the pulsar has a pulse period of 0.71 s @xcite . the companion to the x - ray pulsar , , is a b0 i supergiant located in the wing " of the small magellanic cloud ( smc ) at a distance of 60 kpc . its mass was estimated to be @xmath7 @xcite , @xmath8@xcite , and @xmath9 @xcite , while more recent work by @xcite gives @xmath10 . the x - ray source exhibits an eclipse duration of @xmath11 day @xcite in a 3.892 day orbit . timing studies of the x - ray pulsations @xcite give @xmath12 ltsec for the projected semi - major axis and indicate a circular orbit with an eccentricity @xmath13 . a decay in the orbital period at a rate @xmath14 was found @xcite , probably due to tidal interaction between the orbit and the rotation of the companion star , which is supposed to be in the hydrogen shell burning phase . the observed period change may also be explained by the self - sustaining mass loss through the outer lagrangain ( @xmath15 ) point @xcite . the x - ray emission from has also been found to exhibit a long quasi - stable super - orbital period of @xmath16 days , which is believed to be a result of obscuration of the ns by a warped , precessing accretion disk @xcite . in x - rays both low- and high - intensity states have been observed with an x - ray luminosity @xmath17 varying from @xmath18 to @xmath19 @xcite . the mass transfer in probably has significant contribution from rlof @xcite , as the stellar winds from are not strong enough to power the x - rays @xcite . the x - ray pulsar in the system has a short spin period ( seconds ) compared to those in wind - fed systems ( minutes ) , as the mass and angular - momentum accretion rate in rlof systems is much higher than in the latter . since its discovery , observations with various x - ray telescopes show a steady spin - up of the ns . this makes an exceptional x - ray pulsar in which no spin - down episode has been observed @xcite . @xcite suggested that the magnetic moment in may be as low as that of the bursting pulsar , i.e. @xmath20 . recent analysis by @xcite and @xcite , like several earlier investigations , found a low value for the ns mass @xmath21 . both theories and statistical analyses suggest that the initial masses of nss might have a bimodal underlying distribution , probably originating from two different formation mechanisms : core collapse supernovae ( ccsne ) and electron - capture supernovae ( ecsne ) ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? a ccsn occurs when a high - mass star develops a degenerate iron core that exceeds the chandrasekhar limit , while an ecsn is associated with the collapse of a lower - mass onemg core as it loses pressure support owing to the sudden capture of electrons by the mg and ne nuclei @xcite . numerical simulations indicate that ecsne are under - energetic compared to ccsne by at least an order of magnitude . however they are , at the same time , relatively fast , neutrino - driven , delayed explosions . the short time available for the propagation of the shock is expected to prevent a strong build - up of core asymmetries , leading to typical kicks of @xmath22 , lower than those in ccsn events @xcite . as ecsne should produce somewhat less massive nss ( @xmath23 ) than ccsne @xcite , the low mass of the ns in would be consistent with its formation by an ecsn in a degenerate onemg core , which is is expected to form at the end of the evolution of stars with masses of @xmath24 @xcite . thus , assuming an explosive mass loss during the formation of the ns of about @xmath25 and a few solar mass of stellar wind mass loss from the ns progenitor , @xcite concluded that a @xmath26 main - sequence progenitor system would be consistent with the present configuration of . the more massive star ( i.e. , the progenitor of the ns ) first evolves to fill its rl , triggering mass transfer to the less massive star . since the initial mass ratio is not extreme , the binary is unlikely to undergo common envelope ( ce ) evolution before the sn that cause its orbit to shrink rapidly ( for reviews of ce evolution , see * ? ? ? * ; * ? ? ? however , a potential problem with such a scenario is that quasi - conservation of mass and orbital angular momentum during the mass transfer would lead to a fairly wide pre - sn system , such that the present orbital period of @xmath27 days would be hard to understand , unless a large amount of orbital angular momentum has been lost with relatively small mass ( at most a few solar masses ) from the system . in this paper , we will try to search plausible evolutionary tracks for , taking into account various kinds of orbital angular momentum loss mechanisms . this paper is organized as follows . in section 2 we describe the stellar evolution code and the angular momentum loss mechanisms used . we present the calculated results of the binary evolution in section 3 . discussion and conclusions are given in section 4 . we adopt an updated version of the stellar evolution code originally developed by @xcite to compute the binary evolution . the binary system is initially composed of two zero - age main - sequence ( zams ) stars with an orbital period @xmath28 . we set the more massive star as the primary ( of mass @xmath29 ) , which fills the rl first and is the progenitor of the ns , and the other one as the secondary ( of mass @xmath30 ) . low metallicities ( @xmath31 and 0.01 ) are taken for each star according to the environment of the smc @xcite . the effective radius @xmath32 of the rl for the primary star is calculated from the @xcite s equation , @xmath33 where @xmath34 is the mass ratio of the binary components and @xmath35 is the orbital separation . the rate of mass transfer via rlof is calculated with @xmath36m_\odot { \rm yr^{-1}}$ ] in the code , where @xmath37 is the radius of the primary , and we adopt @xmath38 in the calculation . we assume that the primary star rotates synchronously with the binary orbital revolution , since the timescale of tidal synchronization is generally much shorter than the characteristic evolutionary timescale of the binaries considered here . we consider two kinds of mechanisms of angular momentum loss . the first is the angular momentum loss due to gravitational radiation . this becomes important when the orbital period is short . the rate of the angular momentum loss is given by @xcite @xmath39 where @xmath40 , @xmath41 , and @xmath42 are the orbital angular momentum , gravitational constant , and speed of light , respectively . the second and more important angular momentum loss mechanism is nonconservative mass transfer . during the mass - transfer processes , part of the transferred mass from the primary star may escape the binary system , carrying away the orbital angular momentum @xcite . this mass loss might be related to the rapid rotation of the accreting star . since the accreting matter carries a certain angular momentum that will be transferred to the secondary @xcite , this angular momentum spins up the top layers of the secondary star , and is transferred further into the star due to rotationally induced mixing processes . when the secondary is spun up to close to critical rotation it starts losing mass due to the influence of centrifugal force ( e.g. * ? ? ? * ) , although it is not clear in which ways the material leaves the binary . considering the complicated processes of mass loss , we adopt four kinds of mass loss as follows . in the first case we assume that a fraction @xmath43 of the transferred mass is ejected out of the binary as isotropic winds from the secondary , carrying away the specific angular momentum of the secondary , @xmath44 where @xmath45 is the angular velocity of the binary . alternatively , the transferred mass might be lost at the inner lagrangian ( @xmath46 ) point and the corresponding rate of angular momentum loss is @xmath47 where @xmath48 is the fraction of mass loss , and @xmath49 is the distance from the @xmath46 point to the center of mass of the binary system . in some cases , the matter surrounding the two components may expand to the outer lagrangian ( @xmath15 ) point , and part of the material lost from the donor star may escape the system through the @xmath15 point @xcite . assuming that a fraction @xmath50 of the matter flow leaves the binary , the angular momentum loss rate due to the @xmath15 point outflow is given by @xmath51 where @xmath52 is the distance between the mass center of binary and the @xmath15 point . in our calculation , we take into account the @xmath15 point outflow after the mass ratio inverts , to ensure that the outflows is on the side of the lower - mass star that fills the rl . the last case is the angular momentum loss due to a circumbinary ( cb ) disk . cb disks were first proposed by @xcite when investigating the evolution of x - ray binaries . recently , cb disks are considered in a wide variety of astrophysical objects , e.g. , young binary stars , protoplanetary systems , cataclysmic variables ( cvs ) and massive binary black hole systems in active galactic nuclei ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . it has been argued that during mass exchange in binary systems , some of the lost matter which possesses high orbital angular momentum may form a disk surrounding the binary system rather than leave the binary system @xcite . as shown by @xcite and @xcite , cb disks can efficiently extract orbit angular momentum from the binary , and enhance the mass transfer rates . in this work , we assume that a fraction @xmath53 of the mass lost from the donor feeds into the cb disk . at the inner edge @xmath54 of the disk tidal torques are then exerted on the binary via gravitational interaction , and the corresponding angular momentum loss rate is @xcite @xmath55 where @xmath56 , @xmath57 @xcite , @xmath58 is the time since the onset of the cb disk formation , and @xmath59 is the viscous timescale at the inner edge of the cb disk . for the standard @xmath43-viscosity disk @xcite , @xmath60 where @xmath61 is the scale height of the disk at the inner edge . in the following calculation we set the viscosity parameter @xmath62 , and assume that the disk is hydrostatically supported , and geometrically thin with @xmath63 @xcite . the formation of a hmxb like smc x-1 requires two initially massive stars . when it evolves to become a supergiant and fill its rl , the primary star transfers material to the secondary on a thermal timescale , and the binary orbit shrinks rapidly at first . because the mass ratio is not far from unity , the mass transfer process is dynamically stable without the occurrence of a ce . a large fraction of the transferred matter is expected to be accreted by the secondary star . when the mass ratio inverts , further mass transfer leads to increase of the orbital period . finally the primary star develops an onemg core , which collapses to be a ns . at this time the primary star is much less massive than the secondary . considering the fact that a small fraction of the binary mass is lost and a small kick is imparted to the newborn ns during the sn , the binary orbit should be nearly circular or mildly eccentric after the sn . during the subsequent evolution , since stellar winds of the main - sequence secondary is not strong , there is hardly any mass transfer until the secondary starts to fill its rl . so we expect that the masses of the ns and the optical companion has changed little since the sn , and will use them to constrain the binary parameters before the sn . it is noted that , after the onset of the secondary rlof , the large mass ratio between the companion star and the ns will cause the orbit to shrink in response to mass transfer , most likely leading to a ce event . when a star explodes as a sn , the ns formed receives a velocity kick due to any asymmetry in the explosion @xcite . this also imparts an impulse to the companion star resulting in an eccentric binary orbit . as mentioned earlier , the low mass of the ns in suggests that its formation is likely to be due to an electron - capture collapse with a low kick velocity , which results in a small eccentricity . tidal interaction between the ns and the companion star can also circularize the binary orbit if it is narrow enough . we first try to estimate the possible distributions of the orbital period before the sn and the eccentricity after the sn from current binary parameters . the ns mass ( @xmath64 ) , the secondary mass ( @xmath65 ) and the orbital period ( @xmath66 d ) have already been estimated from the observations of . the mass of the secondary is assumed to be constant during the sn , and the primary s mass before the sn can be constrained by the range of the helium core mass ( @xmath67 ) that leads to an ecsn @xcite . the sn is thought to take place in a circular orbit because of the previous mass transfer , and the kick direction is assumed to be uniform over all directions . we take the kick velocity @xmath68 from a maxwellian distribution @xmath69 where the velocity dispersion @xmath70 is usually taken to be @xmath71 @xcite or @xmath72 @xcite based on the analysis of various pulsar proper motion samples for ccsn events . the kick velocities must be much lower for ecsne ( e.g. , * ? ? ? * ) . a small kick value ( @xmath73 of the standard ns kick , i.e. , @xmath74 kms@xmath75 ) was adopted by @xcite , in order to estimate the overconcentration of ecsn sources in the smc bar . here we set @xmath76 , 50 , and 70 kms@xmath75 to examine the effect of ecsn kicks . figure 1 shows the derived distributions of the pre - sn orbital period ( top panel ) and the post - sn eccentricity ( middle panel ) for a @xmath77 pre - sn system , with the assumption that the post - sn period is @xmath78 days . the results with the ccsn kicks are also presented for comparison . with the increase of @xmath70 , a more extended range of the pre - sn orbital period can result in the 4 day post - sn period , and the post - sn eccentricity tends to be larger . for sufficiently large kicks the eccentricity can be larger than unity so that the binary is disrupted . for ccsne this occurs at the high - end in the distribution of the pre - sn period , providing an upper limit of @xmath79 days . this means that a @xmath80 day pre - sn binary would not evolve to a 4 day post - sn binary with the mass parameters . from the contours in the orbital period vs. eccentricity plane ( the bottom panel ) , one can see the trend of pre - sn period and post - sn eccentricity distributions . there are two wings " with the center close to the post - sn period . the most probable regions vary from the center to the wings when @xmath81 is increasing . for ccsn kicks , to form the expected post - sn systems like requires a pre - sn period @xmath82 days , and the post - sn eccentricity is high ( @xmath83 ) . for ecsne a lower eccentricity is expected and a relatively longer pre - sn period ( @xmath84 ) is required . following the above analysis , we simplify the formation of to be the evolution of a zams binary system to a pre - sn binary with the orbital period @xmath84 , the primary ( he star ) s mass @xmath85 , and the secondary s mass @xmath86 . we select two sets of mass parameters for the progenitor binary to illustrate the possible evolutionary tracks of . they are @xmath87 and @xmath88 zams binaries with similar initial total masses for the conservative evolution . we adjust the initial mass of the secondary if the evolutionary model is nonconservative , in order to keep the secondary mass at the sn within the suitable range . we first consider the evolution of the progenitor system with the initial primary s mass @xmath90 . to examine the possible evolutionary tracks and the influence of different angular momentum loss mechanisms , we construct five models with various kinds of mass and angular momentum loss mentioned in section 2.2 : ( 1 ) conservative mass transfer ; ( 2 ) nonconservative mass transfer with mass loss from the secondary ; ( 3 ) nonconservative mass transfer with mass loss from the @xmath46 point ; ( 4 ) nonconservative mass transfer with mass loss from the @xmath15 point ; and ( 5 ) nonconservative mass transfer with a small fractional mass feeding into the cb disk . the initial orbit periods are taken to be @xmath91 , 2.0 , 2.5 , 3.0 , 3.5 , 4.0 , 5.0 , 7.0 , and 10.0 days . at the smallest orbital period the zero - age primary star just fits the size of the rl . the metallicities are adopted to be @xmath92 and 0.004 . the results of the pre - sn evolution for different initial orbital periods are summarized in tables 1 ( for models [ 1]-[4 ] ) , 2 , and 3 ( for model [ 5 ] ) . in the tables @xmath93 , @xmath94 , and @xmath95 are the primary ( he star ) s mass , the secondary s mass and the orbital period close to the end of the primary s evolution , respectively . in table 1 we distinguish cases a and b mass transfer , which occurs when the primary burns hydrogen in its core and evolves off the main - sequence but before helium core ignition , respectively . the case b mass transfer is most likely to be unstable if the value of @xmath50 is high for model ( 4 ) . this is what divergent " means in the tables . so we only present the results of model ( 2)-(4 ) with case a and stable case b mass transfer . the results of the first three models in table 1 show little distinction . the orbital angular momentum loss from the binaries is all inefficient , so that the final orbital periods are usually larger than 40 days ( for @xmath92 ) or 25 days ( for @xmath97 ) , inconsistent with those for . it is also noted that a larger @xmath98 leads to longer @xmath99 because of longer mass transfer time . for each model , the higher @xmath100 , the heavier the evolved he star and the lighter the secondary . it is interesting to note that @xmath99 decreases with increasing @xmath100 for case a mass transfer , but increases with @xmath100 for case b mass transfer . this means the lowest pre - sn period @xmath99 corresponds to the @xmath100 around the boundary between case a and case b evolution . different from the first three models , in model ( 4 ) the mass loss from the @xmath15 point can cause secular orbital shrinking . if the mass loss fraction @xmath101 , @xmath99 becomes shorter than 10 days , which satisfies what is needed for the formation of . although the @xmath15 point outflow provides a way for the formation of relatively compact systems . this mechanism requires a considerable mass loss rate , and sometimes the primary s mass at the end of the evolution is lower than the range for an ecsn . thus in model ( 5 ) we take into account the influence of a cb disk formed during the mass transfer . in this case we adopt the progenitor binary system of @xmath102 with the secondary s mass slightly higher than that in model ( 1 ) , and we only calculate the evolution with @xmath100 near the the boundary between case a and case b mass transfer . similar as in model ( 4 ) , the case b mass transfer becomes unstable if @xmath53 is large , so we mainly focus on case a evolution . the results are presented in tables 2 and 3 with @xmath92 and @xmath97 , respectively . we can always find that it is able to form a pre - sn binary system with a relatively small @xmath53 ( a few hundredth ) , which meets the requirements . the wider the initial system , the higher @xmath53 needed . we then select some representative scenarios from the tables to demonstrate the detailed evolution of the binary . figure 2 shows the pre - sn evolutionary tracks of the primary of initial mass @xmath103 in the h - r diagram . the panels from top to bottom correspond to models ( 1)-(5 ) , respectively . in the left , middle and right columns the initial orbital period @xmath100 is taken to be 2.0 , 2.5 , and 3.0 days , respectively . for model ( 1 ) , the initial secondary s mass is @xmath104 ; for models ( 2)-(4 ) , we adopt @xmath105 ; for model ( 5 ) , @xmath106 , and @xmath107 . the solid and dotted lines are obtained with @xmath108 and 0.004 , respectively . in fig . 3 , we compare the mass transfer rates between different models with different @xmath109 . the figure is organized same as fig . 2 . the only case b evolution is for @xmath110 days and @xmath111 , and in other scenarios are all case a evolutions . we can see that there are not any significant differences in the evolutionary tracks and in the mass transfer rates among the first three models with the same @xmath109 , while the other two models are distinct from the others . the binaries generally experience three ( or two , the first two phases are sometimes degenerated ) phases of mass transfer in case a evolution in fig . 3 . in the first phase the mass transfer proceeds on a thermal timescale ( see * ? ? ? * for further discussions on thermally unstable mass transfer ) . the outcome in this situation is that the donor star overfills its rl even more , leading to further mass loss . although the star is driven out of thermal equilibrium during this phase , it manages to retain hydrostatic equilibrium and the system can in this case avoid a so - called delayed dynamical instability @xcite , which would have resulted in a ce stage . the second and third phases proceed on nuclear timescales determined by the core burning of the remaining hydrogen and , later on , the hydrogen shell burning , respectively . obviously the wider the initial system , the later the rlof mass transfer begins , because there is no effective angular momentum loss mechanism before the onset of mass transfer , and the mass transfer timescale becomes shorter . when @xmath109 is so long that there is case b mass transfer , the three phases merge into one . furthermore , lower metallicity causes later onset of rlof and shorter mass transfer time due to a larger initial stellar radius and ending the red giant stage earlier . for models ( 4 ) and ( 5 ) , if @xmath50 or @xmath53 is high enough , the mass transfer can be dynamically unstable , so that the binary enters the ce evolution when the orbital period is as low as @xmath112 days . the orbital period evolution is shown in fig . 4 , which is also organized similar as fig . 2 , but in each subgraph we add the cases with different values of the mass loss fraction in different colored lines for comparison . the dashed horizontal lines indicate the possible range of the pre - sn periods expected to lead to the formation of . it is clearly seen that only in model ( 4 ) or ( 5 ) with suitable values of @xmath50 or @xmath53 can the binary evolve into the proper orbital period . in this subsection we investigate the evolution of the possible progenitor of with a @xmath114 main - sequence primary . tables 4 , 5 and 6 present the results of the binary parameters for the primary evolving from zams to pre - sn in different models , similar as those for the @xmath103 primary . the orbital period evolution is shown in fig . 5 . compared with the case of @xmath90 , the higher mass ratio for @xmath115 leads to shorter @xmath99 for similar total mass , initial orbital period and metallicity . usually models ( 4 ) and ( 5 ) are more preferred , while in the case of conservative mass transfer , the @xmath99 sometimes nearly fits in the pre - sn period range for the formation of , but the he star mass exceeds the upper limit of the he core mass for an ecsn . combining the calculated results of binary evolutions with different initial parameters , we can roughly constrain the mass of the ns progenitor in to be @xmath116 . since the pioneering work of @xcite the formation and evolution of hmxbs with ns components has been intensively studied @xcite . the generally accepted evolutionary picture is that , the originally more massive star becomes the less massive as a result of mass transfer or mass loss , leaving behind a system composed of an evolved core and a ms companion . the subsequent evolution of the core to the sn leads to the formation of a ns / massive star binary system . however , there are big uncertainties associated with the effects of mass and angular momentum loss on the evolution of these systems ( in the context of the mass transfer and ce evolution ) . detailed investigations on the formation of individual hmxbs are also lacking because it is difficult to constrain the initial binary parameters from current observations . the low mass of the ns in is suggestive of its formation through an ecsn , which provides useful information on the properties of the ns progenitor and the formation scenarios . ecsne are thought to occur in stars of mass @xmath117 . combining with the total mass of the current binary , this suggests that the initial masses of the primary and secondary stars are comparable so that ce evolution is not likely to occur , and the orbital evolution is mainly dominated by the mass transfer / loss processes and tidal interaction . if the mass transfer is conservative before the sn , the ratio between the final and initial orbital periods is @xmath118 this gives an orbital period at least tens of days when the sn occurs ( the orbit may shrink to some extent due to the wind from the primary ) , and in the further evolution it would be difficult for the orbital period to decrease to @xmath78 days by tidal torques within the lifetime of the secondary star . this point has already been notified by @xcite to explain why be / x - ray binaries are in wide ( @xmath119 days ) and eccentric orbits . for , as the orbital change during the post - sn evolution is limited , through numerical calculation of the change in the orbital period during the sn , we simplify the evolutional problem to evolving a progenitor zams binary system up to the pre - sn epoch , when it has developed an onemg core approximately fitting the range expected to lead to an ecsn , with the orbital period @xmath120 days , and the secondary mass @xmath121 . obviously these requirements rule out the conservative mass transfer scenario , and the key problem is to find an effective orbital angular loss mechanism with relatively small mass loss to cause secular orbital shrinking . according to our calculation , it could be either mass loss through the @xmath15 point or or a cb disk ( note that the formation of a cb disk might also originate from the outflow from the @xmath15 point ) . the difference is that , in the former case some of the material leaves the system carrying away the orbital angular momentum , while in the latter the lost mass accumulates in a disk surrounding the binary , draining the orbital angular momentum from the binary through tidal torques . because of the difference in the efficiency of angular momentum loss , the fraction of lost mass in the total transferred mass is significantly higher for the @xmath15 point outflow than for the cb disk . our calculated results indicate that the evolution with the @xmath15 point mass loss or a cb disk could lead to the formation of through case a mass transfer ( @xmath122 days ) with @xmath123 . although our analysis focuses on the formation of , the conclusion that mass transfer is likely to be associated with significant angular momentum loss may be a general feature in the evolution of various kinds of binaries ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? previous investigations emphasized the necessity of nonconservative mass transfer to account for the properties of the related binaries . however , the ways of mass loss and the corresponding efficiency of angular momentum loss have not been well constrained . in this work we show that the outflow from the @xmath15 point or a circumbinary disk is more preferred than mass loss from the accreting star or the @xmath46 point , at least for the formation of . similar suggestions can be found in the literature . for example , in order to address the observed lower mass limit ( @xmath124 ) for the be stars in be / x - ray binaries , @xcite proposed that there may be mass loss from the binary systems at the @xmath15 point when the primary star transfers mass to the secondary ( who s rotation will be accelerated such that it will become a be star ) . the escaped matter takes away about six times the specific angular momentum of the binary system , so that those with small initial mass ratios would undergo spiral - in and evolve towards a ce phase . in the other extreme of low - mass binaries , @xcite recently reconstructed the complete evolutionary path followed by cataclysmic variables ( cvs ) , using the observed mass - radius relationship of the secondary stars . the best - fit revised model of cv evolution indicates that the angular momentum loss rate below the period gap is @xmath125 times the rate induced by gravitational radiation , suggesting the existence of some other angular momentum loss mechanisms . @xcite considered several kinds of consequential angular momentum loss mechanisms including isotropic wind from the accreting white dwarfs , outflows from the langrangian points , and the formation of a cb disk . they showed that neither isotropic wind from the white dwarf nor outflow from the @xmath46 point can explain the extra angular momentum loss rate , while outflow from the @xmath15 point or a cb disk can satisfy the extra angular momentum loss provided that @xmath126 of the transferred mass is lost from the binary , or @xmath127 of the transferred mass goes into the cb disk ( see also * ? ? ? these results are in general line with what we have obtained for . mass transfer critically affects the evolutionary paths of binaries . depending on whether it is stable or not , a binary may either survive the initial mass transfer phase to become a semi - detached system or end up merging completely . the stability of mass transfer depends sensitively on the angular momentum transport mechanisms . including the mass loss from the @xmath15 point ( or a cb disk ) can significantly destabilize the mass transfer processes , enhance the merging rate , and influence the birth rate of compact star binaries . obviously a thorough investigation on this subject is needed before it can be implemented into future population synthesis calculation . this work was supported by the natural science foundation of china under grant numbers 11133001 and 11333004 , the national basic research program of china ( 973 program 2009cb824800 ) , and the qinglan project of jiangsu province . artymowicz , p. , & lubow , s. h. 1994 , , 421 , 651 barziv , o. , kaper , l. , van kerkwijk , m. h. , telting , j. h. , & van paradijs , j. 2001 , , 377 , 925 belle , k. e. , sanghi , n. , howell , s. b. , holberg , j. b. , & williams , p. t. 2004 , , 128 , 448 chen , w .- c . & li , x .- d . 2006 , , 373 , 305 clarkson , w. i. , charles , p. a. , coe , m. j. , et al . 2003 , , 339 , 447 dessart , l. , burrows , a. , ott , c. d. , et al . 2006 , , 644 , 1063 dalton , w. w. & sarazin , c. l. , 1995 , , 440 , 280 drechsel , h. , haas , s. , lorenz , r. , & gayler , s. 1995 , , 294 , 723 eggleton , p. p. 1971 , , 151 , 351 eggleton , p. p. 1972 , , 156 , 361 eggleton , p. p. 1983 , , 268 , 368 eggleton , p. 2000 , , 44 , 111 haensel , p. , zdunik , j. l. , & douchin , f. 2002 , , 385 , 301 hammerschlag - hensberge , g. , kallman , t. r. , & howarth , i. d. 1984 , , 283 , 249 han , z. , podsiadlowski , p. , & eggleton , p. p. 1994 , , 270 , 121 hansen , b. m. s. , & phinney , e. s. 1997 , , 291 , 569 hayasaki , k. , & okazaki , a. t. 2009 , , 691 , l5 hilditch , r. w. , howarth , i. d. , & harries , t. j. 2005 , , 357 , 304 hjellming , m. s. , & webbink , r. f. 1987 , , 318 , 794 hobbs , g. , lorimer , d. r. , lyne , a. g. , & kramer , m. 2005 , , 360 , 974 iben , i. , jr , tutukov , a. v. , & yungelson , l. r . , 1995 , , 100 , 217 ivanova , n. , justham , s. , chen , x. , et al . 2013 , , 21 , 59 kahabka , p. , & li , x .- d . 1999 , , 345 , 117 kalogera , v. , & webbink , r. f. 1996 , , 458 , 301 kitaura , f. s. , janka , h .- t . , & hillebrandt , w. 2006 , , 450 , 345 kiziltan , b. , kottas , a. , & thorsett , s. e. 2010 , arxiv:1011.4291 knigge , c. , baraffe , i. , & patterson , j. 2011 , , 194 , 28 landau , l. d. , & lifshitz , e. m. 1975 , course of theoretical physics - pergamon international library of science , technology , engineering and social studies , oxford : pergamon press , 1975 , 4th rev.engl.ed . langer , n. 1998 , , 329 , 591 langer , n. , deutschmann , a. , wellstein , s. , hflich , p. 2000 , , 362 , 1046 lattimer , j. m. , & prakash , m. 2004 , science , 304 , 536 levine , a. , rappaport , s. , deeter , j. e. , boynton , p. e. , & nagase , f. 1993 , , 410 , 328 li , x .- d . , & van den heuvel , e. p. j. 1997 , , 322 , l9 liller , w. 1973 , , 184 , l37 linden , t. , sepinsky , j. f. , kalogera , v. , & belczynski , k. 2009 , , 699 , 1573 lipunov , v. m. , postnov , k. a. , prokhorov , m. e. , & osminkin , e. yu . 1994 , , 423 , l121 l , g. , zhu , c. , & wang , z. 2011 , raa , 11 , 327 lucke , r. , yentis , d. , friedman , h. , fritz , g. , & shulman , s. 1976 , , 206 , l25 lyne , a. g. , & lorimer , d. r. 1994 , , 369 , 127 massevitch , a. & yungelson , l. 1975 , mmsai , 46 , 217 meurs , e. j. & van den heuvel , e. p. j. , 1989 , , 226 , 88 nieuwenhuijzen , h. , & de jager , c. 1990 , , 231 , 134 nomoto , k. 1984 , , 277 , 791 zel , f. , psaltis , d. , narayan , r. , & santos villarreal , a. 2012 , , 757 , 55 packet , w. 1981 , , 102 , 17 petrovic , j. , langer , n. , & van der hucht , k. a. 2005 , , 435 , 1013 pfahl , e. , rappaport , s. , podsiadlowski , p. , & spruit , h. 2002 , , 574 , 364 phillips , s. n. , & podsiadlowski , p. 2002 , , 337 , 431 piatti , a. e. 2012 , , 422 , 1109 podsiadlowski , p. , dewi , j. d. m. , lesaffre , p. , et al . 2005 , , 361 , 1243 podsiadlowski , p. , langer , n. , poelarends , a. j. t. , et al . 2004 , , 612 , 1044 pols , o. r. , cote , j. waters , l. b. f. m. , & heise , j. 1991 , , 241 , 419 pols , o. r. , tout , c. a. , eggleton , p. p. , & han , z. 1995 , , 274 , 964 portegies zwart , s. f. , 1995 , , 296 , 691 price , r. e. , groves , d. j. , rodrigues , r. m. , et al . 1971 , , 168 , l7 primini , f. , clark , g. w. , lewin , w. , et al . 1976 , , 210 , l71 quaintrell , h. , norton , a. j. , ash , t. d. c. , et al . 2003 , , 401 , 313 raguzove , n. v. & lipunov , v. m. 1998 , , 340 , 85 rappaport , s. & van den heuvel , e. p. j. 1982 , in the be stars , ed . m. jaschek & h. groth ( dordrecht : reidel ) , 327 rawls , m. l. , orosz , j. a. , mcclintock , j. e. , et al . 2011 , , 730 , 25 reynolds , a. p. , hilditch , r. w. , bell , s. a. , & hill , g. 1993 , , 261 , 337 refsdal , s. , roth , m. , & weigert , a. 1974 , , 36 , 113 sarna , m. 1993 , , 262 , 534 schreier , e. , giacconi , r. , gursky , h. , kellogg , e. , & tananbaum , h. 1972 , , 178 , l71 schwab , j. , podsiadlowski , p. , & rappaport , s. 2010 , , 719 , 722 shakura , n. i. & sunyaev , r. a. , 1973 , , 24 , 337 shao , y. & li , x .- d . 2012 , , 745 , 165 spruit , h. c. , & taam , r. e. 2001 , , 548 , 900 taam , r. e. , & sandquist , e. l. 2000 , , 38 , 113 terman , j. l. , taam , r. e. , & savage , c. o. 1998 , , 293 , 113 val baker , a. k. f. , norton , a. j. , & quaintrell , h. 2005 , , 441 , 685 vanbeveren , d. , de loore , c. , & van rensbergen , w. 1998 , , 9 , 63 van den heuvel , e. p. j. 1994 , saas - fee advanced course 22 : interacting binaries , 263 van den heuvel , e. p. j. 2004 , 5th integral workshop on the integral universe , 552 , 185 van den heuvel , e. p. j. , & de loore , c. 1973 , , 25 , 387 van den heuvel , e. p. j. & heise , j. , 1972 , nature phys . , 239 , 67 van der meer , a. , kaper , l. , van kerkwijk , m. h. , heemskerk , m. h. m. , & van den heuvel , e. p. j. 2007 , , 473 , 523 van paradijs , j. , & kuiper , l. 1984 , , 138 , 71 van rensbergen , w. , de greve , j. p. , de loore , c. , & mennekens , n. 2008 , , 487 , 1129 warner , b. 1978 , , 28 , 303 wellstein , s. , langer , n. , & braun , h. 2001 , , 369 , 939 wojdowski , p. , clark , g. w. , levine , a. m. , woo , j. w. , & zhang , s. n. 1998 , , 502 , 253 zhang , c. m. , wang , j. , zhao , y. h. , et al . 2011 , , 527 , a83 cccccccccccc 1.5 & 8.02 & 0 & 1 & a & 1.53 & 16.47 & 48.38 & a & 1.85 & 16.15 & 28.70 + & 8.62 & 0.10 & 2 & a & 1.57 & 16.19 & 51.42 & a & 1.90 & 15.89 & 30.82 + & & & 3 & a & 1.59 & 16.17 & 47.63 & a & 1.92 & 15.87 & 29.40 + & & & 4 & a & 1.38 & 16.43 & 24.93 & a & 1.69 & 16.16 & 15.94 + & 9.02 & 0.15 & 2 & a & 1.60 & 16.14 & 53.29 & a & 1.93 & 15.86 & 32.11 + & & & 3 & a & 1.63 & 16.12 & 47.55 & a & 1.96 & 15.84 & 29.86 + & & & 4 & a & 1.29 & 16.48 & 17.68 & a & 1.59 & 16.22 & 11.64 + & 9.62 & 0.25 & 2 & a & 1.65 & 15.87 & 56.23 & a & 1.98 & 15.62 & 34.06 + & & & 3 & a & 1.69 & 15.83 & 47.02 & a & 2.03 & 15.58 & 30.33 + & & & 4 & a & 1.07 & 16.39 & 8.50 & a & 1.36 & 16.16 & 5.82 + 2.0 & 8.02 & 0 & 1 & a & 1.75 & 16.25 & 44.30 & a & 2.15 & 15.85 & 25.82 + & 8.62 & 0.10 & 2 & a & 1.79 & 15.99 & 48.25 & a & 2.19 & 15.64 & 28.48 + & & & 3 & a & 1.81 & 15.98 & 45.95 & a & 2.20 & 15.63 & 27.98 + & & & 4 & a & 1.63 & 16.21 & 23.06 & a & 2.00 & 15.87 & 14.74 + & 9.02 & 0.15 & 2 & a & 1.81 & 15.96 & 50.81 & a & 2.21 & 15.63 & 30.17 + & & & 3 & a & 1.84 & 15.94 & 47.06 & a & 2.22 & 15.61 & 29.29 + & & & 4 & a & 1.55 & 16.26 & 16.38 & a & 1.91 & 15.95 & 10.92 + & 9.62 & 0.25 & 2 & a & 1.85 & 15.72 & 54.55 & a & 2.24 & 15.43 & 32.93 + & & & 3 & a & 1.89 & 15.69 & 48.04 & a & 2.24 & 15.43 & 32.19 + & & & 4 & a & 1.36 & 16.14 & 7.66 & a & 1.69 & 15.88 & 5.51 + 2.5 & 8.02 & 0 & 1 & a & 1.96 & 16.04 & 41.43 & a & 2.39 & 15.61 & 24.77 + & 8.62 & 0.10 & 2 & a & 1.98 & 15.82 & 45.94 & a & 2.41 & 15.43 & 27.71 + & & & 3 & a & 1.99 & 15.81 & 44.63 & a & 2.41 & 15.43 & 27.67 + & & & 4 & a & 1.84 & 16.01 & 21.94 & a & 2.28 & 15.62 & 14.02 + & 9.02 & 0.15 & 2 & a & 2.00 & 15.80 & 48.84 & a & 2.42 & 15.44 & 29.64 + & & & 3 & a & 2.02 & 15.79 & 46.61 & a & 2.43 & 15.44 & 29.47 + & & & 4 & a & 1.78 & 16.07 & 15.72 & a & 2.18 & 15.73 & 10.71 + & 9.62 & 0.25 & 2 & a & 2.03 & 15.58 & 53.53 & a & 2.44 & 15.27 & 32.86 + & & & 3 & a & 2.06 & 15.56 & 49.25 & a & 2.46 & 15.26 & 32.34 + & & & 4 & a & 1.59 & 15.96 & 7.53 & a & 2.01 & 15.73 & 5.47 + 3.0 & 8.02 & 0 & 1 & a & 2.15 & 15.85 & 38.63 & b & 2.63 & 15.37 & 23.27 + & 8.62 & 0.10 & 2 & a & 2.16 & 15.65 & 43.83 & b & 2.67 & 15.20 & 25.54 + & & & 3 & a & 2.17 & 15.65 & 43.33 & b & 2.67 & 15.20 & 25.89 + & & & 4 & a & 2.05 & 15.82 & 20.90 & b & 2.61 & 15.32 & 12.83 + & 9.02 & 0.15 & 2 & a & 2.18 & 15.65 & 46.92 & b & 2.68 & 15.23 & 27.50 + & & & 3 & a & 2.18 & 15.65 & 45.92 & b & 2.68 & 15.22 & 27.94 + & & & 4 & a & 1.99 & 15.90 & 15.34 & b & 2.53 & 15.43 & 9.84 + & 9.62 & 0.25 & 2 & a & 2.19 & 15.46 & 52.20 & b & 2.69 & 15.09 & 30.82 + & & & 3 & a & 2.21 & 15.45 & 49.82 & b & 2.69 & 15.09 & 31.39 + & & & 4 & a & 1.83 & 15.79 & 7.42 & b & & @xmath96 divergent & + 3.5 & 8.02 & 0 & 1 & a & 2.36 & 15.64 & 35.83 & b & 2.68 & 15.32 & 26.00 + & 8.62 & 0.10 & 2 & a & 2.36 & 15.48 & 41.01 & b & 2.69 & 15.18 & 29.44 + & & & 3 & a & 2.36 & 15.48 & 40.94 & b & 2.69 & 15.18 & 29.86 + & & & 4 & a & 2.28 & 15.63 & 19.70 & b & 2.63 & 15.30 & 14.73 + & 9.02 & 0.15 & 2 & a & 2.36 & 15.50 & 44.28 & b & 2.69 & 15.22 & 31.71 + & & & 3 & a & 2.37 & 15.49 & 43.99 & b & 2.69 & 15.21 & 32.24 + & & & 4 & a & 2.23 & 15.68 & 14.33 & b & 2.59 & 15.38 & 11.01 + & 9.62 & 0.25 & 2 & a & 2.36 & 15.33 & 49.63 & b & 2.70 & 15.08 & 35.55 + & & & 3 & a & 2.37 & 15.33 & 49.11 & b & 2.70 & 15.08 & 36.24 + & & & 4 & a & 2.10 & 15.51 & 7.19 & b & & @xmath96 divergent & + 4.0 & 8.02 & 0 & 1 & b & 2.52 & 15.48 & 34.49 & b & 2.69 & 15.31 & 29.41 + & 8.62 & 0.10 & 2 & b & 2.53 & 15.33 & 39.16 & b & 2.70 & 15.17 & 33.32 + & & & 3 & b & 2.53 & 15.33 & 39.51 & b & 2.70 & 15.17 & 33.80 + & & & 4 & b & 2.47 & 15.44 & 18.99 & b & 2.64 & 15.29 & 16.67 + & 9.02 & 0.15 & 2 & b & 2.53 & 15.35 & 42.26 & b & 2.70 & 15.21 & 35.89 + & & & 3 & b & 2.53 & 15.35 & 42.63 & b & 2.70 & 15.20 & 36.50 + & & & 4 & b & 2.44 & 15.50 & 13.97 & b & 2.61 & 15.36 & 12.47 + & 9.62 & 0.25 & 2 & b & 2.54 & 15.20 & 47.52 & b & 2.71 & 15.07 & 40.24 + & & & 3 & b & 2.54 & 15.20 & 47.81 & b & 2.71 & 15.07 & 41.06 + & & & 4 & b & & @xmath96 divergent & & b & & @xmath96 divergent & + 5.0 & 8.02 & 0 & 1 & b & 2.54 & 15.46 & 42.47 & b & 2.71 & 15.29 & 36.16 + 7.0 & 8.02 & 0 & 1 & b & 2.56 & 15.44 & 58.20 & b & 2.73 & 15.27 & 49.43 + 10.0 & 8.02 & 0 & 1 & b & 2.58 & 15.42 & 81.51 & b & 2.76 & 15.24 & 69.03 + cccccc 2.0 & 5 & 0.01 & 1.61 & 16.42 & 15.02 + & & 0.015 & 1.51 & 16.48 & 6.76 + & & 0.02 & 1.34 & 16.60 & 1.91 + 2.5 & 5 & 0.015 & 1.81 & 16.18 & 13.58 + & & 0.02 & 1.75 & 16.21 & 8.06 + & & 0.025 & 1.66 & 16.26 & 4.13 + & & 0.03 & 1.53 & 16.34 & 1.52 + 3.0 & 5 & 0.02 & 2.03 & 15.93 & 14.03 + & & 0.025 & 1.99 & 15.93 & 10.15 + & & 0.03 & 1.94 & 15.95 & 6.98 + & & 0.035 & 1.87 & 15.98 & 4.48 + & & 0.04 & 1.79 & 16.02 & 2.49 + 3.5 & 5 & 0.03 & 2.27 & 15.63 & 13.55 + & & 0.04 & 2.23 & 15.60 & 10.06 + & & 0.05 & 2.18 & 15.58 & 6.70 + & & 0.06 & 2.12 & 15.57 & 4.20 + & & 0.07 & 2.04 & 15.58 & 2.42 + cccccc 2.0 & 5 & 0.005 & 2.09 & 15.98 & 18.44 + & & 0.01 & 2.01 & 16.02 & 12.36 + & & 0.015 & 1.92 & 16.08 & 7.66 + & & 0.02 & 1.80 & 16.15 & 4.13 + & & 0.025 & 1.64 & 16.27 & 1.88 + 2.5 & 5 & 0.01 & 2.32 & 15.72 & 15.85 + & & 0.02 & 2.23 & 15.74 & 9.17 + & & 0.03 & 2.10 & 15.79 & 4.54 + & & 0.04 & 1.89 & 15.92 & 1.52 + 3.0 & 5 & 0.04 & 2.62 & 15.22 & 13.57 + & & 0.05 & 2.60 & 15.18 & 11.77 + & & 0.06 & 2.58 & 15.14 & 10.14 + & & 0.07 & 2.55 & 15.09 & 8.64 + & & 0.08 & 2.54 & 15.04 & 7.31 + & & 0.09 & & @xmath96 divergent & + cccccccccccc 2.0 & 5.60 & 0 & 1 & a & 2.11 & 16.05 & 17.79 & a & 2.70 & 15.46 & 9.98 + & 7.00 & 0.10 & 2 & a & 2.25 & 16.28 & 23.14 & a & 2.79 & 15.79 & 13.38 + & & & 3 & a & 2.33 & 16.21 & 22.14 & a & 2.83 & 15.75 & 13.72 + & & & 4 & a & 2.09 & 16.71 & 11.82 & a & 2.56 & 16.26 & 7.50 + & 8.00 & 0.15 & 2 & a & 2.37 & 16.66 & 25.97 & a & 2.88 & 16.23 & 15.85 + & & & 3 & a & 2.42 & 16.62 & 25.81 & a & 2.92 & 16.19 & 16.37 + & & & 4 & a & 2.05 & 17.27 & 9.57 & a & 2.51 & 16.88 & 6.61 + & 10.0 & 0.25 & 2 & a & 2.51 & 17.54 & 33.54 & a & 3.00 & 17.17 & 21.06 + & & & 3 & a & 2.59 & 17.48 & 31.86 & a & 3.07 & 17.11 & 21.26 + & & & 4 & a & 1.87 & 18.36 & 5.69 & a & 2.30 & 18.10 & 4.20 + & 11.50 & 0.50 & 2 & a & 2.64 & 16.46 & 39.75 & a & 3.13 & 16.21 & 25.43 + & & & 3 & a & 2.80 & 16.38 & 36.58 & a & 3.29 & 16.14 & 25.72 + & & & 4 & a & & @xmath96 divergent & & a & & @xmath96 divergent & + 2.5 & 5.60 & 0 & 1 & a & 2.46 & 15.70 & 15.14 & a & 2.93 & 15.23 & 9.34 + & 7.00 & 0.10 & 2 & a & 2.60 & 15.96 & 19.94 & a & 3.17 & 15.45 & 12.24 + & & & 3 & a & 2.63 & 15.93 & 20.52 & a & 3.20 & 15.43 & 12.89 + & & & 4 & a & 2.41 & 16.41 & 10.90 & a & 2.95 & 15.92 & 7.20 + & 8.00 & 0.15 & 2 & a & 2.66 & 16.42 & 24.20 & a & 3.22 & 15.94 & 14.96 + & & & 3 & a & 2.70 & 16.38 & 24.82 & a & 3.25 & 15.92 & 16.01 + & & & 4 & a & 2.38 & 17.00 & 9.11 & a & 2.88 & 16.57 & 6.51 + & 10.00 & 0.25 & 2 & a & 2.77 & 17.34 & 32.39 & a & 3.31 & 16.93 & 20.61 + & & & 3 & a & 2.83 & 17.30 & 32.27 & a & 3.35 & 16.90 & 21.73 + & & & 4 & a & 2.20 & 18.90 & 5.58 & a & 2.72 & 17.73 & 4.31 + & 11.50 & 0.50 & 2 & a & 2.87 & 16.34 & 39.90 & a & 3.39 & 16.09 & 25.94 + & & & 3 & a & 2.99 & 16.28 & 39.39 & a & 3.48 & 16.04 & 28.42 + & & & 4 & a & & @xmath96 divergent & & a & & @xmath96 divergent & + 3.0 & 5.60 & 0 & 1 & a & 2.75 & 15.41 & 13.67 & a & 3.23 & 14.93 & 8.93 + & 7.00 & 0.10 & 2 & a & 2.88 & 15.71 & 18.60 & a & 3.43 & 15.22 & 12.19 + & & & 3 & a & 2.90 & 15.69 & 19.50 & a & 3.44 & 15.20 & 13.02 + & & & 4 & a & 2.71 & 16.15 & 10.39 & a & 3.28 & 15.63 & 7.21 + & 8.00 & 0.15 & 2 & a & 2.92 & 16.20 & 22.93 & a & 3.45 & 15.74 & 15.19 + & & & 3 & a & 2.95 & 16.17 & 24.07 & a & 3.47 & 15.72 & 16.43 + & & & 4 & a & 2.67 & 16.75 & 8.93 & a & 3.24 & 16.26 & 6.43 + & 10.00 & 0.25 & 2 & a & 3.00 & 17.17 & 31.77 & a & 3.51 & 16.79 & 21.35 + & & & 3 & a & 3.03 & 17.15 & 32.91 & a & 3.54 & 16.77 & 23.09 + & & & 4 & a & 2.50 & 17.88 & 5.69 & a & 3.07 & 17.46 & 4.43 + & 11.50 & 0.50 & 2 & a & 3.05 & 16.25 & 40.53 & a & 3.55 & 16.00 & 27.55 + & & & 3 & a & 3.16 & 16.20 & 41.99 & a & 3.62 & 15.97 & 31.20 + & & & 4 & a & & @xmath96 divergent & & a & & @xmath96 divergent & + 3.5 & 5.60 & 0 & 1 & a & 3.01 & 15.15 & 12.74 & b & 3.48 & 14.68 & 8.81 + & 7.00 & 0.10 & 2 & a & 3.11 & 15.51 & 17.96 & b & 3.64 & 15.03 & 12.35 + & & & 3 & a & 3.13 & 15.49 & 18.99 & b & 3.66 & 15.01 & 13.25 + & & & 4 & a & 2.98 & 15.90 & 10.04 & b & & @xmath96 divergent & + & 8.00 & 0.15 & 2 & a & 3.14 & 16.00 & 22.27 & b & 3.68 & 15.54 & 15.19 + & & & 3 & a & 3.17 & 15.98 & 23.78 & b & 3.70 & 15.53 & 16.65 + & & & 4 & a & 2.94 & 16.52 & 8.81 & b & & @xmath96 divergent & + & 10.00 & 0.25 & 2 & a & 3.18 & 17.03 & 31.75 & b & 3.72 & 16.63 & 21.56 + & & & 3 & a & 3.21 & 17.01 & 33.56 & b & 3.74 & 16.62 & 23.81 + & & & 4 & a & 2.78 & 17.66 & 5.78 & b & & @xmath96 divergent & + & 11.50 & 0.25 & 2 & a & 3.24 & 16.16 & 40.82 & b & 3.75 & 15.90 & 28.11 + & & & 3 & a & 3.30 & 16.13 & 44.71 & b & 3.77 & 15.89 & 33.43 + & & 0.25 & 4 & a & & @xmath96 divergent & & b & & @xmath96 divergent & + 4.0 & 5.60 & 0 & 1 & b & 3.41 & 14.75 & 10.91 & b & 3.51 & 14.65 & 9.94 + & 7.00 & 0.10 & 2 & b & 3.46 & 15.19 & 15.87 & b & 3.66 & 15.01 & 13.94 + & & & 3 & b & 3.48 & 15.17 & 16.99 & b & 3.68 & 14.99 & 14.94 + & & & 4 & b & & @xmath96 divergent & & b & & @xmath96 divergent & + & 8.00 & 0.15 & 2 & b & 3.51 & 15.70 & 19.52 & b & 3.70 & 15.53 & 17.19 + & & & 3 & b & 3.52 & 15.69 & 21.30 & b & 3.71 & 15.52 & 18.88 + & & & 4 & b & & @xmath96 divergent & & b & & @xmath96 divergent & + & 10.00 & 0.25 & 2 & a & 3.39 & 16.88 & 31.03 & b & 3.74 & 16.62 & 24.46 + & & & 3 & a & 3.41 & 16.86 & 33.43 & b & 3.75 & 16.61 & 27.03 + & & & 4 & a & 3.11 & 17.42 & 5.78 & b & & @xmath96 divergent & + & 11.50 & 0.50 & 2 & a & 3.41 & 16.07 & 40.85 & b & 3.76 & 15.90 & 31.91 + & & & 3 & a & 3.44 & 19.50 & 46.49 & b & 3.78 & 15.89 & 37.99 + & & & 4 & a & & @xmath96 divergent & & b & & @xmath96 divergent & + 5.0 & 5.60 & 0 & 1 & b & 3.44 & 14.72 & 13.35 & b & 3.54 & 14.62 & 12.17 + 7.0 & 5.60 & 0 & 1 & b & 3.49 & 14.67 & 18.15 & b & 3.59 & 14.57 & 16.55 + 10.0 & 5.60 & 0 & 1 & b & 3.53 & 14.63 & 25.24 & b & 3.63 & 14.53 & 23.01 + cccccc 2.0 & 5 & 0.005 & 2.12 & 16.41 & 13.18 + & & 0.01 & 2.06 & 16.43 & 8.63 + & & 0.015 & 1.98 & 16.47 & 5.17 + 2.5 & 5 & 0.005 & 2.46 & 16.07 & 12.74 + & & 0.01 & 2.41 & 16.08 & 9.54 + & & 0.015 & 2.35 & 16.10 & 6.88 + & & 0.02 & 2.29 & 16.13 & 4.67 + 3.0 & 5 & 0.01 & 2.72 & 15.78 & 10.31 + & & 0.015 & 2.68 & 15.78 & 8.32 + & & 0.02 & 2.63 & 15.79 & 6.61 + & & 0.025 & 2.59 & 15.81 & 5.11 + & & 0.03 & 2.53 & 15.83 & 3.82 + 3.5 & 5 & 0.01 & 3.01 & 15.49 & 11.12 + & & 0.015 & 2.99 & 15.48 & 9.84 + & & 0.02 & 2.97 & 15.47 & 8.67 + & & 0.025 & 2.94 & 15.46 & 7.63 + & & 0.03 & 2.91 & 15.46 & 6.61 + & & 0.035 & 2.88 & 15.45 & 5.76 + & & 0.04 & 2.86 & 15.45 & 5.02 + cccccc 2.0 & 5 & 0.005 & 2.60 & 15.93 & 8.80 + & & 0.01 & 2.54 & 15.96 & 6.67 + & & 0.015 & 2.46 & 16.00 & 4.86 + 2.5 & 5 & 0.01 & 2.97 & 15.53 & 7.23 + & & 0.015 & 2.92 & 15.54 & 6.05 + & & 0.02 & 2.87 & 15.57 & 4.97 + & & 0.025 & 2.81 & 15.59 & 4.02 + 3.0 & 5 & 0.015 & 3.30 & 15.14 & 8.18 + & & 0.02 & 3.30 & 15.14 & 7.71 + & & 0.025 & 3.28 & 15.13 & 7.27 + & & 0.03 & 3.27 & 15.11 & 6.84 + & & 0.035 & 3.26 & 15.09 & 6.43 +

this paper is concerned with the coherent quantum filtering ( cqf ) problem , where a quantum observer is cascaded in a measurement - free fashion with a linear quantum plant so as to minimize a mean square error of estimating the plant variables of interest . both systems are governed by markovian hudson - parthasarathy quantum stochastic differential equations driven by bosonic fields in vacuum state . these quantum dynamics are specified by the hamiltonians and system - field coupling operators . we apply a recently proposed transverse hamiltonian variational method to the development of first - order necessary conditions of optimality for the cqf problem in a larger class of observers . the latter is obtained by perturbing the hamiltonian and system - field coupling operators of a linear coherent quantum observer along linear combinations of unitary weyl operators , whose role here resembles that of the needle variations in the pontryagin minimum principle . we show that if the observer is a stationary point of the performance functional in the class of linear observers , then it is also a stationary point with respect to the weyl variations in the larger class of nonlinear observers . in loving memory of my mother , raisa dmitrievna vladimirova ( 10.05.1933 30.05.2015 )

1506.07653

the dynamics of a wide class of open quantum systems , interacting with the environment , can be described in the framework of the hudson - parthasarathy quantum stochastic calculus @xcite . this approach employs quantum stochastic differential equations ( qsdes ) which are driven by a quantum mechanical analogue of the classical wiener process @xcite . the quantum wiener process models a heat reservoir of external fields and acts on a boson fock space @xcite . the drift vector and the dispersion matrix of the qsdes depend on the system hamiltonian and the system - field coupling operators . these energy operators specify the evolution of the system as a result of its internal dynamics influenced by the interaction with the environment . the hamiltonian and the coupling operators are usually modelled as functions of the system variables . in particular , such functions can be polynomials or weyl quantization integrals @xcite , which affects the complexity of the resulting quantum system . an important role in the linear quantum control and filtering theory @xcite is played by open quantum harmonic oscillators ( oqhos ) @xcite with quadratic hamiltonians and linear system - field coupling operators . the linear - quadratic dependence of the energy operators on the system variables , in combination with the canonical commutation relations ( ccrs ) between the variables , makes the oqho dynamics linear ( and gaussian in the case of vacuum fields and gaussian initial states @xcite ) . despite some similarities to the classical linear sdes , the coherent ( that is , measurement - free ) quantum counterparts @xcite to the classical lqg control and filtering problems @xcite for oqhos are complicated by the physical realizability ( pr ) constraints . the latter are associated with the state - space matrices of the qsdes for fully quantum controllers or filters and are related , in particular , to the ccr preservation . we mention one of the existing variational approaches @xcite to the coherent quantum lqg ( cqlqg ) control and coherent quantum filtering ( cqf ) problems which develops optimality conditions using the frechet differentiation of the lqg cost with respect to the state - space matrices . the quantum nature of the underlying problem enters this approach only through the pr constraints , with all the other aspects of the method being essentially `` classical '' . the latter has certain advantages , such as practical applicability to the numerical optimization algorithms @xcite . however , this approach is limited to linear controllers and filters , and the resulting optimality conditions do not provide insights into whether nonlinear quantum controllers or filters can outperform the linear ones for linear quantum plants . in the present paper , we consider a cqf problem , similar to @xcite , where a quantum observer is cascaded in a measurement - free fashion with a linear quantum plant so as to minimize a mean square error with which the observer variables approximate linear combinations of plant variables of interest . both systems are governed by markovian hudson - parthasarathy qsdes driven by bosonic fields in vacuum state . we employ a recently proposed fully quantum variational method of @xcite based on using a transverse hamiltonian , an auxiliary time - varying operator which encodes the propagation of perturbations through the unitary system - field evolution . we apply the transverse hamiltonian approach to the development of first - order necessary conditions of optimality for the cqf problem in a larger class of observers . the latter is obtained by perturbing the hamiltonian and system - field coupling operators of a linear coherent quantum observer along linear combinations of the unitary weyl operators @xcite . similar trigonometric polynomials of quantum variables have recently been used in @xcite for modelling the uncertainty in system hamiltonians . in the present paper , the weyl variations play a different role which resembles that of the needle variations in the proof of the pontryagin minimum principle @xcite . we show that if the observer is a stationary point of the cost functional in the class of linear observers , then it is also a stationary point with respect to the weyl variations ( with the latter leading to nonlinear observers ) . therefore , in the mean square optimal cqf problem for linear quantum plants , linear coherent quantum observers are locally sufficient at least in the sense of the weyl variations of the energy operators . the paper is organised as follows . section [ sec : not ] outlines notation used in the paper . section [ sec : oqho ] specifies the class of open quantum stochastic plants being considered . section [ sec : cqf ] formulates the mean square optimal cqf problem . section [ sec : var ] employs the transverse hamiltonian approach in order to study the sensitivity of the performance criterion to perturbations in the energy operators of observers . section [ sec : linopt ] applies these results to liner - quadratic perturbations of the observer energy operators and obtains the first - order necessary conditions of optimality in the class of linear observers . section [ sec : weyl ] introduces the weyl variations of the energy operators and establishes the main result of the paper that the stationarity of the quadratic cost functional with respect to linear - quadratic perturbations of a linear observer implies the stationarity with respect to the weyl variations . section [ sec : conc ] provides concluding remarks . in what follows , @xmath0:= ab - ba$ ] denotes the commutator of linear operators @xmath1 and @xmath2 on a common space . as a linear superoperator , the commutator with a fixed operator @xmath1 , is denoted by @xmath3 $ ] . this extends to the commutator @xmath4-matrix @xmath5 : = xy^{\rt } - ( yx^{\rt})^{\rt } = ( [ x_j , y_k])_{1\ < j\ < n,1\ < k\ < m } $ ] for a vector @xmath6 of operators @xmath7 and a vector @xmath8 of operators @xmath9 . vectors are organized as columns unless indicated otherwise , and the transpose @xmath10 acts on matrices of operators as if their entries were scalars . in application to such matrices , @xmath11 denotes the transpose of the entry - wise operator adjoint @xmath12 . for complex matrices , @xmath13 is the usual complex conjugate transpose @xmath14 . the subspaces of real symmetric , real antisymmetric and complex hermitian matrices of order @xmath15 are denoted by @xmath16 , @xmath17 and @xmath18 , respectively , where @xmath19 is the imaginary unit . the real and imaginary parts of a complex matrix are denoted by @xmath20 and @xmath21 . these extend to matrices @xmath22 with operator - valued entries as @xmath23 and @xmath24 which consist of self - adjoint operators . also , @xmath25 denotes the symmetrizer of square matrices . positive ( semi- ) definiteness of matrices and the corresponding partial ordering are denoted by ( @xmath26 ) @xmath27 . also , @xmath28 and @xmath29 denote the sets of positive semi - definite real symmetric and complex hermitian matrices of order @xmath15 , respectively . the tensor product of spaces or operators ( in particular , the kronecker product of matrices ) is denoted by @xmath30 . the tensor product @xmath31 of operators @xmath1 and @xmath2 acting on different spaces will sometimes be abbreviated as @xmath32 . the identity matrix of order @xmath15 is denoted by @xmath33 , while the identity operator on a space @xmath34 is denoted by @xmath35 . the frobenius inner product of real or complex matrices is denoted by @xmath36 . also , @xmath37 denotes the euclidean ( semi-)norm of a real vector @xmath38 associated with a real positive ( semi-)definite symmetric matrix @xmath39 . the expectation @xmath40 of a quantum variable @xmath41 over a density operator @xmath42 extends entrywise to vectors and matrices of such variables . we consider a quantum plant which is modelled as a quantum stochastic system interacting with @xmath43 external boson fields . the plant has @xmath15 dynamic variables @xmath44 which evolve in time @xmath45 . the plant variables are self - adjoint operators on a composite plant - field hilbert space @xmath46 , where @xmath47 is the initial complex separable hilbert space of the plant which provides a domain for @xmath48 , and @xmath49 is a boson fock space @xcite for the action of quantum wiener processes @xmath50 . the latter are self - adjoint operators which model the external boson fields . the energetics of the plant - field interaction is specified by the plant hamiltonian @xmath51 and the plant - field coupling operators @xmath52 which are self - adjoint operators , representable as time - invariant functions ( for example , polynomials with constant coefficients or weyl quantization integrals @xcite ) of the plant variables @xmath44 . therefore , both @xmath53 and @xmath54 act on the initial space @xmath47 . omitting the time arguments , we assemble the plant and field variables , and the coupling operators into vectors : @xmath55 for what follows , the plant is assumed to be an oqho @xcite . more precisely , the plant variables satisfy the weyl ccrs @xmath56 for all @xmath57 , where @xmath58 is the unitary weyl operator @xcite on @xmath46 which is associated with ( and inherits time dependence from ) the plant variables . the heisenberg infinitesimal form of the weyl ccrs is described by @xmath59 = 2i \theta\ ] ] where the ccr matrix @xmath60 represents @xmath61 and remains unchanged . the hamiltonian of the oqho is a quadratic polynomial of the plant variables , and the plant - field coupling operators in ( [ xwl ] ) are linear functions of the variables : @xmath62 where @xmath63 is the energy matrix , and @xmath64 is the coupling matrix . the plant and the external fields form an isolated quantum system whose evolution is described by a unitary operator @xmath65 on @xmath46 driven by the fields and their interaction with the plant according to the qsde @xcite @xmath66 with initial condition @xmath67 . the subscript @xmath68 indicates the initial values of time - varying operators ( or vectors and matrices thereof ) , so that @xmath69 , @xmath70 and @xmath71 , and the time arguments will often be omitted for brevity . due to the continuous tensor product structure of the fock space @xcite , the future - pointing increments @xmath72 commute with adapted processes ( including @xmath73 ) taken at the same ( or an earlier ) moment of time . the matrix @xmath74 in ( [ duhl ] ) is the ito matrix of the quantum wiener process @xmath75 : @xmath76 and the matrix @xmath77 specifies the cross - commutations between the entries @xmath78 of @xmath75 : @xmath79 = 2ij\rd t , \quad j : = \bj\ox i_{m/2 } , \quad \bj : = { \begin{bmatrix } 0 & 1\\ -1 & 0 \end{bmatrix}},\ ] ] where the dimension @xmath43 is assumed to be even . the qsde ( [ duhl ] ) corresponds to an important particular case of open quantum dynamics when the scattering matrix is the identity matrix , and there is no photon exchange between the fields , thus eliminating the gauge processes @xcite from consideration . a plant operator @xmath80 on the initial space @xmath47 ( which can be identified with its extension @xmath81 to the plant - field space @xmath46 ) evolves to an operator @xmath82 on @xmath46 at time @xmath83 according to the flow @xmath84 when it is applied to vectors and matrices of operators , the flow @xmath85 acts entrywise . in view of the identity @xmath86 , which follows from ( [ flow ] ) and the unitarity of @xmath73 , the second equality in ( [ duhl ] ) employs the representation of the hamiltonian and the coupling operators in terms of the flow : @xmath87 note that the flow @xmath85 depends on the energy operators @xmath88 and @xmath89 ( or the energy and coupling matrices @xmath90 and @xmath91 in ( [ rn ] ) in the case of oqhos ) . more precisely , any perturbation of @xmath88 and @xmath89 ( as functions of the fixed set of system variables @xmath92 ) modifies the flow . now , the quantum adapted process @xmath93 in ( [ flow ] ) satisfies the following hudson - parthasarathy qsde @xcite : @xmath94\rd w , \quad \cg(\sigma ) : = i[h,\sigma ] + \cd(\sigma).\ ] ] here , @xmath95 is the gorini - kossakowski - sudarshan - lindblad ( gksl ) decoherence superoperator @xcite which acts on @xmath93 as @xmath96 + [ l_j,\sigma]l_k ) \\ % \nonumber % & = \frac{1}{2 } \big ( l^{\rt}\omega [ \sigma , l ] + [ l^{\rt},\sigma]\omega l \big)\\ \nonumber & = -[\sigma , l^{\rt}]\omega l - \frac{1}{2 } [ l^{\rt}\omega l , \sigma]\\ \label{cd } & = -[\sigma , x^{\rt}]n^{\rt}\omega nx - \frac{1}{2 } [ x^{\rt}n^{\rt}\omega nx , \sigma].\end{aligned}\ ] ] the last two equalities in ( [ cd ] ) are convenient for the entrywise evaluation of @xmath95 at vectors of operators . the superoperator @xmath97 in ( [ dsigma ] ) is referred to as the gksl generator . in application to the vector @xmath6 of plant variables , the flow ( [ flow ] ) acts entrywise as @xmath98 in view of ( [ theta ] ) and ( [ rn ] ) , the corresponding qsde ( [ dsigma ] ) takes the form @xmath99\rd w = ax\rd t + b\rd w,\ ] ] with the @xmath15-dimensional drift vector @xmath100 and the dispersion @xmath4-matrix @xmath101=-i[x , x^{\rt}]n^{\rt}= b$ ] , where the matrices @xmath102 and @xmath103 are given by @xmath104 the interaction of the input field with the plant produces an @xmath43-dimensional output field @xmath105 where the plant - field unitary evolution is applied to the current input field variables ( which is closely related to the innovation role of the quantum wiener process @xmath75 in the qsdes ) . the output field satisfies the qsde @xmath106 where @xmath107 is the matrix from ( [ j ] ) , @xmath108 is the vector of plant - field coupling operators in ( [ xwl ] ) and ( [ rn ] ) , and the matrix @xmath109 is given by @xmath110 consider a measurement - free cascade connection of the quantum plant , described in section [ sec : oqho ] , with another open quantum system . the latter plays the role of a coherent quantum observer and is driven by @xmath111 ( which is part of the plant output @xmath8 in ( [ y ] ) ) and a quantum wiener process @xmath112 of even dimension @xmath113 on a boson fock space @xmath114 ; see fig . [ fig : filtering ] . = 0.8 mm ( 50.00,50.00 ) ( 20,20)(20,10)[cc]observer ( 20,40)(20,10)[cc]plant ( 20,25)(-1,0)10 ( 50,25)(-1,0)10 ( 30,40)(0,-1)10 ( 9,25)(0,0)[rc]@xmath115 ( 52,25)(0,0)[lc]@xmath112 ( 52,45)(0,0)[lc]@xmath75 ( 50,45)(-1,0)10 ( 32,35)(0,0)[lc]@xmath111 -15 mm the matrix @xmath116 is formed from conjugate pairs of rows of a permutation matrix of order @xmath43 , with @xmath117 . the observer has its own initial hilbert space @xmath118 , dynamic variables @xmath119 with a ccr matrix @xmath120 , and a @xmath121-dimensional output field @xmath115 : @xmath122 we denote the observer hamiltonian by @xmath123 , while the vectors of operators of coupling of the observer with the selected plant output @xmath111 and the quantum wiener process @xmath112 are denoted by @xmath124 respectively . the hamiltonian @xmath123 and the coupling operators @xmath125 and @xmath126 are functions of the dynamic variables @xmath119 of the observer and hence , commute with functions of the plant variables , including the plant hamiltonian @xmath34 and the plant - field coupling operators in @xmath108 . the plant and the observer form a composite open quantum stochastic system , whose vector @xmath127 of dynamic variables satisfies the ccrs @xmath128 = 2i\btheta , \qquad \btheta : = { \begin{bmatrix } \theta & 0\\ 0 & \vartheta \end{bmatrix } } , \qquad \cx : = { \begin{bmatrix } x\\ \xi \end{bmatrix}}\ ] ] and is driven by a combined quantum wiener process @xmath129 with the ito table @xmath130 here , @xmath131 is the ito matrix of the quantum wiener process @xmath112 of the observer which is defined similarly to @xmath132 in ( [ omega ] ) and ( [ j ] ) : @xmath133 the hamiltonian @xmath134 of the plant - observer system and the vector @xmath135 of operators of coupling with @xmath129 can be computed by using the quantum feedback network formalism @xcite as @xmath136 while the plant dynamics , governed by ( [ dx ] ) and ( [ dy ] ) , remains unaffected by the observer , the dynamic variables of the latter in ( [ xiomlam ] ) are governed by the qsde @xmath137 \rd \cw\\ \label{dxi } & = ( i[\gamma,\xi ] + \delta(\xi))\rd t - i[\xi , \phi^{\rt } ] \pi \rd y - i[\xi , \psi^{\rt } ] \rd \omega.\end{aligned}\ ] ] here , @xmath138 + \bd(\zeta)\ ] ] denotes the gksl generator for the plant - observer system , and @xmath139\bomega \bl - \frac{1}{2 } [ \bl^{\rt}\bomega \bl , \zeta]\ ] ] is the corresponding decoherence superoperator , similar to ( [ cd ] ) . in ( [ dxi ] ) , use is also made of the partial gksl decoherence superoperator @xmath140 which acts on the observer variables as @xmath141\pi\omega \pi^{\rt}\phi -[\xi , \psi^{\rt}]\mho \psi\\ \label{delta } & - \frac{1}{2 } [ \phi^{\rt}\pi \omega \pi^{\rt } \phi + \psi^{\rt}\mho\psi , \xi]\end{aligned}\ ] ] in view of ( [ cx])([bhbl ] ) , with @xmath142 being the quantum ito matrix of @xmath143 . now , similarly to @xcite , we formulate a cqf problem as the minimization of the steady - state mean square discrepancy @xmath144 between @xmath145 linear combinations of the plant variables of interest and observer variables as specified by given matrices @xmath146 and @xmath147 ( with @xmath148 being interpreted as the estimation error ) . here , the quantum expectation @xmath149 is taken over the tensor product @xmath150 of the initial quantum state @xmath151 of the plant - observer system and the vacuum state @xmath152 in the composite boson fock space @xmath153 for the external fields . also , it is assumed that the plant - observer system is ergodic , whereby the limit in ( [ cqf ] ) reduces to averaging over the invariant state of the system , provided the plant and observer variables satisfy an appropriate version of the uniform mean square integrability condition . the criterion process @xmath154 in ( [ cqf ] ) is a quadratic function of the vector @xmath127 from ( [ cx ] ) : @xmath155 the minimization in ( [ cqf ] ) is carried out over the observer hamiltonian @xmath123 and the vector @xmath156 of the observer - plant coupling operators in ( [ phipsi ] ) , while @xmath157 and all the dimensions are fixed . this problem extends @xcite in that we do not restrict attention to linear observers even though the plant is an oqho . in order to develop first - order necessary conditions of optimality for the cqf problem ( [ cqf ] ) in an extended class of observers , we will apply the transverse hamiltonian variational method of @xcite to the infinitesimal perturbation analysis of the performance criterion @xmath158 . to this end , suppose @xmath159 and @xmath160 depend smoothly ( for example , linearly ) on a small scalar parameter @xmath161 and are perturbed in the directions @xmath162 consisting of self - adjoint operators on the observer initial space @xmath118 , representable as functions of the observer variables , with @xmath163 . the corresponding perturbations of the plant - observer hamiltonian and the coupling operators in ( [ bhbl ] ) are @xmath164 the propagation of these perturbations of the energy operators through the unitary evolution @xmath165 of the plant - observer - field system on the space @xmath166 is encoded by the transverse hamiltonian @xcite . the latter is a time - varying self - adjoint operator defined by @xmath167 , so that @xmath168 for all @xmath83 , with zero initial condition @xmath169 . in view of ( [ cw ] ) , ( [ km0cqf ] ) and ( [ kmcqf ] ) , the general qsde obtained in ( * ? ? ? * theorem 1 ) for the transverse hamiltonian , takes the form : @xmath170 here , @xmath171 and @xmath172 are the evolved versions of the initial perturbations from ( [ km0cqf ] ) under the unperturbed flow @xmath173 of the plant - observer - field system ( which is applied here to observer operators @xmath174 on @xmath118 ) . in ( [ dqcqf ] ) , use is also made of the relation @xmath175 which follows from ( [ omega ] ) , the commutativity @xmath176 = 0 $ ] and the antisymmetry of @xmath107 . since the criterion process @xmath154 in ( [ cqf ] ) does not depend explicitly on the energy operators , the corresponding formal gateaux derivative @xmath177 of the cost functional can be computed by using ( * ? ? ? * theorem 2 , section vii ) as @xmath178.\ ] ] here , @xmath179 is the derivative process @xcite associated with @xmath154 . its expectation satisfies the integro - differential equation @xmath180 + \be \chi(z),\ ] ] where @xmath181 is the unperturbed plant - observer gksl generator given by ( [ bhbl ] ) , ( [ bg ] ) , ( [ bd ] ) , and @xmath182 is an auxiliary linear superoperator acting on plant - observer system operators @xmath93 as @xmath183\\ \label{chiz } & -2\re ( [ \sigma,\ , ( l+\pi^{\rt}\phi)^{\rt}]\omega \pi^{\rt}m).\end{aligned}\ ] ] note that @xmath184 depends linearly on the perturbations @xmath39 and @xmath22 . now , in addition to the plant being an oqho , suppose the unperturbed observer is also an oqho with energy matrix @xmath185 and coupling matrices @xmath186 and @xmath187 . the corresponding observer hamiltonian @xmath123 and the coupling operators in ( [ phipsi ] ) are @xmath188 in this case , in view of ( [ cw ] ) and ( [ delta ] ) , the qsde ( [ dxi ] ) becomes linear : @xmath189 where the matrices @xmath190 , @xmath191 and @xmath192 are computed as @xmath193 therefore , the plant - observer system is governed by a linear qsde @xmath194 for what follows , both matrices @xmath1 and @xmath195 are assumed to be hurwitz , and hence , so is @xmath196 . this implies that the plant - observer system is ergodic and has a unique invariant state which is gaussian @xcite with zero mean and quantum covariance matrix @xmath197 . here , @xmath198 is the controllability gramian @xcite of the pair @xmath199 which is a unique solution of the ale @xmath200 [ lem : dash ] suppose the plant and the unperturbed observer are oqhos described by ( [ rn ] ) , ( [ dx])([c ] ) and ( [ gamma_phi_psi])([bb ] ) , with hurwitz matrices @xmath1 and @xmath195 . also , let the observer hamiltonian and the observer - plant coupling operators are perturbed according to ( [ km0cqf ] ) . then the corresponding formal gateaux derivative of the cost functional in ( [ cqf ] ) can be computed as @xmath201 here , @xmath202 is the observability gramian of the pair @xmath203 satisfying the ale @xmath204 where the matrices @xmath205 and @xmath196 are given by ( [ zcqf ] ) and ( [ dcx ] ) . also , @xmath206 denotes the quantum expectation over the invariant gaussian state of the plant - observer system ( with @xmath207 and @xmath208 found from ( [ pale ] ) ) . furthermore , the superoperator @xmath182 from ( [ chiz ] ) acts on the operator @xmath93 in ( [ czcqf ] ) as @xmath209\\ \label{chisigma } & + 8\im \big(\cx^{\rt } \cq \bit{\theta}{\begin{bmatrix } n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \omega \pi^{\rt}m\big).\end{aligned}\ ] ] since @xmath154 in ( [ zcqf ] ) is a quadratic form of @xmath127 , then ( [ cz ] ) and ( [ bephidot ] ) imply that @xmath210,\ ] ] where @xmath211 is the limit mean value of the derivative process @xmath212 $ ] associated with @xmath213 . the gksl generator @xmath181 of the unperturbed linear plant - observer system acts on @xmath214 as @xmath215 , and hence , similarly to ( * ? ? ? * example 3 , section vii ) , the matrix @xmath216 satisfies the ale @xmath217 where the superoperator @xmath182 in ( [ chiz ] ) is applied to @xmath214 entrywise . by combining the integral representations of the solutions of the ales ( [ qale ] ) , ( [ upsale ] ) and using duality , it follows from ( [ zups ] ) that @xmath218 which establishes ( [ czcqf ] ) since @xmath219 . indeed , @xmath93 enters @xmath184 in ( [ chiz ] ) only through the commutators @xmath220 $ ] and hence , @xmath221 for any @xmath222 . the representation ( [ chisigma ] ) is obtained by substituting @xmath108 from ( [ rn ] ) , @xmath156 , @xmath157 from ( [ gamma_phi_psi ] ) and @xmath93 from ( [ czcqf ] ) into ( [ chiz ] ) , and using the relation @xmath223 = 2\btheta \cq \cx$ ] . in particular , the second line of ( [ chisigma ] ) is established by @xmath224\omega \pi^{\rt}m)\\ & = -2\re \big([\sigma , \cx^{\rt}]{\begin{bmatrix}n^{\rt}\\ n_1^{\rt } \pi \end{bmatrix}}\omega \pi^{\rt}m\big)\\ & = -2\re \big((-4i\btheta \cq \cx)^{\rt}{\begin{bmatrix}n^{\rt}\\ n_1^{\rt } \pi \end{bmatrix}}\omega \pi^{\rt}m\big)\\ % & = % -8\re \big(i\cx^{\rt } \cq \btheta { \begin{bmatrix}n^{\rt}\\ n_1^{\rt } \pi \end{bmatrix}}\omega \pi^{\rt}m\big)\\ & = 8\im \big(\cx^{\rt } \cq \btheta { \begin{bmatrix}n^{\rt}\\ n_1^{\rt } \pi \end{bmatrix}}\omega \pi^{\rt}m\big).\end{aligned}\ ] ] although the unperturbed observer in lemma [ lem : dash ] is an oqho , the perturbations ( [ km0cqf ] ) are not assumed to be linear - quadratic . therefore , the lemma provides a perturbative tool to develop conditions for such an observer to be a stationary point of the cqf problem ( [ cqf ] ) in a wider class of observers . we will now apply lemma [ lem : dash ] to the first - order necessary conditions of optimality among linear observers . to this end , associated with the gramians @xmath225 and @xmath226 from ( [ pale ] ) and ( [ qale ] ) is the hankelian @xmath227 which , together with @xmath225 and @xmath226 , is split into appropriately dimensioned blocks @xmath228 according to the partitioning of @xmath127 in ( [ cx ] ) , with @xmath229 the @xmath230th block - row and @xmath231 the @xmath232th block - column . [ th : optlin ] under the assumptions of lemma [ lem : dash ] , the linear observer , described in the lemma , is a stationary point of the cqf problem ( [ cqf ] ) in the class of linear observers if and only if @xmath233 let the matrices @xmath234 and @xmath235 in ( [ gamma_phi_psi ] ) be smooth functions of the parameter @xmath161 , and hence , the corresponding perturbations of the linear observer in ( [ km0cqf ] ) take the form @xmath236 where the matrices @xmath237 and @xmath238 can be arbitrary . substitution of ( [ kmlin ] ) into ( [ chisigma ] ) leads to @xmath239\\ \label{chixx } & + 8\im \big(\cx^{\rt } \cq \bit{\theta}{\begin{bmatrix } n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \omega \pi^{\rt}n_1 ' \xi\big),\end{aligned}\ ] ] where @xmath93 is the quantum variable from ( [ czcqf ] ) . the identity @xmath240 = 2\cx^{\rt } ( r_2\btheta r_1 - r_1\btheta r_2)\cx$ ] , which holds for any @xmath241 due to the ccrs ( [ cx ] ) , implies that @xmath242 & = 2 \cx^{\rt } \big ( \cq \btheta { \begin{bmatrix } 0 & 0\\ 0 & r ' \end{bmatrix } } - { \begin{bmatrix } 0 & 0\\ 0 & r ' \end{bmatrix } } \btheta \cq \big ) \cx\\ & = 2 \cx^{\rt } \big ( { \begin{bmatrix } 0 & \cq_{\bullet 2}\vartheta r ' \end{bmatrix } } - { \begin{bmatrix } 0\\ r'\vartheta \cq_{2 \bullet } \end{bmatrix } } \big ) \cx.\end{aligned}\ ] ] the averaging of the latter representation over the invariant quantum state leads to @xmath243 & = 2 \big\bra { \begin{bmatrix } 0 & \cq_{\bullet 2}\vartheta r ' \end{bmatrix } } - { \begin{bmatrix } 0\\ r'\vartheta \cq_{2 \bullet } \end{bmatrix}},\ , \cp \big\ket\\ & = -4 \bra \vartheta \cq_{2\bullet } \cp_{\bullet 2},\ , r ' \ket = -4 \bra \bs(\vartheta \ce_{22}),\ , r ' \ket,\end{aligned}\ ] ] where the equality @xmath244 follows from ( [ ce ] ) , and the symmetry of @xmath245 is used . therefore , in view of ( [ czcqf ] ) , the corresponding formal frechet derivative @xmath246 of the cost functional vanishes if and only if the matrix @xmath247 satisfies ( [ stat1 ] ) . by a similar reasoning , in view of ( [ chixx ] ) , the gateaux derivative of @xmath158 along @xmath22 in ( [ kmlin ] ) takes the form @xmath248\\ \nonumber = & 8\im \bra \cq \bit{\theta}{\begin{bmatrix } n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega } \pi^{\rt}n_1 ' , \ , \cp_{\bullet 2 } + { \begin{bmatrix } 0\\ i\vartheta \end{bmatrix } } \ket \\ \nonumber & + 4 \re \bra { \begin{bmatrix } 0 & { \begin{bmatrix } 2n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega}\pi^{\rt } n_1 ' \end{bmatrix}},\ , i(\ce^{\rt } \btheta-\btheta \ce ) \ket \\ \nonumber = & 8 \big\bra \pi j { \begin{bmatrix } n & \pi^{\rt}n_1 \end{bmatrix } } \btheta \ce_{\bullet 2 } - \pi { \begin{bmatrix } n & \pi^{\rt}n_1 \end{bmatrix } } \btheta \cq_{\bullet 2}\vartheta\\ \nonumber & + \pi jn(\ce_{21}^{\rt } \vartheta - \theta \ce_{12 } ) - \pi j \pi^{\rt } n_1 \bs(\vartheta \ce_{22}),\ , n_1 ' \big\ket\\ \label{chain1 } = & 4 \bra \pi ( ( c\ce_{21}^{\rt } + b^{\rt}\cq_{12}+b_1^{\rt } \cq_{22 } ) \vartheta -j b_1^{\rt } \ce_{22}),\ , n_1 ' \ket,\end{aligned}\ ] ] where use is also made of ( [ ab ] ) , ( [ c ] ) , ( [ bb ] ) and ( [ ce ] ) . here , @xmath249 under the condition ( [ stat1 ] ) , in which case , the formal frechet derivative @xmath250 vanishes if and only if ( [ stat2 ] ) holds . therefore , ( [ stat1 ] ) and ( [ stat2 ] ) are indeed equivalent to the stationarity in the class of linear observers . now , consider the following bounded perturbations of the observer hamiltonian and the observer - plant coupling operators in ( [ km0cqf ] ) : @xmath251 with parameters @xmath252 , @xmath253 , @xmath254 , @xmath255 and @xmath256 . here , @xmath257 is the unitary weyl operator associated with the observer variables . we will refer to the perturbations ( [ kex ] ) and ( [ mex ] ) as the weyl variations ( of the observer energy operators ) . similar trigonometric polynomials of quantum variables have recently been used in @xcite to model uncertainties in system hamiltonians . however , in what follows , the weyl variations play a different role which resembles that of the needle variations in the proof of the pontryagin minimum principle @xcite . [ th : weyl ] under the assumptions of lemma [ lem : dash ] , the linear observer satisfies the first - order necessary conditions of optimality in the cqf problem ( [ cqf ] ) among linear observers if and only if it is a stationary point with respect to the weyl variations in ( [ kex ] ) and ( [ mex ] ) for any @xmath252 , @xmath258 , @xmath256 . the stationarity of the cost functional @xmath158 with respect to arbitrary weyl variations of the observer implies the stationarity with respect to the linear - quadratic perturbations ( [ kmlin ] ) because @xmath259 and @xmath260 . in fact , arbitrary polynomial perturbations can be reproduced by the weyl variations which form a larger class of perturbations . therefore , it remains to prove that ( [ stat1 ] ) and ( [ stat2 ] ) imply the stationarity with respect to arbitrary weyl variations . by lemma [ lem : dash ] , the linear response of the cost functional @xmath158 to the weyl variation @xmath39 of the observer hamiltonian in ( [ kex ] ) takes the form @xmath261 = - \im ( \alpha \be _ * [ \sw_u , \sigma ] ) .\ ] ] let @xmath262 denote a linear superoperator , which is parameterized by @xmath263 and carries out a unitary similarity transformation of quantum variables as @xmath264 , where use is made of a well - known identity for operator exponentials @xcite . its application to @xmath127 from ( [ cx ] ) leads to @xmath265 since @xmath266 = 0 $ ] ( whence @xmath267 ) and @xmath268 = -[\xi , u^{\rt}\xi ] = -[\xi,\xi^{\rt}]u = -2i\vartheta u$ ] ( whereby @xmath269 ) . therefore , in view of ( [ czcqf ] ) , @xmath270 , and hence , @xmath271 & = ( \ss_u(\sigma ) - \sigma)\sw_u\\ \nonumber & = -4u^{\rt}\vartheta ( \cq_{21}x + \cq_{22}(\xi+ \vartheta u ) ) \sw_u\\ \label{wsig } & = 4iu^{\rt}\vartheta ( \cq_{21}\d_w + \cq_{22}\d_u ) \re^{i(w^{\rt } x + u^{\rt}\xi)}\big|_{w=0}.\end{aligned}\ ] ] here , @xmath272 in view of the commutativity @xmath273=0 $ ] between the weyl operators for all @xmath274 , @xmath256 , and hence , @xmath275 also , the relation @xmath276 follows from @xmath277 due to the weyl ccrs @xmath278 which hold for all @xmath279 . by averaging ( [ wsig ] ) over the invariant gaussian quantum state , whose quasi - characteristic function @xcite is @xmath280 for any @xmath281 , it follows that @xmath282 & = 4iu^{\rt}\vartheta \cq_{2\bullet } \d_{\lambda } \re^{-\frac{1}{2}\|\lambda\|_{\cp}^2 } \big|_{w=0}\\ \nonumber & = -4iu^{\rt}\vartheta \cq_{2\bullet } \cp { \begin{bmatrix}0\\ u\end{bmatrix } } \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2}\\ \label{kvargauss } & = -4iu^{\rt } \bs(\vartheta \ce_{22 } ) u \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2},\end{aligned}\ ] ] where use is made of the hankelian from ( [ ce ] ) . by substituting ( [ kvargauss ] ) into ( [ kvar ] ) , the gateaux derivative of @xmath158 along @xmath39 in ( [ kex ] ) takes the form @xmath283 = 4u^{\rt } \bs(\vartheta \ce_{22 } ) u \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2 } \re \alpha $ ] . therefore , the fulfillment of ( [ stat1 ] ) makes this derivative vanish for any @xmath252 and @xmath263 . similar arguments ( see also ( [ chain1 ] ) ) lead to the following gateaux derivative of @xmath158 along @xmath22 in ( [ mex ] ) : @xmath284\\ \nonumber = & 8\im \bra \cq \bit{\theta}{\begin{bmatrix } n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega } \pi^{\rt } , \ , \be_*(\cx \re ( \beta^{\rt}\sw_u ) ) \ket \\ \nonumber & + \re \bra { \begin{bmatrix } 2n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega}\pi^{\rt } , \ , \be_*[\sigma , \cx \re ( \beta^{\rt}\sw_u ) ] \ket \\ \nonumber = & -8 \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2}\im \bra \cq \bit{\theta}{\begin{bmatrix } n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega } \pi^{\rt } , \big(\cp_{\bullet 2 } + { \begin{bmatrix } 0\\ i\vartheta \end{bmatrix}}\big)u\im \beta^{\rt } \ket \\ \nonumber & + 4 \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2 } \re \bra { \begin{bmatrix } 2n^{\rt}\\ n_1^{\rt}\pi \end{bmatrix } } \overline{\omega}\pi^{\rt } , \ , i(\theta \ce_{12}-\ce_{21}^{\rt } \vartheta)u\im \beta^{\rt } \ket \\ % \nonumber % = & % 8 % \big\bra % \pi j { \begin{bmatrix } % n & % \pi^{\rt}n_1 % \end{bmatrix } } % \btheta \ce_{\bullet 2 } % - % \pi { \begin{bmatrix } % n & % \pi^{\rt}n_1 % \end{bmatrix } } % \btheta \cq_{\bullet 2}\vartheta\\ % \nonumber % & + % \pi jn(\ce_{21}^{\rt } \vartheta - \theta \ce_{12 } ) - \pi j \pi^{\rt } n_1 \bs(\vartheta \ce_{22}),\ , % n_1 ' % \big\ket\\ \label{chain2 } = & 4 \im \beta^{\rt } \pi(j b_1^{\rt } \ce_{22}- ( c\ce_{21}^{\rt } + b^{\rt}\cq_{12}+b_1^{\rt } \cq_{22 } ) \vartheta)u \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2}.\end{aligned}\ ] ] here , @xmath285=0 $ ] follows from ( [ kvargauss ] ) under the condition ( [ stat1 ] ) . also , in view of ( [ wsig])([xiw ] ) , use is made of the relations @xmath286 & = -i\d_{\lambda } \be _ * [ \sigma , \re^{i\lambda^{\rt}\cx}]\big|_{w=0 } -{\begin{bmatrix}0\\ \vartheta u\end{bmatrix}}\be _ * [ \sigma , \sw_u]\\ & = -4\d_{\lambda } ( \lambda^{\rt}\btheta \cq \d_{\lambda}\be _ * \re^{i\lambda^{\rt}\cx})\big|_{w=0}\\ & = 4\d_{\lambda } \big(\lambda^{\rt}\bs(\btheta \ce)\lambda \re^{-\frac{1}{2}\|\lambda\|_{\cp}^2}\big)\big|_{w=0}\\ & = 4 \re^{-\frac{1}{2}\|u\|_{\cp_{22}}^2 } ( \theta \ce_{12}-\ce_{21}^{\rt } \vartheta)u.\end{aligned}\ ] ] the fulfillment of ( [ stat1 ] ) and ( [ stat2 ] ) makes the gateaux derivative ( [ chain2 ] ) of @xmath158 along the weyl variation @xmath22 in ( [ mex ] ) also vanish for any @xmath258 and @xmath263 . therefore , ( [ stat1 ] ) and ( [ stat2 ] ) indeed imply the stationarity with respect to the arbitrary weyl variations ( [ kex ] ) and ( [ mex ] ) . note that the proof of theorem [ th : weyl ] ( which employs the gaussian averaging of quasi - polynomials of system variables , such as @xmath287 ) is closely related to the integro - differential identities @xcite for expectations of weyl quantization integrals over gaussian states . we have applied the transverse hamiltonian variational method @xcite to the mean square optimal cqf problem for a linear observer cascaded with a linear quantum plant . it has been shown that if such an observer is a stationary point of the problem among linear observers , then it also satisfies the first - order necessary conditions of optimality with respect to a wider class of weyl variations of the energy operators of the observer . in this sense , linear observers are locally sufficient for linear quantum plants as far as the mean square performance criteria are concerned . similar ideas can be developed for the cqlqg control problem @xcite and an optimal control theory for classical port - hamiltonian systems @xcite and their stochastic versions . c.d.cushen , and r.l.hudson , a quantum - mechanical central limit theorem , _ j. appl _ , vol . 8 , no . 3 , 1971 , pp . 454469 . d.dong , and i.r.petersen , quantum control theory and applications : a survey , _ iet contr . 4 , no . 12 , 2010 , pp . 26512671 . s.c.edwards , and v.p.belavkin , optimal quantum filtering and quantum feedback control , arxiv : quant - ph/0506018v2 , august 1 , 2005 . g.b.folland , _ harmonic analysis in phase space _ , princeton university press , princeton , 1989 . c.w.gardiner , and p.zoller , _ quantum noise _ , 3rd ed . , springer , berlin , 2004 . v.gorini , a.kossakowski , and e.c.g.sudarshan , completely positive dynamical semigroups of @xmath91-level systems , _ j. math 17 , no . 5 , 1976 , pp . 821825 . j.gough , and m.r.james , quantum feedback networks : hamiltonian formulation , _ commun . _ , vol . 287 , 2009 , pp . 11091132 a.s.holevo , quantum stochastic calculus , _ j. math . 56 , no . 5 , 1991 , pp . 26092624 . r.l.hudson , and k.r.parthasarathy , quantum ito s formula and stochastic evolutions , _ commun . 93 , 1984 , pp . k.jacobs , and p.l.knight , linear quantum trajectories : applications to continuous projection measurements , _ phys . 57 , no . 4 , 1998 , pp . m.r.james , h.i.nurdin , and i.r.petersen , @xmath288 control of linear quantum stochastic systems , _ ieee trans . 53 , no . 8 , 2008 , pp . 17871803 . i.karatzas , and s.e.shreve , _ brownian motion and stochastic calculus _ , 2nd ed . , springer - verlag , new york , 1991 . h.kwakernaak , and r.sivan , _ linear optimal control systems _ , wiley , new york , 1972 . g.lindblad , on the generators of quantum dynamical semigroups , _ commun . math _ , vol . 48 , 1976 , pp . 119130 . e.merzbacher , _ quantum mechanics _ , 3rd ed . , wiley , new york , 1998 . z.miao , and m.r.james , quantum observer for linear quantum stochastic systems , proc . 51st ieee conf . decision control , maui , hawaii , usa , december 10 - 13 , 2012 , pp . 16801684 . h.i.nurdin , m.r.james , and i.r.petersen , coherent quantum lqg control , _ automatica _ , vol . 45 , 2009 , pp . 18371846 . k.r.parthasarathy , _ an introduction to quantum stochastic calculus _ , birkhuser , basel , 1992 . k.r.parthasarathy , what is a gaussian state ? 4 , no . 2 , 2010 , pp . 143160 . k.r.parthasarathy , and k.schmidt , _ positive definite kernels , continuous tensor products , and central limit theorems of probability theory _ , springer - verlag , berlin , 1972 . i.r.petersen , quantum linear systems theory , proc . networks syst . , budapest , hungary , july 59 , 2010 , pp . l.s.pontryagin , v.g.boltyanskii , r.v.gamkrelidze , and e.f . mishchenko , _ the mathematical theory of optimal processes _ , wiley , new york , 1962 . a.k.sichani , i.g.vladimirov , and i.r.petersen , robust mean square stability of open quantum stochastic systems with hamiltonian perturbations in a weyl quantization form , proc . australian control conference canberra , 17 - 18 november 2014 , pp . 8388 . a.k.sichani , i.g.vladimirov , and i.r.petersen , a gradient descent approach to optimal coherent quantum lqg controller design , _ accepted at american control conference , 2015 _ , preprint : arxiv:1502.00274 , 1 february 2015 . a. van der schaft , port - hamiltonian systems : an introductory survey , proc . congress of math . , madrid , spain , 2006 . i.g.vladimirov , and i.r.petersen , a quasi - separation principle and newton - like scheme for coherent quantum lqg control , _ syst . 62 , no . 7 , 2013 , pp . 550559 . i.g.vladimirov , and i.r.petersen , coherent quantum filtering for physically realizable linear quantum plants , proc . european control conference , ieee , zurich , switzerland , 17 - 19 july 2013 , pp . i.g.vladimirov , a quantum mechanical version of price s theorem for gaussian states , proc . australian control conference , aucc 2014 , ieee , canberra , australia , 17 - 18 november 2014 , pp . 118123 , preprint : arxiv:1409.4377 [ quant - ph ] , 15 september 2014 . i.g.vladimirov , a transverse hamiltonian variational technique for open quantum stochastic systems and its application to coherent quantum control , _ submitted _ , preprint : arxiv:1506.04737 [ quant - ph ] , 15 june 2015 . r.m.wilcox , exponential operators and parameter differentiation in quantum physics , _ j. math . _ , vol . 8 , no . 4 , 1967 , pp .

we report our detailed analysis of a table - top system for the measurement of the weak - force - induced electric dipole moment of a ground state hyperfine transition carried out in an atomic beam geometry . we describe an experimental configuration of conductors for application of orthogonal r.f . and static electric fields , with cavity enhancement of the r.f . field amplitude , that allows confinement of the r.f . field to a region in which the static fields are uniform and well - characterized . we carry out detailed numerical simulations of the field modes , and analyze the expected magnitude of statistical and systematic limits to the measurement of this transition amplitude in atomic cesium . the combination of an atomic beam with this configuration leads to strong suppression of magnetic dipole contributions to the atomic signal . the application of this technique to the measurement of extremely weak transition amplitudes in other atomic systems , especially alkali metals , seems very feasible .

1601.02690

laboratory measurements of very weak atomic transitions that violate the usual parity selection rules are a means of determining the weak force at low collision energies @xcite . the component of this electric dipole transition moment @xmath0 that is induced by the weak - force coupling between nucleons has become of great interest in recent years @xcite . these nuclear spin dependent ( nsd ) contributions to @xmath0 are expected to result from the nuclear anapole moment of the nucleus , with additional smaller contributions from the weak neutral axial - vector nucleon vector electron ( @xmath1,@xmath2 ) current , and the combined effect of the hyperfine interaction and the ( @xmath3,@xmath4 ) current @xcite . to date , the only non - zero determination of nsd contributions to @xmath0 in _ any _ element was based upon the difference between measurements of @xmath5 in atomic cesium @xcite , where @xmath6 is the vector polarizability for the transition , on two different hyperfine components of the @xmath7 transition ; the @xmath8 and the @xmath9 lines . @xmath5 on these lines differed by @xmath105% of their average value . this nsd factor was much larger than was expected , and theoretical efforts @xcite to understand this result have not been successful . meson exchange coupling constants of the so - called ddh model @xcite derived from this result do not agree well with results derived from measurements of the asymmetry in the high - energy scattering of light nuclei @xcite . while the applicability of the ddh model to such a large atom is questionable , there is none - the - less strong interest in understanding the nsd of large nuclei , as evidenced by the many efforts underway worldwide in a variety of systems . laboratory efforts have sought , or are currently underway , to determine the anapole moment of other nuclei , including tl @xcite , yb @xcite , fr @xcite , ba@xmath11 @xcite , ra@xmath11 @xcite , and yb@xmath11 @xcite , and several molecular systems as well @xcite . differences between @xmath0 on various hyperfine lines for these systems could reveal the nuclear anapole moment of these systems . comparison between different isotopes of the same species could remove the dependence of the determination on precise atomic theory , subject to the ability to correct for variations in the nuclear structure among the isotopes @xcite . measurements performed on a hyperfine transition between components of an atomic ground state present an attractive alternative to the above schemes for determining the nsd contributions to @xmath0 . this moment contains only the nsd contribution , simplifying the measurement , and in many cases , the value of @xmath0 on ground state transitions is predicted to be larger than the weak amplitude between different electronic states @xcite . of particular interest is a large program on francium @xcite , one goal of which is to measure @xmath0 on transitions between hyperfine levels of the ground state of this unstable heavy element at triumf . to carry out these measurements , development of techniques for cooling and trapping these species in a magneto - optical trap ( mot ) and carrying out the measurements in this restricted space is necessary . the measurement in atomic cesium that we have under development in our laboratory , which we describe in this work , has several features in common with those of the francium effort . as a ground state transition , atomic coherences are long - lived , and we exploit the interference between the direct transition driven by a radio frequency ( r.f . ) field and the raman process driven by a two - frequency cw laser field , in a derivation of the two - pathway coherent control techniques that we have developed for similar measurements @xcite . atomic cesium offers several benefits over the francium system that are derived from an atomic beam geometry : that is , a greater atomic density , the capability of sequential preparation , interaction , and detection , and a less restrictive experimental environment . furthermore , the beam geometry allow us to spatially separate the interaction regions for the different coherent fields , and to highly suppress the magnetic dipole contributions to the atomic signal , a primary challenge in ground - state measurements of weak signals . in this work , we discuss how the two - pathway interference method can be used to determine the ratio of the pnc amplitude to the stark vector polarizability @xmath6 . while our primary interest is in atomic cesium , the technique is generally applicable in any of the stable alkali metal species . we describe in detail the measurement requirements , and the capability of our technique . the optimal arrangement uses r.f . and static electric fields that are oriented in perpendicular directions , and the r.f . field should be confined to a space within which the static field is uniform . these requirements can be satisfied by a parallel plate transmission line ( pptl ) configuration to which cylindrical reflectors ( to form an r.f . resonant cavity ) and isolated conducting pads ( for application of the orthogonal d.c . field ) have been added . we report the results of our detailed numerical analysis of the electric and magnetic fields supported by this structure , and we use the magnitudes of the field components to estimate the residual systematic effects that one should expect in a determination of @xmath0 in atomic cesium . this paper is organized as follows . in sec . [ sec : cohcontschm ] , we discuss the transition probability of a two - level atom interacting with a resonant r.f . field and a two - frequency optical field through a raman interaction . we show that , when a variable d.c . electric field is applied , this coherent control process allows one to determine @xmath12 . we then discuss in sec . [ sec : expmagnitudes ] the various transition amplitudes , including the magnetic dipole , stark - induced electric dipole , and weak - induced electric dipole , for the transition between hyperfine levels of the ground state of an alkali metal atom . we present an estimate of the signal size in section [ sec : signalsize ] , with an estimate of the statistical uncertainty , and review the benefits of carrying out the measurement in a standing wave cavity for suppression of magnetic dipole contributions in section [ sec : stwvcav ] . in the following section , we introduce the pptl structure , and describe the field modes supported by it . finally , we analyze the magnitudes of the dominant residual contributions to the measurement of @xmath0 , and consider the effects of the distribution of atomic velocities in the beam . we employ the two - pathway coherent control scheme for sensitive measurement of weak moments . this technique is based on the interference between various optical interactions driven by two or more coherently - related fields . we developed and employed this technique on measurements of the magnetic dipole transition moment @xmath13 on the @xmath14 transition in atomic cesium @xcite . the fr collaboration bases their measurements on this technique also @xcite . in this section , we describe the principles behind this technique , with particular attention paid to a transition between hyperfine components of a ground state system , in which both states are long lived . we show how this measurement can yield a determination of @xmath12 , independent of the profile or amplitude of the r.f . field that drives the transition . we consider a sinusoidal wave of amplitude @xmath15 and frequency @xmath16 , incident upon a two - level atom with hyperfine components @xmath17 and @xmath18 of the ground state , of energy @xmath19 and @xmath20 , respectively . we choose the field to be continuous wave , but spatially varying , such that as the atoms move across the interaction region , they effectively see a time - varying field . when the atoms are initially prepared in a single hyperfine component @xmath21 , and when the field components are chosen so as to couple the initial state to a single final state @xmath22 , the atomic system is very closely described as a two - level system , and we can write the state of the atoms using the time - varying amplitudes @xmath23 and @xmath24 as @xmath25 the time evolution of the system is described in terms of the hamiltonian @xmath26 , where @xmath27 is the atomic hamiltonian and @xmath28 describes the interaction between the atom and the field . in this work , we consider the weak - force induced electric dipole interaction @xmath29 , the stark - induced electric dipole interaction @xmath30 , and the magnetic dipole interaction @xmath31 of the atom with the r.f . field , plus a raman interaction @xmath32 of the atom with a two - frequency laser field , all of which we describe in more detail later , and write @xmath28 as the sum of the individuals @xmath33 we illustrate these interactions schematically in fig . [ fig : cs_energy_levels_hfs_ground ] . , where @xmath34 . through the interactions with the r.f . field and the optical field , some of the atoms are transferred to the level @xmath35 . in this figure , we show @xmath36 as the initial state , and @xmath37 as the final state . , title="fig:",width=207 ] + when the atoms exit the interaction region , the probability that they are in state @xmath22 is @xmath38 where the @xmath39 are the integrated interaction strengths of any of the individual interactions @xmath40 the rabi frequencies of the various interactions are @xmath41 , and @xmath42 represents the reduction in amplitude when the r.f . frequency is detuned from the resonant frequency by @xmath43 . @xmath42 depends on the temporal shape of the ` pulse ' as the atoms travel through the interaction region in a non - trivial way , and we will limit our discussion to resonant excitation , @xmath44 , for which @xmath45 . in an atomic beam , collisions are infrequent , and the atoms travel through the interaction region with a constant velocity @xmath46 . in this case , the interaction strength can be rewritten @xmath47 we use notation similar to that of gilbert and wieman @xcite for each of the various interactions , and show the optimal field geometry for this measurement in fig . [ fig : e_pnc_hfs_expgeom_1015 ] . on the @xmath48 , @xmath49 ground state hyperfine transition . the static electric and magnetic fields are oriented in the @xmath50-direction , while the polarization of the r.f . field is in the @xmath51-direction . the polarizations of the laser field components that drive the raman interaction , not shown , are aligned with the @xmath51- and @xmath50-axes . the r.f . and raman fields propagate parallel to one another , shown as the direction @xmath52 , as required to maintain a uniform phase difference between interactions . , title="fig:",width=188 ] + that is , the r.f . and raman fields propagate in the @xmath53-direction , the d.c . electric @xmath54 and magnetic @xmath55 fields are oriented in the @xmath50-direction , and the electric field @xmath56 of the r.f . field is directed in the @xmath51-direction . ( parallel propagation of the r.f . and raman fields is necessary to maintain a uniform phase difference between interactions . ) not shown in this figure are the two components of the laser electric field that drive the raman transition , each linearly polarized , one in the @xmath51-direction , the other in the @xmath50-direction . in this geometry , the primary r.f . and raman fields each independently drive a @xmath48 , @xmath49 transition , the magnetic dipole contribution on this transition is suppressed , and the stark - induced and the pnc interactions are in quadrature - phase with one another . the primary contributions here , under the precise ( idealized ) conditions specified in fig . [ fig : e_pnc_hfs_expgeom_1015 ] , are @xmath57 and @xmath58 in eq . ( [ eq : astprimary ] ) , @xmath6 is the vector polarizability and @xmath59 is a factor related to the clebsch - gordon coefficients , defined in detail in ref . note that we have explicitly included the phase of the r.f . field in these expressions . in addition to these primary amplitudes , extra contributions due to magnetic dipole transitions and field misalignments can arise . the largest of these is @xmath60 c_{fm}^{f^{\prime}m\pm 1 } \right . \\ & & \hspace{0.1 in } + h_z^{\rm rf } \left ( \frac{\pm b_x^0 + ib_y^0}{b_z^0 } \frac{c_{fm}^{f^{\prime } m } c_{f^{\prime } m \pm 1}^{f^{\prime}m}}{g_{f^{\prime } } } \right . \nonumber \\ & & \left . \left . \hspace{-0.3 in } + \frac{\mp b_x^0 + ib_y^0}{b_z^0 } \frac{c_{fm}^{f m\pm 1 } c_{f m \pm 1}^{f^{\prime}m \pm 1}}{g_{f } } \right ) \right\ } e^{i ( \omega^{\rm rf } t - k y - \phi^{\rm rf } ) } \nonumber \end{aligned}\ ] ] for @xmath49 transitions , where the @xmath61 are the components of the magnetic field of the r.f . wave , @xmath13 is the magnetic dipole transition moment , @xmath62 is the impedance of vacuum , and @xmath63 and @xmath64 are the gyromagnetic ratio of the initial and final states . for cesium , @xmath63 is @xmath65 for the @xmath66 level and @xmath67 for the @xmath68 level of the ground state . the first terms in eq . ( [ eq : amdelmpm1 ] ) are the magnetic dipole amplitude driven by the @xmath69 and @xmath70 field components , while the last terms in @xmath71 and @xmath72 or @xmath73 arise from zeeman mixing of the hyperfine components by the static magnetic field . to investigate possible interferences from @xmath74 transitions , we also present the magnetic dipole transition amplitude for these transitions @xmath75 \right . \nonumber \\ & & \hspace{-0.3 in } \times \left [ \left ( \frac{\mp b_x^0 + ib_y^0}{b_z^0 } \right ) \frac{c_{fm}^{f^{\prime } m \pm1 } c_{f^{\prime } m } ^{f^{\prime } m \pm 1 } } { g_{f^{\prime } } } \right . \\ & & \hspace{-0.5 in } + \left . \left ( \frac{\pm b_x^0 + ib_y^0}{b_z^0 } \right ) \frac{c_{fm}^{f m \mp 1 } c_{f m \mp 1}^{f^{\prime } m } } { g_{f } } \right ] \right\ } e^{i ( \omega^{\rm rf } t - k y - \phi^{\rm rf } ) } . \nonumber \end{aligned}\ ] ] in addition to these transitions driven by the r.f . field , we consider the raman transition of the form @xmath76 where @xmath77 and @xmath78 are the electric field amplitudes of the two laser components , and @xmath79 , where @xmath80 and @xmath81 are the optical frequencies . the phase @xmath82 is the phase difference between the phases of the two components @xmath83 . the raman polarizability @xmath84 depends on the detuning @xmath85 of these field components from the d@xmath86 transition frequency , and the raman transition can be enhanced by making @xmath87 small . we will analyze these r.f . transition amplitudes later using electric and magnetic field amplitudes that we expect to encounter for our parallel plate structure to place limits on unwanted magnetic dipole contributions to the pnc signal . before we do this , we return to eq . ( [ eq : cfeqsinsq ] ) , which we examine in the limit of the raman interaction strength @xmath88 being much greater than any of the interactions driven by the r.f . field @xmath89 , @xmath90 , and @xmath91 . under these conditions , and with the detuning @xmath44 , eq . ( [ eq : cfeqsinsq ] ) can be expanded to the form @xmath92 . \nonumber\end{aligned}\ ] ] @xmath93 is the controllable phase difference between the r.f . field and the phase difference @xmath82 , and @xmath94 is the phase shift introduced by the quadrature combination of @xmath0 and @xmath95 . ( in writing this phase shift , we presume that the magnetic dipole contributions are suppressed , as we show later . ) we see from this expression a feature that is similar to that of the coherent control scheme on a short - lived state @xcite ; that is , that the signal consists of a d.c . term resulting from the raman interaction alone , plus a sinusoidally - varying contribution that varies with the phase difference @xmath96 between the raman field and the one - photon r.f . field . furthermore , the amplitude of the modulating term is the magnitude of the sum of interaction angles @xmath97 , where we have omitted the small magnetic dipole integrated angle in the final step . a laboratory measurement of this population modulation amplitude as a function of the d.c . electric field @xmath98 yields @xmath99 . we see this as follows . @xmath100 \ : dz \right|,\ ] ] which , using eqs . ( [ eq : astprimary ] ) and ( [ eq : apncprimary ] ) becomes @xmath101 valid when @xmath98 is uniform in the interaction region . since the stark and pnc moments add in quadrature , the amplitude of the sinusoidal modulation of the signal scales as @xmath102 at small d.c . field , the modulation amplitude is proportional to @xmath103 alone , while at large field , the modulation amplitude is nearly proportional to @xmath95 . by measuring this amplitude of the population modulation as a function of the d.c . field , therefore , one can determine the ratio @xmath104 . to optimize the amplitude of the signal modulation in eq . ( [ eq : cfeqdcpmodterm ] ) , one should adjust the amplitude of the raman interaction to @xmath105 . at this value , the factor @xmath106 is equal to 1 , and the atomic population due to the raman interaction alone is equal to 1/2 , i.e. equal probability in the initial and final states . any additional interactions of the atom with the r.f . field add ( slightly ) to the population in the @xmath22 state when this interaction is in phase with the raman interaction , and subtract when out - of - phase . we can gain some insight into the interference by following the evolution of the amplitudes @xmath107 ( red solid ) and @xmath108 ( blue dashed ) as the atoms move across the interaction region , which we show in fig . [ fig : cevstime ] . for this illustration , the atoms move from left to right , and encounter the raman field first , centered at @xmath50 = -4 cm , which prepares them in a coherent superposition state . the atoms then enter the broad r.f . field . we use gaussian profiles for the r.f . and raman fields . for the former , the peak amplitude is @xmath109 and beam radius @xmath110 in the interaction region , @xmath111 we show this for two values of the phase @xmath96 in fig . [ fig : cevstime ] . ( red solid ) and @xmath112 ( blue dashed ) versus @xmath50 as the atoms pass through the interaction region from left to right . the atoms are prepared by the raman beams in a superposition state before entering the broad r.f . both fields are gaussian in shape , with peak rabi frequency and beam radii of @xmath113 = 23.9 ms@xmath114 and 0.5 cm for the raman beam , and @xmath115 = 0.61 ms@xmath114 and 2.5 cm for the r.f .- driven interaction . in ( a ) , the raman and r.f . interactions are in phase with one another , while in ( b ) , the interactions are out of phase . in either case , the duration of the interaction is @xmath116 19 @xmath117s for the raman beam , and @xmath118 93 @xmath117s for the r.f . field . , title="fig:",width=294 ] + fig . [ fig : cevstime](a ) shows the magnitudes of the state amplitudes when the raman and r.f .- driven interactions are in phase with one another , while fig . [ fig : cevstime](b ) shows the state amplitudes when the interactions are @xmath119 out of phase with one another . the peak rabi frequency , center position , and beam radius are @xmath113 = 23.9 ms@xmath114 , @xmath120 = -4 cm , and @xmath121 = 0.5 cm for the raman beam , and @xmath115 = 0.61 ms@xmath114 , @xmath120 = 0 , and @xmath110 = 2.5 cm for the r.f .- driven interaction . we use 270 m / s , the peak velocity of the atoms in our atomic beam for @xmath46 . the duration of the interaction is @xmath116 19 @xmath117s for the raman beam , and @xmath118 93 @xmath117s for the r.f . field . when the amplitudes are in phase with one another , @xmath122 grows monotonically , while when the interactions are out of phase , the amplitude decreases after its initial preparation by the raman beam . the value of @xmath123 after the atoms have exited the interaction region is @xmath124 for in - phase interactions and @xmath125 for out - of - phase interactions . when the pnc and stark - induced terms are driven by the r.f . field , then @xmath126 is @xmath127 , where the pnc interaction angle is @xmath128 similarly , the integrated area of the stark - induced interaction angle for this gaussian - shaped profile is @xmath129 . the term @xmath130 in the expressions for @xmath123 comes from @xmath131 with @xmath132 . the weak signal strength is @xmath133 in this example is 0.10 . any interaction of the atoms with the r.f . field therefore is evident as a modulation of this signal as we vary the phase difference between the fields . we illustrate this in fig . [ fig : sigvsphase ] , which shows the sinusoidal modulation of the final state population as a function of @xmath96 . = 0.061 ms@xmath114 for this plot . other parameters are as given in the caption to fig . [ fig : cevstime ] . , title="fig:",width=294 ] + here the parameters are as they were in fig . [ fig : cevstime ] , with the exception of @xmath115 which we have decreased to 0.061 ms@xmath114 for this figure . the amplitude of the modulation of @xmath134 is @xmath135 , in agreement with the numerical data in the figure . in our simulations , the amplitude of the modulation scales linearly with the weak amplitude . important conditions and features of this measurement technique include : 1 . mutual coherence of the different time - varying fields is required . this can be implemented in the laboratory by using non - linear mixing , injection locking of diode lasers , or frequency modulation techniques . 2 . the coherent beams that drive the interactions must propagate in the same direction in order to maintain a uniform phase difference for all atoms in the interaction region . the raman and the r.f . field distribution need not overlap one another . since the ground state is long lived , the final level retains its coherence , and the net excitation of the final state depends on the accumulated effect across the interaction region . we control the phase difference between the transition amplitudes with r.f . devices , completely external to the interaction region . we select the particular interactions that contribute to the measurement by choosing the orientation of the various fields in the interaction region . the measurement uses only modest d.c . electric fields , @xmath136 100 v / cm . this allows flexibility in the experimental configuration . since the interactions @xmath137 and @xmath138 are @xmath139 out of phase with one another , these amplitudes add in quadrature . this indicates that the amplitude of the modulating signal is at a minimum when the static electric field is turned off , and increases when a static field of either polarity is applied . using different field orientations , this coherent control technique may be used to determine @xmath140 . this may be a useful means of determining the vector polarizability @xmath6 , but we defer any further discussion of this to a future report . in the following sections , we will discuss the expected magnitudes of the different interactions , and present an experimental assembly of conductors for such a measurement in an atomic beam configuration . finally , we will analyze the effect of expected magnetic dipole contributions to the measurement . in order to design a measurement system and understand the effect of stray fields and the magnitude of unwanted contributions to the signal , we must first know the expected magnitudes of the pnc moment , @xmath0 , the vector polarizability @xmath6 , and the magnetic dipole moment @xmath13 for the transition . the pnc amplitude for this transition is calculated @xcite to be @xmath141 where @xmath142 and @xmath143 are the electron charge and the bohr radius , respectively . this is larger than @xmath0 for the moment on the @xmath7 transition in cesium by a factor of 2.2 . the vector polarizability has not previously been calculated , but we can estimate its approximate magnitude using the sum - over - states expansion of refs . @xcite and @xcite , @xmath144,\end{aligned}\ ] ] where @xmath145 represents the reduced dipole matrix elements @xmath146 for @xmath147 or @xmath148 , and @xmath149 are the energy differences @xmath150 for the two hyperfine states @xmath66 or @xmath151 of the ground @xmath152 and the excited @xmath153 states . the @xmath154 term dominates this sum , and the ground state hyperfine splitting @xmath155 is small compared to the energy of the 6p states , so the polarizability is approximately @xmath156 . \nonumber\end{aligned}\ ] ] we use @xmath157 and @xmath158 @xcite to estimate the vector polarizability for this transition as @xmath159 . based on these expected magnitudes of @xmath6 and @xmath0 , the ratio @xmath99 is about 27 v / cm ; upon application of a static electric field of this magnitude , the magnitudes of the stark - induced amplitude and the pnc amplitude are equivalent . since @xmath6 is so small for this transition , we conclude that systematic errors due to uncontrolled electric fields in the interaction region , due to surface contamination and patch effects and estimated to be @xmath1360.1 v / cm , are inconsequential in these ground state measurements . this is in strong contrast to measurements of @xmath0 on the @xmath7 transition @xcite , for which uncontrolled electric fields were of major concern . in addition to these two relatively weak amplitudes driven by the r.f . field , the magnetic dipole moment is active on this transition . the amplitude for this transition is @xmath160 , where @xmath161 is the magnetic moment of the atom , @xmath162 is the bohr magneton , and @xmath163 is the magnetic flux density of the r.f . wave . @xmath164 , @xmath165 and @xmath166 are the usual orbital , spin , and nuclear angular momenta , and @xmath167 , @xmath168 , and @xmath169 are the respective gyromagnetic ratios . for the transition of this work , the orbital angular momentum is zero , and @xmath169 is much less than @xmath168 ( which is @xmath170 ) due to the heavy mass of the nucleus . for the ground state transition , the spatial parts of @xmath21 and @xmath22 are the same , and using @xmath171 , the magnetic dipole amplitude is @xmath172 . but @xmath173 , where @xmath174 is the fine structure constant , so @xmath175 , and the ratio @xmath176 . the magnetic dipole contributions to the signal must be suppressed for a successful measurement of @xmath0 , representing the primary challenge of these measurements . the orientations of the field components that we have shown in fig . [ fig : e_pnc_hfs_expgeom_1015 ] are an important first step in meeting this challenge . in this section , we will use the results of the analysis of sec . [ sec : cohcontschm ] , in particular eqs . ( [ eq : cfeqdcpmodterm ] ) and ( [ eq : theta_pnc ] ) , and the calculated value of @xmath0 given in eq . ( [ eq : epnc ] ) , to estimate the magnitude of the pnc signal , and from this the integration time required to achieve a useful statistical uncertainty of the measurement . to calculate the signal size , we will use @xmath177 , @xmath178 = 250 v / cm , and @xmath110 = 2.50 cm . the value of @xmath179 is valid for cesium ground state transitions @xmath180 or @xmath181 , and we will show in sec . [ sec : pptl ] that the values of the peak field amplitude and radius are reasonable . then using the cesium atomic beam peak velocity @xmath46 = 270 m / s , we estimate that the interaction angle for the pnc interaction is @xmath182 to measure this amplitude , one can drive the interfering raman and pnc interactions , and count the transition rate as a function of the phase difference between the transitions . a minimal measurement may consist of @xmath183 , the total count of atomic excitations when the r.f . and raman interactions are in phase with one another ( @xmath184 ) , and @xmath185 the total count of excitations when the r.f . and raman interactions are @xmath119 out of phase with one another ( @xmath186 ) . then @xmath187 to use this result to determine @xmath0 , however , one must also have an accurate determination of the r.f . beam profile and field amplitude . alternatively , one can apply a d.c . electric field to the atoms , and measure the amplitude of the modulation as a function of the field amplitude @xmath188 , as suggested in eqs . ( [ eq : thetastplpnc ] ) and ( [ eq : quadsumtplpnc ] ) . when the precision of @xmath183 and @xmath185 is limited by counting statistics , then the uncertainty in either of these counts is @xmath189 , where @xmath190 represents either @xmath183 or @xmath185 ( which are essentially the same ) . the uncertainty in @xmath91 is @xmath191 , and to achieve a 3% measurement of @xmath91 , one must count @xmath192 atoms for each individual measurement . in a counting interval @xmath193 , the number of counts is @xmath194 , where @xmath195 is the average excitation probability , @xmath196 is the number density of the atomic beam ( @xmath197 @xmath198 ) , @xmath199 is the cross sectional area of the atomic beam ( 1 mm@xmath200 ) , and @xmath46 is the peak velocity of the atoms in the beam . the counting time t to achieve the required statistics is 20 seconds per data point . during the course of a measurement , one must repeat the process at many different phases , not just two , and one must vary the d.c . electric field strength @xmath98 and repeat the measurement . regardless , the estimate of the integration time shows that the measurement is feasible in the beam geometry . we conclude this section with an estimate of the maximum value of the d.c . field amplitude @xmath201 needed . as discussed in the previous section , we expect that the ratio @xmath202 is approximately 27 v / cm . in carrying out the measurements , we must vary the stark - induced angle @xmath89 over the range from zero to @xmath203 . this requires a variable field strength of maximum value @xmath20480 v / cm . in the previous section , we estimated the magnitude of the hyperfine ground state pnc coherent control signal , based on expected atomic parameters and reasonable field strengths that can be generated in the laboratory . among the latter was an r.f . field amplitude @xmath178 of 250 v / cm . this field amplitude can be achieved either inside a resonant power build - up cavity , or by using a very large r.f . use of a resonant cavity also helps to suppress the magnetic dipole contributions to the measured signal , as we now discuss . this approach is also discussed in ref . @xcite . as we discussed earlier , the large magnetic dipole amplitude is suppressed to first order by the choice of orientations of the primary fields . ( the @xmath71 component drives a @xmath205 transition , whereas the interference that we have discussed takes place on a @xmath206 transition . ) still , due to the large magnitude of the ratio @xmath207 and reasonable limits in the field uniformity and experimental alignment , additional measures are required to suppress this interaction further . this additional suppression can be achieved by working in a standing wave configuration , in which the nodes of the magnetic field coincide with the anti - nodes of the electric field , as we illustrate in fig . [ fig : beeps_orientations_swp ] . and magnetic field @xmath71 , with the atomic beam located at the node of the magnetic field . , title="fig:",width=207 ] + at this point , the interactions @xmath29 and @xmath30 are maximized , and @xmath31 is minimized . to take best advantage of this , one should ( 1 ) use a cavity geometry in which the amplitudes of the traveling waves propagating in the @xmath208 and @xmath209 directions , @xmath210 and @xmath211 , respectively , are equal , and ( 2 ) keep the radius @xmath212 of the atomic beam small . the first requires either that the cavity is symmetric ( the reflectivities of the two end reflectors are equal , and the cavity is excited by equal amplitude inputs on each side ) , or that one of the reflectors has unit reflectivity . the choice of beam radius @xmath212 is a compromise between large atom number , improving the counting statistics , or small magnetic dipole amplitude for atoms at the edge of the beam , which scales as @xmath213 , where @xmath214 = 3.2 cm is the wavelength of the 9.2 ghz wave . for @xmath212 = 0.5 mm , this reduction factor is @xmath100.1 . furthermore , the sign of the magnetic dipole amplitude is opposite on the two sides of the node , further reducing this contribution . we will return to this reduction in section [ sec : systerrors ] . in the next section , we will discuss the design and analysis of a symmetric r.f . power build - up cavity based on a parallel plate transmission line structure , which allows spatial confinement of the r.f . field and generation of a transverse d.c . electric field . the measurement that we have described presents several experimental challenges . first , we must apply r.f . and static electric fields that are oriented in directions that are perpendicular to one another . second , we require that the r.f . field is in a standing wave configuration for suppression of the magnetic dipole contributions . and third , we must minimize the unwanted field components of the r.f . field , as these also lead to systematic magnetic dipole contributions to the signal . in this section , we describe an electrode configuration that allows us to meet these requirements . and the static electric field @xmath98 . , title="fig:",width=298 ] + in fig . [ fig : pptl_structure_w_ref_rotated ] , we show a section of a parallel plate transmission line , with waves propagating in the @xmath215 directions , that is modified in two regards . first , we have isolated several conducting pads on the top and bottom conductors for application of a d.c . bias , and secondly , we have inserted cylindrical reflectors to either side of the interaction region to form an r.f . cavity , open on the @xmath50 faces , allowing power build - up of the cavity mode at the resonant frequency . when we have biased the d.c . pads progressively , at a voltage @xmath216 on one side to @xmath217 on the other , we can generate an electric field @xmath201 in the central region between the plates that is primarily directed in the @xmath218-direction . we capacitively couple each of the bias pads to the transmission line structure so that they carry the a.c . components without any significant perturbation . for a transmission line characteristic impedance @xmath219 = 50 @xmath220 , this requires a coupling capacitance of @xmath221 30 pf . we can model the cavity modes that are supported by the parallel - plate structure in the region between the cylindrical reflectors approximately using the elliptical hermite - gaussian modes as described in yariv @xcite . these modes are nearly gaussian in shape in the @xmath50-direction , but uniform in the @xmath51-dimension , in the limit of an infinite beam size in this dimension . within the cavity , the spatial mode is described by the superposition of waves traveling in the @xmath208 and @xmath209 directions , @xmath222 and @xmath223 where @xmath224 \hspace{0.2 in } \right . \\ & & \left . \hspace{0.9 in } -z^2 \left [ \frac{1}{w^2(y ) } + \frac{ik}{2r(y ) } \right ] \right\ } , \nonumber\end{aligned}\ ] ] in these expressions , @xmath225 is the @xmath226 ( intensity ) beam radius at the focus , the beam profile radius a distance @xmath53 from the focus is @xmath227 @xmath228 is the confocal parameter @xmath229 @xmath230 is the radius of curvature of the wavefronts @xmath231,\ ] ] and @xmath232 @xmath233 is the slow phase shift ( the guoy phase ) through the focal region . for a symmetric cavity constructed of cylindrical reflectors of radius of curvature @xmath234 separated by a distance @xmath235 , the confocal parameter is @xmath236 , the beam radius at the center is @xmath237 , and the beam radius at the reflectors is @xmath238 . the cavity mode has an electric field anti - node ( and magnetic field node ) at the center when the cavity length @xmath235 is approximately @xmath239 , where @xmath240 is an integer . the r.f . beam radius @xmath241 at the reflectors is minimized when the reflector spacing is confocal , i.e. @xmath242 . by adjusting the reflector slightly away from the confocal spacing , one can retain the small mode size @xmath243 at the reflectors , but shift the frequencies of the transverse modes away from the frequency of the lowest order mode , improving the selectivity of cavity modes . we calculate that for @xmath244 cm and @xmath245 cm , the cavity has a resonance at the cesium hyperfine transition frequency ( 9.2 ghz ) , its free spectral range ( fsr ) is @xmath246 ghz , the beam radius at the waist is 2.50 cm , the beam radius at the reflectors is 3.53 cm , and the transverse mode spacing is 0.2487 times the fsr , or about 313 mhz . we estimate the field amplitude at the interaction region as follows . we choose the spacing between the parallel plates of the transmission line to be 1 cm , and the conductor width 7.5 cm . these dimensions yield a characteristic impedance of the transmission line of 50 @xmath220 , and allow for a reasonable clearance of the atomic beam in the space between the conductors . with a copper thickness on the reflectors of 170 nm , we calculate a reflection coefficient of 0.9992 . note that this thickness is smaller than the skin depth @xmath247 = 680 nm of copper at this frequency , so the transmission losses are small , but not negligible . with this reflectivity , the cavity losses due to reflection are of the same magnitude as the losses @xmath248 due to other mechanisms , primarily conduction losses in the upper and lower conducting plates , and diffraction losses due to the finite size of the conductor . ( these results come from our numerical analysis of the cavity modes , which we discuss next . ) for an r.f . input power of 250 mw incident on the cavity from either side , the incident voltage amplitude is 5.0 v , and the electric field of the traveling wave incident on the cavity is @xmath249 = 5.0 v / cm . the amplitude of the traveling wave inside the cavity is @xmath250 where we use @xmath251 = 0.04 for the transmission coefficient of the reflector and @xmath252 . the factor 2 results from symmetric inputs from the two sides . at the anti - node of the field , where the amplitudes of the two traveling waves inside the cavity add in phase , the field amplitude is twice this value , or 250 v / cm . this is the value of the r.f . field amplitude that we used in sec . [ sec : signalsize ] to estimate the signal size . in making this estimate , we have not included the absorption of the copper reflector , which reduces the amplitude , or the increase of the wave amplitude as the wave propagates to the waist of the gaussian profile . in order to determine more - detailed field parameters , we have carried out a series of numerical simulations of the cavity mode using comsol multiphysics . these simulations allow us to determine the effects of resistive losses of the parallel plates , the thickness of the reflective copper layers , and the finite width of the cavity on the cavity @xmath253 ; the effect of the gaps in the conductor between the d.c . bias pads ; and the uniformity of the static electric field in the interaction region . we show the three primary components , @xmath254 $ ] , @xmath255 $ ] , and @xmath256 $ ] , of the simulated r.f . field mode in fig . [ fig : mode_patterns ] . we note very close agreement of the components @xmath257 with the analytic result in eq . ( [ eq : exswfoc ] ) and @xmath258 with eq . ( [ eq : hzswfoc ] ) . + ( a ) ( b ) ( c ) the component @xmath256 $ ] would be negligible for a weakly focused beam , but since in our geometry , @xmath259 , this component survives . for this figure , the separation between the upper and lower conducting planes of the pptl and the width of the conductors are as before , 1.0 cm and 7.5 cm , respectively , as are the radius of curvature of the cylindrical reflectors @xmath234 = 12.0 cm , and the reflector separation @xmath235 = 11.9 cm . with the thickness of the copper reflector layers equal to 200 nm , we determine a cavity @xmath253 of @xmath260 , while for a 1.5 @xmath117 m @xmath261 layer , the @xmath253 increases to 13,000 . in the latter case , the @xmath253 is limited primarily by the resistive losses in the conductors and diffraction losses of the finite width of the reflectors . for a cavity @xmath253 of 9000 , the linewidth of the transmission peak of the cavity is @xmath262 1 mhz . we show the computed gaussian r.f . field amplitude , @xmath263 across the interaction region as the solid blue curve in fig . [ fig : field_profiles ] . ( red dashed ) and @xmath264 ( black dotted ) , each normalized to the maximum magnitude of @xmath265 , at the center of the region between the parallel conducting plates . also shown is the gaussian r.f . field amplitude , @xmath263 ( blue solid ) . , width=264 ] the diameter of the cavity mode agrees well with @xmath266 = 5.0 cm that we determined analytically earlier . we used the eigenfrequency module and frequency domain analysis to carry out these calculations , and determined the quality factor of the cavity as the ratio of the energy stored inside the cavity to the diffraction and dissipation losses . we obtained the field patterns by launching a 9.2 ghz plane - wave - like electric field on the parallel plate transmission line towards the cavity , exciting a te@xmath267 cavity mode , where indices @xmath268 and @xmath240 label the transverse and longitudinal modes . the mode spacing between the te@xmath267 and te@xmath269 mode agrees well with the 1.26 ghz fsr that we determined earlier . we used a trial - and - error approach to reduce the diffraction losses by varying the cavity parameters , such as the width and height of the cavity , while maintaining the resonant mode frequency close to 9.2 ghz . in order to calculate the r.f . field distributions in a more refined manner in the interaction region , we added about ten thousand times more mesh points in the vicinity of the interaction region . higher mesh point density helped to reduce errors that are present in the interpolation schemes , without compromising the eigenfrequency calculations . we used ten bias pads , with the spacing between the pads about one tenth the width of the pads . as long as the transmission lines are thin ( less than 0.1 mm ) , the gaps have little impact on the r.f . fields . we found that neither horizontal nor vertical misalignment of the cylindrical reflectors affects the field patterns or the @xmath253 factor , for misalignment less than 1 degree . we have also modeled all components of the static electric field @xmath54 , and shown that with an array of 10 bias pads and @xmath270 = 100 v between pads , we can generate a relatively uniform field of magnitude @xmath271140 v / cm . we show this field , normalized to its maximum value , as the red dashed line in fig . [ fig : field_profiles ] . we also show @xmath264 in the plot ( black dotted line ) , which is small in magnitude , and has an average value of zero . the non - uniform part of @xmath265 , seen in fig . [ fig : field_profiles ] as a nearly sinusoidal modulation of amplitude @xmath107% of the constant part , has little impact on the measurement . we can see this by integrating the product @xmath272 across the interaction region in @xmath50 . for the case of ten bias pads , as shown , the correction to the signal due to the sinusoidal modulation is less the 0.7% of the signal . we can also see in this figure that the width of the gaussian shaped r.f . field profile is somewhat less than the width of the d.c . field , allowing us to avoid fringe effects of the d.c . field near the edges of the conductors . we have used these simulations of the field amplitudes , and their variation through the interaction region , to estimate systematic contributions to the pnc signal . we discuss these contributions in the next section . in this section , we will make use of the field simulations of the previous section in order to estimate the expected systematic contributions to the pnc signal . the primary contributions that must be considered are the magnetic dipole terms , due to the relatively large magnetic dipole moment @xmath13 on this transition . as we have shown , the primary magnetic field components of the r.f . field are @xmath273 and @xmath274 , where we use primed ( double - primed ) variables for the real ( imaginary ) part of the field quantities , and omit the superscript ` rf ' . by setting up the geometry of the experiment to make the atomic beam cross the r.f . field at the center of the cavity , where the component @xmath275 is minimal , the magnetic dipole contributions to the signal from any individual atom can be reduced . furthermore , the contributions from atoms on one side of the node are of opposite sign to those on the other side of the node , and the net magnetic dipole contribution can be suppressed even further . in this section , we use the numerical simulations of the fields supported by the resonant cavity to explore the magnitude of magnetic dipole contributions to the pnc signal . the net contribution of the @xmath273 term can be minimized by adjusting the relative position @xmath276 of the center of the atomic beam relative to the node of the magnetic field . ( no control of the @xmath51-position is necessary , since the fields are uniform in this direction . ) for a uniform beam density with beam radius @xmath212 , the magnitude of the @xmath273 contribution grows as @xmath277 ( independent of @xmath212 ) . with piezoelectric adjustment and feedback control of the distance @xmath276 , such that @xmath276 is maintained to a value less than , say , 10 @xmath117 m , then the magnetic dipole signal is suppressed by an additional factor of @xmath278 . this approach to suppression of this magnetic dipole contribution is applicable even with less than ideal beam symmetries , such as an asymmetric beam density , a misalignment of the beam with the z - axis , or divergence of the atomic beam . due to the sign reversal of @xmath273 across the node , there must always exist a beam location for which this signal averaged over all the atoms in the beam vanishes . for the @xmath274 contributions , the zero is also guaranteed , but for a different reason . as can be seen in fig . [ fig : mode_patterns ] , this field component is maximal along the atom beam at the center of the cavity , @xmath279 , but is an odd function of @xmath50 . since the magnetic dipole contribution scales as the integral of @xmath274 across the interaction region , and since @xmath280 , by faraday s law , is proportional to @xmath281 , the path integral of @xmath274 across the interaction region depends only on @xmath282 at the beginning and end of the path . but @xmath282 is zero far from the center of the pptl , so this contribution also vanishes . for these various reasons , we find that the average value of @xmath283 ( including effects of the divergence of the atomic beam ) is equal to zero for all components of @xmath284 , where the average is computed over all atoms in the beam , and presuming that we have successfully adjusted the node of @xmath285 to be co - located with the center of the atomic beam . @xmath90 for individual atoms may have non - zero values , but when averaged over all atoms , every one of these terms vanishes . in table [ table : gs_systematics ] , we have listed the magnetic dipole contributions that appeared in eq . ( [ eq : amdelmpm1 ] ) . ' '' '' comp . & @xmath286 & @xmath287_{\rm rms}$ ] & magnetic dipole contribution & @xmath288_{\rm rms } / \sqrt{n}$ ] + + @xmath289 & 0 & 40 @xmath117a & @xmath290_{\rm rms } / \hbar v \sqrt{n}$ ] & @xmath291 + @xmath292 & 0 & 7 na & @xmath293_{\rm rms } / \hbar v \sqrt{n}$ ] & @xmath294 + @xmath295 & 0 & 0.1 a & @xmath296_{\rm rms } / \hbar v \sqrt{n } \times ( b_x^0/b_z^0)$ ] & @xmath297 + @xmath298 & 0 & 8 na & @xmath299_{\rm rms } / \hbar v \sqrt{n } \times ( b_y^0/b_z^0)$ ] & @xmath300 + + @xmath301 & 0 & 5 na & @xmath302_{\rm rms } / \hbar v \sqrt{n}$ ] & @xmath303 + @xmath304 & 0 & 50 @xmath117a & @xmath305_{\rm rms } / \hbar v \sqrt{n}$ ] & @xmath306 + @xmath307 & 0 & 8 na & @xmath308_{\rm rms } / \hbar v \sqrt{n } \times ( b_x^0/b_z^0)$ ] & @xmath300 + @xmath309 & 0 & 0.1 a & @xmath296_{\rm rms } / \hbar v \sqrt{n } \times ( b_y^0/b_z^0)$ ] & @xmath297 + these terms can potentially contribute to @xmath49 transitions , and are therefore candidates for obscuring the @xmath0 signal . in the second column of this table , we list the average value of each of the @xmath283 terms , which we have already argued must vanish in each case . a better metric for comparison is therefore the standard deviation of the distributions of the relevant path integrals of @xmath310 , which we list in the third column of table [ table : gs_systematics ] . for this calculation , we computed the distribution of the integrals @xmath283 ( separately for the real and imaginary parts ) over various straight - line paths through the interaction region , and determined the width of this distribution by calculating the root - mean - square value , @xmath311_{rms}$ ] . after @xmath190 atoms have traveled through the interaction region , the standard deviation of the mean of these integrals is @xmath311_{rms}/\sqrt{n}$ ] . we use @xmath312 for this purpose , the number of atoms that we computed would be necessary to produce a shot - noise limited measurement with an uncertainty of 3% in section [ sec : signalsize ] . multiplying by @xmath313 ( which numerically is equal to 380 a@xmath114 , which we compute using @xmath314 cm and @xmath46 = 270 m / s ) yields the rms magnitude of each magnetic dipole term . for the zeeman mixing contributions , we use a value of @xmath315 for the fractional transverse magnetic field amplitudes @xmath316 and @xmath317 . as can be seen in table [ table : gs_systematics ] , the greatest of any of these magnetic dipole terms is of magnitude @xmath306 . for comparison , the pnc term integrated across the interaction region is @xmath318 , which we compute as @xmath319 for the same set of field parameters . since these rms magnetic dipole contributions are much less than the pnc term in each case , we conclude that the magnetic dipole contributions can be sufficiently reduced , and that the pnc measurement in this geometry can be successfully executed . finally , we must consider excitation of the @xmath74 transition , which , if present at sufficient rates , could obscure the signal . excitation of this transition is strongly driven by the @xmath285 field component . this transition is suppressed , however , by the application of @xmath320 , which produces a zeeman shift between the magnetic levels . for @xmath321 7 gauss , the zeeman shift is @xmath103 mhz . when the frequencies of the r.f . and raman beams are tuned to resonance with the @xmath206 transition , the @xmath74 transition is off resonant . to gauge the degree of this excitation , we must estimate the linewidth of the transition . lifetime broadening of ground states and the collisional linewidth in an atomic beam are negligible . there is a small doppler broadening of the transition due to divergence of the atomic beam . @xmath322 , where @xmath323 is the 9.2 ghz transition frequency , and @xmath324 is the transverse velocity spread of the atomic beam . using 1 mm apertures separated by 40 cm to form the atomic beam gives a divergence angle of @xmath102.5 mrad , and with a mean atomic velocity of 270 m / s , @xmath325 0.7 m / s , leading to a doppler width of @xmath326 20 hz . the broadening due to the finite interaction time of the atoms with the r.f . field as they pass through the interaction region is @xmath327 = 2 khz . this transit time broadening and the linewdith of the r.f . source are the primary broadening mechanisms that we expect in this atomic beam geometry . since the linewidth for the transition is much less than the zeeman shift , excitation of this line is very weak . furthermore , when we use phase modulation of the r.f . field and phase - sensitive detection of the signal , only the signal resulting from the interference between the raman transition and the r.f . excitation is detected . the raman transition on the @xmath205 transition is highly suppressed by careful orientation of these polarizations and by the zeeman shift , and so the net interference by excitation of the @xmath205 line is expected to be below detection limits . to reduce systematics in this measurement , we have several tools and metrics available . these include : translation of the r.f . cavity about the atomic beam , reversals of @xmath54 , @xmath55 , and the initial projection state ( @xmath328 , @xmath329 ) in which the atoms are prepared , and application of transverse @xmath72 or @xmath73 field components to intentionally introduce systematic effects . we expect that these tests and studies will likely require the bulk of our attention as we carry out these measurements . in this section we examine two effects of the distribution of velocities in the atomic beam on the signal , as well as the variation of the intensity of the raman coupling laser beams . the interaction time for an individual atom depends upon its velocity , and we show the distribution of the atomic velocities in fig . [ fig : vel_dist_at_beam ] . c. the peak velocity is 270 m / s , and the standard deviation is 106 m / s . , title="fig:",width=283 ] + when the oven temperature is 120 @xmath330c , the peak velocity of the atoms is 270 m / s , and the standard deviation of the distribution is 106 m / s . the interaction time will vary among the atoms due to the distribution of their velocities , resulting in a variation of the interaction angle @xmath331 for each of the interactions , as shown in eq . ( [ eq : thetaeq1ovintomegadz ] ) . the interaction affected most significantly is @xmath88 , since an increase or decrease of this term tends to reduce the signal gain factor @xmath332 in eq . ( [ eq : cfeqdcpmodterm ] ) . we have analyzed this reduction of signal by numerically averaging @xmath333 over the velocity distribution shown in fig . [ fig : vel_dist_at_beam ] , and find that this reduction is @xmath1023% . since the stark and pnc amplitudes are affected in the same way , the only effect is a reduction of the signal magnitude ; it does not otherwise affect the accuracy of the measurement . the distribution of atomic velocities also reduces the signal through the variation in the flight time of atoms from the interaction region to the detection region in the measurement apparatus . this effect is much less significant in our apparatus , for the parameters of our measurement . in our system , this distance from the interaction region to the detection region is @xmath1013 cm , so the average time required to reach the detection region is only 0.5 ms . due to the width of the velocity distribution , the spread in arrival times is @xmath334 ms . this satisfies the requirement that this @xmath335 is much smaller than the inverse of the modulation frequency modulation of the signal , which in our previous measurements was 150 hz , and the loss of signal is a few percent . once again , this loss only affects the signal size , but does not affect the accuracy of the measurement . finally , we examine the effect of the gaussian intensity distribution of the laser beams that drive the raman transition in the atoms . since for an optimized measurement , the interaction angle @xmath88 is adjusted to a value of @xmath336 , if we set the raman beam amplitudes to this value on the axis , the angle for any off - axis atoms is less , decreasing our signal . to ensure that all atoms experience the same field amplitude to within 10% of the peak value , an atom beam diameter of 1 mm requires a beam radius of the raman beam of 7 mm . ( this dimension is consistent with the value of @xmath121 that we used in our simulation shown in fig . [ fig : cevstime ] , since the raman interaction scales as the product of the laser field amplitudes @xmath77 and @xmath78 . ) in this paper , we have reported our design of an experimental configuration for the measurement of the weak - force induced transition moment on the ground - state hyperfine transition of atomic cesium , or any of the alkali metal species , in an atomic beam configuration . we have introduced an r.f . resonant cavity based upon a parallel - plate transmission line structure that allows one to generate the very uniform , well - characterized , high - amplitude r.f . and static fields required for this measurement . this cavity design could be applied to any measurement that requires similar levels of controlled fields . we have carried out detailed numerical simulations of the various field amplitudes of the standing wave , and used these results to estimate the magnitude of the magnetic dipole contributions to the atomic signal . the atomic beam geometry is especially well - suited to these purposes , allowing suppression of all magnetic dipole contributions , and we estimate that a precision measurement of @xmath0 on the ground state hyperfine transition can be achieved within an integration time of a few tens of seconds . there are several individuals with whom we have benefited through very useful conversations and exchanges , including s. g. porsev , l. orozco , m. safronova , and a. derevianko . we also acknowledge the contributions of m. y. shalaginov , s. bogdanov , m. swabey , k. webb , and the group of d. peroulis , who were very helpful through their advice with the numerical simulations . s. aubin , e. gomez , j. a. behr , m. r. pearson , d. sheng , j. zhang , r. collister , d. melconian , v. v. flambaum , g. d. sprouse , l. a. orozco , and g. gwinner , 19th particles and nuclei international conference , aip conf . * 1441 * , 555 - 557 ( 2012 ) . s. aubin , j. a. behr , r. collister , v. v. flambaum , e. gomez , g. gwinner , k. p. jackson , d. melconian , l. a. orozco , m. r. pearson , d. sheng , g. d. sprouse , m. tandecki , j. zhang , and y. zhao , hyperfine interact . * 214 * , 163 - 171 ( 2012 ) . o. o. versolato , g. s. giri , l. w. wansbeck , j. e. van den berg , d. j. van der hoek , k. jungmann , w. l. kruithof , c. j. g. onderwater , b. k. sahoo , b. santra , p. d. shidling , r. g. e. timmermans , l. willmann , and h. w. wilschut , phys . a * 82 * , 010501 ( 2010 ) . o. o. versolato , l. w. wansbeck , g. s. giri , j. e. van den berg , d. j. van der hoek , k. jungmann , w. l. kruithof , c. j. g. onderwater , b. k. sahoo , b. santra , p. d. shidling , r. g. e. timmermans , l. willmann , and h. w. wilschut , can . j. phys . * 89 * , 65 - 68 ( 2011 ) . m. nuez portela , j. e. van den berg , h. bekker , o. bll , e. a. dijck , g. s. giri , s. hockstra , k. jungmann , a. mohanty , c. j. g. onderwater , b. santra , s. schlesser , r. g. e. timmermans , o. o. versolato , l. w. wansbeck , l. willmann , and h. w. wilschut , hyperfine interact . * 214 * , 157 - 162 ( 2013 ) . m. nuez portela , e. a. dijck , a. mohanty , h. bekker , j. e. van den berg , g. s. giri , s. hockstra , c. j. g. onderwater , s. schlesser , r. g. e. timmermans , o. o. versolato , l. willmann , h. w. wilschut , and k. jungmann , appl . b * 114 * , 173 - 182 ( 2014 ) .

the gj876 system was among the earliest multi - planetary detections outside of the solar system , and has long been known to harbor a resonant pair of giant planets . subsequent characterization of the system revealed the presence of an additional neptune mass object on an external orbit , locked in a three body laplace mean motion resonance with the previously known planets . while this system is currently the only known extrasolar example of a laplace resonance , it differs from the galilean satellites in that the orbital motion of the planets is known to be chaotic . in this work , we present a simple perturbative model that illuminates the origins of stochasticity inherent to this system and derive analytic estimates of the lyapunov time as well as the chaotic diffusion coefficient . we then address the formation of the multi - resonant structure within a protoplanetary disk and show that modest turbulent forcing in addition to dissipative effects is required to reproduce the observed chaotic configuration . accordingly , this work places important constraints on the typical formation environments of planetary systems and informs the attributes of representative orbital architectures that arise from extended disk - driven evolution .

1504.00051

the realization that planets form in gaseous protoplanetary disks dates back to the collective works of swedenborg , kant and laplace . despite several alterations , this nebular hypothesis has survived the test of time @xcite . nevertheless , a quantitative characterization of the consequences of planet - disk interactions @xcite and the associated sculpting of orbital architectures of planetary systems @xcite has only become an active field of research comparatively recently . large - scale orbital migration of giant planets was first recognized as a theoretical possibility by @xcite , in an effort to quantify the orbital evolution of satellites embedded in circum - planetary disks . however , it was nt until the discovery of the first close - in extrasolar planets @xcite that this idea gained wide - ranging traction @xcite . accordingly , over the last two decades , disk - driven migration has been repeatedly invoked as a unifying mechanism to explain the various orbital properties of extrasolar giant planets @xcite as well as low - mass compact multi - planetary systems alike @xcite . qualitatively distinct modes of migration , characterized by different timescales and directions , can arise depending on the physical properties and structure of the disk as well as the embedded planet @xcite . because of this intrinsic diversity in physical behavior , simultaneous migration of multiple planets residing within the same disk may cause the orbits to approach each other . convergent orbital evolution can result in capture of objects into mean - motion resonances ( orbital states characterized by rational period ratios ) and evidence for this process is plentiful throughout the satellite population of the solar system @xcite . moreover , the existence of a substantial number of ( near-)resonant exoplanetary systems ( see @xcite ) suggest that resonant locking is not limited to satellites and is also active among planets @xcite . despite their non - negligible count , resonant systems comprise a minority within the current observational aggregate ( @xcite ; see also @xcite ) . instead , giant planets often reside on dynamically excited orbits that are believed to be a result of planet - planet scattering @xcite . the connection between processes that occur within the nebular epochs of planetary systems and the onset of large - scale orbital instabilities responsible for sculpting the observed semi - major axis - eccentricity distribution are at present poorly understood @xcite . nevertheless , it is entirely plausible that planet - planet scattering originates within compact systems assembled by disk - driven migration @xcite . it is worthwhile to note that the presently favored view of the solar system s early dynamical history ( see @xcite for a review ) is consistent with the picture delineated above . that is , an evolutionary sequence where the giant planets emerged from the solar nebula in a multi - resonant configuration that subsequently became unstable , causing the planets to scatter onto their current orbits is broadly consistent with the available observations @xcite . numerous insights into the physical processes that operate within birth environments of planetary systems could be obtained if the orbital states at the time of nebular dispersion could be inferred . in practice , this is difficult to do for dynamically relaxed systems because planet - planet scattering is a highly stochastic forward " process that tends to erase all memory of the system s initial state . in other words , the characterization of the vast majority of the observational sample can yield only limited information about the primordial nature of the associated planetary systems because their dynamical histories have been chaotically eliminated . even in the solar system , where the modeling efforts have enjoyed a multitude of observational restrictions , strong degeneracies with respect to the initial state persist @xcite . on the contrary , resonant planetary systems ( which have managed to escape the onset of instability ) may in fact yield tangible constraints on the environment within which they formed . as such , they constitute exceptionally high value targets for theoretical inquiries . arguably the most exotic resonant exoplanetary system detected to date is gj876 @xcite , and this remarkable collection of planets will be the primary subject of this paper s study . with three objects locked in resonance , gj876 comprises the only well - characterized extrasolar multi - resonant chain , although additional multi - resonant systems exist within the _ kepler _ data set ( e.g. kepler-16 , kepler-79 , koi-730 ; @xcite ) . while this system exhibits some similarity to the galilean satellites @xcite , it differs from the io - europa - ganemede system in a crucial aspect : the resonant arguments of gj876 exhibit vigorous , yet bounded ( on multi - gyr timescales ) chaos @xcite . the architecture of gj876 is fully consistent with the picture of conventional disk - driven migration @xcite . as a result , the dynamical characterization of gj876 provides a rare window into the description of orbital properties of planetary systems , as they appear when they emerge from their natal disks . with this notion in mind , performing a theoretical analysis of the system s dynamical state with an eye towards obtaining some insights into the nature of the disk from which this system formed is the goal of this work . the paper is structured as follows . in section 2 , we examine the qualitative features of the system s orbital architecture and setup the basis for theoretical analysis . in section 3 , we construct a perturbative model for the resonant dynamics of the system and elucidate the origins of chaotic motion in an analytical fashion . in section 4 , we consider the assembly of the multi - resonant chain in a dissipative and turbulent protoplanetary nebula . we conclude and discuss our results in section 5 . the observational saga of the gj876 system effectively spans the entire active history of exoplanetary science . the initial detection of a @xmath0 giant planet b " in a 61 day orbit dates back to the infancy of large - scale dedicated radial velocity surveys @xcite . the discovery of a @xmath1 companion c " on a 30 day orbit followed shortly thereafter @xcite , rendering the gj876 c - b " pair the first mean motion resonance to be identified outside the solar system . taking advantage of the observational imprint of resonant coupling , @xcite and @xcite presented independent analyses of the radial velocity data that accounted for planet - planet interactions and were able to break the @xmath2 degeneracy inherent to radial velocity detections , deriving the system s inclination with respect to the line of sight of @xmath3 deg . concurrently , the signal of planet b " was confirmed astrometrically by @xcite ( see also @xcite ) . detections within the system continued , as an additional close - in @xmath4 planet d " , residing on a 2 day orbit was announced by @xcite . under the assumption of a 3-planet system , @xcite re - analyzed the available data and with extensive modeling derived a mutual inclination between the resonant planets b " and c " of @xmath5 deg . moreover , the study of @xcite confirmed the existence of planet d " and strongly hinted at the eccentric nature of its orbit ( see also @xcite ) . .adopted orbital fit of the gj876 system [ cols= " < , < , < , < , < , < , < " , ] [ frequencies ] the latest advancement in the observational characterization of gj876 arose from the work of @xcite , who uncovered yet another resonant @xmath6 planet e , " occupying a 124 day orbit . further dynamical analysis revised the system inclination to @xmath7 deg and more importantly , showed that the evolution of the multi - resonant configuration undergoes bounded , yet chaotic variations . over the last decade and a half , the orbital state of the system and its origin have been studied by a substantial number of authors , employing a wide variety of methods . specifically , in addition to the studies quoted above , orbital characterization and stability have been explored by : @xcite and @xcite . meanwhile the assembly of the particular resonant configuration has been simulated and studied by : @xcite and @xcite . additionally , @xcite examined the possibility of extra bodies being locked in the multi - resonant configuration . thanks to the long time - span covered by the radial velocity observations and the aforementioned fitting efforts , the orbital properties of planets b " and c " are well constrained @xcite . however , substantial uncertainties exist in the orbital solutions of the lower mass components d " and e " @xcite . the desired precision of the knowledge of the dynamical state of a system is largely dictated by the purpose of the calculation one wishes to perform . here , our aim is not to create an ephemeris for gj876 , but rather to shed light on the origins of chaotic motion within the multi - resonant system and place rough constraints on the properties of the nebula within which the system was born . to this end , we note that once a 4-planet system is adopted , chaos ( highlighted by the stochastic evolution of planet e " ) is more or less ubiquitous throughout the parameter space restricted by the data @xcite . consequently , for the purposes of this work , we shall simply adopt the best - fit co - planar orbital solution of @xcite at face - value , keeping in mind that quantitatively different evolutions stemming from nearby initial conditions can be equally representative of the system s dynamical behavior . to this end , we further remark that while the orbital fit of @xcite clearly favors a chaotic solution , there may exist quasi - periodic islands in phase space that reside within parameter space covered by observational uncertainties . for completeness , the adopted orbital solution is presented in table ( [ f1 ] ) . note that in this work , we have adopted a host stellar mass estimate of @xcite , which differs from that of @xcite by @xmath8 , meaning that our results will differ from previous works on a detailed level . for the purposes of understanding the evolution of the c - b - e " multi - resonant chain in the simplest terms possible , the perturbations arising from the close - in planet d " can be neglected . broadly speaking , this is justified by the fact that while each member of the chain interacts with its nearest neighbor(s ) via first order resonant terms that scale as @xmath9 ( where @xmath10 is the eccentricity ) , the averaged gravitational coupling between planet d " and the multi - resonant system is secular in nature and to leading order scales as @xmath11 @xcite . to confirm this argument , we numerically integrated the @xmath12-body evolution of the system using the ` mercury6 ` software package @xcite with and without planet d " and found no meaningful differences in the obtained solutions . it is further noteworthy that the mass of the outermost planet is more than an order of magnitude smaller than the two inner members of the chain . it is thus tempting to further simplify the system and treat planet e " as a test particle , subject to perturbations arising from the c - b " resonant pair . numerical examination of such a setup reveals that qualitative aspects of the dynamical evolution of planet e " are largely insensitive to this assumption . on the contrary , under this simplification the evolution of the massive resonant pair ceases to be chaotic and instead exhibits quasi - periodic motion @xcite . figure ( [ f1 ] ) shows the results of the simplified integration where planet e " is treated as a test particle , and planet d " is neglected . evidently , the stochastic features of the multi - resonant chain arise entirely as a result of the perturbations of planets c " and b " onto planet e , " and the chaotic diffusion of the massive pair is communicated exclusively via a back - reaction . this attribute has important consequences for the purposes of constructing a simple model aimed at elucidating the dynamical structure that underlies chaotic motion . specifically , in this work we will take advantage of this characteristic and derive the chaotic properties of the system by considering the evolution of a massless planet e , " subject to periodic perturbations from planet b , " whose orbit is in turn modulated by interactions with planet c. " panel a of figure ( [ f2 ] ) shows the numerically obtained time evolution of the longitude of perihelion of planet b " ( blue ) as well as difference between the longitudes of perihelia of planets c " and b " ( red ) . clearly , the c - b " resonant pair resides near a dynamical equilibrium characterized by alignment and co - precession of the apsidal lines of the orbits @xcite . such a state is actually quite peculiar for resonant orbits and is only possible when the orbital eccentricities are not small . indeed , the classical first - order expansion of the resonant hamiltonian @xcite does not capture the existence of this fixed point . to this end , @xcite developed an alternative expansion of the planetary three - body hamiltonian and showed that the resulting perturbative ( first - principles ) solution matches the numerically computed evolution well . taking a somewhat alternative approach , @xcite utilized synthetic perturbation theory to construct a lagrange - laplace like periodic solution for the resonant pair which also shows excellent agreement with @xmath12-body calculations . because the dynamics of the c - b " resonant pair have been studied extensively , here we shall not duplicate the published results and instead refer the interested reader to the aforementioned studies . as already briefly mentioned above , in this work we shall examine the consequences of gravitational excitation of planet e " facilitated by planet b. " in order to perform this analysis within a perturbative framework , we must first delineate an approximate functional form for the dynamical behavior of planet b. " let us begin by defining the following scaled cartesian canonical coordinates : @xmath13 where @xmath14 is the longitude of perihelion . following @xcite , we note that in the ( @xmath15 ) plane the trajectory of planet b " is well described by a circle of radius @xmath16 that is off - set from the origin by @xmath17 ( see also the discussion on free and forced elements in ch . 7 of @xcite ) . accordingly , we parameterize the evolution of planet b " in the following manner : @xmath18 in the above expression , @xmath19 rad / yr is the ( retrograde ) apsidal precession rate and @xmath20 is the phase offset , while the constants are set to @xmath21 and @xmath22 . panels b and c in figure ( [ f2 ] ) depict a comparison between results obtained with an @xmath12-body simulation and the analytical prescription ( [ hkb ] ) . needless to say , the observed agreement is excellent . it is further noteworthy that the numerically averaged semi - major axis of planet b " does not deviate from its nominal value , @xmath23_{\rm{b}}$ ] by more than a few parts in a thousand . thus , for the purposes of the following calculation we may readily neglect the semi - major axis evolution all together and set @xmath24_{\rm{b}}$ ] . with a simple analytical model for the dynamical evolution of the perturbing planet at hand , we are now in a position to perform a perturbative analysis of the system s chaotic behavior . before we proceed , let us begin by defining restricted poincar@xmath25 action - angle variables in terms of standard orbital elements : @xmath26 e^2/2 \ \ \ \ \ \ \ \ \gamma = -\varpi\end{aligned}\ ] ] where @xmath27 is the gravitational constant , @xmath28 is the stellar mass , @xmath29 is the mean anomaly and the nominal action @xmath30 $ ] is evaluated at @xmath23 = 2^{2/3 } [ a]_{\rm{b}}$ ] . for the remainder of the paper , quantities not labeled with an index shall correspond to planet e. " generally , unlike secular perturbations ( see e.g. @xcite ) , mean - motion resonances modulate both the eccentricity and semi - major axis . hence , keplerian motion can not be averaged over , and the corresponding part of the hamiltonian must be retained : @xmath31 expanding @xmath32 around the nominal 2:1 resonant semi - major axis quoted above , equation ( [ hkep ] ) takes the form : @xmath33 ^ 2 } + \frac{\mathcal{g}^2 m_{\star}^2}{[\lambda]^3 } \big(\lambda - [ \lambda]\big ) - \frac{3\mathcal{g}^2 m_{\star}^2}{2[\lambda]^4 } \big(\lambda - [ \lambda]\big)^2 \nonumber \\ & = - \frac{3 \mathcal{g}^2 m_{\star}^2}{[\lambda]^2 } + 4 [ n ] \lambda - \frac{3}{2 } [ h ] \lambda^2 , \label{hkep2}\end{aligned}\ ] ] where @xmath34=\sqrt{\mathcal{g } m_{\star}/[a]^3}$ ] is the mean motion and @xmath35=[n]/[\lambda]$ ] . the first term on the second line of equation ( [ hkep2 ] ) is constant and can thus be dropped from the hamiltonian . upon averaging out short - periodic terms ( see ch . 2 of @xcite ) , the component of the hamiltonian that governs first order ( in @xmath10 ) resonant planet - planet interactions ( within the framework of the elliptic restricted three body problem ) reads @xcite : @xmath36 } \bigg ( f_{\rm{res}}^{(1 ) } e_{\rm{b } } \cos(2\lambda - \lambda_{\rm{b } } - \varpi_{\rm{b } } ) \nonumber \\ & + f_{\rm{res}}^{(2 ) } \sqrt{\frac{2 \gamma}{[\lambda ] } } \cos(2 \lambda - \lambda_{\rm{b } } + \gamma ) \bigg ) , \label{hres}\end{aligned}\ ] ] where @xmath37 and @xmath38 are interaction coefficients that depend exclusively on the semi - major axis ratio @xcite . the presence of @xmath39 and @xmath40 in the hamiltonian ( [ hres ] ) gives rise to explicit time - dependence in expression ( [ hres ] ) . thus , the quoted model constitutes a non - autonomous dynamical system with two degrees of freedom . the hamiltonian can be made autonomous by extending the phase - space to three degrees of freedom @xcite : @xmath41 where @xmath42 is the newly introduced action conjugate to time , @xmath43 . accordingly , substituting the solution ( [ hkb ] ) into equation ( [ hres ] ) , we obtain the following expression : @xmath44 \lambda - \frac{3}{2 } [ h ] \lambda^2 + \mathbb{t } - \frac{\mathcal{g } m_{\rm{b}}}{[a ] } \nonumber \\ & \times \bigg ( f_{\rm{res}}^{(1 ) } \big ( ( \epsilon + \delta \cos(g t + \varphi_0 ) ) \cos(2 \lambda - n_{\rm{b } } t - \lambda_{\rm{b}0 } ) \nonumber \\ & + \delta \sin(gt - \varpi_0 ) \sin(2 \lambda - n_{\rm{b } } t - \lambda_{\rm{b}0 } ) \big ) \nonumber \\ & + f_{\rm{res}}^{(2 ) } \sqrt{\frac{2 \gamma}{[\lambda ] } } \cos(2 \lambda - n_{\rm{b } } t - \lambda_{\rm{b}0 } + \gamma ) \bigg ) , \label{ht}\end{aligned}\ ] ] where @xmath45 is an initial phase of planet b , " and @xmath46 is its mean motion . equation ( [ ht ] ) is rather cumbersome and can be made more succinct . first , let us define the constants : @xmath47 } f_{\rm{res}}^{(1 ) } \nonumber \\ \beta & = - \frac{\mathcal{g } m_{\rm{b}}}{[a ] } \frac{f_{\rm{res}}^{(2)}}{\sqrt{[\lambda]}}. \label{resconst}\end{aligned}\ ] ] next , let us perform a canonical transformation of coordinates , arising from the following type-2 generating function @xcite : @xmath48 an application of the transformation equations yields the new action - angle variables : @xmath49 in terms of the new coordinates , the hamiltonian reads : @xmath50 \theta - 6 [ h ] \theta^2 + \alpha \epsilon \cos(\theta ) + \alpha \delta \cos(\theta ) \nonumber \\ & \times \cos(g \xi + \varphi_0 ) + \alpha \delta \sin(\theta ) \sin(g \xi + \varphi_0 ) \nonumber \\ & + \beta \sqrt{2 \phi } \cos(\theta + \phi ) - n_{\rm{b } } \theta + \xi . \label{hxi}\end{aligned}\ ] ] although somewhat less unwieldily than equation ( [ ht ] ) , the hamiltonian ( [ hxi ] ) is still characterized by three degrees of freedom , precluding straightforward analytical treatment . to remedy this issue , let us define canonical cartesian coordinates related to the @xmath51 degree of freedom : @xmath52 after some algebraic manipulation , the hamiltonian obtains the form : @xmath53 - n_{\rm{b } } ) \theta - 6 [ h ] \theta^2 + \alpha \delta \cos(\theta - g \xi - \varphi_0 ) \nonumber \\ & + \alpha \epsilon \cos(\theta ) + \beta x \cos(\theta ) - \beta y \sin(\theta ) + \xi . \label{hxy1}\end{aligned}\ ] ] to reduce the number of harmonics present in @xmath54 , we follow @xcite and define the following canonical translation : @xmath55 this transformation morphs the first and second terms on the second line of equation ( [ hxy1 ] ) into a single term . accordingly , upon defining implicit action - angle variables @xmath56 the hamiltonian takes on a form characterized by only two harmonics : @xmath53 - n_{\rm{b } } ) \theta - 6 [ h ] \theta^2 + \alpha \delta \cos(\theta - g \xi - \varphi_0 ) \nonumber \\ & + \beta \sqrt{2 \psi } \cos(\theta + \psi ) + \xi . \label{hxy}\end{aligned}\ ] ] from here , the reduction of the hamiltonian to a two - degree of freedom system is straightforward . particularly as in equation ( [ transone ] ) , define a contact transformation arising from a type-2 generating function : @xmath57 the transformation equations yield : @xmath58 noting that the resonant condition implies that @xmath59 $ ] the hamiltonian takes the form : @xmath60 ( \mathcal{w } + \mathcal{z } ) - 6 [ h ] ( \mathcal{w } + \mathcal{z})^2 - g \mathcal{w } \nonumber \\ & + \alpha \delta \cos(w ) + \beta \sqrt{2 \mathcal{z } } \cos(z ) + \mathcal{k}. \label{haaone}\end{aligned}\ ] ] because the angle @xmath61 ( which corresponds to time ) is now absent from the hamiltonian , the action @xmath62 is a constant of motion and can thus be dropped from equation ( [ haaone ] ) . the reduction of the hamiltonian to an autonomous two - degree of freedom system is now complete . qualitatiely , the hamiltonian ( [ haaone ] ) represents a system of two momentum - coupled pendulums . it is noteworthy that in isolation , the @xmath63 degree of freedom possesses the dalembert characteristic and thus has the dynamics corresponding to the second fundamental model for resonance @xcite whereas the isolated @xmath64 degree of freedom has the phase - space structure of a simple pendulum @xcite . the extent to which the perturbative model governed by hamiltonian ( [ haaone ] ) successfully captures the dynamical behavior of gj876 e " can be tested numerically . however prior to applying hamilton s equations , note that in terms of action - angle coordinates , hamiltonian ( [ haaone ] ) possesses a coordinate singularity at @xmath65 . this nuisance can be circumvented by transforming to canonical cartesian coordinates : @xmath66 accordingly , the hamiltonian is rewritten as follows : @xmath60 \left(\mathcal{w } + \frac{u^2 + v^2}{2 } \right ) - 6 [ h ] \left(\mathcal{w } + \frac{u^2 + v^2}{2 } \right)^2 \nonumber \\ & - g \mathcal{w } + \alpha \delta \cos(w ) + \beta u. \label{haa}\end{aligned}\ ] ] the corresponding equations of motion take the form : @xmath67 - g - 12 [ h ] \left ( \mathcal{w } + \frac{u^2 + v^2}{2 } \right ) \nonumber \\ \frac{du}{dt } & = - 6 [ n ] v + 12 [ h ] v \left ( \mathcal{w } + \frac{u^2 + v^2}{2 } \right ) \nonumber \\ \frac{dv}{dt } & = \beta + 6 [ n ] u - 12 [ h ] u \left ( \mathcal{w } + \frac{u^2 + v^2}{2 } \right ) \label{eom}\end{aligned}\ ] ] after numerous variable changes , it is useful to explicitly relate the final variables ( [ finangles],[uv ] ) to keplerian orbital elements . working back through the transformations delineated above , it can be shown that the orbital semi - major axis and eccentricity are expressed as follows : @xmath68}}}. \label{var2kep}\end{aligned}\ ] ] meanwhile , the angles present in the original formulation of the hamiltonian ( [ hres ] ) are related to the variables ( [ finangles ] ) in a unembellished way : @xmath69 the latter equality is inexact , but is nevertheless a good approximation because @xmath70 . in the same spirit , the following relationships approximately hold : @xmath71 } \simeq \sqrt{\frac{a}{[a ] } } - e^2 . \label{var2kepref}\end{aligned}\ ] ] note that the last of the above expressions is closely related to the well - known tisserand parameter . orbital evolution obtained by numerical integration of equations ( [ eom ] ) is shown with black curves in figure ( [ f3 ] ) , where the results of an @xmath12-body simulation are also depicted with blue lines . evidently , the system described by hamiltonian ( [ haa ] ) provides an excellent perturbative representation of the real system , meaning that the dynamical behavior within the chaotic layer is well captured by a first - order expansion of the disturbing function . it is noteworthy that the semi - analytical solution initially tracks the @xmath12-body solution , but the two evolutionary sequences lose coherence after @xmath72 years . this is indicative of a lyapunov time that is a factor of a few shorter than @xmath73 years year lyapunov time reported by @xcite . ] . this places gj876 into the same category of rapidly chaotic systems as _ @xcite . despite rapid chaotic diffusion , @xmath12-body calculations reported by @xcite suggest that the system is stable on multi - gyr timescales . in other words , the chaos exhibited by the multi - resonant system is simultaneously vigorous and bounded . both of these characteristics can be qualitatively understood within the context of hamiltonian ( [ haa ] ) by examination of surfaces of section . as a starting point , let us examine the evolution of the angles depicted in figure ( [ f3 ] ) . note that the angle @xmath74 librates around @xmath75 while @xmath76 switches between libration around @xmath77 and circulation . preliminary intuition about the amplitude of orbital excursions can be obtained by constructing a double surface of section of the hamiltonian . in accord with the evolution of angles denoted in figure ( [ f3 ] ) , we fix @xmath78 and @xmath79 and plot level curves of @xmath54 on panel a of figure ( [ f4 ] ) . suitably , on the @xmath80-axis , we plot the scaled action @xmath81 $ ] while on the @xmath82-axis we plot the scaled momentum @xmath83}$ ] . because the hamiltonian ( [ haa ] ) is autonomous , any ( potentially chaotic ) trajectory it describes will be constrained to map onto a corresponding level curve ( given by the value of @xmath54 ) every time the orbital state crosses the section condition . a red curve corresponding to the nominal semi - major axis @xmath84 $ ] is also shown on this panel . crudely speaking , the horizontal deviation away from the nominal resonance curve on the double section is indicative of changes in semi - major axis whereas vertical deviation corresponds to eccentricity modulation . thus , the area occupied by a given constant @xmath54 curve on the double section serves as a proxy for the amplitude of orbital variations associated with resonant motion . this point is of some importance to understanding the long - term stability of the system . particularly , this simple analysis suggests that to the extent that hamiltonian ( [ haa ] ) provides an adequate representation of the dynamics , the conservation of @xmath54 restricts the maximal deviation from nominal resonance of the trajectory , no matter how vigorously chaotic it may be . qualitatively , this explains how a rapidly stochastic system such as gj876 can remain stable on multi - gyr timescales . in addition to delineating energy contours on a double section , it is further useful to explore the actual dynamical behavior of the system on the contours ( i.e. provided specific values of @xmath54 ) . to this end , we have constructed poincar@xmath25 surfaces of section for the five highlighted curves shown on the double section , labeled @xmath85 . the sections are taken with respect to @xmath78 at @xmath86 and depict the @xmath87 degree of freedom . the surfaces are shown as panels b - f on figure ( [ f4 ] ) and are labeled according to the level of @xmath54 they represent . the value of @xmath54 that corresponds to the initial condition depicted in table ( [ f1 ] ) is @xmath88 and is shown on panel d. quasi - periodic trajectories , shown as purple and brown curves dominate the surfaces of section that correspond to energy levels that yield the most limited orbital variations ( i.e. @xmath89 and @xmath90 , shown on panels e and f respectively ) . on the contrary , the remaining surfaces of section feature substantial chaotic seas , shown with black points . the location of a given point on the surface of section is related to the orbital eccentricity ( evaluated at the section conditions ) through expression ( [ var2kep ] ) . accordingly , the range of chaotic eccentricity excursions at a given energy level can be gathered from examining the extent of the phase - space occupied by the irregular region . note further that in all cases , the dynamics resides on a well - defined domain , which is restricted by the conservation of @xmath54 and the requirement for the actions to have a null imaginary component . the physical characters of the periodic points that reside at the centers of the nested quasi - periodic trajectories is not uniform across the plotted energy levels . particularly , on panels b and c , these orbits correspond to limit cycles that are characterized by rapid circulation of @xmath91 and a modulated circulation of @xmath92 , such that the majority of time spent on the circulation cycle remains in the vicinity of @xmath93 . conversely , fixed points in panels d , e , and f are true equilibria that are characterized by stationary evolution of the angles at @xmath94 and @xmath93 . despite these differences , as long as the system is initiated within the extensive chaotic region ( irrespective of whether the value of the hamiltonian is set to @xmath95 , @xmath96 , or @xmath88 ) , the qualitative behavior of the dynamics resembles that observed in figure ( [ f3 ] ) . in light of this , it is tempting to obtain an estimate for the characteristic rate of chaotic decoherence of the system , that is independent of the exact value of @xmath54 . the simplest approach to such an estimate requires further reduction of the model . recall that the only assumptions inherent to the hamiltonian ( [ haa ] ) are the truncation of the disturbing function at first order in @xmath10 @xcite and the adoption of equations ( [ hkb ] ) for the dynamical evolution of the perturbing planet . the subsequent conversion of the hamiltonian to an autonomous two degree of freedom system was possible because of the reducing transformation ( [ reducing ] ) @xcite . because the hamiltonian ( [ haa ] ) can not be simplified further with canonical transformations , in order to convert the system into a more tractable non - autonomous one degree of freedom system ( see e.g. @xcite ) , we must prescribe a functional form to one of the degrees of freedom . an examination of panels b and c of figure ( [ f3 ] ) shows that while the oscillations of the angle @xmath91 are approximately limited to the range @xmath97 , the angle @xmath92 recurrently transitions between libration and circulation . this implies that the repeated separatrix crossing associated with the @xmath87 degree of freedom acts as the primary driver of stochasticity @xcite . moreover , recall that the equations of motion ( [ eom ] ) dictate that the interactions between the degrees of freedom are facilitated exclusively by action coupling . consequently , for further reduction of the model , it is sensible to parameterize the time - evolution of @xmath98 . in absence of coupling , hamiltonian ( [ haa ] ) describes simple pendulum - like dynamics for the @xmath64 degree of freedom . under the assumption of small - amplitude libration of @xmath91 ( which allows one to approximate the dynamics of a pendulum with that of a harmonic oscillator ) , @xmath98 will exhibit sinusoidal variations @xcite . accordingly , let us adopt the following functional form : @xmath99}{2 } \left(\eta + \mu \cos(\sigma t ) \right ) . \label{wt}\end{aligned}\ ] ] empirically , the newly introduced constants and frequency are set to @xmath100 , @xmath101 , and @xmath102 rad / year , to provide a suitable match to the numerical calculations . to this end , panel a of figure ( [ f5 ] ) shows a comparison between the evolution of @xmath98 , computed within the framework of an @xmath12-body simulation ( blue ) , numerical integration of the perturbative model governed by hamiltonian ( [ haa ] ) ( black ) and the prescription ( [ wt ] ) ( green ) . for consistency , a sinusoidal wave with the same frequency ( corresponding to the envisioned evolution of @xmath91 , given the approximation [ wt ] ) is also depicted on panel b of figure ( [ f3 ] ) . we note that even though the adopted parameterization does not respect the physical requirement for the quantity @xmath103 to not exceed unity ( see equation [ var2kepref ] ) , it suffices for the purposes of an illustrative model we aim to construct . adopting equation ( [ wt ] ) , equations of motion ( [ eom ] ) can again be integrated numerically to yield an approximate dynamical evolution for the @xmath87 degree of freedom . the resulting solutions for the eccentricity as well as the critical angle @xmath104 are over - plotted as green curves in panels a and c of figure ( [ f3 ] ) respectively . although the shown solutions lose coherence ( as chaotic systems must ) after a few tens of years of evolution , the simplified model clearly captures the stochastic behavior exhibited by the system in a satisfactory manner . thus , it can be sensibly employed to further characterize the dynamics in a rudimentary fashion . to understand the origins of chaos observed in the numerical solutions , note that under the assumption ( [ wt ] ) , the equations of motion ( [ eom ] ) correspond to a non - autonomous one degree of freedom hamiltonian : @xmath105 \left(\frac{u^2 + v^2}{2 } \right ) + \beta u \nonumber \\ & - 6 [ h ] \left(\frac{[\lambda]}{2 } \left(\eta + \mu \cos(\sigma t ) \right ) + \frac{u^2 + v^2}{2 } \right)^2 . \label{honedof}\end{aligned}\ ] ] this hamiltonian possesses a single critical curve that sweeps across a region of phase - space with every circulation of the angle @xmath106 @xcite . accordingly , repeated encounters with the critical curve ( which comprises an infinite - period orbit ) render the motion on the separatrix - swept region of phase - space irregular @xcite . panel b of figure ( [ f5 ] ) depicts the critical curve of hamiltonian ( [ honedof ] ) at various phases of @xmath106 . shown on the same panel , is a stroboscopic surface of section arising from the same hamiltonian , where the green points represent a chaotic sea while quasi - periodic trajectories are shown with gray curves . note that as expected , the size of the chaotic region approximately conforms to the maximal phase - space area occupied by the separatrix . indeed , this figure is quite similar to the poincar@xmath25 surface of section shown in panel d of figure ( [ f4 ] ) . moreover , the differences in the locations of families of quasi - periodic trajectories embedded in the chaotic sea , ( shown in panels b , c and d of figure [ f4 ] ) , can now be understood as changes in the extent of separatrix sweeping that result from alteration of the modulation at different values of @xmath54 . it is noteworthy that the picture delineated in panel b of figure ( [ f5 ] ) is not exactly one of adiabatic chaos @xcite . this is made evident by the fact that quasi - periodic resonant trajectories that are not swept by the separatrix do not exist in this panel . this is likely due to the fact that the modulation frequency , @xmath107 , is so high , that the appearance and overlap of secondary resonances acts to wipe out this family of orbits @xcite . for reference , panel c of figure ( [ f5 ] ) shows a similar stroboscopic surface of section where the modulation frequency has been artificially reduced by a factor of 10 . with a simplified picture at hand , we may now analytically derive the stochastic properties of the system . first and foremost , the above analysis allows us to obtain an estimate of the lyapunov time . crudely speaking , the lyapunov time can be understood as a characteristic decoherence time of the system @xcite . accordingly , within the framework of hamiltonian ( [ honedof ] ) , it can be approximated as the time interval between successive encounters with the separatix , or half the modulation period : @xmath108 this simple functional form is in agreement with the estimates obtained by @xcite for the asteroid belt . given that the modulation period is slightly shorter than 15 years , the above relation suggests that the lyapunov time associated with gj876 e " should be of order @xmath109 years . to check this estimate numerically , we evaluated the lyapunov time by integrating the full linearized variational equations ( see @xcite , ch . 5 of @xcite , @xcite ) in parallel with a direct @xmath12-body simulation years and the initial tangent vector was randomly oriented . ] . the numerical calculation yielded a lyapunov time of @xmath110 years , in excellent agreement with equation ( [ lyap ] ) . as already mentioned above , this strongly suggests that in some similarity with the @xmath111 - 36 system @xcite , the gj876 system exhibits rapid dynamical chaos . over timescales longer than a lyapunov time , it is not sensible to treat any one trajectory as being representative . instead , a statistical description of irregular trajectories is more appropriate @xcite . in a uniformly chaotic region , the transport of actions obeys the fokker - plank equation @xcite , which reduces to the diffusion equation for hamiltonian systems @xcite . accordingly , the chaotic diffusion coefficient , @xmath112 , quantifies the essential attributes of dynamical evolution . in the quasi - linear approximation @xcite , the diffusion coefficient is given by the square of the typical change in action , @xmath113 , that takes place over a decoherence ( i.e. lyapunov ) time , divided by the lyapunov time . in the non - adiabatic regime ( which is evidently representative of gj876 ) , correlations can be neglected ( see @xcite for a discussion ) and the trajectory can be envisioned to explore the chaotic layer uniformly . accordingly , the average jump in action is of order half the size of the chaotic region . we have already mentioned that within the context of hamiltonian ( [ honedof ] ) , the extent of the chaotic layer approximately corresponds to the maximal phase - space area attained by the separatrix over a modulation cycle . thus , a rough estimate for @xmath113 is given by : @xmath114 where @xmath115 corresponds to the hyperbolic fixed point of the separatrix when @xmath116 ( see figure [ f5 ] ) and is the solution to the equilibrium equation : @xmath117 u - 12 [ h ] u \left ( \frac{[\lambda]}{2 } \left(\eta - \mu \right ) + \frac{u^2}{2 } \right ) . \label{hyp}\end{aligned}\ ] ] combining equations ( [ lyap ] ) , ( [ dz ] ) and ( [ hyp ] ) , the explicit form of the diffusion coefficient becomes : @xmath118 ( [ n]+[h][\lambda](\mu-\eta ) ) + ( 6)^{1/3}(3 [ h]^2 \beta \nonumber \\ & + \sqrt{3[h]^3(3[h ] \beta^2 - 16([n]+[h][\lambda](\mu-\eta))^3 ) } ) ^{2/3 } \big)^4 \nonumber \\ & \times \big(\big ( \sqrt{3[h]^3(3[h ] \beta^2 - 16([n]+[h][\lambda](\mu-\eta))^3)}\nonumber \\ & + 3[h]^2 \beta \big)^{4/3}(1296 [ h]^4 ) \big)^{-1}. \label{diff}\end{aligned}\ ] ] quantitatively , equation ( [ diff ] ) evaluates to @xmath119 [ \lambda]^2)/(2\pi)$ ] . this suggests that the diffusive progress of the eccentricity of planet e " over an orbital period is @xmath120 . we can compare this result with a numerical estimate of the diffusion coefficient , calculated on a poincar@xmath25 surface of section . specifically , we utilized the perturbative model ( [ haa ] ) to compute the average of the square of the change in action divided by the time - span between successive section intersections , using parameters corresponding to panel d of figure ( [ f4 ] ) . this procedure yielded @xmath121 [ \lambda]^2)/(2\pi)$ ] . thus , the analytical estimate underestimates the numerically computed diffusion coefficient by a factor of @xmath122 ; an acceptable ( and perhaps expected ) error given the crudeness of the approximation involved in deriving equation ( [ diff ] ) . as already discussed in the introduction , assembly of the c - b " resonance has been studied extensively in the literature ( @xcite and the references therein ) . had the discovery of the additional planet e " implied quasi - periodic behavior of the multi - resonant system ( in some similarity to the case of the galilean satellites ) , the assembly of the laplace resonance would have been a straightforward extension of previous results @xcite . however , the chaotic nature of the orbits raises questions regarding the specific nature of the natal disk that is both sufficiently laminar to not preclude smooth migration ( see e.g. @xcite ) and is simultaneously turbulent enough to prevent the system from settling to quasi - periodic depths of the resonant potential well @xcite . for coherence , let us perform our investigation sequentially and first consider only dissipative effects . accordingly , envisage the c - b - e " resonant chain embedded in a perfectly laminar protoplanetary nebula . once locked in a mean - motion resonance , gap - opening giant planets ( b " and c " ) tend to carve out a vast mutual gap , greatly reducing the disk - facilitated migration rate @xcite . the same argument however does not apply to planet e " , which likely experiences much stronger coupling with the disk . thus , in line with the approximations invoked in the previous section , for the purposes of the perturbative model we can take the dynamics of the massive resonant pair to be isolated and periodic , while studying the dissipative evolution of the outermost planet in the test - particle limit . following @xcite , we shall parameterize the effects of planet - disk interaction using the following simple relationships : @xmath123 where @xmath124 is the characteristic eccentricity damping timescale , and @xmath125 is a constant that parameterizes the semi - major axis damping timescale in terms of @xmath124 . because we are interested in understanding the long - term evolution of a chaotic multi - resonant state within the disk , we shall adopt the present ( observed ) state as an adequate initial condition . in line with the approximations quoted above , we have performed a series of @xmath12-body simulations where only the outermost planet in the system is affected by the fictitious dissipative forces ( [ nebula ] ) . recall that the resonant modulation time invoked in the previous section is @xmath126 years , much shorter than the typically quoted estimates for @xmath124 . therefore , in accordance with adiabatic theory @xcite , the final outcome of our simulations is insensitive to the exact value of @xmath124 ( which we safely set to @xmath127 years ) . we note further that the adopted adiabatic regime can be used to replace the effective unit of time associated with dissipative evolution , allowing the obtained results to be scaled to other values of @xmath124 . ] is consistent with the observed orbital state , since resonant capture probabilities diminish significantly in non - adiabatic systems . panel a of figure ( [ f6 ] ) shows the eccentricity evolution of planet e " observed in the dissipated @xmath12-body simulations . specifically , the calculations suggest that following an initial transient period of order @xmath128 , the system settles onto a quasi - periodic limit cycle . moreover , the evolutionary sequence appears to be independent of the assumed value of @xmath125 , as long as it exceeds @xmath129 . to this end , we note that @xcite favor a value of @xmath130 for the natal disk of gj876 ( which we adopt for subsequent calculations ) , rendering our results rather robust . as with chaotic diffusion , the quoted results of dissipative @xmath12-body simulations can be understood within the framework of the perturbative model delineated in the previous section . that said , care must be taken in implementing equations ( [ nebula ] ) . explicitly , the definition of variables ( [ finangles ] ) implies that both degrees of freedom will be affected by dissipative evolution . however , the perturbative model differs from the @xmath12-body system in a crucial manner : the test particle approximation employed in the former impedes resonant transport of the chain . that is , although realistically the application of semi - major axis migration to planet e " affects the radial evolution of the whole system , this effect is not captured in the formulation of the hamiltonian ( [ haa ] ) introduces an unphysical dependence on @xmath125 into the perturbative model . ] . consequently , given that the action @xmath98 is directly proportional to @xmath131 , a more physical representation of the real system can be attained by only applying dissipative effects to the @xmath87 degree of freedom . suitably , the revised equations of motion take the form : @xmath132 where the subscript @xmath54 signifies the hamiltonian contribution . the equations corresponding to the ( @xmath133 ) degree of freedom remain unchanged from ( [ eom ] ) . using the modified perturbative model , we have simulated the dissipative evolution of the system , starting with different values of @xmath54 depicted in figure ( [ f4 ] ) . the calculated eccentricity is shown as a function of time in panel b of figure ( [ f6 ] ) , where each curve is labeled by the starting value of the hamiltonian . the corresponding evolution of @xmath54 is shown in panel c of figure ( [ f6 ] ) . evidently , two distinct modes of dissipative evolution exist : the approach to a limit cycle ( as shown by the blue curve , @xmath95 ) and the approach towards a fixed point . as long as the damping is comparatively slow ( as discussed above ) , at each point along the evolutionary track , a purely hamiltonian snap - shot of the dynamical portrait yields a good representation of the orbital state . thus , the dichotomy inherent to the behavior of the dissipative evolution can be understood by examining the surfaces of section depicted in figure ( [ f4 ] ) . the addition of dissipative forces into the model alters the dynamical behavior into two ways : it leads to a gradual reduction of the phase - space area enclosed by a given orbit in phase - space ( see e.g. @xcite for a related discussion ) and an increase in the value of the hamiltonian . consequently , if the orbit is initialized somewhere within the chaotic layer ( see panels b , c and d of figure [ f4 ] ) , as time proceeds the orbit will tend to exit the chaotic sea and settle onto the center of the corresponding quasi - periodic island . simultaneously , the dynamical portraits will change in such a manner that the area enclosed by the constant - energy curves on the double section ( panel a of figure [ f4 ] ) will also decrease . the two processes are interrelated as the rate of change of @xmath54 grows as the action @xmath134 ( associated with the @xmath87 degree of freedom ) increases . if the starting condition of a @xmath135 orbit is chosen such that it is relatively close to the center of the quasi - periodic island depicted in panel b of figure ( [ f4 ] ) , the orbit will rapidly decay onto a limit - cycle that intersects the center of the associated poincar@xmath25 surface of section . concurrently , because such a limit cycle is characterized by a consistently low eccentricity , the rate of dissipative increase of @xmath54 will be reduced , rendering such a state of the system quasi - steady . an example of such an evolution is shown by the blue curves in panels b and c of figure ( [ f6 ] ) . the @xmath12-body solution shown in panel a of figure ( [ f6 ] ) also exhibits such behavior . if the system is initialized at a higher level of @xmath54 ( e.g. @xmath88 , @xmath89 , @xmath90 ) , the corresponding poincar@xmath25 surfaces of section depicted in figure ( [ f4 ] ) show that the limit cycle ( which turns out to correspond to a fixed point ) resides at a high eccentricity meaning that the evolution of @xmath54 will also proceed at an unhindered rate . consequently , on a timescale not much grater that @xmath136 , the system will settle onto an equilibrium described by a maximal attainable value of @xmath54 ( which also corresponds to a null area enclosed by the orbit in the double - section shown on figure [ f4 ] ) . solutions of this kind are shown as pink ( @xmath88 ) , orange ( @xmath89 ) , and brown ( @xmath90 ) curves on panels b and c of figure ( [ f6 ] ) . the value of @xmath54 that separates the two regimes lies between @xmath137 . thus , it is possible to have a trajectory that first evolves onto a limit cycle , but upon crossing the critical value of @xmath54 , breaks out of the limit cycle and begins its approach to the fixed point . an example of such an evolution is shown in purple ( @xmath96 ) on panels b and c of figure ( [ f6 ] ) . irrespective of the details of the solution , a common feature observed in all evolutionary tracks is the approach to quasi - periodicity on a timescale comparable to @xmath124 . numerical simulations ( see e.g. @xcite ) suggest that @xmath124 is considerably shorter than the lifetime of a disk . consequently , in absence of additional perturbations , the formation of a chaotic laplace resonance in a perfectly laminar protoplanetary disk appears unlikely . accordingly , in the following discussion we shall invoke turbulent forcing as a means of preventing the system from reaching dynamical equilibration . angular momentum transport within protoplanetary disks ( that facilitates accretion of disk material onto the host stars ) is typically attributed to turbulence @xcite that is expected to stem from the magneto - rotational instability @xcite or some other process . turbulent effects can have important consequences for resonant coupling and have generally been shown to lead to a breakup of resonant libration @xcite . therefore , provided sufficiently vigorous stochastic diffusion , it is reasonable to expect that turbulent perturbations may overcome dissipative interactions , allowing for the formation of a chaotic multi - resonant chain . as is with laminar disk - star interactions , the most direct way to calculate the evolution of embedded planets is using global 3d magnetohydrodynamic simulations ( see for example @xcite ) . however , such simulations can be computationally expensive and will likely preclude a statistically sound exploration of the relevant parameter regime . as a result , here we shall again opt for a parametrized treatment of stochastic forcing and perform the simulations within the framework of the perturbative model . the simplest approach to mimicking stochastic forcing is by introducing gaussian white noise , @xmath138 , into the equations of motion @xcite . following the purely dissipative treatment outlined above , we shall only apply these effects to the @xmath87 degree of freedom . moreover , as we will seek to examine the evolved outcome of these simulations , we shall additionally introduce a time - dependence , @xmath139 , to the damping and driving terms which will ( over a time considerably greater than @xmath124 ) cause these effects to vanish slowly . accordingly , the relevant equations of motion become : @xmath140 \upsilon \nonumber \\ \frac{dv}{dt } & = \left ( \frac{dv}{dt } \right)_{\mathcal{h } } + \left [ \mathscr{f}_v - \frac{v}{\tau_e } \right ] \upsilon . \label{dampdriven}\end{aligned}\ ] ] equations of the form ( [ dampdriven ] ) are typically referred to as langevin equations and yield solutions that should be interpreted in a statistical sense @xcite . it is well known that the integration of white noise , @xmath138 , yields the weiner process , @xmath141 ( a scaling limit of a random walk ) . with a zero expectation value , the progress of a pure weiner process is @xmath142 , where the @xmath143 is the effective diffusion coefficient . @xmath143 is directly related to the variance of @xmath138 , which we take as an adjustable parameter . if the stochastic term in the equations of motion is augmented with dissipative effects ( i.e. the terms within the brackets of equations [ dampdriven ] ) , the progress of the associated quantity becomes limited from above and upon saturation , approaches @xmath144 . consequently , for the system at hand , we can define a dimensionless number : @xmath145 } } \leqslant 1 , \label{dimentsionless}\end{aligned}\ ] ] that approximately characterizes the maximal eccentricity that the orbit would attain exclusively due to interactions with the nebula by unity from above , for bound systems . ] . note that the physical meaning of @xmath146 roughly parallels that of the turbulent schmidt number : it represents a ratio of turbulent forcing to viscous dissipation . thus , a value of @xmath146 that leads to the formation of a system that resembles the real gj876 resonant chain is informative of the properties of the protoplanetay disk in which the system was assembled . of course , the dynamical behavior of the actual damped , driven two degree of freedom system is quite complex ( in part because the value of @xmath54 changes substantially ) , and it is not clear what value of @xmath146 leads to favorable evolution a - priori . thus , we have performed a series of monte - carlo numerical experiments in an effort to survey the characteristic outcome as a function of @xmath146 . for all simulations , a starting value of @xmath135 , corresponding to the poincar@xmath25 surface of section depicted in panel b of figure ( [ f4 ] ) , was adopted and the initial condition was chosen randomly on the section . indeed , the choice of the starting value of @xmath54 is somewhat arbitrary and is generally unimportant because ( with the exception of a finely tuned set of parameters ) the system quickly loses memory of its starting state . each integration spanned @xmath147 years and the functional form for the time - dependence quoted in equations ( [ dampdriven ] ) was chosen to be @xmath148 , where @xmath149 . a range of @xmath150 was explored and 30 realizations of the system were integrated per choice of @xmath146 . upon completion of the integrations , the end state of each simulation was used as an initial condition for a purely hamiltonian integration ( equations [ eom ] ) and a surface of section corresponding to the particular orbit was constructed . the stochastic properties of the orbit were then examined . the results of the monte - carlo survey are shown in figure ( [ f7 ] ) . most importantly , panel a shows the probability of capture into a chaotic laplace resonance as a function of @xmath146 . it is striking that the probability of reproducing the observed state can be substantial within the context of our model , however this remains the case only for a rather restricted range of parameters . particularly , the probability of success is nearly @xmath151 for @xmath152 but drops to zero for @xmath153 and @xmath154 . the reason for a relatively narrow range of @xmath146 that allows for chaotic resonances as an outcome has to do with the evolution of @xmath54 . in the case of the over - damped system ( where @xmath146 is low ) , the evolution essentially proceeds as described in the preceding discussion of a purely dissipative setup . that is , the trajectory collapses onto a nearby quasi - periodic island and @xmath54 evolves to a stationary value characterized by an equilibrated orbital state . in the case of an over - driven system , a somewhat different picture emerges . as turbulent forcing allows the orbit to explore the phase - space stochastically , the system eventually exits the resonance all together , and settles onto an orbit dominated by secular interactions ( not accounted for in our model ) . such evolutions are marked by a substantial decrease in the value of the hamiltonian . the evolution of the eccentricity and the value of @xmath54 for representative trajectories observed in our simulations are plotted in panels b and c of figure ( [ f7 ] ) respectively . the evolutionary track shown in brown , exhibits some similarity to that discussed within the context of purely dissipative evolution . specifically , @xmath54 increases to its maximal value and the system settles to the vicinity of a fixed point . note however , that owing to turbulent forcing , the system retains a finite libration amplitude at the end of the simulation , shown by non - negligible oscillation of the eccentricity . the purple evolutionary track is also similar to the purely dissipative case , but lies in the limit - cycle category . correspondingly , the eccentricity evolution resembles the results of @xmath12-body integrations , depicted in panel a of figure ( [ f6 ] ) . an over - driven simulation , where the system breaks out of resonance and ultimately settles onto a quasi - periodic orbit ( see also @xcite ) is shown in red . note that in this simulation , the evolution of the hamiltonian occurs in the opposite sense compared to the over - damped simulation . that is , the value of @xmath54 decreases relative to its initial condition . finally , an orbit that settles onto a chaotic laplace resonance is shown in blue . here , the value of @xmath54 remains close to the @xmath155 range where the poincar@xmath25 surfaces of section depicted in figure ( [ f4 ] ) show a considerable fraction of phase - space that is occupied by chaotic seas . as such , in this simulation the system successfully acquires a chaotic end - state , facilitated by stochastic elimination of stable quasi - periodic trajectories throughout the evolutionary sequence . in this paper , we have examined the dynamical state as well as the formation scenario of a multi - resonant chain of planets residing in the gj876 system . in particular , we began our investigation by constructing a simplified model , based on hamiltonian perturbation theory , that broadly captures the dynamical behavior of the system . upon a detailed examination of this model , we considered the assembly of the multi - resonant system in presence of dissipative and stochastic forces . with the tally of exoplanetary detections now in excess of a thousand @xcite , it is tempting to question the value held in detailed examination of the orbital architecture of a particular system . while such criticism may be appropriate for bodies whose overall state is reminiscent of other well - characterized exoplanets , the gj876 system easily escapes such scrutiny as it clearly stands out as a unique member of the observational aggregate . to this end , the gj876 c - b - e " system represents the only known extrasolar laplace - like resonance . unlike the the galilean satellites however , the gj876 multi - resonant system is vigorously chaotic , with a characteristic decoherence time on the order of a decade . the primary difference between the two configurations stems from the high eccentricities attained by the massive planets in the gj876 system and the fact that the resonant libration amplitudes associated with planet e " never approached near - null values . while interesting in its own right , the nature of the gj876 resonance plays an important additional role as a means of placing much - needed constraints on the nature of the protoplanetary disk from which the system emerged . the characterization of the stability , dynamical structure , and origins of the laplace resonance inherent to the galilean satellites ( within a tangible framework ) is among the most widely - celebrated achievements of celestial mechanics of the latter half of the 20th century ( see @xcite for reviews ) . as we have shown here , the fiercely chaotic laplace - like resonance of the gj876 system can also be understood within the context of a simple time - independent perturbative model , characterized by two degrees of freedom . specifically , the mathematical formulation of the governing hamiltonian is analogous to a pair of momentum - coupled pendulums . while in agreement with direct numerical integration ( @xcite ) , this model clearly illuminates the origins of stochastic motion and allows one to derive simple analytic estimates of the lyapunov time and action diffusion coefficient related to the outer - most planet s eccentricity @xcite . given the system s short ( @xmath156 decadal ) timescale for stochastic excursions ( implied by the analytical estimates and observed in the numerical simulations ) , one may readily expect the associated quasi - random signal to become evident in the observational radial velocity time series of the system . specifically , one may naively expect that the individual lines observed on the periodogram will be chaotically broadened @xcite . this consideration may place fundamental limits on any dynamical model s ability to match the observed signal , even provided an outstanding signal to noise ratio ( see @xcite ) . in practice , however , the chaotic nature of the orbits is probably not the dominant source of error in the data @xcite and likely leads to a limited reduction in the goodness of fit @xcite . a crucial feature that arises naturally within the context of the developed framework is the simultaneous rapidity of the chaotic loss of dynamical memory and long - term stability . this characteristic contrasts the de - populated mean motion resonances ( kirkwood gaps ) within the asteroid belt @xcite but bears some resemblance to the long - term stability of the planets of the solar system @xcite . the analogy is in fact surprisingly robust , especially for the case of mercury . that is , much like gj876 e , " mercury is characterized by a lyapunov time that is orders of magnitude shorter than the solar system s lifetime and has only a slim chance of escaping the solar system within the remaining main - sequence lifetime of the sun @xcite . in a related effort , @xcite have shown that mercury s secular evolution can be understood within the context of an autonomous hamiltonian corresponding to a pair of momentum - coupled pendulums ( see also @xcite ) , much like the case of the resonant evolution of gj876 e " . furthermore , in direct correspondence to the stability of gj876 , @xcite have recently shown that mercury s long - term stability also arises from a topological boundary associated with the approximate conservation of the hamiltonian itself ( see poincar@xmath25 surfaces of section depicted in figure [ f4 ] ) . the formation of the gj876 multi - resonant chain almost certainly requires convergent approach of the orbits facilitated disk - driven migration . while a perfectly laminar evolutionary track had previously been invoked to explain the ( nearly periodic ) c - b " mean motion resonance ( see @xcite as well numerous other references quoted above ) , our modeling suggests that a chaotic configuration is incompatible with formation in a purely dissipative nebular environment . instead , limited stochastic perturbations arising from turbulent forcing @xcite appear to be needed to explain the observed configuration . accordingly , we utilized our perturbative model to survey a range of assumed disk parameters with an eye towards identifying a regime that leads to a successful formation of a chaotic multi - resonant chain . this allowed us to statistically infer ( albeit within the framework of an approximate model ) the likely combination of relative strengths of turbulent perturbations and dissipative effects inherent to the gj876 protoplanetary disk . the corresponding dimensionless quantity derived in section ( 4.2 ) is related to the turbulent schmidt number of the disk . although parameters of this sort are of great importance for the quantification of angular momentum transport within disks @xcite , their observational characterization is at present scarce @xcite . thus , theoretical considerations such as that undertaken in this work are required to inform the relevant characteristics . systems such as gj876 provide a rare opportunity to perform such an analysis . accordingly , a viable and consequential extension of the presented work would involve direct magnetohydrodynamic modeling of the assembly of the gj876 multi resonant system within its natal disk ( see e.g. @xcite ) . such an investigation would no doubt further our understanding of typical formation environments of planetary systems and place additional constraints on the dominant processes involved in sculpting their orbital architectures .

we consider the problem of factorization of permutations . we begin with a discussion of some simple mathematical ideas behind a matching game known as amidakuji concerning the assignment of jobs to people . it is shown that the ideas are related to the study of other topics including sorting algorithms in computer science , permutations in abstract algebra , and research in genomics . we then extend the ideas to obtain new results and discuss new problems on factorization of permutations into special type of transpositions . .85 * keywords . * amidakuji , bubble sort , genomics , permutation , symmetric groups , cayley graphs .

1303.3776

we consider the problem of factorization of permutations . we begin with a discussion of some simple mathematical ideas behind a matching game known as amidakuji concerning the assignment of @xmath0 jobs to @xmath0 people . it is shown that the ideas are related to the study of other topics including sorting algorithms in computer science , permutations in abstract algebra , and research in genomics . in section 2 , we extend the ideas to obtain new results and discuss new problems on factorization of permutations into special type of transpositions . amidakuji , also known as ghost leg drawing , is a matching game ( with a certain degree of randomness ) for creating a one - one correspondence between @xmath0 people and @xmath0 jobs , or between @xmath0 boys and @xmath0 girls . this game is quite popular in asian regions ( japan , korea , hong kong , taiwan ) , and has even been created into an iphone app . @xcite . to do the matching , one draws @xmath0 vertical lines connecting persons @xmath1 to jobs @xmath2 , which yields the natural assignment that @xmath3 corresponds to @xmath4 for @xmath5 . one then randomly adds some horizontal line segments between any pair of adjacent vertical lines so that no two horizontal line segments meet . now , @xmath3 will find their job assignment by going down the @xmath6th vertical line and making a turn ( to left , right , or going down ) whenever an intersection of a horizontal and vertical line segments is encountered . this will set up a one - one correspondence between @xmath1 and @xmath2 . one can see the example in figure 1 ( from @xcite ) below . first time players are often amazed by the fact that the procedure always leads to a one - one correspondence between @xmath0 people and @xmath0 jobs . that is , at the end of the game , no two people will be assigned the same job . also , it is natural to ask whether one can get all possible job assignments , and how to draw the horizontal lines to arrange the specific job assignment . we list several natural questions in the following . 1 . why do we always get a one - one correspondence ? is it possible to get every possible job assignments ? 3 . what is the minimum number of horizontal lines one has to add to achieve a certain specific job assignment ? 4 . what is the job assignment so that the minimum number of horizontal lines needed for the assignment is maximum ? the answer of question 1 is easy if one approaches it in the `` right '' way . one can interpret the procedure as follows . the people @xmath1 sit in a row of chairs labeled by @xmath2 . adding a line segment corresponds to switching an adjacent pair of people . of course , no matter how many horizontal line segments are added , @xmath0 people will still occupy the @xmath0 seats without any conflicts . formally , one can prove the result by induction on the number of horizontal line segments . if there is no horizontal line segment , clearly one has a one - one correspondence between the people and jobs . suppose @xmath7 horizontal line segments will give rise to a one - one correspondence . then adding one more line segment will only add an additional switch of the job assignment of an adjacent pair of people . using the above interpretation , one can readily answer questions 2 4 . we can represent the original sitting arrangement by @xmath8 $ ] and the final sitting arrangement by @xmath9 $ ] , which is referred to as a _ permutation _ of @xmath10 we can always restore the order of @xmath9 $ ] by doing the following . survey the list from left to right , whenever we encounter two consecutive number so that the left one is larger than the right one , we switch the two numbers . we * claim * that after a number of such switches , one will not see any pair of consecutive terms such that the left number is larger than the right one . then the permutation must be @xmath10 $ ] at that stage . reversing the above switching steps , we will generate the permutation @xmath9 $ ] from @xmath11 $ ] . hence , we can generate any permutation if we can prove our * claim*. to this end , we define the number of inversions of a permutation @xmath9 $ ] as the sum of the numbers inv@xmath12 , where inv@xmath12 is the numbers of integers smaller than @xmath6 lying on the right side of @xmath6 in @xmath9 $ ] . for example , the number of inversions in @xmath13 $ ] is @xmath14 . note that if we exchange two consecutive numbers we either increase or decrease the inversion numbers by @xmath15 or @xmath16 depending on whether the left number is smaller or larger . thus , each step of our proposed scheme will reduce the number of inversions by 1 . as a result , after a number of steps ( equal to the number of inversions ) , we will have no more inversions . * figure 1 * the above analysis does not only show that one can get all permutations of @xmath17 $ ] . it also tells us the minimum number of switches needed to get to a special permutation @xmath9 $ ] , namely , the number of inversions in the permutation . now , it is clear that the permutation with the maximum number of inversions is @xmath18 $ ] requiring @xmath19 switches . so , we have the answers for questions 2 4 . * sorting algorithm * it is easy to see that the mathematical idea behind the amidakuji game can be used to determine the number of steps required to restore the order of a permutation @xmath9 $ ] under the restriction that one can only switch two consecutive numbers in each step . in fact , this is the well known bubble sort algorithm in computer science study ; see @xcite and its references . the basic idea is : to restore a permutation @xmath9 $ ] to the identity permutation @xmath17 $ ] by exchanging a consecutive pair of numbers in each step , the minimum number of steps needed is the number of inversions in @xmath9 $ ] . as long as one exchanges a pair of numbers @xmath20 for @xmath21 in each step to reduce the total number of inversions in the permutation , the procedure will bring @xmath9 $ ] to @xmath17 $ ] in the minimum number of steps . there are many other efficient sorting algorithms . we refer the readers to the excellent reference @xcite for more details . * genome rearrangements * it turns out that the study is related to the study of genome rearrangement . in nature , some species have similar genetic make up and differ only in the order of their genes . finding the shortest rearrangement path between two related bacteria or viruses is useful in drug discovery and vaccine development . the study is also useful in the study of mutations . in fact , a slight change of the genetic sequence may have significant effect , and it is more likely to see a change ( permutation ) of the positions of the nucleotides close to each other in the genetic sequences . that is why there are keen interest in studying such permutations in genetic sequences ; for example , see @xcite and their references . * group theory and cayley graph * the problem is clearly related to the study of group theory and cayley graph in abstract algebra . denote by @xmath22 the set of permutations of @xmath10 $ ] , i.e. , symmetric group of degree @xmath0 , and regard every permutation as a vertex . we connect two vertices @xmath23 and @xmath24 if @xmath25 for a permutation @xmath26 which exchanges two consecutive numbers at the @xmath27th and @xmath28st positions for some @xmath29 . then the bubble sort algorithm amounts to moving a permutation @xmath9 $ ] to @xmath17 $ ] in the cayley graph most efficiently ( using the minimum number of steps ) . moreover , @xmath30 is the maximum distance from @xmath17 $ ] to another permutation @xmath9 $ ] , which is @xmath31 $ ] . one easily shows that the value @xmath30 is actually the maximum distance between any two vertices , and is known as the diameter of the cayley graph . it also indicates that a permutation is a product of no more than @xmath30 transpositions of the form @xmath32 . a more general version of the above problem is how to represent a permutation as a product of a minimal number of permutations selected from a certain set . the study has a long history , and attracted many authors ; see @xcite and their references . we will consider this type of problem in the next section . the study of this problem has a long history . d@xmath33nes @xcite determined that the minimum number is @xmath34 , where @xmath35 is the number of disjoint cycles in the given permutation on @xmath0 letters . other approaches have been used to get the minimum number in @xcite . later on , people found it natural and interesting to determine the number of minimal factorization of a permutation and the minimal product with factors from a given set of transpositions @xcite . * graph algorithms * graph algorithms concern the study of efficient computational methods in solving pure and applied problems which can be modeled using graph theory ; see @xcite and their references . formulating our problem in terms of the permutation group and cayley graph naturally lead to the use of graph algorithms to search the shortest path from a given permutation @xmath36 $ ] to the identity @xmath17 $ ] , and the search of the diameter of the cayley graph . in our study , we also develop some computer programs in our study ; see section 2.3 . in the following we consider a variation of the sorting algorithm , namely , in each step one can switch two numbers which are separated by fewer than @xmath7 numbers . in other words , we can switch @xmath37 and @xmath38 in @xmath9 $ ] as long as @xmath39 . we will see that it is more challenging to determine the minimum number of such moves to reduce a permutation to the identity permutation , and which permutation requires the maximum number of moves for the reduction . mathematically , for @xmath40 , let @xmath41 be the set of transpositions @xmath42 in @xmath22 with @xmath43 . then @xmath41 generates @xmath22 , i.e. , every permutation in @xmath22 is a product of transpositions in @xmath41 . we are interested in finding the smallest number of transpositions in @xmath41 needed to convert a given permutation to the identity , and the permutation requires the maximum number of transposition to do the reduction . in the context of cayley graph , one can construct the graph @xmath44 so that the vertices are permutations in @xmath22 , and two vertices @xmath23 and @xmath24 if @xmath45 . denote by * 1 * the identity permutation @xmath11 $ ] . we are interested in the shortest path joining the identity permutation @xmath46 to a given permutation @xmath47 . the length of this path is denoted by @xmath48 . also , we are interested in the permutation @xmath49 with a maximum distance to the identity @xmath46 . clearly , the maximum distance will be the same as the maximum distance between any two vertices in the cayley graph , and we will denote this quantity by @xmath50 . the result in section 1 shows that @xmath51 . in what follows we always use @xmath52 to specify that the transpositions are from @xmath41 . denote by @xmath53 the permutation @xmath54 such that @xmath55 and @xmath56 for other @xmath6 , which is referred to as a length @xmath35 cycle . also denote by @xmath57 the smallest integer greater than or equal to @xmath58 and @xmath59 the largest integer less than or equal to @xmath58 respectively . the following lemma will be used frequently in our discussion . [ le1 ] let @xmath54 . a transposition in a cycle of @xmath47 splits the cycle into two disjoint cycles . a transposition between two disjoint cycles of @xmath47 joints the two cycles into one . _ proof . let @xmath60 and @xmath61 be two disjoint cycles in @xmath47 . let @xmath62 and @xmath63 be two transpositions with @xmath64 , @xmath65 and @xmath66 . then @xmath67and @xmath68 -.3 in _ the result for @xmath69 was discussed in section 1 . the other extreme is when @xmath70 , i.e. , @xmath41 is the set of all transpositions . we have the following known result , see @xcite . here give a short proof of it for completeness . [ 2.2 ] suppose @xmath71 . then @xmath72 , where @xmath35 is the number of cycles in the disjoint cycle representation of @xmath47 under the convention that each fixed point is counted as a 1-cycle . thus , @xmath73 is attained at a @xmath0-cycle . _ proof . writing @xmath74 for a minimum number of transpositions @xmath75 is the same as finding the minimum number of transpositions @xmath76 such that @xmath77 . using lemma [ le1 ] , to convert @xmath47 to the product of @xmath0 disjoint cycles , i.e. , back to the identity permutation @xmath17 $ ] most efficiently by composing @xmath47 with transpositions , the most efficient way is to choose transposition @xmath42 in each step such that @xmath27 and @xmath6 lie in the same cycle . so , using @xmath78 transpositions @xmath79 to convert @xmath47 to the identity permutation will be the most efficient scheme . _ [ 2.3 ] suppose @xmath80 . then @xmath81 . a permutation @xmath71 attains @xmath82 if and only if it is of the form @xmath83 where @xmath84 is an @xmath85-cycle with terms @xmath86 . _ if @xmath47 in @xmath22 has disjoint cycle decomposition @xmath87 without @xmath88 , then one can do it in @xmath89 steps because we do not use the transposition @xmath88 by proposition [ 2.2 ] . if one of the cycles , say , @xmath90 is @xmath88 , then at least one of the transpositions @xmath91 used to write @xmath92 $ ] must involve @xmath93 or @xmath0 . that particular step will delay the process of decomposing @xmath47 into @xmath0 singleton cycle in at least one step by the argument in the proof of proposition [ 2.2 ] . by even and odd permutation rule , we need at least two more transpositions . but we can always handle @xmath94 using @xmath95 transpositions . so , the minimum number needed for @xmath47 is @xmath96 if @xmath88 is not a cycle . it is @xmath97 if @xmath88 is one of the cycle . [ 2.4 ] let @xmath98 . suppose @xmath54 has a disjoint cycle decomposition @xmath99 . then one of the following holds . 1 . if one of following appear in @xmath100 : @xmath88 , @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , then @xmath47 is a product of @xmath107 elements in @xmath108 . 2 . if all @xmath100 are not the cycles stated in ( a ) , then @xmath47 is a product of @xmath78 elements in @xmath108 . consequently , @xmath109 and for any @xmath54 , @xmath110 if and only if @xmath111 such that @xmath23 is one of the two - cycles or three - cycles in ( a ) . _ proof . _ suppose first that all @xmath100 are not the cycles stated in ( a ) , then we need not use @xmath112 in the decomposition . by proposition [ 2.2 ] , the statement ( b ) holds . now suppose the condition in ( a ) holds . if only one of @xmath100 is equal to the cycles mentioned in ( a ) , we can assume that @xmath90 is the problematic cycle . suppose @xmath113 contains @xmath114 elements . then by proposition [ 2.3 ] , we can decompose @xmath113 into product of @xmath114 length one cycles , i.e. , @xmath115 , using @xmath116 transpositions in @xmath108 . notice that the cycle @xmath117 is in this decomposition . we have the following cases . @xmath118 for the first three cases , @xmath119 and @xmath99 can be composited into @xmath120 transpositions in @xmath108 . the the remaining four cases , @xmath121 and @xmath99 can also be composited into @xmath122 transpositions in @xmath108 . finally , if two of the cycles @xmath100 are the problematic cycles , say @xmath90 and @xmath123 , then we must have @xmath124 . notice that @xmath125 on the other hand , by proposition [ 2.3 ] , @xmath126 can be decomposed into product of @xmath127 using @xmath128 transpositions in @xmath108 . therefore , @xmath47 can be decomposition into a product of @xmath107 cycles in @xmath108 . if @xmath129 then @xmath130 if @xmath131 then @xmath132 if @xmath133 then @xmath134 if @xmath135 then @xmath136 if @xmath137 , then @xmath138 if @xmath139 , then @xmath140 if @xmath141 , then @xmath142 the result follows . suppose @xmath143 and @xmath144 with @xmath145 . assume that @xmath146 . equivalently , @xmath147 is greater than or equal to @xmath93 . then we can use the following transpositions to move @xmath93 and @xmath0 back to the first position and the last position , respectively @xmath148 the number of these transpositions is @xmath149 now it remains to consider the case when both @xmath150 . then @xmath151 implies @xmath152 and @xmath153 . furthermore , @xmath154 implies @xmath155 . suppose @xmath156 with @xmath157 . then @xmath158 the number of the above transpositions is @xmath159 -.3 in if @xmath160 , let @xmath161 and @xmath162 . then we can use the following transpositions to move @xmath93 and @xmath0 back to the first position and the last position , respectively @xmath163 the number of these transpositions is at most @xmath164 suppose @xmath165 with @xmath166 be integers and @xmath167 . then @xmath168 . if @xmath169 , let @xmath170 be the smallest integer such that @xmath171 , and @xmath172 be the largest integer such that @xmath173 . then we can use the following transpositions to move @xmath93 and @xmath0 back to their positions @xmath174 the number of these transpositions is at most @xmath175 -.3 in given @xmath54 . denote by @xmath177 the set of transpositions in @xmath178 splitting a cycle of @xmath47 into two and by @xmath179 the set of transpositions in @xmath178 jointing two cycles of @xmath47 into one . we call @xmath177 and @xmath179 type one and type two transpositions , respectively . by direct verification we have [ le7 ] let @xmath54 , @xmath180 and @xmath181 . then @xmath182 let @xmath114 and @xmath0 be integers with @xmath183 and @xmath184 . then @xmath185 _ proof . notice that @xmath186 . we claim that there exists a positive integer @xmath187 such that @xmath188 let @xmath71 . set @xmath189 and @xmath190 . then one can choose two sequences of distinct integers @xmath191 in @xmath192 and @xmath193 in @xmath194 with @xmath195 such that @xmath196 then we can use the following transpositions to move @xmath191 and @xmath193 back to their original positions respectively . @xmath197 where @xmath198 if @xmath199 and @xmath200 and @xmath201 if @xmath202 and @xmath203 . notice that after the transpositions @xmath204 , @xmath205 , and @xmath206 , @xmath207 and @xmath208 are in their original positions and @xmath209 and @xmath210 are in the positions @xmath211 and @xmath212 respectively . then one can move @xmath213 entries to their original positions in at most @xmath214 steps . further , one can fix the remaining @xmath215 entries in at most @xmath216 steps . therefore , @xmath217 thus , the claim holds . now we use induction on @xmath114 to show @xmath218 by propositions [ 2.3 ] and [ 2.4 ] , the inequality holds for @xmath219 and @xmath220 . now suppose @xmath221 and for any @xmath222 , @xmath223 for all @xmath224 . by proposition [ 2.6 ] and the above claim , we have @xmath225 _ _ proof . we use induction on @xmath114 to show @xmath218 by propositions [ 2.3 ] and [ 2.4 ] , the inequality holds for @xmath219 and @xmath220 . now suppose @xmath221 and for any @xmath222 , @xmath223 for all @xmath224 . by proposition [ 2.6 ] we have @xmath226 _ on the other hand , let @xmath227 and @xmath228\\ & = & \left(1,n - p+1\right ) \left(2,n - p+2\right ) \cdots \left(p , n\right ) \left(p+1,p+2 , \ldots , n - p\right).\end{aligned}\ ] ] @xmath229\\ & = & \left(1,n-\lfloor\frac{k}{2}\rfloor+1\right ) \left(2,n-\lfloor\frac{k}{2}\rfloor+2\right ) \cdots \left(\lfloor\frac{k}{2}\rfloor , n\right ) \left(\lfloor\frac{k}{2}\rfloor+1,\lfloor\frac{k}{2}\rfloor+2 , \ldots , n-\lfloor\frac{k}{2}\rfloor\right).\end{aligned}\ ] ] it suffices to verify @xmath230 . then ( [ eq3.1 ] ) and @xmath231 ensures that @xmath232 let @xmath233 be @xmath170 transpositions in @xmath234 such that @xmath235 assume that @xmath236 for @xmath237 with @xmath238 and @xmath239 for other @xmath240 . since @xmath241 are not in @xmath234 and @xmath242 , one needs at least @xmath243 distinct transpositions of @xmath244 from @xmath179 that joining the mentioned cycles with other cycles . thus , @xmath245 . notice that @xmath246 by proposition [ 2.2 ] and lemma [ le7 ] , @xmath247 then @xmath248 therefore , @xmath249 -.3 in @xmath250 therefore , @xmath251 -.3 in for any cycle @xmath252 in @xmath22 , we say @xmath84 is in the set @xmath253 if for each term @xmath254 in @xmath84 , there is a term @xmath255 such that @xmath256 . on the other hand , a cycle @xmath257 if there is some @xmath258 such that @xmath259 for all @xmath260 . [ 2.9 ] let @xmath0 and @xmath7 be positive integers and @xmath261 . suppose @xmath262 is a length @xmath243 cycle in @xmath22 . then one of the following holds . 1 . if @xmath263 , then @xmath84 can be written as a product of @xmath264 transpositions in @xmath265 . if @xmath266 , then @xmath84 can be written as a product of @xmath267 transpositions in @xmath41 . furthermore , suppose @xmath268 and @xmath269 are two cycles in @xmath253 if there are @xmath258 and @xmath270 such that @xmath271 can be written as a product of @xmath272 transpositions in @xmath41 . suppose @xmath263 . notice that @xmath273 for all @xmath172 . then one can write @xmath274 then the result ( a ) holds . _ suppose now @xmath266 . that is , there is some @xmath258 such that @xmath259 for all @xmath260 . without loss of generality , we may assume that @xmath275 . then @xmath84 can be written as @xmath276 thus , the result ( a ) holds . suppose now @xmath100 and @xmath277 are disjoint cycles and satisfying ( [ rs ] ) . we may assume that @xmath275 and @xmath278 in ( [ rs ] ) . then @xmath279 can be written as @xmath280 thus , the result follows . [ 2.10 ] let @xmath0 and @xmath7 be positive integers and @xmath261 . suppose @xmath71 has a disjoint cycle decomposition @xmath281 under the convention that each fixed point is counted as a @xmath93-cycle . if @xmath170 of the cycles @xmath100 belongs to @xmath253 and among them , @xmath172 disjoint pairs of @xmath100 and @xmath277 satisfy condition ( [ rs ] ) in lemma [ 2.9 ] , then @xmath282 . suppose @xmath71 has a disjoint cycle decomposition @xmath281 such that @xmath283 for @xmath284 , @xmath285 for @xmath286 . further , the cycles @xmath287 and @xmath288 satisfy condition ( [ rs ] ) for @xmath289 . _ assume that @xmath277 has length @xmath290 . then by lemma [ 2.9 ] , @xmath291 -.3 in let @xmath0 and @xmath7 be positive integers and @xmath261 . then @xmath292 set @xmath293 . a permutation @xmath71 attains @xmath294 if and only if one of the following holds . 1 . @xmath295 is even and @xmath47 is a product of @xmath296 disjoint cycles of the form @xmath297 where @xmath298 , @xmath299 , and @xmath300 . @xmath295 is odd and @xmath47 is a product of @xmath296 disjoint cycles of the form 1 . @xmath301 where @xmath302 , @xmath303 , and @xmath304 such that @xmath305 . 2 . @xmath306 where @xmath307 , @xmath299 , and @xmath308 . 3 . @xmath309 where @xmath310 , @xmath311 , and @xmath312 . 4 . @xmath313 where @xmath307 , @xmath311 , and @xmath314 such that @xmath315 . + @xmath316 { \rm ( ii ) } & & ( i_1,j_1 ) ( i_2,j_2 ) \cdots ( i_{p-1},j_{p-1 } ) ( i_p , j_p , k_1 ) ( k_2,\dots , k_{n-2p}),\end{aligned}\ ] ] where @xmath317 , @xmath318 , @xmath319 , and @xmath320 such that @xmath321 . let @xmath0 and @xmath7 be positive integers and @xmath261 . suppose @xmath71 has a disjoint cycle decomposition @xmath281 under the convention that each fixed point is counted as a @xmath93-cycle . if @xmath170 of the cycles @xmath100 belongs to @xmath253 and among them , @xmath172 disjoint pairs of @xmath100 and @xmath277 satisfy condition ( [ rs ] ) in lemma [ 2.9 ] , then @xmath282 . consequently , @xmath292 further , a permutation @xmath71 attains @xmath322 if and only if @xmath47 is a product of @xmath35 disjoint cycles @xmath323 with @xmath324 such that @xmath325 of @xmath100 belong to @xmath253 and @xmath172 disjoint pairs of them satisfy ( [ rs ] ) . _ proof . suppose @xmath47 has a disjoint cycles decomposition @xmath281 . assume @xmath326 are disjoint cycles in @xmath253 while @xmath327 are not in @xmath328 . notice that @xmath329 as there must exist at least one cycle that is not in @xmath253 . for @xmath330 , let @xmath331 and @xmath332 be the smallest team and largest term of the cycle @xmath277 respectively . since @xmath333 , @xmath334 and @xmath335 are all distinct , @xmath336 , and @xmath337 we claim that @xmath100 and @xmath277 satisfy ( [ rs ] ) if @xmath338 . the claim clearly follows by the fact that @xmath339 . then there are @xmath340 such that @xmath341 and @xmath342 . it follows that @xmath343 thus , @xmath100 and @xmath277 are two distinct cycles in @xmath253 . furthermore , as @xmath344 the pair of cycle @xmath100 and @xmath277 satisfies the condition ( [ rs ] ) . now assume there are @xmath172 disjoint pairs of cycles satisfying ( [ rs ] ) . without loss of generality , we assume that no any pair satisfying ( [ rs ] ) can be found among the cycles @xmath345 with @xmath346 . by the claim , we must have @xmath347 for all @xmath348 . then @xmath349 thus , @xmath350 and hence @xmath351 . so lemma [ 2.10 ] and by the fact that @xmath352 , @xmath353 furthermore , the equality holds only if @xmath354 and @xmath355 . assume now that @xmath47 attains the upper bound . them @xmath354 . as any two cycles of @xmath345 do not satisfy ( [ rs ] ) , we must have @xmath356 further , at most one of the inequalities holds if @xmath357 is odd while both two inequalities are strictly inequalities if @xmath357 is even . if @xmath358 , the union of two sets @xmath359 contains at most one element . therefore , there is @xmath114 such that @xmath360 and @xmath361 . but then @xmath362 , which contradicts that @xmath363 . therefore , @xmath364 . thus , @xmath365 and @xmath47 has a @xmath296 disjoint cycle decomposition @xmath366 with @xmath367 and @xmath368 . assume now that @xmath47 attains the upper bound . if @xmath358 , there exist distinct indices @xmath369 and @xmath370 such that @xmath371 it follows that @xmath372 by the claim , we have @xmath373 disjoint pairs of cycles satisfying ( [ rs ] ) , which contradicts to the assumption that only @xmath172 disjoint pair exist . therefore , @xmath364 . thus , @xmath365 and @xmath47 has a @xmath296 disjoint cycle decomposition @xmath366 with @xmath367 and @xmath368 . if @xmath357 is even , then @xmath374 and @xmath375 it follows that @xmath376 thus , each @xmath100 has exactly one element in @xmath377 and one element in @xmath378 . suppose any of @xmath379 has length greater than @xmath380 . without loss of generality , assume @xmath381 has length @xmath243 , @xmath382 , with @xmath383 . let us assume that @xmath384 . let @xmath385 be the only element of @xmath90 that lies in @xmath378 . then one can see that @xmath386 on the other hand , there is another cycle , say @xmath123 , such that @xmath387 . let @xmath388 . then @xmath389 we assume that @xmath390 . then @xmath391 can be written as @xmath392 which is a product of @xmath272 transpositions in @xmath265 . by lemma [ 2.9 ] , @xmath393 can be written as a product of @xmath394 transpositions in @xmath41 . thus , @xmath47 is a product of @xmath395 transpositions , which contradicts that @xmath47 attains the upper bound . therefore , all @xmath100 have length @xmath380 and the case ( a ) holds by ( [ uv0 ] ) . now if @xmath396 is odd , then @xmath397 and either @xmath398then either @xmath399 or @xmath400 suppose any of @xmath100 contains an element in @xmath401 . without loss of generality , assume @xmath381 is the cycle and @xmath402 . then there is another length @xmath403 cycle , say @xmath123 , such that @xmath404 . following the same above argument , we conclude that @xmath391 is a product of @xmath272 transpositions in @xmath41 . by a similar argument as above , one can conclude that this contradicts to our assumption . therefore , all elements of @xmath100 lie @xmath405 . furthermore , at most two cycles have length greater than @xmath380 . suppose all cycles have length @xmath380 . then by ( [ uv1 ] ) and ( [ uv2 ] ) , the case ( b.i ) follows . suppose now exactly one cycle has length @xmath220 . if @xmath406 and @xmath407 , then the length @xmath220 cycle can contain either @xmath296 or @xmath408 but not both . here ( b.ii ) holds if the cycle contains @xmath296 and ( b.iii ) holds otherwise . suppose @xmath409 and @xmath407 . we assume that @xmath410 with @xmath411 and @xmath412 . we claim that @xmath408 does not lie in any of @xmath379 . suppose not , then @xmath408 must lie in the length @xmath220 cycle , say @xmath413 with @xmath414 . then @xmath415 in both cases , @xmath391 can be written as @xmath416 transpositions in @xmath41 . therefore , ( b.ii ) holds . if @xmath406 and @xmath417 , then by a similar argument , it can be shown that @xmath296 is not in any of @xmath379 and so ( b.iii ) holds . suppose exactly one cycle has length @xmath418 . by ( [ uv1 ] ) and ( [ uv2 ] ) , the length @xmath418 cycle must contain at least one of @xmath419 and @xmath408 . suppose the cycle @xmath420 contains only one of them , say @xmath408 . let @xmath421 . then there is another length @xmath380 cycle , say @xmath422 with @xmath423 . if @xmath424 , then @xmath425 is a length @xmath220 cycle containing @xmath408 . then by the above discussion , @xmath426 can be written as a product of @xmath416 transpositions and so @xmath391 is a product of @xmath427 transpositions . similarly , one can show that @xmath428 is a product of @xmath427 transpositions if @xmath429 or @xmath430 . now if all @xmath431 , @xmath432 , and @xmath433 are greater than @xmath7 , one must have @xmath434 and @xmath435 . thus , ( b.iv ) follows similarly , the result holds if @xmath90 contains only @xmath296 . finally , suppose @xmath90 contains both @xmath296 and @xmath408 . we show that it is impossible to have @xmath436 and @xmath437 . if @xmath438 , then @xmath439 which is not in @xmath440 . then @xmath90 can be written as a product of @xmath220 transpositions in @xmath41 . the same result holds if @xmath421 . suppose @xmath411 and @xmath441 . then @xmath442 . therefore , @xmath90 must have the form @xmath443 wih @xmath444 and @xmath445 . thus , ( b.iv ) holds . finally suppose exactly two cycles have length greater than @xmath380 . then they must have length @xmath220 . let @xmath446 and @xmath413 be the two cycles . by the above argument , @xmath296 and @xmath408 can not appear in any cycle other than @xmath90 and @xmath123 . moreover , @xmath296 and @xmath408 can not be in the same cycle . so we may assume that @xmath438 and @xmath414 . then @xmath447 in all cases , @xmath428 is a product of @xmath427 transpositions in @xmath41 , which contradicts to our assumption . therefore , at most one cycle has length greater than @xmath380 . we claim that @xmath296 and @xmath448 can not appear in two disjoint cycles . otherwise suppose @xmath296 and @xmath448 appear in @xmath90 and @xmath123 , respectively , which implies @xmath90 and @xmath123 can not be both 2-cycles . we have 3 cases for @xmath90 and @xmath123 : they are both 3-cycles ; one of them is 2-cycle and the other is 3-cycle ; one of them is 2-cycle and the other is 4-cycle . for the first case , we may assume @xmath446 and @xmath413 with @xmath438 and @xmath414 . then @xmath447 in all cases , @xmath428 is a product of @xmath427 transpositions in @xmath41 , which contradicts to our assumption . for the second case , we may assume @xmath410 and @xmath413 with @xmath411 , @xmath414 and @xmath412 . then @xmath415 in both cases , @xmath391 can be written as @xmath416 transpositions in @xmath41 , a contradiction . for the last case , we may assume @xmath449 and @xmath450 with @xmath451 and @xmath452 . if @xmath453 , then @xmath454 is a product of a 1-cycle and a length @xmath220 cycle containing @xmath408 . then by the above discussion , @xmath455 can be written as a product of @xmath416 transpositions and so @xmath391 is a product of @xmath427 transpositions . similarly , one can show that @xmath428 is a product of @xmath427 transpositions if @xmath456 or @xmath457 . now if all @xmath458 , @xmath459 , and @xmath460 are greater than @xmath7 , one must have @xmath461 and @xmath462 . furthermore , at most two cycles have length greater than @xmath380 . suppose exactly two cycles have length greater than @xmath380 . then they must have length @xmath220 . let @xmath446 and @xmath413 be the two cycles . in particular , we may assume that @xmath438 and @xmath414 . then @xmath447 in all cases , @xmath428 is a product of @xmath427 transpositions in @xmath41 , which contradicts to our assumption . therefore , at most one cycle has length greater than @xmath380 . suppose all cycles have length @xmath380 . then by ( [ uv1 ] ) and ( [ uv2 ] ) , the case ( b.i ) follows . suppose now exactly one cycle has length @xmath220 . if @xmath406 and @xmath407 , then the length @xmath220 cycle can contain either @xmath419 or @xmath408 but not both . here ( b.ii ) holds if the cycle contains @xmath296 and ( b.iii ) holds otherwise . suppose @xmath409 and @xmath407 . we assume that @xmath410 with @xmath411 and @xmath412 . we claim that @xmath408 does not lie in any of @xmath379 . suppose not , then @xmath408 must lie in the length @xmath220 cycle , say @xmath413 with @xmath414 . then @xmath415 in both cases , @xmath391 can be written as @xmath416 transpositions in @xmath41 . therefore , ( b.ii ) holds . if @xmath406 and @xmath417 , then by a similar argument , it can be shown that @xmath296 is not in any of @xmath379 and so ( b.iii ) holds . finally suppose exactly one cycle has length @xmath418 . by ( [ uv1 ] ) and ( [ uv2 ] ) , the length @xmath418 cycle must contain at least one of @xmath419 and @xmath408 . suppose the cycle @xmath420 contains only one of them , say @xmath408 . let @xmath421 . then there is another length @xmath380 cycle , say @xmath422 with @xmath423 . if @xmath424 , then @xmath425 is a length @xmath220 cycle containing @xmath408 . then by the above discussion , @xmath426 can be written as a product of @xmath416 transpositions and so @xmath391 is a product of @xmath427 transpositions . similarly , one can show that @xmath428 is a product of @xmath427 transpositions if @xmath429 or @xmath430 . now if all @xmath431 , @xmath432 , and @xmath433 are greater than @xmath7 , one must have @xmath434 and @xmath435 . thus , ( b.iv ) follows similarly , the result holds if @xmath90 contains only @xmath296 . finally , suppose @xmath90 contains both @xmath296 and @xmath408 . we show that it is impossible to have @xmath436 and @xmath437 . if @xmath438 , then @xmath439 which is not in @xmath440 . then @xmath90 can be written as a product of @xmath220 transpositions in @xmath41 . the same result holds if @xmath421 . suppose @xmath411 and @xmath441 . then @xmath442 . therefore , @xmath90 must have the form @xmath443 with @xmath444 and @xmath445 . thus , ( b.iv ) holds . it remains to show that all the permutations mentioned in ( a ) and ( b ) attain the upper bound . suppose @xmath357 is even and @xmath47 has the required form in ( a ) . it suffices to verify @xmath463 . since @xmath464 are not in @xmath265 and @xmath465 , one needs at least one term in @xmath466 and @xmath187 distinct type two transpositions to fix those @xmath187 2-cycles . by lemma [ le7 ] , those @xmath187 type two transpositions will lead to @xmath187 extra type one transpositions and @xmath467 when @xmath357 is odd and @xmath47 has the required form in ( b.i ) , ( b.ii ) or ( b.iii ) , one can use the same above argument to deduce that @xmath47 attains the upper bound . now suppose @xmath47 has the required form in ( b.iv ) . denote by @xmath468 for @xmath469 , @xmath470 and @xmath471 . note that for each of @xmath472 , we need a type two transpositions containing a term in @xmath473 to fix it . without loss of generality , we may assume the first @xmath170 cycles @xmath474 are fixed by using a term in @xmath475 and @xmath476 are fixed by using the terms @xmath296 and @xmath408 . therefore , we need at least @xmath477 transpositions to fix @xmath478 and @xmath479 transpositions to fix @xmath480 . note that @xmath481 and @xmath482 . the only possible way to move any term in @xmath483 back to its place is using a transposition @xmath484 . therefore , to move the terms @xmath485 and @xmath486 back to their positions , we need at least @xmath487 transpositions @xmath484 , @xmath488 transpositions of the form @xmath489 and @xmath488 transpositions of the form @xmath490 . therefore , we need at least @xmath491 transpositions to fix @xmath47 . on the other hand , by the former arguments , @xmath47 can be written as a product of @xmath492 transpositions from @xmath41 . by even and odd permutation rule , we have @xmath493 -.3 in ' '' '' * old stuff : * _ proof . we claim that for any disjoint cycles @xmath494 in @xmath253 with @xmath495 , there exists a pair of @xmath496 and @xmath497 satisfying ( [ rs ] ) . if the claim holds , then @xmath498 . also since any cycle contains the term @xmath499 does not belong to @xmath253 , one must have @xmath500 be at least one . @xmath501 to prove the claim , suppose @xmath494 are disjoint cycles in @xmath253 with @xmath495 . for @xmath502 , let @xmath331 and @xmath332 be the smallest team and largest term of the cycle @xmath497 respectively . since @xmath503 @xmath504 and @xmath505 . note also that @xmath506 and @xmath507 for all @xmath508 . then there are @xmath340 such that @xmath341 and @xmath342 . it follows that @xmath509 and hence @xmath510 as @xmath496 and @xmath497 are in @xmath253 . also @xmath511 therefore , the pair @xmath496 and @xmath497 satisfy the condition ( [ rs ] ) . _ to show the equality , let @xmath512 and @xmath228\\ & = & \left(1,n - p+1\right ) \left(2,n - p+2\right ) \cdots \left(p , n\right ) \left(p+1,p+2 , \ldots , n - p\right).\end{aligned}\ ] ] it suffices to verify @xmath513 . then ( [ eq3.1 ] ) and @xmath514 ensures that @xmath515 let @xmath233 be @xmath170 transpositions in @xmath265 such that @xmath235 assume that @xmath236 for @xmath237 with @xmath238 and @xmath239 for other @xmath240 . since @xmath241 are not in @xmath265 and @xmath516 , one needs at least @xmath243 distinct transpositions of @xmath244 from @xmath179 that joining the mentioned cycles with other cycles . thus , @xmath245 . notice that @xmath517 by proposition [ 2.2 ] and lemma [ le7 ] , @xmath247 then @xmath518 therefore , @xmath519 -.3 in ' '' '' _ remarks on the proof of the second part . _ suppose @xmath54 . to obtain equality in ( [ e43 ] ) , one must have @xmath520 and @xmath521 . so no two of @xmath522 satisfy ( [ rs ] ) . next we will argue that @xmath523 . if @xmath357 is even , using the above notation , we claim that @xmath524 and @xmath525 for @xmath526 . in fact , assume there is some @xmath527 . then there are at most @xmath528 terms @xmath529 such that @xmath530 , a contradiction . similarly , one can argue that there is no @xmath531 . therefore , we can not find two cycles in @xmath253 all whose terms belongs to @xmath532 . it follow that @xmath523 . similarly , if @xmath357 is odd , we can deduce that @xmath533 and @xmath534 for @xmath526 . moreover , only one of @xmath419 and @xmath535 can appear in those @xmath334 and @xmath335 . denote by @xmath536 . we can not find two cycles in @xmath253 all whose terms belongs to @xmath537 . thus @xmath523 . _ now * if * we can claim that no term in @xmath532 or @xmath538 can appear in those @xmath243 cycles , then we can use the same argument as in the case @xmath539 to prove that @xmath540 . and when @xmath357 is even , the permutations attaining the maximum should be @xmath541 with @xmath524 , @xmath525 and @xmath84 being a long cycle with terms @xmath542 . when @xmath357 is odd , the maximum permutations may more complicated . _ ' '' '' we are not able to determine the exact value of @xmath50 when @xmath543 . nevertheless , we have the following upper bounds . [ 2.5 ] let @xmath544 be integers with @xmath545 and @xmath546 . then @xmath547 _ proof . let @xmath54 with @xmath548 . suppose @xmath549 with @xmath550 . take the transpositions @xmath551 thus we move @xmath0 to the last position by @xmath552 transpositions and get a new permutation @xmath553 $ ] . note that @xmath554 $ ] is a permutation in @xmath555 . we have @xmath556 -.3 in _ _ proof . let @xmath54 with @xmath548 . suppose @xmath557 with @xmath558 . if @xmath523 , take the transpositions @xmath559 otherwise for @xmath358 , take @xmath560 thus we move @xmath0 to the last position by @xmath561 transpositions and get a new permutation @xmath553 $ ] . note that @xmath554 $ ] is a permutation in @xmath555 . we have @xmath556 -.3 in _ [ 2.6 ] let @xmath544 be integers with @xmath545 and @xmath562 . then @xmath563 _ proof let @xmath54 with @xmath548 and @xmath564 . it suffices to verify that we can move 1 and @xmath0 back to their positions in @xmath565 steps and get a new permutation @xmath566 $ ] . since @xmath567 $ ] is a permutation in @xmath568 , we can fix it in at most @xmath569 steps . therefore @xmath570 _ suppose @xmath165 with @xmath166 be integers and @xmath167 . since @xmath571 , we have @xmath572 . let @xmath170 and @xmath172 be positive integers such that @xmath573 notice that @xmath574 . suppose @xmath575 . then we can use the following transpositions to move @xmath93 and @xmath0 back to the first position and the last position , respectively @xmath576 the number of these transpositions is @xmath577 suppose @xmath578 . then we can use the following transpositions to move @xmath93 and @xmath0 back to the first position and the last position , respectively @xmath579 the number of these transpositions is @xmath580 -.3 in by the results in the previous sections and numerical computation , we have the following table for @xmath50 . @xmath581&6&6&5 & & & & & & \cr 7&21&[10]&8&7&7&6 & & & & & \cr 8&28&[14]&[10]&9&8&8&7 & & & & \cr 9&36&[16]&[11]&10&10&9&9&8 & & & \cr 10&45&[19]&[14]&[12 ] & 11&11&10&10&9&&\cr 11&55&24?&18?&15?&13&12&12&11&11&10&\cr 12&66&30?&22?&19?&16?&14&13&13&12&12&11\cr \end{array}\ ] ] @xmath582 here , the number marked with `` @xmath583 '' are upper bounds for @xmath50 . note that the upper bounds of the numbers in the table are obtained by propositions [ 2.5 ] and [ 2.6 ] . for example , by proposition [ 2.5 ] , @xmath584 ; by proposition [ 2.6 ] , @xmath585 . the values in square bracket @xmath586 $ ] were computed by a java program written by the third author . the program uses breadth first search to generate permutations from @xmath10 \in s_n$ ] using elements in @xmath41 , and identifies the minimum number of steps needed to generate all permutations in @xmath22 , and also the permutations require the maximum number of steps . one can download the program and source code from the links `` http://acm.jhu.edu/@xmath587sharonli/permutation/permutation.java '' and `` http://acm.jhu.edu/@xmath587sharonli/permutation/permutation.class '' . when the command `` permutation '' is executed , one will be asked to input @xmath0 and @xmath7 . one will also be asked whether the program should show all the permutations generated in each step . if one says no , only the permutations generated in the final step will be displayed . because of memory limitations , this program can handle the problem up to @xmath588 . it is easy to modify the program to determine the diameter of cayley graphs with vertices connected by other sets of permutations . here are some numerical results obtained by the program . the same 4 permutations attain @xmath589 and @xmath590 . + @xmath591 , [ 6 , 7 , 4 , 5 , 3 , 1 , 2 ] , [ 6 , 7 , 5 , 3 , 4 , 1 , 2 ] , [ 7 , 5 , 6 , 3 , 4 , 1 , 2].$ ] \2 . the same unique permutation attain @xmath592 and @xmath593 , namely , @xmath594.$ ] \3 . there is a unique permutation attaining @xmath595 , namely , @xmath596.$ ] \4 . there are @xmath597 permutations attaining @xmath598 . there are 39 permutations attaining @xmath599 . + @xmath600 , [ 9,10,8,7,6,5,3,4,2,1 ] , [ 10,9,8,7,6,5,2,4,3,1 ] , [ 9,10,7,8,5,6,3,4,2,1],$ ] + @xmath601 , [ 10,9,8,7,4,6,5,3,2,1 ] , [ 10,8,9,6,7,4,5,2,3,1 ] , [ 10,9,6,8,7,5,4,3,2,1],$ ] + @xmath602 , [ 10,9,7,8,5,6,3,4,1,2 ] , [ 10,9,8,7,6,5,4,1,3,2 ] , [ 10,9,8,6,7,5,4,2,3,1],$ ] + @xmath603 , [ 10,9,8,7,5,6,3,4,2,1 ] , [ 9,10,7,8,6,5,4,3,2,1 ] , [ 9,10,8,7,6,5,4,3,1,2],$ ] + @xmath604 , [ 10,9,8,7,6,3,5,4,2,1 ] , [ 9,8,10,7,6,5,4,3,2,1 ] , [ 9,10,7,8,5,6,4,3,1,2],$ ] + @xmath605 , [ 8,10,9,7,6,5,4,3,2,1 ] , [ 10,9,8,7,6,5,3,4,1,2 ] , [ 10,9,8,7,5,6,4,3,1,2],$ ] + @xmath606 , [ 10,8,9,7,6,5,4,2,3,1 ] , [ 10,9,7,6,8,5,4,3,2,1 ] , [ 10,8,9,6,7,5,4,3,2,1],$ ] + @xmath607 , [ 10,9,7,8,6,5,4,3,1,2 ] , [ 10,9,8,7,6,5,4,3,2,1 ] , [ 9,10,8,7,5,6,3,4,1,2],$ ] + @xmath608 , [ 10,9,8,5,7,6,4,3,2,1 ] , [ 10,9,8,7,6,5,3,2,4,1 ] , [ 10,9,8,6,5,7,4,3,2,1],$ ] + @xmath609 , [ 9,10,8,7,5,6,4,3,2,1 ] , [ 10,9,8,7,6,4,3,5,2,1].$ ] \6 . there are 8 permutations attaining @xmath610 . + @xmath611 , [ 10,9,7,8,6,5,4,3,2,1 ] , [ 10,9,8,7,6,5,3,4,2,1 ] , [ 9,10,8,7,6,5,3,4,1,2],$ ] + @xmath612 , [ 8,9,10,5,6,7,3,4,2,1 ] , [ 8,9,10,6,7,5,3,4,2,1 ] , [ 10,9,7,8,4,5,6,1,2,3].$ ] \7 . there are 38 permutations attaining @xmath613 . + @xmath614 , [ 8,9,10,7,4,5,6,1,2,3 ] , [ 10,7,8,9,6,5,4,3,2,1 ] , [ 9,8,10,6,7,5,4,2,1,3],$ ] + @xmath615 , [ 10,8,9,5,7,4,6,2,3,1 ] , [ 8,10,9,6,4,7,5,1,3,2 ] , [ 10,9,8,7,6,4,5,3,2,1],$ ] + @xmath616 , [ 8,10,9,5,7,4,6,1,3,2 ] , [ 8,9,10,5,7,4,6,1,2,3 ] , [ 8,10,9,7,6,4,5,1,3,2],$ ] + @xmath617 , [ 10,8,9,7,4,5,6,2,3,1 ] , [ 10,8,9,6,4,7,5,2,3,1 ] , [ 9,8,10,5,6,7,4,2,1,3],$ ] + @xmath618 , [ 9,10,8,6,4,7,5,3,1,2 ] , [ 9,10,8,6,7,5,4,3,1,2 ] , [ 9,8,10,7,6,4,5,2,1,3],$ ] + @xmath619 , [ 10,9,8,5,7,4,6,3,2,1 ] , [ 10,9,8,7,6,5,2,3,4,1 ] , [ 9,8,10,6,4,7,5,2,1,3],$ ] + @xmath620 , [ 8,9,10,5,6,7,4,1,2,3 ] , [ 9,8,10,5,7,4,6,2,1,3 ] , [ 9,10,8,7,4,5,6,3,1,2],$ ] + @xmath621 , [ 9,10,8,5,6,7,4,3,1,2 ] , [ 9,10,8,7,6,4,5,3,1,2 ] , [ 8,10,9,6,7,5,4,1,3,2],$ ] + @xmath622 , [ 10,8,9,5,6,7,4,2,3,1 ] , [ 9,10,8,5,7,4,6,3,1,2 ] , [ 10,9,8,7,4,5,6,3,2,1],$ ] + @xmath623 , [ 10,9,8,6,7,5,4,3,2,1].$ ] the study of the problem in the paper began when c.k . li was visiting the hong kong university of science and technology supported by a fulbright fellowship in 2011 . the problem arose in a general education course `` mathematics in daily life '' conducted by c.k . he gracefully acknowledge the support of fulbright foundation , the hong kong university of science and technology , and some helpful discussion with colleagues and students at the hong kong university of science and technology . the research of huang and sze were supported by a hong kong rgc grant polyu 502411 . j. d@xmath33nes , the representation of a permutation as the product of a minimal number of transpositions , and its connection with the theory of graphs . magyar tud . akad . mat . kutat int . k@xmath624zl . 4 ( 1959 ) , 63 - 71 . ( z. huang ) department of applied mathematics , the hong kong polytechnic university , hung hom , hong kong . ( [email protected] ) + ( c.k . li ) department of mathematics , college of william and mary , williamsburg , virginia 23187 - 8795 , usa . ( [email protected] ) + ( s.h . li ) department of computer science , the johns hopkins university , baltimore , md 21218 , usa . ( [email protected] ) + ( n.s . sze ) department of applied mathematics , the hong kong polytechnic university , hung hom , hong kong . ( [email protected] )

we present newly acquired images of the near - infrared counterpart of 3cr radio sources . all the sources were selected to have a redshift of less than 0.3 to allow us to obtain the highest spatial resolution . the observations were carried out as a snapshot program using the near - infrared camera and multiobject spectrograph ( nicmos ) on - board the hubble space telescope ( hst ) . in this paper we describe 69 radio galaxies observed for the first time with nicmos during hst cycle 13 . all the objects presented here are elliptical galaxies . however , each of them has unique characteristics such as close companions , dust lanes , unresolved nuclei , arc - like features , globular clusters and jets clearly visible from the images or with basic galaxy subtraction .

astro-ph0603239

extragalactic radio sources are often associated with massive elliptical galaxies . these so called radio galaxies are one of the most extraordinary astrophysical phenomena , powered , it is generally believed , by supermassive black holes in the galaxy nuclei . radio galaxies are found in a variety of environments and across a wide range of redshifts . the revised third cambridge catalogue ( 3cr ) ( bennett 1962a , 1962b ) is the best studied sample of radio loud galaxies and quasars . spinrad et al . ( 1985 ) confirmed 298 extragalactic radio sources of the 3cr catalogue . this catalogue was made based on the radio properties of the sources , and therefore its selection criteria are mostly independent with respect to orientation and hst wavelengths . the 3cr catalogue has , thus , excellent attributes for a survey . here we describe all the observations taken with our snapshot program in the hst cycle 13 . we are using nicmos to obtain high spatial resolution , deep images in the near infrared @xmath0-band of 3cr sources at low redshift , @xmath1 . the primary goal of this project is to characterize the radio galaxy hosts free from the obscuring effects of dust , or at least , substantially reduced relative to the optical and uv . we also want to establish how radio galaxy hosts compare to the hosts of the most powerful high - z agn , to qso , and to quiescent ellipticals . the high linear resolution of hst grants us the possibility to observe details of the galaxies such as point - like nuclei , jets , and hot spots . these near - infrared observations are aimed at providing a zero redshift comparison sample for observations at high redshift . this nicmos snapshot program is a major enhancement to the dataset of hst observations of the 3cr sources . successful snapshot programs have been carried out during past cycles in the optical and in the ultraviolet by sparks and collaborators . the wfpc2 observations were presented by martel et al . ( 1999 ) for @xmath2 , by de koff et al . ( 1996 ) for @xmath3 , and by mccarthy et al . ( 1997 ) for @xmath4 . uv observations with stis were published by allen et al . these successive hst programs have generated a database with a sample completeness comparable , for some range of redshift , to the radio catalogue . this database provides an excellent foundation for statistical studies as a vast number of observations of these sources are available for comparison at other wavelengths and with other instruments . the content of this paper is as follows : section 2 describes the observation strategy , section 3 discusses the different steps of the data reduction , section 4 highlights results obtained from this survey and gives near - infrared photometry , section 5 presents three galaxies with jets visible with galaxy subtraction , in section 6 we make comments on the image of each individual galaxy . we use throughout this paper h@xmath5=71 km.s@xmath6mpc@xmath6 , @xmath7 , and @xmath8 . there is a total of 115 objects in the 3cr catalogue with @xmath1 , 18 of them have been observed with nicmos during previous cycles as part of other hst programs . we present 69 sources that have never been observed before , this corresponds to 71@xmath9 of the sample . these new observations are associated with program 10173 , pi : sparks . this nicmos survey uses snapshot exposures , a capability developed by the stsci to maximize the observing efficiency of hst . the observations obtained during this survey are taken at irregular intervals that fill the scheduling gaps between other accepted go programs . the remaining 28 sources were not scheduled for observation during cycle 13 . the observation log is presented in table 1 . since the observed targets are randomly chosen based on the constraints of the observing schedule of hst , the 69 galaxies we present here are not biased towards any particular characteristic . thus , they are suitable for statistical analysis . the radio properties of the observed sample are given in table 2 . we present the flux density and radio power at 178 mhz , the radio spectral index , largest angular size , position angle , and fanaroff and riley ( 1974 ) morphological classification . the sources are representative of large , steep spectrum , high power radio sources . the sources span three decades in radio power from log@xmath10 25 to 28 . the edge - darkened , lower luminosity fri sources make up 26% of the sample . all the observations presented here were carried out with nicmos camera 2 ( nic2 ) on multiaccum observing mode . we use the multiaccum sequence step32 , which consists of rapid reads up to 32 seconds followed by 32 second steps . nic2 has a field of view of @xmath11 and a projected pixel size of @xmath12 . the measured fwhm for a bright source is 0.14 . the field size allows us to detect both the host galaxy and nuclear regions at the redshifts we aim . the cooresponding physical scale over our redshift range varies from 0.335 kpc / arcsecond to 4.376 kpc / arcsecond . we use the f160w filter , the analog of the @xmath0 band . this filter is centered at 1.6037@xmath13 m , covering a wavelength range from 1.4@xmath13 m to 1.8@xmath13 m , and includes the paschen @xmath14 emission line for objects at @xmath2 . for further details the reader is referred to the nicmos instrument handbook ( noll et al . 2004 ) . all images have the same total exposure time of 1152 seconds , split into four exposures of 288s . we perform sub - pixel dithering to improve the psf sampling and remove bad pixels . we use a squared dither pattern , with 7.4 pixels offsets . we obtain the data from the multimission archive at space telescope ( mast ) . the data are processed by the standard `` on - the - fly '' reprocessing calibration pipeline . the pipeline separates the data into science and engineering data , creates fits files and populates its headers . the pipeline also calibrates the raw data using calnica ( noll et al . 2004 ) . we do the subsequent reduction of the data with the noao image processing software image reduction analysis facility ( iraf ) . the most important anomalies that we encounter in nic2 images are : the pedestal effect , the amplifier glow , the coronographic spot , cosmic rays and bad pixels . in this section we describe the steps we follow to get rid of these anomalies on the nicmos images using different iraf tasks . nic2 is divided into four quadrants of @xmath15 pixels each . on visual inspection the raw images appear to have one quadrant darker than the others . this is particularly apparent for images with low background , e.g. small galaxies . this is known as the pedestal effect and is part of the instrumental signature for nicmos . the origin of this effect is an additional offset introduced during the detector reset . the pedestal effect is stochastic and time variable , which makes it impossible to remove with the standard pipeline offered by the stsci ( noll et al . 2004 ) . we remove the pedestal effect with the tasks pedsky or pedsub from the nicmos package under the hst calibration of the space telescope science data analysis system ( stsdas ) . these two tasks determine the pedestal offset of each quadrant with a similar technique . both pedsky and pedsub assume that the image pixel value is the sum of the astronomical signal ( source + sky ) , and the pedestal effect modulated by the flatfield . these tasks use the flatfield as input and loop over a range of trial values for the pedestal and find the best fit to the science image , for greater details see bushouse et al . ( 2000 ) . pedsky does the fit of the pedestal only with those pixels near the background level . this task dismisses pixels containing signal from a source during the fitting process . pedsky removes the pedestal effect and also subtracts the sky background . we use pedsky for galaxies that cover only a small fraction of the detector . this task is more effective when a large area of the image is free from sources in order to allow a better estimate of the background . for sources covering a large portion of the detector we use the task pedsub to remove the pedestal effect . pedsub has a series of filters to remove unwanted features such as large galaxies . contrary to pedsky , pedsub does not remove the background . in the case where the background is not removed , we use msky2 , a task written by mark dickinson ( private communication ) , that can fit the sky interactively , and subtract the sky mode from each image to obtain a zero level background . given the small projected size of the detector , a few galaxies cover the whole chip , for these images we adopt the median background of the other frames as the sky level . this is determined by measuring the sky level on 152 single exposures of the dither pattern . in these 152 exposures the source covers only a small fraction of the field of view . we find a median background of 0.046 counts per pixel with a standard deviation of 0.01806 . for the galaxies covering most of the detector , the subtracted sky level amounts , on average , to 1.3% of the flux inside an aperture of 1kpc radius . some images have visible noise in the corners due to the amplifier glow of nic2 . this glow is caused by the readout amplifier situated close to each corner of the detector . each time the detector is read out the amplifier warms up and emits infrared radiation that is detected by the chip ( noll et al . this anomaly does not affect the overall appearance of the image . in order to obtain images of a better quality we create individual masks for each object to cover the small blemishes of the detector such as residual cosmic rays and bad pixels . furthermore , nic2 has a hole through the camera to allow coronographic observations , which creates a circular spot on the images of about ten pixels in diameter . we treat this area with a mask to conceal the coronographic spot . we also create cosmic rays masks , with the task drizzle@xmath16cr , that we incorporate into the individual masks for each object . the pixels of column 128 in nic2 contains the first pixels read out in each quadrant and have an incorrect bias subtracted from them . this is called the `` photometrically challenged '' column . we applied a bad pixel mask to the pixels of this column . we recover , thus , information on the image . the second extension of the nicmos fits files contains the error image . we convert these error files into weightings which are the inverse square of the noise per pixel . the mask files are multiplied into the weight files described above , giving zero weight to bad pixels in the individual files but leaving the others unchanged . we combine the four calibrated , background subtracted images of the dither pattern with the task drizzle under the package dither ( fruchter & hook 2002 ) . the final weight files are called when dithering the four exposures of each object . the projected pixel size in the final image is @xmath17 . drizzle is also used to apply the necessary rotations to the images to obtain a north up , east left orientation . these final images are presented in figures 1 - 77 . the high spatial resolution provided by hst allows us to obtain images of the 3cr radio galaxies with unprecedented details . all galaxies imaged during this program are ellipticals but each of them has its own peculiar characteristics and environment . in our sample of 69 galaxies 36 of them ( 52@xmath9 ) have companions present in the field of view of nic2 . more than one third of this sample , i.e. 25 galaxies , have an unresolved nucleus and nine galaxies have dust lanes . nineteen galaxies , primarily the closest , show off nuclear point - like sources likely to be globular clusters . four galaxies have intriguing arc - like structures , which are perhaps lensed arcs or merger remnants . we use the iraf task radprof to measure the flux of the galaxies at five different radii . the radii of the apertures we use correspond to : 1 kpc , 5 kpc , 10 kpc , 15 kpc , 20 kpc . given the small size of the detector we can not always measure the flux at all five different radii . the count rate ( cr ) is converted into flux using photflam , the inverse sensitivity , in the header of the calibrated nicmos images : photflam=1.74779.10@xmath18 erg @xmath19 dn@xmath6 we determine the magnitudes in the space telescope ( st ) system with the expression : m@xmath20 = - 2.5log(photflam @xmath21 cr ) + photzpt where photzpt is the st magnitude zero point i.e. -21.1 , see the hst data handbook for nicmos ( dickinson et al . 2002 ) . the limiting magnitude for the images presented here is about 25.3 . we present the results of the near - infrared photometry in table 3 . note that these magnitudes correspond to the total observed flux inside the radii set above . we did not perform photometry for the nuclei , which will be analyzed in detail in a forthcoming paper ( chiaberge et al . in preparation ) we have used the ellipse task in iraf stsdas.analysis.isophote , based on the algorithm of jedrzejewski ( 1987 ) , to fit elliptical isophotes to the nicmos images . each isophote is fitted at a pre - defined , fixed semi - major axis length . starting from a first guess , the image is sampled along an elliptical path ( defined by central coordinates @xmath22 , position angle @xmath23 , and ellipticity @xmath24 ) to produce a 1-dimensional intensity distribution as a function of @xmath23 . modeling with ellipse allows us to track radial changes in the ellipticity and position angle of the isophotes of galaxies . models based on the best - fitting ellipse parameters are constructed in iraf using the task bmodel . a complete sample of model - subtracted residual images of this survey will be presented by floyd et al . ( in preparation ) . here we show two examples where jets are clearly visible with galaxy subtraction . this is the case for 3c66b , 3c133 figures 10,23 . figure 62 shows the residual image for 3c401 , its jet is also visible in the nicmos image ( chiaberge at al . we provide here a short description of the most prominent features of the nicmos images for each object . 3c20 : : the near - infrared image reveals a circular galaxy . one source is visible approx . @xmath25 to the northwest of the nucleus . a bright unresolved source is present @xmath26 to the southeast . this galaxy has a known radio hotspot detected by de koff et al . ( 1996 ) with wfpc2 . this hotspot falls outside the smaller field of view of nic2 . 3c28 : : 3c28 is an elliptical elongated on the northwest to southeast direction . at least two other sources are present in the field of view : one is on the edge of the chip to the northwest and the second one is @xmath27 to the southwest . this galaxy is the cd of the x - ray cluster abell 115 ( mccarthy et al . 3c29 : : a round , undisturbed elliptical with a bright central unresolved component ( nucleus ) . several globular clusters are present in the field of view . this galaxy has a faint compact nucleus in the optical and in the uv . the uv nucleus was first detected by allen et al . 3c31 : : this galaxy covers most of the area of the detector . a face on dust ring around the galaxy s nucleus can be seen on the infrared image . the wfpc2 image shows that this dust ring is composed of several dust strands with a larger absorption on the southwestern side ( martel et al . ( 1999 ) and de koff et al . ( 2000 ) ) . 3c33.1 : : elliptical galaxy with no obvious disturbance with an unresolved nucleus and a single close companion about @xmath28 to the south - southwest . a bright unresolved source lies @xmath29 to the southwest , three other faint unresolved sources are also visible as well as another double system @xmath29 to the northwest . 3c35 : : this flattened galaxy is elongated northwest to southeast . the wfpc2 image of this galaxy shows an unremarkable elliptical ( martel et al . 1999 ) while the stis image exhibits a faint compact nucleus ( allen at al . 3c52 : : the near - infrared image reveals an elongated galaxy in a crowded field . one unresolved source is present @xmath30 to the northwest and a second unresolved source is located @xmath31 to the north - northeast of the center of the galaxy . the near - infrared image does not show the dust lane , perpendicular to the radio axis , visible in the optical ( de koff et al . ( 1996 , 2000 ) ) . 3c61.1 : : the galaxy lies in the middle of the field of view which shows three prominent resolved companions . 3c61.1 is elliptical and compact . we do not see the tails of emission to the south and east detected by de koff et al . ( 1996 ) with wfpc2 . one unresolved source is present to the north of the galaxy . 3c66b : : this nicmos image shows a round galaxy with a bright unresolved nucleus , several globular clusters and two unresolved sources to the south . we notice a small asymmetry in the shape of this galaxy with its northwest side more sharply defined than the southeast one . zirbel & baum ( 1998 ) claim that this slight asymmetry may be caused by a neighbor to the southeast . the jet of 3c66b is clearly visible in this nicmos image after basic galaxy subtraction as shown in section 5 . 3c75n : : 3c75n is slightly elongated on the east - southeast to the west - northwest direction . this is the northern component of a system of two elliptical galaxies , the edge of the southern companion is visible at the bottom of the image . some faint globular clusters are also seen . 3c76.1 : : 3c76.1 is an elliptical with a very elongated galaxy to the north , it is unclear from this image if they are interacting or not . 3c79 : : a galaxy with elliptical morphology , elongated north - south . the image shows two companions : one bright unresolved source less than one arcsecond south of the nucleus , and the other one to the north . 3c79 shows a complex morphology in the optical , with no compact nucleus ( de koff et al . 1996 ) 3c83.1 : : the near - infrared image shows an elliptical elongated north to south and covering most of the detector area . the image reveals a thin dust lane wrapping around the nucleus . a bright nearby star generates diffraction spikes crossing the detector . some faint globular clusters are also visible . 3c88 : : elliptical elongated northwest to southeast , the radio - axis is perpendicular to the elongation of the galaxy . globular clusters are visible on the image and there is no sign of disturbance or merger . 3c105 : : highly flattened elliptical , faintly disturbed . a faint unresolved source is visible 4@xmath32 to the west - southwest off the nucleus . 3c111 : : this galaxy has a very bright nucleus with the host clearly visible . 3c111 is an elliptical elongated north to south . 3c123 : : this elliptical galaxy is a member of a cluster . at least three diffuse companions and two point - like sources are present in the field of view . the optical image of de koff et al . ( 1996 ) shows also a faint diffuse source . 3c129 : : an elongated arc - shaped source is clearly visible @xmath33 east of the nucleus of the galaxy . globular clusters in this galaxy are also detected . several unresolved foreground stars are present in the field of view . 3c129.1 : : this very round elliptical galaxy is found in a region of high foreground stellar density . it is the cd of the cluster 4u0446 + 44 . globular clusters are also possibly present in the image . 3c130 : : this radiogalaxy is a round elliptical without peculiarities . it covers a large portion of the detector and lies in a region of high foreground stellar density . thus a large number of point sources are visible . some of them may be globular clusters . 3c133 : : we have discovered a new optical - ir jet in 3c133 , see floyd et al . ( 2006 ) for further discussion . the jet and eastern hotspot are well resolved , and visible at both optical and ir wavelengths , in spite of the low galactic latitude . the infrared jet follows the morphology of the inner part of the radio jet , with three distinct , aligned , bright knots east of the unresolved nucleus . the host is a round elliptical . 3c135 : : this frii radio galaxy is a cluster member ( mccarthy et al . 1995 ) with a close companion to the southwest . 3c165 : : the near infrared image reveals an elliptical elongated north to south in a crowded field . a source is present approx @xmath34 to the northeast . 3c165 is a cluster member . 3c171 : : this galaxy of slightly disturbed elliptical morphology has particularly strong extended emission line regions . these emission line regions are aligned with the radio emission ( heckman et al . 1984 , blundell 1996 ) . the nicmos image shows faint tails , most likely pa@xmath14 emission , to the east and west overlapping the optical emission line regions , see zoom image . 3c173.1 : : slightly boxy elliptical with faint companion @xmath35 south . an asymmetric , low surface brightness component , is visible to the northwest , perhaps evidence of recent merger activity . 3c173.1 is a member of a group . 3c180 : : our nicmos image shows a boxy giant elliptical . 3c180 is a member of a cluster according to mccarthy et al . five unresolved sources are present in the field of view . 3c184.1 : : this galaxy is a cluster member with several faint companions . it is an elongated elliptical with noticeable disturbed isophotes . an arc - shaped structure is present @xmath36 northeast of the unresolved nucleus . 3c192 : : this object appears as a round , undisturbed elliptical galaxy in the near - infrared , while it shows a compact core with diffuse emission structures in the uv ( allen et al . several globular clusters are also detected . 3c196.1 : : the near - infrared image shows an elliptical galaxy elongated northeast to southwest , which is the same direction of the radio emission . the same morphology is seen in the optical ( de koff et al . 1996 and baum et al . five other sources populate the field of view . 3c197.1 : : this galaxy has a round , slightly disturbed morphology and an unresolved nucleus . 3c198 : : the nicmos image shows an elongated elliptical galaxy with a bright unresolved nucleus . the circumnuclear compact sources seen in the uv image ( allen et al . 2002 ) do not appear in the near - infrared . 3c213.1 : : this radio galaxy has an extended area of emission to the southeast . the near - infrared counterpart of the northern radio hot spot is clearly visible . the southern hot spot is also visible but is much dimmer . the two hotspots are also present on the wfpc2 image ( de koff et al . 1996 , de vries et al . 3c219 : : a member of a cluster of galaxies , this source shows five different companions of different types and a bright unresolved nucleus . as noted by de koff et al . ( 1996 ) 3c219 appears to be interacting with the large companion to the southeast , however , mclure et al . ( 1999 ) modeled the wfpc2 image and found no evidence of such interaction . 3c223 : : our near infrared image shows an asymmetric galaxy , slightly elongated northeast to southwest . 3c223.1 : : the image shows a galaxy with a very elongated elliptical shape and a thin dust disk to the southeast , see zoom image with different intensity scale . the dust disk is best seen in the optical ( de koff et al . a tail of emission is present to the northeast . 3c227 : : this is a broad line radio galaxy . we can clearly see the host galaxy and its bright nucleus on this near - infrared image . both the optical and uv images show a very bright unresolved nucleus ( martel et al . 1999 , allen et al . 2000 ) . a close companion is also visible 5@xmath32 southeast off the nucleus . 3c236 : : elliptical elongated from the northeast to the southwest . a dust lane is notable on the wfpc2 images and was detected by martel et al . the absorption map of de koff et al . ( 2000 ) clearly shows this dust disk too . odea at al . ( 2001 ) found four very blue regions of star formation on the edge of the dust lane . 3c277.3 : : the host galaxy of coma a is a round elliptical . a source elongated northeast to southwest is present approximately @xmath37 to the southeast . this source is the radio knot k1 as defined by miley et al . ( 1981 ) and van breugel et al . the knot is also visible in the optical ( capetti et al . 2000 ) . an unresolved source , likely to be associated to knot k2 , is visible @xmath38 southwest of k1 . 3c285 : : irregular clumps of emission linked to star formation are detected in the ultraviolet image of this galaxy ( allen et al . the near - infrared image shows a flattened galaxy elongated southeast to northwest . the dust lane perpendicular to the elongation of the galaxy is visible in both the wfpc2 and nicmos images . a faint arc - like source is present approx . @xmath39 to the southeast , see zoom image . 3c287.1 : : this broad line radio galaxy has a bright unresolved nucleus with visible diffraction spikes , the host is an elongated elliptical . a companion is present @xmath39 to the northeast . 3c288 : : an e0 elliptical galaxy with round isophotes , a faint unresolved nucleus , and two compact nearby companions . one more extended companion is visible to the west . two fainter companions are also present , one to the northwest and the other one to the southwest . 3c303 : : this galaxy is characterized by a strong x - ray emission ( hardcastle & worrall 1999 ) and has a bright unresolved nucleus in the near - infrared image . a companion is visible @xmath40 northeast of the nucleus . 3c310 : : as noted by martel et al . ( 1999 ) for the optical image , this galaxy is flattened east - west , on an almost perpendicular direction to the radio jet axis . the edge of a companion to the east is visible on the corner of the chip . there is a bright foreground star to the south of the galaxy . globular clusters and a faint diffuse companion @xmath34 northwest of the nucleus are also seen . 3c314.1 : : the nicmos image shows an elliptical galaxy elongated east - northeast to west - southwest . the isophotes are elongated in the near - infrared as well as in the optical ( de koff et al . 3c315 : : this is a very elongated galaxy associated to an `` x''-shaped radio source . a faint arc like feature is visible at the southern end of the galaxy , see zoom image . 3c315 has a very round large companion galaxy to the south . a small faint source is visible to the southwest of the chip . 3c319 : : the host galaxy associated to 3c319 is close to the southwest edge of the image . it appears as a slightly elongated elliptical . three companions are also visible in the image . 3c321 : : this galaxy shows a clearly visible dust lane on our near - infrared image . a companion in the process of merging ( roche & eales , 2000 ) is located @xmath41 to the northwest and is elongated towards the central source along the radio axis . bright knots of emission are present along the northern border of the dust lane on the uv image ( allen et al . 3c346 : : the northwestern galaxy of the double system corresponds with the radio core and has a very bright nucleus . the radio jet of this galaxy bends at a very bright knot , unresolved in the nicmos image . this knot is located @xmath42 to the east of the nucleus . the jet and its knots are visible in this near - infrared image and in the optical as well ( de koff et al . 1996 ) . the main companion galaxy is highly asymmetric and possibly merging with 3c346 . a galaxy , likely to be an edge - on spiral , is visible to the southwest of the image . 3c348 : : this source , also known as hercules a , is a cluster cd galaxy . 3c348 is a double - galaxy system in the process of merging ( sadun & hayes , 1993 ) . the radio core coincides with the southeastern galaxy . this galaxy has faint dust rings around the core in this near - infrared image . these rings are discussed in detail by baum et al . ( 1996 ) based on wfpc2 observations . 3c349 : : the nicmos image shows a boxy , elongated , slightly asymmetric elliptical galaxy with an unresolved nucleus . 3c353 : : a round giant elliptical with no signs of disturbance . martel et al . ( 1999 ) find very circular isophotes ( @xmath43 ) on the optical image . a few small sources , most likely globular clusters , are visible around this galaxy . 3c371 : : this object is often classified as a bl lac in the literature . in our image it appears as a round and smooth galaxy with a very bright nucleus that causes marked diffraction spikes . 3c379.1 : : slightly elongated elliptical galaxy . we notice a curious arc - like feature @xmath44 to the west . this arc like feature might be a lensed arc or a merger remnant , see zoom image . 3c381 : : smooth elliptical with a close projected companion galaxy to the east . roche & eales ( 2000 ) , with a u - band image taken with the wide field camera on the isaac newton telescope , find an apparent tidal distortion that confirms that the two galaxies are interacting . there is an additional faint unresolved source @xmath36 south . 3c382 : : this galaxy has a very bright unresolved nucleus with marked diffraction spikes . this galaxy lies on a field populated by unresolved sources . 3c386 : : this elliptical galaxy fills a large portion of the detector . new spectroscopic evidence shows that what appears to be a bright nucleus is a star superimposed on the center of the galaxy ( marchesini et al . private communication ; chiaberge et al . several point sources are discernible on the field of view , some of them are possibly globular clusters . 3c388 : : this is the cd galaxy of an abell class 2 cluster ( prestage & peacock 1988 ) with a close bright and round companion to the southwest . two possible nearby companions are visible within the halo of the galaxy , about 1 north of the nucleus . these two possible companions were also detected by martel et al . ( 1999 ) in their wfpc2 image . in the near - infrared , they appear unresolved . 3c390.3 : : the host galaxy of this blrg is clearly visible in our nicmos image . a faint arc - like component is visible @xmath45 southeast of the nucleus . 3c401 : : this elliptical shows an unresolved nucleus . the image shows a faint jet about 6@xmath32 southwest of the nucleus identified by chiaberge et al . ( 2005 ) as the infrared counterpart of the brightest region of the radio jet . chiaberge et al . find that the infrared emission dominates the spectral energy distribution of this jet . an elliptical companion is visible @xmath28 to the north . a bright unresolved source is present @xmath46 north of the core . this object lies on the radio jet axis but is not associated with any of the radio features . in the optical image of de koff et al . ( 1996 ) it appears as a diffuse source . it is possible that this object is an infrared - bright background galaxy . 3c402 : : a large , elongated , smooth elliptical with several point like sources which might be globular clusters . 3c403 : : elliptical galaxy with a companion to the southeast . this galaxy appears to be in a region of high foreground stellar density . 3c430 : : a dust lane around the nucleus of this elongated elliptical is weakly visible on this image . several foreground stars are present in the field of view . some of this compact sources may be globular clusters . 3c433 : : this cluster member ( mccarthy , 1995 ) has a peculiar radio jet oriented north - south . we see an elongated elliptical galaxy with a bright nucleus . the nucleus is totally absent in the optical band ( chiaberge et al . 1999 ) . a patch of emission detected by de koff et al . ( 1996 ) in the optical is also present to the northwest of the nucleus on our nicmos image . two companion galaxies lie to the north - northeast . a bright foreground star is visible on the northern corner of the image . 3c436 : : the galaxy has an elongated shape , somewhat aligned with the radio image . an elongated companion is present @xmath47 to the east - northeast . 3c438 : : this cluster member has at least three companions , the brightest of which lies approx . @xmath28 to the northeast . this galaxy appears to be in a region of high foreground stellar density given the presence of eight unresolved sources on the hst image . 3c445 : : this blrg appears as a round elliptical with a very bright unresolved nucleus and clearly visible diffraction spikes . some faint globular clusters are also visible . 3c449 : : the nicmos image of this galaxy shows the dust lane detected with wfpc2 by martel et al . ( 1999 ) although with less dramatic features , see zoom image . the dust lane has been modeled as a warped disk by tremblay et al . we also detect the abundant globular clusters seen in the optical . the galaxy is elongated north - south and occupies a large portion of the nic2 chip . 3c452 : : an elliptical elongated east to west showing no signs of disturbance . a faint compact source is visible @xmath48 southwest of the nucleus . 3c465 : : the nicmos image shows an elongated galaxy with its close companion visible to the north . globular clusters surround the galaxy . this galaxy , the brightest galaxy in the cluster abell 2634 , shows a faint unresolved nucleus from the near - infrared to the uv ( allen et al . 2002 , chiaberge et al . 1999 ) and a visible dust lane in the optical ( martel et al . we have presented 69 hst nicmos images of the near - infrared counterparts of radio sources of the 3cr catalogue at low redshift . these nicmos images allow us to charaterize the host galaxies and their environment . we also detect special features such as jets , dust disks , nuclei , and hot spots . this database of high - quality near - infrared images presented in this paper provide an excellent foundation for future statistical studies of radio galaxies and galaxy evolution . this research has made use of the nasa astrophysics data system bibliographic services . this research also made use 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 . we are grateful to eddie bergeron and santiago arribas for helping us with the nicmos data reduction . allen , m. g. , et al . 2002 , apjs , 139 , 411 baum , s. a. , heckman , t. , bridle , a. , van breugel , w. , & miley , g. 1988 , apjs , 68 , 643 baum , s. a. , odea , c. p. , de koff , s. , sparks , w. , hayes , j. j. e. , livio , m. , & golombek , d. 1996 , apj,465 , l5 bennett , a. s. 1962a , mem . ras , 68 , 163 bennett , a.s . 1962b , mnras , 125 , 75 blundell , k. m. 1996 , mnras , 283 , 538 bushouse , h. , dickinson , m. , van der marel , r. p. 2000 , astronomical data analysis software and systems ix , asp conference series , vol . 216 p. 531 capetti , a. , de ruiter , h.r . , fanti , r. , morganti , r. , parma , p. , & ulrich , m .- h . 2000 , a&a , 362 , 871 chiaberge , m. , capetti , a. , celotti , a. 1999 , a&a , 349 , 77 chiaberge , m. , capetti , a. , celotti , a. 2002 , a&a , 394 , 791 chiaberge , m. , et al . 2005 , apj , 629 , 100 de koff , s. , baum , s. a. , sparks , w. b. , biretta , j. , golombek , d. , macchetto , f. , mccarthy . , p. , & miley , g. 1996 , apjs , 107 , 621 de koff , s. , et al . 2000 , apjs , 129 , 33 de vries , w.h . , et al . 1997 , apjs , 110 , 191 dickinson , m.e . , et al . 2002 , hst nicmos data handbook v5.0 , ed . b. mobasher , baltimore , stsci fanaroff , b. l. & riley , j. m. 1974 , mnras , 167 , 31 floyd , d. j. e. , et al . 2006 , astro - ph/0602021 fruchter , a.s . & hook , r. n. 2002 , pasp , 114 , 144 hardcastle , m.j . & worrall , d.m . 1999 , mnras , 309 , 969 hardcastle , m.j . 2003 , mnras , 339 , 360 heckman , t. m. , van breugel , w. j. m. , miley , g. k. 1984 , apj , 286 , 509 jedrzejewski , r. 1987 , mnras , 226 , 747 martel , a. , et al . 1999 , apjs , 122 , 81 mccarthy , p. j. , spinrad , h. , van breugel , w. 1995 , apjs , 99 , 27 mccarthy , p. j. , miley , g. k. , de koff , s. , baum , s. a. , sparks , w. b. , golombek , d. , biretta , j. , & macchetto , f. 1997 , apjs , 112 , 415 mclure , r.j . , et al . 1999 , mnras , 308 , 377 miley , g. k. , heckman , t. m. , butcher , h. r , van breugel , w. 1981 , apj , 247 , l5 noll , k. , et al . 2004 , nicmos instrument handbook " , version 7.0 , ( baltimore : stsci ) odea , c.p . , et al . 2001 , aj , 121,1915 prestage , r.m . & peacock , j.a . 1988 , mnras , 230 , 131 roche , n. , & eales , s.a . 2000 , mnras , 317 , 120 sadun , a. c. , & hayes , j. j. e. 1993 , pasp , 105 , 379 spinrad , h. , djorgovski , s. , marr , j. , & aguilar , l. 1985 , pasp , 97 , 932 tremblay , g. r. et al . 2005 , astro - ph/0510650 van breugel , w. , miley , g. , heckman , t. , butcher , h. , bridle , a. 1985 , apj , 290 , 496 zirbel , e. l. & baum , s.a . 1998 , apjs , 114 , 177 3c20 & 2005 feb 27 & 00 43 09.27 & + 52 03 36.66 & -10.79 + 3c28 & 2005 jun 13 & 00 55 50.65 & + 26 24 36.93 & -36.45 + 3c29 & 2004 dec 4 & 00 57 34.88 & -01 23 27.55 & -64.22 + 3c31 & 2005 jun 17 & 01 07 24.99 & + 32 24 45.02 & -30.34 + 3c33.1 & 2004 aug 15 & 01 09 44.27 & + 73 11 57.2 & + 10.38 + 3c35 & 2005 mar 16 & 01 12 02.29 & + 49 28 35.33 & -13.26 + 3c52 & 2005 mar 11 & 01 48 28.90 & + 53 32 27.9 & -8.38 + 3c61.1 & 2004 aug 9 & 02 22 36.00 & + 86 19 08.0 & + 23.73 + 3c66b & 2004 nov 5 & 02 23 11.46 & + 42 59 31.34 & -16.77 + 3c75n & 2004 nov 11 & 02 57 41.55 & + 06 01 36.58 & -44.93 + 3c76.1 & 2005 feb 6 & 03 03 15.0 & + 16 26 19.85 & -35.96 + 3c79 & 2004 oct 30 & 03 10 00.1 & + 17 05 58.91 & -34.46 + 3c83.1 & 2005 mar 12 & 03 18 15.8 & + 41 51 28.0 & -13.13 + 3c88 & 2004 nov 6 & 03 27 54.17 & + 02 33 41.82 & -42.02 + 3c105 & 2004 oct 26 & 04 07 16.46 & + 03 42 25.68 & -33.62 + 3c111 & 2004 dec 8 & 04 18 21.05 & + 38 01 35.77 & -8.82 + 3c123 & 2004 dec 7 & 04 37 04.4 & + 29 40 13.2 & -11.66 + 3c129 & 2004 dec 8 & 04 49 09.07 & + 45 00 39.0 & + 0.14 + 3c129.1 & 2004 nov 22 & 04 50 06.7 & + 45 03 06.0 & + 0.30 + 3c130 & 2005 apr 19 & 04 52 52.78 & + 52 04 47.53 & + 5.12 + 3c133 & 2004 dec 13 & 05 02 58.4 & + 25 16 28.0 & -9.91 + 3c135 & 2005 apr 8 & 05 14 08.3 & + 00 56 32.0 & -21.04 + 3c165 & 2005 apr 26 & 06 43 06.6 & + 23 19 03.0 & + 8.67 + 3c171 & 2004 nov 14 & 06 55 14.72 & + 54 08 58.27 & + 22.23 + 3c173.1 & 2004 nov 22 & 07 09 24.34 & + 74 49 15.19 & + 27.27 + 3c180 & 2005 feb 20 & 07 27 04.77 & -02 04 30.97 & + 06.96 + 3c184.1 & 2004 nov 26 & 07 43 01.28 & + 80 26 26.3 & + 28.86 + 3c192 & 2005 jan 8 & 08 05 35.0 & + 24 09 50.0 & + 26.40 + 3c196.1 & 2005 feb 1 & 08 15 27.73 & -03 08 26.99 & + 17.07 + 3c197.1 & 2005 apr 19 & 08 21 33.7 & + 47 02 37.0 & + 34.48 + 3c198 & 2005 may 3 & 08 22 31.9 & + 05 57 7.0 & + 22.95 + 3c213.1 & 2005 feb 12 & 09 01 05.3 & + 29 01 46.0 & + 39.67 + 3c219 & 2004 sep 14 & 09 21 8.64 & + 45 38 56.49 & + 44.76 + 3c223 & 2005 feb 10 & 09 39 52.76 & + 35 53 59.12 & + 48.66 + 3c223.1 & 2005 jan 18 & 09 41 24.04 & + 39 44 42.39 & + 48.92 + 3c227 & 2005 mar 28 & 09 47 45.14 & + 07 25 20.33 & + 42.29 + 3c236 & 2004 nov 2 & 10 06 01.7 & + 34 54 10.0 & + 53.98 + 3c277.3 & 2005 mar 24 & 12 54 12.06 & + 27 37 32.66 & + 89.21 + 3c285 & 2004 dec 5 & 13 21 17.8 & + 42 35 15.0 & + 73.39 + 3c287.1 & 2005 jul 16 & 13 32 53.27 & + 02 00 44.73 & + 62.99 + 3c288 & 2004 oct 31 & 13 38 50.0 & + 38 51 10.7 & + 74.66 + 3c303 & 2004 dec 26 & 14 43 02.74 & + 52 01 37.5 & + 57.50 + 3c310 & 2004 aug 13 & 15 04 57.18 & + 26 00 56.87 & + 60.21 + 3c314.1 & 2005 feb 24 & 15 10 23.12 & + 70 45 53.4 & + 42.18 + 3c315 & 2004 dec 30 & 15 13 40.0 & + 26 07 27.0 & + 58.30 + 3c319 & 2004 dec 29 & 15 24 05.5 & + 54 28 14.6 & + 51.05 + 3c321 & 2004 dec 27 & 15 31 43.4 & + 24 04 19.0 & + 53.88 + 3c346 & 2005 may 19 & 16 43 48.69 & + 17 15 48.09 & + 35.77 + 3c348 & 2005 may 9 & 16 51 08.16 & + 04 59 33.84 & + 28.95 + 3c349 & 2005 mar 23 & 16 59 28.84 & + 47 02 56.8 & + 38.20 + 3c353 & 2004 sep 9 & 17 20 28.16 & -00 58 47.06 & + 19.65 + 3c371 & 2005 jan 29 & 18 06 50.6 & + 69 49 28.0 & + 29.17 + 3c379.1 & 2004 nov 5 & 18 24 32.53 & + 74 20 58.64 & + 27.85 + 3c381 & 2004 nov 11 & 18 33 46.29 & + 47 27 02.9 & + 22.48 + 3c382 & 2005 jun 22 & 18 35 03.45 & + 32 41 46.18 & + 17.45 + 3c386 & 2005 jun 15 & 18 38 26.27 & + 17 11 49.57 & + 10.55 + 3c388 & 2004 oct 19 & 18 44 02.4 & + 45 33 30.0 & + 20.22 + 3c390.3 & 2004 sep 17 & 18 42 09.0 & + 79 46 17.0 & + 27.07 + 3c401 & 2004 aug 11 & 19 40 25.14 & + 60 41 36.85 & + 17.77 + 3c402 & 2004 dec 10 & 19 41 46.0 & + 50 35 44.9 & + 13.27 + 3c403 & 2004 nov 6 & 19 52 15.81 & + 02 30 24.4 & -12.31 + 3c430 & 2005 jan 27 & 21 18 19.15 & + 60 48 06.88 & + 7.96 + 3c433 & 2004 aug 18 & 21 23 44.6 & + 25 04 28.5 & -17.69 + 3c436 & 2004 nov 9 & 21 44 11.74 & + 28 10 18.67 & -18.77 + 3c438 & 2004 nov 18 & 21 55 52.3 & + 38 00 30.0 & -12.98 + 3c445 & 2005 jun 24 & 22 23 49.57 & -02 06 13.08 & -46.71 + 3c449 & 2004 nov 11 & 22 31 20.63 & + 39 21 30.07 & -15.92 + 3c452 & 2004 nov 28 & 22 45 48.9 & + 39 41 14.47 & -17.06 + 3c465 & 2004 sep 28 & 23 38 29.41 & + 27 01 53.03 & -33.07 + 3c20 & 0.174 & 42.9 & 27.43 & 0.67 & 51 & 101 & ii + 3c28 & 0.1952 & 16.3 & 27.11 & 1.06 & 30 & 166 & ii + 3c29 & 0.04481 & 15.1 & 25.8 & 0.50 & 139 & 160 & i + 3c31 & 0.0167 & 16.8 & 25.0 & 0.57 & 1833 & 159 & i + 3c33.1 & 0.1809 & 13.0 & 26.95 & 0.62 & 216 & 45 & ii + 3c35 & 0.0670 & 10.5 & 26.0 & 0.77 & 704 & 12 & ii + 3c52 & 0.2854 & 13.5 & 27.37 & 0.62 & 51 & 20 & ii + 3c61.1 & 0.184 & 31.2 & 27.35 & 0.77 & 186 & 2 & ii + 3c66b & 0.0215 & 24.6 & 25.4 & 0.62 & 330 & 54 & i + 3c75n & 0.02315 & 25.8 & 25.5 & 0.71 & 692 & 111 & i + 3c76.1 & 0.0324 & 13.3 & 24.33 & 0.77 & 200 & & i + 3c79 & 0.25595 & 30.5 & 27.63 & 0.92 & 86 & 105 & ii + 3c83.1 & 0.0255 & 26.0 & 25.6 & 0.64 & 680 & 96 & i + 3c88 & 0.03022 & 15.3 & 25.5 & 0.52 & 259 & 60 & ii + 3c105 & 0.089 & 17.8 & 25.74 & 0.58 & & 335 & ii + 3c111 & 0.0485 & 64.6 & 26.5 & 0.73 & 220 & 62 & ii + 3c123 & 0.2177 & 189.0 & 28.28 & 0.70 & 23 & 115 & ii + 3c129 & 0.0208 & 46.9 & 25.66 & 0.92 & & & i + 3c129.1 & 0.0222 & 10.5 & 25.07 & 0.89 & & & i + 3c130 & 0.1090 & 15.5 & 26.67 & 0.89 & & & i + 3c133 & 0.2775 & 22.3 & 27.57 & 0.70 & 12 & 107 & ii + 3c135 & 0.1253 & 17.3 & 26.75 & 0.92 & 130 & & ii + 3c165 & 0.2957 & 13.5 & 27.41 & 0.71 & & 159 & + 3c171 & 0.2384 & 19.5 & 27.37 & 0.87 & 30 & 100 & ii + 3c173.1 & 0.2921 & 15.4 & 27.45 & 0.88 & 58 & 17 & ii + 3c180 & 0.22 & 15.1 & 27.19 & 0.84 & & 7 & + 3c184.1 & 0.1182 & 13.0 & 26.56 & 0.68 & 167 & 157 & ii + 3c192 & 0.0598 & 23.0 & 25.10 & 0.79 & 200 & & ii + 3c196.1 & 0.198 & 18.6 & 27.18 & 1.16 & 4 & 43 & ii + 3c197.1 & 0.1301 & 8.1 & 26.44 & 0.69 & 14 & 2 & ii + 3c198 & 0.0815 & 9.7 & 26.1 & 0.69 & 338 & 38 & ii + 3c213.1 & 0.194 & 6.6 & 26.71 & 0.55 & 43 & 162 & ii + 3c219 & 0.1744 & 41.2 & 27.41 & 0.81 & 184 & 40 & ii + 3c223 & 0.1368 & 14.7 & 26.75 & 0.74 & 300 & 164 & ii + 3c223.1 & 0.107 & 6.0 & 26.14 & 0.56 & 117 & 15 & ii + 3c227 & 0.0861 & 30.4 & 26.7 & 0.67 & 246 & 86 & ii + 3c236 & 0.10050 & 14.4 & 26.5 & 0.51 & 2514 & 122 & ii + 3c277.3 & 0.0857 & 9.0 & 26.1 & 0.58 & 50 & 158 & ii + 3c285 & 0.0794 & 11.3 & 25.46 & 0.95 & 184 & & ii + 3c287.1 & 0.2159 & 8.2 & 26.9 & 0.52 & 112 & 91 & ii + 3c288 & 0.246 & 18.9 & 27.39 & 0.85 & 16 & 146 & i + 3c303 & 0.141 & 11.2 & 26.66 & 0.76 & 38 & 97 & ii + 3c310 & 0.0535 & 55.1 & 26.5 & 0.92 & 323 & 165 & i + 3c314.1 & 0.1197 & 10.6 & 26.49 & 0.95 & 201 & 144 & i + 3c315 & 0.1083 & 17.8 & 26.62 & 0.72 & & 11 & i + 3c319 & 0.192 & 15.3 & 27.07 & 0.90 & 93 & 49 & ii + 3c321 & 0.096 & 13.5 & 26.4 & 0.60 & 309 & 121 & ii + 3c346 & 0.161 & 10.9 & 26.76 & 0.52 & 13 & 71 & ii + 3c348 & 0.154 & 351.0 & 28.23 & 1.00 & 191 & 101 & i + 3c349 & 0.205 & 13.3 & 27.07 & 0.74 & 82 & 142 & ii + 3c353 & 0.03043 & 236.0 & 26.7 & 0.71 & 275 & 85 & ii + 3c371 & 0.0500 & 3.7 & 25.3 & 0.30 & 277 & 59 & ii + 3c379.1 & 0.256 & 7.4 & 27.01 & 0.68 & 76 & 161 & ii + 3c381 & 0.1605 & 16.6 & 26.94 & 0.81 & 69 & 4 & ii + 3c382 & 0.0578 & 19.9 & 26.2 & 0.59 & 179 & 50 & ii + 3c386 & 0.0170 & 23.9 & 25.2 & 0.59 & 288 & 17 & i + 3c388 & 0.091 & 24.6 & 26.6 & 0.70 & 32 & 63 & ii + 3c390.3 & 0.0561 & 47.5 & 26.5 & 0.75 & 231 & 145 & ii + 3c401 & 0.20104 & 20.9 & 27.25 & 0.71 & 19 & 24 & ii + 3c402 & 0.0239 & 10.1 & 25.1 & 0.56 & 528 & 163 & ii + 3c403 & 0.0590 & 17.8 & 26.1 & 0.45 & 230 & 79 & ii + 3c430 & 0.0541 & 33.7 & 26.3 & 0.72 & 99 & 35 & ii + 3c433 & 0.1016 & 56.2 & 27.06 & 0.75 & 58 & 172 & ii + 3c436 & 0.2145 & 17.8 & 27.23 & 0.86 & 105 & 173 & ii + 3c438 & 0.290 & 44.7 & 27.91 & 0.88 & 19 & 136 & i + 3c445 & 0.0562 & 24.8 & 26.2 & 0.85 & 576 & 171 & ii + 3c449 & 0.0171 & 11.5 & 24.9 & 0.58 & 1742 & 10 & i + 3c452 & 0.0811 & 54.4 & 26.9 & 0.78 & 277 & 79 & ii + 3c465 & 0.0303 & 37.8 & 25.8 & 0.75 & 375 & 122 & i + 3c20 & 20.606 & 19.270 & 19.016 & 18.883 & 18.733 & close companion + 3c28 & 21.486 & 19.382 & 18.709 & 18.475 & 18.393 & close companion + 3c29 & 17.966 & 15.741 & & & & unresolved nucleus , globular clusters + 3c31 & 14.795 & & & & & dust lane , globular clusters , unresolved nucleus + 3c33.1 & 20.308 & 19.365 & 18.923 & 18.707 & 18.678 & close companion , unresolved nucleus + 3c35 & 18.410 & 16.908 & 16.365 & & & + 3c52 & 21.348 & 19.436 & 19.006 & 18.842 & 18.779 & close companion + 3c61.1 & 21.873 & 20.613 & 20.325 & 20.270 & & close companion + 3c66b & 15.827 & & & & & globular clusters , unresolved nucleus , jet + 3c75n & 15.469 & & & & & close companion , globular clusters + 3c76.1 & 17.038 & 15.995 & & & & close companion , globular clusters + 3c79 & 20.942 & 19.470 & 19.108 & 18.962 & 18.898 & close companion + 3c83.1 & 15.457 & & & & & dust lane , globular clusters + 3c88 & 17.212 & 15.545 & & & & globular clusters + 3c105 & 19.411 & 18.344 & 18.175 & & & + 3c111 & 17.053 & 16.572 & & & & unresolved nucleus + 3c123 & 21.694 & 19.851 & 19.337 & 19.117 & 19.010 & close companion + 3c129 & 16.313 & & & & & globular clusters , arc ? + 3c129.1 & 16.561 & & & & & globular clusters + 3c130 & 18.271 & 16.297 & 15.704 & 15.373 & & globular clusters + 3c133 & 20.754 & 19.868 & 19.593 & 19.447 & 19.323 & close companion , unresolved nucleus , jet + 3c135 & 19.882 & 18.690 & 18.273 & 18.224 & & close companion + 3c165 & 21.910 & 20.187 & 19.722 & 19.595 & 19.556 & close companion + 3c171 & 21.562 & 19.997 & 19.657 & 19.487 & 19.445 & close companion , emission line tails + 3c173.1 & 21.119 & 19.425 & 18.998 & 18.837 & 18.759 & close companion + 3c180 & 21.534 & 19.572 & 19.009 & 18.790 & 18.716 & + 3c184.1 & 19.261 & 18.432 & 18.284 & 18.275 & & close companion , arc ? , unresolved nucleus + 3c192 & 17.988 & 16.942 & 16.893 & & & globular clusters + 3c196.1 & 21.766 & 19.535 & 18.789 & 18.462 & 18.352 & close companion , globular clusters ? + 3c197.1 & 19.914 & 18.668 & 18.367 & 18.295 & & unresolved nucleus + 3c198 & 19.367 & 18.390 & 18.213 & & & unresolved nucleus + 3c213.1 & 20.879 & 19.523 & 19.190 & 19.109 & 19.066 & hot spots + 3c219 & 20.110 & 18.767 & 18.396 & 18.284 & 18.250 & close companion , unresolved nucleus + 3c223 & 20.118 & 18.761 & 18.416 & 18.351 & & + 3c223.1 & 18.841 & 17.660 & 17.386 & 17.327 & & unresolved nucleus , dust lane + 3c227 & 17.944 & 17.415 & 17.313 & 17.308 & & close companion , unresolved nucleus + 3c236 & 18.857 & 17.502 & 17.210 & 17.140 & & dust lane + 3c277.3 & 19.120 & 17.626 & 17.327 & 17.257 & & close companion , radio knots + 3c285 & 19.167 & 17.624 & 17.248 & 17.163 & & dust lane , arc + 3c287.1 & 19.731 & 18.971 & 18.717 & 18.626 & 18.576 & close companion , unresolved nucleus + 3c288 & 22.113 & 19.476 & 18.657 & 18.367 & 18.227 & close companion , unresolved nucleus + 3c303 & 19.230 & 18.243 & 17.961 & 17.868 & 17.854 & close companion , unresolved nucleus + 3c310 & 18.483 & 17.120 & 16.927 & & & close companion , globular clusters + 3c314.1 & 20.281 & 18.915 & 18.658 & 18.624 & 18.571 & + 3c315 & 19.363 & 18.424 & 18.185 & & & close companion + 3c319 & 21.093 & & & & & close companion + 3c321 & 19.047 & 17.266 & 16.597 & & & close companion , dust lane + 3c346 & 20.162 & 19.060 & 18.345 & 18.019 & 17.932 & close companion , unresolved nucleus , jet + 3c348 & 22.272 & 19.328 & 18.343 & 18.098 & 18.021 & close companion , dust lane + 3c349 & 20.859 & 19.657 & 19.347 & 19.275 & 19.259 & unresolved nucleus + 3c353 & 16.860 & 15.537 & & & & globular clusters + 3c371 & 16.007 & 15.555 & & & & unresolved nucleus + 3c379.1 & 21.227 & 19.407 & 18.891 & 18.717 & & arc + 3c381 & 20.072 & 18.819 & 18.545 & 18.413 & 18.290 & close companion , unresolved nucleus + 3c382 & 15.931 & 15.366 & 15.143 & & & unresolved nucleus + 3c386 & 15.429 & & & & & globular clusters + 3c388 & 19.534 & 17.101 & 16.337 & 15.884 & & close companion + 3c390.3 & 16.806 & 16.249 & 16.187 & & & unresolved nucleus , arc ? + 3c401 & 21.874 & 19.853 & 19.160 & 18.671 & 18.590 & close companion , unresolved nucleus , jet + 3c402 & 15.915 & & & & & globular clusters + 3c403 & 17.848 & 16.485 & & & & close companion + 3c430 & 17.653 & 16.400 & & & & dust lane , globular clusters + 3c433 & 18.922 & 17.500 & 17.057 & 16.615 & & close companion , unresolved nucleus + 3c436 & 20.920 & 19.217 & 18.780 & 18.614 & 18.585 & close companion + 3c438 & 21.851 & 19.807 & 19.240 & 19.015 & 18.839 & close companion + 3c445 & 16.802 & 16.499 & & & & unresolved nucleus , globular clusters + 3c449 & 15.418 & & & & & dust lane , globular clusters + 3c452 & 18.564 & 17.191 & 16.643 & & & + 3c465 & 16.105 & & & & & close companion , globular clusters , unresolved nucleus +

we study consequences of gauge invariance and charge conservation of an electron gas in a strong random potential perturbed by a weak electromagnetic field . we use quantum equations of motion and ward identities for one- and two - particle averaged green functions to establish exact relations between density and current response functions . in particular we find precise conditions under which we can extract the current - current correlation function from the density - density correlation function and vice versa . we use these results in two different ways to extend validity of a formula relating the density response function and the electrical conductivity from semiclassical equilibrium to quantum nonequilibrium systems . finally we introduce quantum diffusion via a response function relating the averaged current with the negative gradient of the averaged charge density . with the aid of this response function we derive a quantum version of the einstein relation and prove the existence of the diffusion pole in the zero - temperature electron - hole correlation function with the long - range spatial fluctuations controlled by the static diffusion constant .

cond-mat0203414

low - energy physics of equilibrium systems with weakly interacting electrons is well understood both qualitatively and quantitatively . the relevant information about the equilibrium system is contained in one- and two - particle green functions , in particular in their behavior around the fermi energy . a number of reliable approximate methods have been developed for the calculation of these functions . among them , systematic renormalized perturbation expansions based on the many - body feynman diagrammatic technique have proved most effective . when impurities or quenched configurational randomness are added we still are able to describe equilibrium properties of such a system quite reliably within diagrammatic schemes and a mean - field - type coherent - potential approximation ( cpa).@xcite the situation gets less straightforward if we come out of equilibrium . this is the case when the system is disturbed by a time - dependent external force but does not manage to reach a new equilibrium within the relaxation times of the experimental setup . we then have to determine the response of the system to the external perturbation to obtain experimentally relevant data . however , we do not have at our disposal many established methods to calculate nonequilibrium response functions . when the perturbation is weak , which is the most common situation in practice , we can use linear response theory to calculate effectively the response functions.@xcite a fundamental tool for the calculation of response functions within the linear - response theory is the kubo formalism.@xcite it determines how the response functions of the disturbed nonequilibrium system can be expressed in terms of two - particle green ( correlation ) functions that are characteristics of an equilibrium state . kubo formulas are means for the description of nonequilibrium systems with equilibrium functions . unlike the response functions , the equilibrium green functions obey equations of motion and are suitable for developing systematic approximations within many - body diagrammatic techniques . kubo formalism , however , provides different representations for different response functions that we have to approximate separately . the problem that emerges with independent approximations for different response functions via different kubo formulas is that we are usually unable to keep exact relations between response functions that may hold due to special symmetries of the system under investigation . most pronounced case with `` hidden '' symmetries is the response to a weak electromagnetic perturbation where we have to satisfy gauge invariance and charge and current conservations . instantaneous field - dependent deviations of charge and current densities from their equilibrium values are described by density and current response functions . within linear - response theory these functions are obtained from independent kubo formulas with density - density and current - current correlation functions , respectively . although both correlation functions are derived from the same two - particle green function , they are normally approximated independently according to the purpose they serve to . when quantum coherence is negligible we can calculate the transport properties from the coherent - potential approximation.@xcite there is no contribution to the homogeneous current - current correlation function beyond the single electron - hole bubble in the single - band coherent - potential approximation.@xcite hence most theories beyond the cpa , either on model or realistic level , use the kubo formula for the electrical conductivity with the current - current correlation function to determine transport properties of disordered solids.@xcite on the other hand , when quantum coherence effects are substantial and we expect the anderson metal - insulator transition , electrical conductivity is usually calculated from the electron - hole correlation function with the aid of the diffusion constant controlling its low - energy behavior near the diffusion pole.@xcite the diffusion pole is crucial and enables application of scaling arguments and the renormalization group approach to anderson localization.@xcite there is a number of more or less heuristic arguments in the literature that relate the density response with the conductivity.@xcite they are essentially based on gauge invariance and charge conservation of a semiclassical equilibrium description of an electron gas exposed to an electromagnetic field . in weakly disordered quantum systems ( described by continuum models ) a relation between the density response and conductivity can formally be derived as follows.@xcite gauge invariance is used to relate the external scalar potential with the electric field @xmath0 . the current density generated by the external field then is @xmath1 where @xmath2 denotes tensor of the electrical conductivity . charge conservation is expressed by a continuity equation . in equilibrium we can use its operator form that follows from the heisenberg equations of motion for the current and density operators . for hamiltonians with quadratic dispersion relation we have @xmath3 energy - momentum representation of the continuity equation in the ground - state solution is @xmath4 we have to use a density variation of the equilibrium density , i. e. , the externally induced density @xmath5 in the continuity equation with averaged values of operators . from the above equations and for linear response @xmath6 we obtain in the isotropic case @xmath7 the derived equality formally holds for complex functions without restrictions on momenta or frequencies . frequencies can , in principle , be even complex . relation is often taken as granted for the whole range of the disorder strength and used for the definition of the zero - temperature conductivity when describing anderson localization transition.@xcite although the above derivation may seem very general it suffers from a few flaws . first , the operator continuity equation can not be directly used out of equilibrium . a new term due to causality of the response functions enters the continuity equation when the equilibrium is disturbed . moreover the nonequilibrium density and current operators in linear - response theory no longer obey heisenberg equations of motion with the perturbed hamiltonian . the perturbation is decoupled from the equilibrium hamiltonian and is treated only to linear order . second , strongly disordered electron systems can be desribed only by lattice models with a nonquadratic dispersion relation . continuity equation is then to be modified even in equilibrium . hence the above , in the literature broadly disseminated reasoning leading to eq . can not be fully trusted in quantitative studies of strongly disordered systems out of equilibrium . presently there is no reliable theory of strongly disordered electrons beyond the mean - field cpa . since mean - field approximations do not include vertex corrections to the electrical conductivity they are not suitable for the investigation of localization effects in three spatial dimensions . unlike low dimensions ( @xmath8 ) , anderson localization in @xmath9 may occur only in strongly disordered systems . anderson localization in bulk systems has not yet been understood or quantitatively described in a satisfactory manner . its quantitative description demands bridging the gap between the mean - field ( cpa ) transport theory and theories for weakly and strongly localized electron states . it is clear that such an interpolating scheme should be based on advanced approximations for two - particle functions . recently a diagrammatic method for summations of classes of two - particle diagrams has been proposed.@xcite this theory has a potential to interpolate between the mean - field and localization theories provided exact relations between the density and current response functions have been proved for this scheme in the whole range of the disorder strength . criteria for validity of relations between density - density and current - current correlation functions in approximate treatments of strongly disordered systems have not yet been established . when we want to go beyond the cpa we still have to establish such relations in the metallic regime of tight - binding models with extended electron states . simultaneously we have to formulate conditions under which the derived relations hold or may be broken . the aim of this paper is threefold . first , we derive various relations between current - current , current - density , and density - density correlation functions for strongly disordered electrons out of equilibrium , i. e. , beyond the reach of the existing derivations . second , we articulate conditions under which these relations hold or to which extent they may be broken in quantitative treatments and show when the conductivity can be calculated from the density response function and vice versa . third , we introduce a generalization of diffusion via a quantum response function , test validity of the einstein relation for this quantum diffusion , and relate quantum diffusion with its classical counterpart and with the diffusion pole in the electron - hole correlation function . to reach this goal we use kubo formalism and averaged many - body green functions . in this approach a weaker form of the continuity equation expressed in terms of one- and two - particle green functions replaces the operator identity . it is derived from equations of motion for green functions and ward identities between them . we use two ward identities that are not fully equivalent . one expresses conservation of probability and the other reflects charge conservation . each ward identity is used in a different manner to relate the conductivity with the density response . the layout of the paper is as follows . in sec . [ sec : density_current ] we summarize the definitions and useful representations of the density and current response functions . in sec . [ sec : einstein_relation ] we show how the current - current correlation function emerges from a momentum expansion of the density response function . then in sec . [ sec : dynamical_conductivity ] we derive continuity equations expressed in green functions and use them to show how the density response function can be revealed from the conductivity . based on the exactly derived relation between the density and current response function we introduce in sec . [ sec : diff_einstein ] quantum diffusion , derive the diffusion pole in the hydrodynamic limit of the density response function and relate the diffusion constant to the conductivity . in sec . [ sec : infinite_d ] we illustrate general results on an exactly solvable limit of infinite spatial dimensions . in appendix we discuss assumptions for and the range of validity of the vollhardt - wlfle - ward identity used in the derivation of the continuity equation for green functions . a simplest description of the electron motion in impure , weakly correlated metals is provided by a tight - binding anderson model . it assumes noninteracting spinless electrons moving in a random , site - diagonal potential @xmath10 . its hamiltonian reads @xmath11 the values of the random potential @xmath10 are site - independent and obey a disorder distribution @xmath12 . i. e. , a function depending on the random potential @xmath10 is averaged as @xmath13 the averaged two - particle propagator ( resolvent ) is defined as the averaged product of one - particle propagators @xmath14^{-1}_{ij } \left[z_2\widehat{1}-\widehat{t } -\widehat{v}\right]^{-1}_{kl } \right\rangle_{av } \ .\end{gathered}\ ] ] fourier transform to momentum space is not uniquely defined , since due to momentum conservation we have only three independent momenta . for our purposes we choose the following notation and definition of the fourier transform @xmath15 the density response function is determined from the following thermodynamic representation of the kubo formula@xcite @xmath16 here @xmath17 are fermionic and @xmath18 bosonic matsubara frequencies at temperature @xmath19 . due to the analytic properties of the two - particle green function we can perform analytic continuation from imaginary matsubara to real frequencies using contour integrals . only the contribution along the real axis remains and we obtain @xmath20 g^{ar}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q } ) \right . \\[2pt ] \left . + \ f(e ) g^{rr}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q } ) - f(e+\omega ) g^{aa}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q})\right\ } \ .\end{gathered}\ ] ] we used the following definition @xmath21 and analogously for the functions @xmath22 and @xmath23 , where both energies are from the upper and lower complex energy half - plane , respectively . we denoted the fermi function@xmath24^{-1}$ ] with the chemical potential @xmath25 . the two - particle green function is generally determined from a bethe - salpeter equation in which the input is a two - particle irreducible vertex . the latter is known only approximately , except for special limits . however , a few specific elements of the full two - particle green function can be evaluated without knowing the two - particle irreducible vertex . it is sufficient to know only the one - particle green function to determine them . determination of two - particle functions from one - particle ones is enabled by ward identities . the ward identity relating the averaged one- and two - particle green functions reads @xmath26\ , .\end{gathered}\ ] ] and was proved for the first time within the coherent - potential approximation by velick.@xcite it is a nonperturbative identity valid quite generally beyond the cpa.@xcite as a consequence of this identity we obtain vanishing of the density response function for a homogeneous global perturbation , i. e. , for @xmath27 . using in we easily find @xmath28 - f(e ) \left[g^a(e ) - g^r(e)\right]\right\ } = 0\ .\end{gathered}\ ] ] another situation where we do not need to know the two - particle irreducible vertex is the static limit , @xmath29 . first , the static density response function is real and reads [ eq : drf_t ] @xmath30 that in the limit @xmath31 goes over to @xmath32 { } n_f \\ ] ] with @xmath33 as the density of states at the fermi level and @xmath34 . note that eqs . hold only in the limit @xmath35 . in the inverse case , @xmath36 , eq . applies . noncommutativity of the limits @xmath37 and @xmath38 indicates that the point @xmath39 is not analytic in the same manner as in the fermi liquid theory.@xcite the current - current correlation function @xmath40 relates the current of the system @xmath41 perturbed by an external vector potential @xmath42 . tensor of the electrical conductivity @xmath43 determines the current response to an external electric field . it can be obtained from the current response function by using a relation @xmath44 . in the fourier representation we obtain@xcite @xmath45\ ] ] with @xmath46 . the compensation term @xmath47 on the right - hand side of eq . is real and was introduced to warrant finiteness of the complex conductivity . this additive term is not generated from the kubo formula with a commutator of current operators but originates from gauge invariance of the electron gas in an external electromagnetic field.@xcite according to eq . there is no current in the system with a static vector potential and the static ( complex ) conductivity can be defined only via a dynamical one from the limit @xmath37 . unlike the density response function the current response function can not be simplified in any limit and to determine the conductivity from eq . we have to know the two - particle irreducible vertex . the current response function alike the density response function is a complex quantity with the same analytic properties . of interest for us are the parallel components that for real frequencies read @xmath48\left[v_\alpha({\bf k}'+{\bf q}/2 ) - v_\alpha({-\bf k}'+{\bf q}/2)\right]\\ \times \int_{-\infty}^{\infty}\frac { d e}{2 \pi\omega } \left\{[f(e+\omega ) - f(e ) ] g^{ar}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q } ) + f(e ) g^{rr}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q}))\right . \\ \left . - f(e+\omega)g^{aa}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf q } ) - f(e ) [ g^{rr}_{{\bf k}{\bf k}'}(e , e;{\bf q})- g^{aa}_{{\bf k}{\bf k}'}(e , e;{\bf q } ) ] \right\}\ .\end{gathered}\ ] ] we denoted the group velocity @xmath49 , where @xmath50 is the dispersion relation of the underlying lattice . as first we show how the dynamical conductivity can be obtained from the long - range behavior of the density response function ( hydrodynamic limit of small transfer momentum ) . to this purpose we use the velick - ward identity , eq . . it enables us to evaluate averaged matrix elements of the two - particle green function with zero transfer momentum in terms of the one - particle propagator . zero transfer momentum is a severe restriction on the applicability and utilization of this ward identity . physically we are interested in the hydrodynamic limit , i. e. , the asymptotics @xmath51 . we are unable to extend identity beyond the homogeneous case . however , if we assume analyticity of the asymptotics @xmath51 , we can use momentum @xmath52 as an expansion parameter and investigate the hydrodynamic limit perturbatively . the hydrodynamic limit is analytic if frequency @xmath53 , i. e. , we are in the regime @xmath54 . we then can expand the density response function in powers of momentum @xmath52 . we define two configuration - dependent resolvent operators @xmath55^{-1}\ ] ] where @xmath56 is a difference in the dispersion relation of the two resolvents and is defined via its matrix elements @xmath57 , @xmath58 . we will be interested in the following function of two energies @xmath59\ ] ] that can be , after averaging over the configurations of the random potential , expressed via the two - particle green function @xmath60 we assume that the expansion in the operators @xmath61 commutes with the configurational averaging . we first expand the quantity @xmath62 and then average the series term by term . it is sufficient for our purposes to expand only to second order in momentum @xmath63 . this precision determines the leading small - momentum behavior . expanding quantity @xmath62 in @xmath61 we have to keep the order of operators in the products , since @xmath61 and the resolvent @xmath64 do not commute . the expansion to second order reads @xmath65 \widehat{g}(z_+ ) \widehat{g}(z_- ) \right . \\ \left . -\ \widehat{g}(z_+)\widehat{g}(z_- ) \left[\delta_{1 } \widehat{t } - \delta_{2 } \widehat{t}\right ] \widehat{g}(z_- ) - \widehat{g}(z_+ ) \delta_{1 } \widehat{t } \widehat{g}(z_+ ) \widehat{g}(z_- ) \delta_{1 } \widehat{t } \widehat{g}(z_-)\right . \\ \left . + \ \widehat{g}(z_+)\left[\delta_{1 } \widehat{t } \widehat{g}(z_+ ) \delta_{1 } \widehat{t}\widehat{g}(z_+ ) + \widehat{g}(z_- ) \delta_{1 } \widehat{t } \widehat{g}(z_- ) \widehat{\delta_1 t } \right ] \widehat{g}(z_-)\right\ } \ .\end{gathered}\ ] ] each direct product of the resolvent operators @xmath66 can be simplified using identity . doing this consequently we end up with a sum of products of two resolvents . we have three different terms to analyze : [ eq : difference ] @xmath67\widehat{g}(z_-)\right\ } = \\ \frac 1{z_+ - z_- } \left\ { \mbox{tr}\left[\left(\widehat{g}'(z_+ ) + \widehat{g}'(z_-)\right ) \delta_{1 } \widehat{t}\right ] + \mbox{tr}\left[\left(\widehat{g}'(z_+ ) - \widehat{g}'(z_-)\right ) \delta_{2 } \widehat{t}\right]\right . \\ \left . -\ \frac 2 { z_+ - z_-}\mbox{tr}\left[\delta_{1 } \widehat{t } \left(\widehat{g}(z_+)- \widehat{g}(z_-)\right)\right]\right\}\ , \end{gathered}\ ] ] @xmath68\right . \\ \left . + \ \mbox{tr}\left[\widehat{g}(z_- ) \widehat{\delta_1 t}\widehat{g}(z_- ) \delta_{1 } \widehat{t}\right ] - 2\mbox{tr}\left[\widehat{g}(z_+)\delta_{1 } \widehat{t } \widehat{g}(z_-)\delta_{1 } \widehat{t}\right ] \right\}\ , \end{gathered}\ ] ] @xmath69 \widehat{g}(z_-)\right\ } = \\ \frac 1{z_+ - z_-}\left\{\mbox{tr}\left[\widehat{g}'(z_+ ) \delta_{1 } \widehat { t}\widehat{g}(z_+ ) \delta_{1 } \widehat{t}\right]%\right . . - \mbox{tr}\left[\widehat{g}'(z_- ) \delta_{1 } \widehat{t}\widehat{g}(z_- ) \delta_{1 } \widehat{t}\right ] \right . \\ \left . -\ \frac 1{z_+ - z_-}\mbox{tr}\left[\left(\widehat{g}(z_+ ) - \widehat{g}(z_-)\right ) \delta_{1 } \widehat{t}\left(\widehat{g}(z_+ ) - \widehat{g}(z_-)\right ) \delta_{1 } \widehat{t}\right]\right\ } \ .\end{gathered}\ ] ] we insert eqs . in expansion and average over the configurations of the random potential @xmath10 . to recover the density response function we have to limit the complex energies to the real axis from above or below with which we distinguish the causality . we define correlation functions that can be represented as a trace of the averaged two - particle green function @xmath70 where @xmath71 stand for @xmath72 depending on whether the imaginary part of the corresponding frequency argument is positive or negative . we then can represent the leading asymptotics in small momenta @xmath73 of the averaged electron - hole correlation function [ eq : g1 ] @xmath74\\ + \frac 1{8\omega}\left[\left\langle \partial_e \left(g^r(e+\omega ) -g^a(e)\right ) ( \mathbf{q}\cdot\nabla)^2\epsilon \right\rangle\right ] + \frac 1{\omega^2 } \left[\left\langle\left(g^r(e+\omega ) - g^a(e)\right ) \mathbf{q}\cdot\mathbf{v}\right\rangle\right ] \\ - \frac 1{4\omega n^2 } \sum_{{\bf k}{\bf k}'}\mathbf{q}\cdot\mathbf{v}({\bf k})\ ; \mathbf{q}\cdot\mathbf{v}({\bf k}')\left\{\partial_\varepsilon \left[g^{aa}_{{\bf k}{\bf k}'}(e+\varepsilon , e;{\bf 0 } ) - g^{rr}_{{\bf k}{\bf k}'}(e+\omega , e+\omega + \varepsilon;{\bf 0})\right]_{\varepsilon=0 } \right . \\ \left . \frac 2{\omega}\left[g^{aa}_{{\bf k}{\bf k}'}(e , e;{\bf 0 } ) + g^{rr}_{{\bf k}{\bf k}'}(e+\omega , e+\omega;{\bf 0 } ) - 2 g^{ar}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf 0})\right ] \right\}\end{gathered}\ ] ] and analogously of the electron - electron ( @xmath75 ) and hole - hole ( @xmath76 ) ones that can be written generically as @xmath77\\ + \frac{1}{8\omega}\left[\left\langle \partial_e \left(g^a(e+\omega ) -g^a(e)\right ) ( \mathbf{q}\cdot\nabla)^2\epsilon \right\rangle\right ] + \frac 1{\omega^2 } \left[\left\langle\left(g^{a}(e+\omega ) - g^{a}(e)\right ) \mathbf{q}\cdot\mathbf{v}\right\rangle\right ] \\ - \frac 1{4\omega n^2 } \sum_{{\bf k}{\bf k}'}\mathbf{q}\cdot\mathbf{v}({\bf k})\ ; \mathbf{q}\cdot \mathbf{v}({\bf k}')\ \left\{\partial_\varepsilon \left[g^{aa}_{{\bf k}{\bf k}'}(e+\varepsilon , e;{\bf 0 } ) - g^{aa}_{{\bf k}{\bf k}'}(e+\omega , e+\omega + \varepsilon;{\bf 0})\right]_{\varepsilon=0 } \right.\\ \left . + \ \frac 2{\omega}\left[g^{aa}_{{\bf k}{\bf k}'}(e , e;{\bf 0 } ) + g^{aa}_{{\bf k}{\bf k}'}(e+\omega , e+\omega;{\bf 0 } ) - 2 g^{aa}_{{\bf k}{\bf k}'}(e , e+\omega;{\bf 0})\right ] \right\ } \ .\end{gathered}\ ] ] in the above equations we used angular brackets to denote summation over fermionic momenta in the first brillouin zone of one - electron functions , that is @xmath78 note that only the term with @xmath79 from the electron - hole correlation function diverges in the limit @xmath37 while in the electron - electron ( hole - hole ) functions it remains finite even in the static limit . this is a manifestation of the diffusion pole missing the latter two correlation functions . in eq . different powers of momentum @xmath63 and frequency @xmath80 appear . there is no restriction on validity of eqs . in frequency but it holds only for small momenta , more precisely only perturbatively up to second order . equations establish relations between the density - density and the current - current correlation functions in the asymptotic limit @xmath81 . generally , however , these two functions are not directly proportional , since the right - hand sides of eqs . contain one - particle contributions . they cancel each other if we combine the electron - hole with electron - electron and hole - hole correlation functions appropriately to build up the density response function . for isotropic situations we define in the hydrodynamic limit [ eq : mobility ] @xmath82 this quantity can be generalized to anisotropic cases as @xmath83 function @xmath84 measures leading long - range correlations of the density response function . its real part can be identified with the diffusive conductivity or mobility of the system . it is now easy to find an explicit representation of this function using eqs . . all contributions except for the last terms on the right - hand sides of eqs . cancel and we obtain equality @xmath85 being just eq . in the limit @xmath86 . note that in the course of the derivation we had to sum contributions from the electron - hole , electron - electron , and hole - hole correlation functions and integrate over energies . if we resort to the low - frequency limit @xmath37 , the diffusion pole in the electron - hole correlation function dominates in the density response and we recover the conductivity directly from @xmath87 . it is interesting to note that in this static limit we derive the mobility ( @xmath88-conductivity ) with contributions from @xmath89 and @xmath22 solely from the electron - hole correlation function @xmath87 . we showed in the preceding section how the conductivity can be revealed from the small - momentum behavior of the density response function . in this section we express the current - current correlation function in terms of the density - density correlation function . the aid we use for this task is a continuity equation and another ward identity @xmath90 \ .\end{gathered}\ ] ] proved for retarded and advanced functions by vollhardt and wlfle in ref . . we denoted @xmath91 . note that identity for @xmath92 was already used in ref . . identity is related to eq . , however , the two ward identities are identical neither in the derivation nor in the applicability and validity domains . the latter holds for nonzero transfer momentum @xmath63 , i. e. , for an inhomogeneous perturbation while the former only for @xmath81 . on the other hand , we show in the appendix that eq . , unlike eq . , holds only perturbatively within an expansion in powers of the random potential . i. e. , we can prove this identity for nonzero transfer momentum @xmath63 only by assuming that perturbation expansions for the self - energy @xmath93 and simultaneously for the two - particle irreducible vertex @xmath94 converge . it was shown earlier that in the homogeneous case , @xmath81 , the vollhardt - wlfle - ward identity follows from the velick - ward one.@xcite to derive a continuity equation for green functions we start with an equation of motion for the two - particle green function . it is a bethe - salpeter equation where the input is a two - particle irreducible vertex . we have three possibilities ( topologically distinct scattering channels ) how to construct a bethe - salpeter equation.@xcite for our purposes the electron - hole channel is the relevant or most suitable one . there the bethe - salpeter equation in momentum representation reads @xmath95\ ] ] where @xmath96 is the two - particle irreducible vertex from the electron - hole channel . one - electron propagators @xmath97^{-1}$ ] are the averaged resolvents . the product of the one - electron propagators from the right - hand side of eq . can be decomposed into @xmath98 where we denoted @xmath99 @xmath100 and analogously the difference @xmath101 . we multiply both sides of the bethe - salpeter equation by the denominator from the right - hand side of eq . and assume validity of eq . relating the two - particle irreducible vertex @xmath94 with the one - particle irreducible self - energy @xmath93 . we then obtain a `` difference '' equation of motion @xmath102 g^{(2)}_{{\bf k}{\bf k}'}(z_+,z_-;{\bf q } ) = \delta_\mathbf{q } g({\bf k};z_+,z_-)\delta(\mathbf{k } - \mathbf{k } ' ) + \frac 1n \sum_{{\bf k}''}\lambda_{{\bf k}{\bf k}''}(z_+,z_-;{\bf q } ) \\ \times \left[\delta_\mathbf{q } g({\bf k};z_+,z_-)g^{(2)}_{{\bf k}''{\bf k}'}(z_+,z_-;{\bf q } ) - g^{(2)}_{{\bf k}{\bf k}'}(z_+,z_-;{\bf q } ) \delta_\mathbf{q } g({\bf k}'';z_+,z_-)\right]\ .\end{gathered}\ ] ] we now define correlation functions generalized to complex frequencies describing density - density and density - current correlations by summing over the fermionic momenta in eq . [ eq : phi_def ] @xmath103 the contributions from the two - particle irreducible vertex @xmath94 in eq . cancel each other provided the two - particle irreducible vertex is symmetric , i. e. , @xmath104 . if so , we end up with a continuity equation relating the generalized density - density and density - current correlation functions [ eq : continuity ] @xmath105 another continuity equation can be derived by multiplying eq . with the energy difference and summing over the fermionic momenta . we obtain an equation relating the generalized current - current and density - current correlation functions @xmath106 combining the above two continuity equations we obtain the resulting relation between the generalized current - current and density - density correlation functions replacing the operator continuity equation @xmath107 \ .\end{gathered}\ ] ] actually , correlation function @xmath108 is strictly speaking not the current - current correlation function , since it is not the momentum or velocity that is summed with the two - particle green function . we used an energy difference @xmath109 from eq . . only in the case of quadratic dispersion relation we have @xmath110 and the energy difference is proportional to the group velocity . otherwise @xmath108 equals the current - current correlation function only in the small - momentum limit , @xmath51 . we use @xmath111 to convert the energy difference in the hydrodynamic limit to a multiple of the group velocity @xmath112 and denote @xmath113 with the aid of eq . we rewrite continuity equation to @xmath114\end{gathered}\ ] ] where the angular brackets stand for the summation over the fermionic momenta from the first brillouin zone as defined in eq . . in the isotropic case we obtain @xmath115\ .\end{gathered}\ ] ] it is the most general relation between the current - current and density - density correlation functions for arbitrary complex frequencies and non - zero momenta . it holds for finite momenta as far as the dispersion law remains quadratic . for general dispersion relations the validity of eq . remains in the asymptotic regime of small momenta . to return to measurable quantities we limit the complex frequencies to the real axis . to derive exact relations we first define a generalization of the conductivity with the correlation functions @xmath116 : @xmath117 \phi_{\epsilon\epsilon}^{ar}(e , e+\omega;\mathbf{q } ) + f(e)\phi_{\epsilon\epsilon}^{rr}(e , e+\omega;\mathbf{q } ) \right . \\ . - f(e+\omega ) \phi_{\epsilon\epsilon}^{aa}(e , e+\omega;\mathbf{q } ) - f(e)\left[\phi_{\epsilon\epsilon}^{rr}(e , e;\mathbf{q } ) - \phi_{\epsilon\epsilon}^{aa}(e , e;\mathbf{q})\right]\right\}\ .\end{gathered}\ ] ] we use continuity equation to relate @xmath118 and @xmath119 and find @xmath120 \\ -\frac { ie^2}{\omega } \int_{-\infty}^{\infty } \frac { de}{\pi}f(e ) \frac 1n \sum_{\mathbf{k}}\delta_\mathbf{q } \epsilon(\mathbf{k } ) \left[\im g^r(\mathbf{k}_+,e ) -\im g^r(\mathbf{k}_-,e)\right ] = 0\ .\end{gathered}\ ] ] vanishing of the right - hand side in eq . can be explicitly manifested when continuity equation is applied to the two - particle correlation functions @xmath121 and @xmath122 . the two terms on the right - hand side of eq . add to zero . we can extract the electrical conductivity from @xmath123 in the limit @xmath51 or for quadratic dispersion law by replacing @xmath124 leading to @xmath125 . in the isotropic case we then reveal eq . from eq . it is important to state that this proof of validity of eq . is strongly based on the ward identity for nonzero momenta . the latter can be proved only perturbatively , see appendix , and hence it is not clear whether eq . holds beyond the perturbative regime near and beyond the anderson localization transition . in the preceding sections we proved validity of eq . for arbitrary frequencies and small momenta from the region where quadratic dispersion relation is accurate in strongly disordered lattice systems . it is a useful relation in particular in approximate theories . it enables one to approximate only either the density response function or the conductivity and to determine the other one from eq . . however , both quantities in eq . are response functions that are not well suited for systematic approximations . we would prefer exact relations or identities for equilibrium correlation or green functions . this can be achieved when we introduce diffusion . diffusion in classical systems is defined by fick s law relating the current with the negative gradient of the charge density.@xcite the diffusion constant introduced in this way , i. e. via fick s law , has in classical physics an intuitive phenomenological character without a proper microscopic backing that should come from quantum physics . the study of diffusion in quantum disordered systems was launched by the seminal work of anderson analyzing destructive effects of quantum coherence on diffusion of disordered electrons.@xcite since then diffusion has become part of transport studies of disordered and interacting quantum itinerant systems.@xcite however even in these quantum approaches diffusion is introduced either via ( self - consistent ) perturbation expansions leading to the diffusion pole in the electron - hole correlation function or via the semiclassical limit where the electron - hole correlation function becomes a green function of the classical diffusion equation.@xcite here we propose a nonperturbative way to define diffusion from first principles . we introduce a quantum generalization of fick s law with a diffusion response function relating the averaged current density with the negative gradient of the charge density of a perturbed system . we hence use [ eq : diffusion ] @xmath126 as a definition of the diffusion response function @xmath127 being an extension of the diffusion constant to which it should reduce in the semiclassical limit . in the above definition we used a perturbed charge density @xmath128 that is not the total change of the charge density @xmath129 . it is a `` dynamical '' density change obeying a condition @xmath130 saying that the total change at any place during the entire time evolution is zero . this restriction is dictated by the same condition obeyed by the averaged current density defined from the conductivity . definition of diffusion is nonperturbative , has quantum character and does not demand any assumption about the behavior in the semiclassical limit or the existence of the diffusion pole in the electron - hole correlation function . moreover , it offers a microscopic determination of the diffusion function from first principles , i. e. , quantum response functions of the perturbed system and via the kubo formalism from correlation and green functions of an equilibrium solution . it is hence directly related to measurable quantities . in fourier representation we obtain for the averaged current @xmath131 the dynamical charge density @xmath132 is induced by an external dynamical scalar potential , which can be expressed in fourier representation as @xmath133 \varphi(\mathbf{q},\omega)\ .\ ] ] it is easy to verify that with this definition eq . is fulfilled . we use eq . together with eqs . and to find an identity between the conductivity , diffusion and density response @xmath134 \ .\ ] ] we have derived a general relation between quantum diffusion and the electrical conductivity . it is now to show that quantum diffusion response function @xmath135 behaves in the semiclassical limit as expected from the diffusion constant in the diffusion equation . namely , we have to show that the homogeneous static diffusion @xmath136 obeys einstein s relation and enters the diffusion equation in the semiclassical limit . to come to it we use eq . to exclude the density response @xmath137 from eq . and eq . to determine @xmath138 doing so we obtain in the isotropic case @xmath139 \sigma(\mathbf{q},\omega)\\ = e^2 d(\mathbf{q},\omega ) \int_{-\infty}^{\infty } \frac{de}{\pi } f(e ) \im\phi^{rr}_e(\mathbf{q},0)\ .\end{gathered}\ ] ] if we define the homogeneous dynamical conductivity @xmath140 ( in the same way also the dynamical diffusion ) and use the velick - ward identity in the limiting case @xmath141 we obtain a dynamical generalization of the well known einstein relation [ eq : einstein_new ] @xmath142 that at zero temperature reduces to @xmath143 a classical version of this formula was proved by einstein for the brownian motion of a particle in a random medium@xcite and later on it was re - derived in the framework of nonequilibrium statistical mechanics by kubo.@xcite here we derived einstein s relation for quantum response functions and showed that it is a consequence of gauge invariance of the system , namely eq . . it is not yet clear from the above reasoning how the diffusion function @xmath135 is related to the classical diffusion equation and to the diffusion pole . the existence of the diffusion pole in quantum systems is usually deduced from the semiclassical limit leading to the diffusion equation . we can prove the existence of the diffusion pole entirely from first quantum principles and from the quantum generalization of fick s law , eq . . we use eq . to exclude now conductivity @xmath144 from eq . . it is easy to come to a representation of the density response function @xmath145 known from other treatments of diffusion.@xcite eq . holds for arbitrary frequencies and momenta within the range of quadratic dispersion relation . this representation directly indicates a singularity in the density response function in the limit @xmath146 . since the order of the limits is relevant for the result , we have to specify it explicitly . to single out the singular contribution in the density response function we have to choose @xmath147 , that is , the inverse situation to sec . [ sec : einstein_relation ] and to eqs . . the singularity in the density response function leads to a pole in the two - particle green function . it is demonstrated in the low - energy and small - momentum asymptotics of the density response function at zero temperature . we find from eq . @xmath148 using this asymptotics in eq . we obtain an explicit manifestation of the diffusion pole in the zero - temperature electron - hole correlation function @xmath149 representing the singular part of the electron - hole correlation function . we denoted a diffusion constant @xmath150 . this representation was derived for zero temperature in the asymptotic limit @xmath151 with the restriction @xmath147 . to extend it to the opposite limit , @xmath54 , we have to show that the diffusion function @xmath135 is analytic in the limit @xmath151 . it is not a priori clear that the diffusion constant in eq . equals the diffusion constant obtained from the inverse order of limits @xmath152 used in the einstein relation , eq . however , the diffusion response function @xmath135 was introduced in such a way that the limits @xmath37 and @xmath51 commute and we have only one definition for the static diffusion constant from the density response function @xmath153 the left equality in eq . holds quite generally while the right one only at zero temperature . note that representation holds only at zero temperature for small momenta and in the low - frequency limit . we can see from eq . that the diffusion function introduced in the quantum version of fick s law reduces to the diffusion constant determining the long - range fluctuations of the electron - hole correlation function and hence in the semiclassical limit also the diffusive behavior in the diffusion equation . however , only the static diffusion constant @xmath154 enters the denominator of the electron - hole correlation function . a generalization of the diffusion constant from the electron - hole correlation function to a frequency - dependent quantity via @xmath155 does not have a rigorous justification . it is nevertheless often used in the literature for a quantitative treatment of anderson localization.@xcite beware that the dynamical diffusion defined in this way does not fulfill the einstein relation and hence deviates from the dynamical conductivity . we show in the next section that in an exactly solvable mean - field limit the left - hand side of eq . related to the conductivity via the einstein relation differs at finite frequencies from the conductivity obtained from the kubo formula with the electrical current . we derived exact relations between density and current correlation functions for a lattice electron gas in a random potential . we now demonstrate the generally derived formulas explicitly on an exactly solvable limit of infinite spatial dimensions . this limit serves as a mathematical tool for the definition of a mean - field theory not only for classical spin systems but also for itinerant disordered and interacting models.@xcite in case of the anderson model of disordered electrons the mean - field theory , i. e. , the limit of infinite spatial dimensions , equals the coherent potential approximation . we hence resort in this section to this solution . equation determining the local self - energy in the coherent - potential approximation is soven s equation that can be written as@xcite [ eq : cpa_equations ] @xmath156^{-1}\right\rangle_{av}\ ] ] where @xmath157 . the two - particle irreducible vertex then is @xmath158\\ = \frac { \sigma(z_+ ) - \sigma(z_- ) } { g(z_+ ) - g(z_- ) } = \frac { \delta\sigma } { \delta g } \ . \end{gathered}\ ] ] the angular brackets were defined in eq . . the second equality in eq is the velick - ward identity . it is straightforward to find an explicit form of the two - particle green function in the coherent - potential approximation @xmath159\end{gathered}\ ] ] where @xmath160 and @xmath161 . using the two - particle green function , eq . , we easily obtain explicit representations for the density - density and density - current correlation functions [ eq : cpa_phi ] @xmath162 we now insert eqs . together with the explicit representation for the two - particle green function , eq . , to the right - hand side of eq . and find @xmath163 that is , continuity equation is fulfilled . analogously we can explicitly verify eqs . and . actually , validity of continuity equations follows from the velick - ward identity , since the two - particle irreducible vertex @xmath164 is momentum independent . in such a case the vollhardt - wlfle and the velick identities are equivalent . as next we show that the @xmath88-conductivity in the coherent - potential approximation can be calculated using the diffusion constant @xmath154 from eq . . to this purpose we need to evaluate @xmath165 at zero transfer frequency @xmath29 . we have @xmath166 where in the last equality we used integration per parts in momentum space . we utilize this result and eq . to obtain the electron - hole correlation function @xmath167 that in the low - energy limit reduces for real frequencies to @xmath168 proving the einstein relation for the @xmath88-conductivity @xmath169 of the coherent - potential approximation . it is evident from eq . that the frequency - dependent coefficient in the denominator of the electron - hole correlation function does not equal the dynamical diffusion from eq . to show this explicitly we use definition and compare the result with the real - part of the conductivity at zero temperature given by [ eq : cpa_conductivity ] @xmath170 = \sigma^{ar}_{\alpha\alpha}(\omega ) + \sigma^{rr}_{\alpha\alpha}(\omega)\ .\end{gathered}\ ] ] we separated contributions to the conductivity from the electron - hole and electron - electron current - current correlation functions , @xmath171 , @xmath172 . it is useful to introduce also a simplified asymptotic form of the low - frequency conductivity @xmath173\end{gathered}\ ] ] where the integrand in the energy integral is replaced by its initial value . this representation is asymptotically exact in the limit @xmath37 for smooth current - current correlation functions @xmath174 near the fermi energy @xmath175 and carries the same frequency dependence as the diffusion constant @xmath176 defined in eq . . for the numerical calculations we use a binary alloy with two values of the random potential @xmath177 weighted with probability @xmath178 and @xmath179 . [ fig : om_01 ] shows the two dynamical conductivities and the diffusion constant from eq . on a simple cubic lattice with parameters @xmath180 , @xmath181 for frequencies @xmath182 and @xmath183 with @xmath184 being the half - bandwidth . note that the split - band value of the disorder strength is @xmath185 . although the three quantities differ for low frequencies only slightly inside the band , they behave differently near the band edges and for higher frequencies . in particular the diffusion constant shows anomalous behavior when the fermi energy approaches a band edge . the compensating terms from the electron - electron ( density - density ) correlation function get relevant there . anomalous behavior of the diffusion constant is more transparent for higher frequencies ( lower pane ) . we can not evidently rely on @xmath186 as a good approximation to the conductivity for finite frequencies except for fermi energies deep inside the energy band . the smoothing impact of the integral over frequencies on the behavior of the dynamical conductivity gets clear from our numerical results . recently a discussion was renewed about the proportion of contributions to the kubo formula for the electrical conductivity from the electron - hole and electron - electron current - current correlation functions , @xmath171 and @xmath172 in eq . , respectively.@xcite in fig . [ fig : sigma_ar_0 ] these contributions are compared for the same setting of the binary alloy on a simple cubic lattice for @xmath29 . the contribution from @xmath171 dominates inside the band far from the band edges . outside the central band and in the satellite impurity band the importance of the compensating effects of @xmath172 is evident . the situation worsens when we go over to the dynamical conductivity , fig . [ fig : sigma_ar_1 ] . we can see that there is no region where the electron - hole contribution would dominate or approximate the full conductivity reliably . it is important to note that conductivity @xmath171 differs from the ( dynamical ) diffusion constant defined from the electron - hole density - density correlation function , eq . . the latter contains even in the static limit both terms , @xmath171 and @xmath172 . the difference is evident in fig . [ fig : sigma_ar_1 ] . the conductivity calculated from @xmath176 is much closer to the full electrical conductivity than @xmath171 . with increasing frequency the role of the term @xmath172 increases . we hence can not interchange contributions to the electrical conductivity from the electron - hole density - density correlation function @xmath187 ( its leading @xmath52-dependent term ) and the conductivity @xmath188 . the former is generally better approximation than the latter except for fermi energies near band edges and in band tails where both the approximations equally fail . in this paper we derived a number general relations between density - density , density - current and current - current correlation functions . with the aid of these relations we proved validity of an equation relating the electrical conductivity to the density response function , eq . , in the hydrodynamic limit of strongly disordered electron lattice systems and derived a dynamical generalization of the einstein relation between the conductivity and diffusion , eq . . we found to what extent eq . can be used in quantitative calculations and when it may be broken . we showed that the dynamical diffusion from the einstein relation can not be related to the diffusion pole beyond the static limit . we used only exact reasoning and equalities derived directly from equations of motion for green functions . we avoided extrapolations of semiclassical results or conclusions derived within equilibrium theory . the electrical conductivity can be related to the density response function only if ward identities are fulfilled . the same holds for the existence of a diffusion pole in the electron - hole correlation function with the static optical conductivity governing its long - range spatial fluctuations . there are two , not fully equivalent , ward identities for noninteracting electrons in a random potential . although they reflect conservation laws they can be proved only if perturbation expansion in the strength of the random potential for the irreducible two - particle vertex converges and results in an analytic function . in sec . [ sec : einstein_relation ] we used velick - ward identity to relate mobility with the density response . this ward identity expresses conservation of probability in the state space and holds only for the homogeneous case , i. e. , zero transfer momentum in the two - particle green function . to be of use in the calculation of the conductivity we had to assume analyticity of the hydrodynamic limit for all frequencies and to use the transfer momentum as an expansion parameter . vollhardt - wlfle - ward identity used in sec . [ sec : dynamical_conductivity ] to reveal the density response function from the conductivity expresses charge conservation . it holds , unlike the velick - ward identity , for arbitrary transfer momenta in the two - particle function , but it can be proved only via a perturbation ( diagrammatic ) expansion for the two - particle vertex function . moreover , to warrant this identity we are not allowed to sum selected classes of relevant diagrams . a single relevant diagram for the two - particle irreducible vertex of order @xmath189 produces @xmath190 irrelevant diagrams in the proof of the ward identity . selection of relevant diagrams may be , however , dictated by the existence of a singularity in the two - particle function or by analyticity ( causality ) of the theory . there is hence no general guarantee that ward identities and consequently the relation between the conductivity and the density response hold beyond the perturbative regime near the anderson localization transition . when the conductivity can be related to the density response via eq . we can derive a quantum version of the einstein relation . to that purpose we introduced a diffusion response function relating the averaged current with the negative gradient of the charge density as a quantum generalization of fick s law . diffusion defined in this way enters the dynamical einstein relation , eq . , but it is not priori clear whether this quantum diffusion reduces to the classical one in the semiclassical limit . the solution of the classical diffusion equation pops up from a quantum theory via the diffusion pole and the long - range asymptotics of the electron - hole correlation function . we showed that the `` classical '' diffusion from the diffusion pole equals the quantum one from the einstein relation and hence the conductivity only in the static limit . a dynamical diffusion constant defined from the electron - hole correlation function is no longer proportional to the conductivity beyond the static limit . this was demonstrated generally as well as explicitly on an exact solution in infinite spatial dimensions . to conclude , we showed that gauge invariance of an electron gas in a random potential responding to an electromagnetic perturbation is guaranteed within linear - response theory if ward identities and are fulfilled . in this case an ( approximate ) averaged two - particle green function generates density and current response functions consistent in the hydrodynamic limit with relation . or , what is more important , we can safely apply the approximate two - particle green function only in either kubo formula and use eq . to determine the other response function . we explicitly demonstrated that ward identities are fulfilled only in the `` diffusive '' regime characterized by convergence of the perturbation expansion for the two - particle irreducible vertex and by analyticity of the hydrodynamic limit for all frequencies . we , however , know that in solutions with weakly or strongly localized electrons a pole in the two - particle irreducible vertex emerges and validity of the ward identities can not be generally proved . one has to bear this aspect in mind when interpolating between the diffusive and localized regimes . this work was supported in part by grant no . 202/01/0764 of the grant agency of the czech republic . vollhardt - wlfle - ward identity is a primary tool for proving the continuity equations relating the current and density correlation functions . understanding its proof and the domain of its validity is hence very important . the original proof of vollhardt and wlfle@xcite is restricted to real frequencies with different small imaginary parts only . however , it is valid for arbitrary complex frequencies . to demonstrate this we summarize the principal assumptions and steps of its proof from which we conclude on the domain of validity of eq . . first , we have to assume that the self - energy @xmath93 is a functional of the full one - electron propagator @xmath191 @xmath192({\bf p},z)\ ] ] so that the perturbation expansion in the random potential consists of irreducible diagrams only . we consider here only noninteracting electrons in a random potential although the reasoning can be extended to correlated electrons as well . perturbation expansion of the self - energy @xmath193 in the random potential reads @xmath194 } \sum^{n+1}_{j_1,\ldots , j_{k}=1 } \sum_{l_1,\ldots , l_k}\frac 1{l_1!\ldots l_k ! } \delta(l_1 + \ldots l_k - n-1 ) \left\langle v_{j_1}^{l_1}\right\rangle_c \cdots \left\langle v_{j_k}^{l_k}\right\rangle_c \\ \sum_{{\bf r}_1,\ldots,{\bf r}_{n+1 } } \sum_{\mathcal{p}\{i_1,\ldots , i_{n+1}\ } } \delta_{j_1,i_1 } \ldots \delta_{j_1,i_{l_1 } } \delta_{j_2,i_{l_1 + 1}}\ldots \delta_{j_k , i_{n+1}}\\ \frac 1{n^n } \sum_{{\bf p}_1,\ldots,{\bf p}_n } e^{i{\bf r}_{i_1}({\bf p } - { \bf p}_1 ) } e^{i{\bf r}_{i_2}({\bf p}_1 - { \bf p}_2 ) } e^{i{\bf r}_{i_{n+1}}({\bf p}_n -{\bf p } ) } g({\bf p}_1,z)\ldots g({\bf p}_n , z)\end{gathered}\ ] ] where @xmath195 $ ] is the integer part of @xmath178 , @xmath196 denotes a permutation of the indices @xmath197 , and the angular brackets stand for cumulant averages defined from @xmath198 only the momentum variables and momentum - dependent functions are of importance . we denote @xmath199 the sum of all diagrams with @xmath189 internal fermionic lines with internal momenta @xmath200 . then representation can be simplified to we use notation @xmath202 and @xmath203 and apply expansion for the self - energy difference . we obtain @xmath204 \ .\end{gathered}\ ] ] the difference of the products of one - electron propagators can further be rewritten to a sum @xmath205 where left ( right ) to the difference @xmath206 only @xmath207 ( @xmath208 ) appear . we sum all diagrams for the fixed difference of the one - electron propagators , being now two - particle irreducible diagrams from the electron - hole channel . we then come to a new representation @xmath209 where @xmath210 is a sum of two - particle irreducible diagrams with @xmath189 internal one - electron lines of which @xmath211 lines carry energy @xmath212 and @xmath213 lines energy @xmath214 . last but very important step in the proof of identity is an assumption that removing the difference @xmath215 from eq . and fixing the internal variable @xmath216 does not change summability of the perturbation expansion . if so , we can write @xmath217 the equality holds if and only if the perturbation expansion in the random potential for the two - particle irreducible vertex @xmath218 converges point - wise for the chosen values of independent variables @xmath219 . this is a rather severe restriction on applicability of the ward identity . it says that the ward identity is valid only in cases where no selective rules for sums of two - particle irreducible diagrams apply , i. e. , all classes of diagrams are equally important . this happens when the irreducible two - particle vertex is regular ( bounded ) . however , physically the most interesting situation occurs when due to backscatterings a cooper pole appears in the irreducible vertex @xmath94 . thanks to this pole only a class of relevant ( crossed electron - hole ) diagrams determines the low - energy behavior of the two - particle vertex . in the asymptotic region of the cooper pole in the electron - hole irreducible vertex , the ward identity _ can not be proved_. in fact it gets violated whenever we take into account a selected series of diagrams dictated by causality of the approximation or leading to a divergence and we have to consider a nonperturbative solution for the two - particle irreducible vertex . each selected ( relevant ) diagram of order @xmath189 from the formal expansion of the singular two - particle vertex generates in the ward identity @xmath190 irrelevant diagrams to complete the sum from the right - hand side of eq . so that the self - energy difference can be represented via the vertex @xmath94 . the irrelevant diagrams are neglected in approximations with selected dominant classes of diagrams violating then identity . hence summability of the perturbation expansion for the two - particle irreducible vertex is an additive assumption in the proof of the ward identity . there is no nonperturbative proof of this identity except for the homogeneous case , @xmath81 , where it is a consequence of the velick - ward identity . r. kubo , j. phys . japan * 12 * , 570 ( 1957 ) , in _ lectures in theoretical physics _ , vol i , ( wiley - interscience , new york 1959 ) , pp . 120 - 203 , and r. kubo , m. toda , n. hashitsume , _ statistical physics ii _ ( springer - verlag , berlin 1985 ) .

a fully antisymmetrized microscopic model is developed for light two - neutron halo nuclei using a hyper - spherical basis to describe halo regions . the many - body wavefunction is optimized variationally . the model is applied to bound by semi realistic minnesota nucleon - nucleon forces . the two - neutron separation energy and the radius of the halo are reproduced in agreement with experiment . antisymmetrization effects between and halo neutrons are found to be crucial for binding of . we also properly extract two - neutron overlap functions and find that there is a significant increase of 30%-70% in their normalization due to microscopic effects as compared to the results of three - body models .

1007.1405

in some light nuclei , the proximity to particle emission thresholds allows loosely bound nucleons to tunnel out into the classically forbidden region and form what is typically referred to as a _ halo_. one neutron halos , such as @xmath2 and @xmath3 , and two - neutron halos , such as @xmath0 and @xmath4 , are the most common , although there are nuclear states with the halo formed by protons or more than two neutrons . all these systems have very small separation energies and unusually large matter radii , when compared to neighboring nuclei , and many of their properties are determined by the long - range part of the many - body wavefunction . halo systems are not specific to nuclear physics ; a review covering a number of fields can be found in @xcite . in this work , we focus on two - neutron halo nuclei in which long - range coulomb effects are not present . traditionally , the theory of halo nuclei has been dominated by simple few - body models . these models place their faith on the fact that the valence particles forming the halo are partially decoupled from the rest of the system , the core . in this frame of mind , one typically assumes that the core degrees of freedom are completely frozen , while meticulously treating the relative motion between the valence nucleons and the core . then , the many - body problem containing a two - neutron halo reduces to a core+n+n three - body one . for an overview of three - body techniques applied to halo nuclei , see for example @xcite . the crucial assumption of few - body models , a macroscopic inert core , is a mixed blessing : on one hand , it allows one to focus on the long - range core - valence correlations , on the other hand , it is undoubtedly a crude simplification of the many - body problem . stemming from the inert core employed , the main drawbacks of three - body models are : i ) unknown properties of the core , i.e. the internal details of the core are left out completely and , only when needed , they are provided in an ad hoc manner , ii ) the improper treatment of antisymmetrization @xcite , iii ) the use of effective core - nucleon interactions constructed case by case in an ambiguous manner , and iv ) the fact that even when the two - body interactions are fitted to ( some ) properties of all two - body subsystems , the resulting three - body separation energy is ( much ) smaller than the experimental value @xcite . this pathological lack of binding is often cured with an empirical three - body force or the two - body forces are refitted to deliver the right three - body binding . furthermore , there are indications that for reaction calculations three - body wavefunctions may require additional renormalization to account for microscopic effects missing in the inert - core approximation @xcite . some improvement over the inert - core picture is provided by few - body models incorporating collective excitations into the core @xcite ; they , however , still suffer from the above - mentioned drawbacks of few - body models , and in addition are limited to applications where the core is a good rotor or vibrator . given all these drawbacks , the predictive power of three - body models of two - neutron halo nuclei is rather limited . the above - mentioned drawbacks of few - body models are eliminated in microscopic ( cluster ) models of light nuclei . over the last decade , there have been tremendous advances in brute force ab - initio methods and due to increasing computational power , these models have improved their accuracy and predictability . amongst these , the green s function monte carlo ( gfmc ) @xcite , the no - core shell model @xcite , and molecular dynamics models @xcite have been successfully applied to light s- and p - shell nuclei . somewhere between few - body and ab - initio models are microscopic cluster models in which some degrees of freedom are frozen to reduce the complexity of the many - body problem . of particular relevance to this work is the stochastic variational method ( svm ) @xcite which inspired several aspects of the model here presented . applications of svm and some other cluster models to two - neutron halo nuclei can be found for example in @xcite . despite the success of existing microscopic ( cluster ) models applied to non - halo nuclei and their ability to reproduce basic properties of some halo species , such as ( three - body ) binding energies and radii , questions may arise about their ability to capture the long - range halo correlations . these correlations are important because the halo nucleons spend a considerable amount of time in regions distant from the core and , generally , it is this part of configuration space that contributes the most to reaction observables . as is known from three - body models , the wavefunction describing a two - neutron halo nucleus ought to fall off exponentially . existing microscopic structure models do not pay close attention ( if any ) to asymptotic regions , instead they use the binding energy to assess the convergence towards an eigenstate . the convergence of the energy , however , does not necessarily guarantee the convergence of the wavefunction in long - range regions . moreover , most microscopic models exploit computationally tractable bases , most commonly gaussians of one sort or the other . in principle , it should be possible to capture the slower exponential decay by using a large gaussian basis , but as argued in @xcite , quality precedes quantity when it comes to halo nuclei ; that is the shape of the basis functions matters more than the size of the basis . in addition , existing microscopic structure models fail to provide input to reaction calculations of two - neutron halo nuclei . to feed reaction calculations , by themselves formulated in a few - body picture , one would have to extract information about halo particles from a full microscopic wavefunction which is a non - trivial task . even though recently we have witnessed some progress in this direction for two - body non - halo projectiles @xcite , most microscopic structure theories are still far from providing such few - body - like information about three - body - like halo nuclei . it is for this reason that in reaction calculations the structure of halo nuclei is taken from few - body models despite all their drawbacks . it is obvious that both few - body and microscopic structure models have their appealing aspects as well as drawbacks . it is the aim of this work to combine the best of the two worlds : to develop a fully microscopic model for light two - neutron halo nuclei bound by nucleon - nucleon interactions that would describe simultaneously short- and especially long - distance regions . the novelty of our model is the integration of few - body and ab - initio methods . we use a basis that describes short - range correlations and at the same time preserves the correct three - body asymptotics . this is achieved at a high computational cost since matrix elements can not be evaluated analytically . our ultimate goal is to provide microscopic structure information to be used in reaction calculations involving two - neutron halo nuclei . in this paper our model is applied to the simplest two - neutron halo nucleus , @xmath0 . preliminary results for this case have been published in @xcite . since neither @xmath5 thought of as nor the subsystem are bound , @xmath0 is a borromean system @xcite . to provide input for reaction calculations involving this nucleus , two - neutron overlap functions are properly computed from our microscopic model and , to our best knowledge , they are for the first time expressed in hyper - spherical coordinates . by doing so , these functions are directly applicable to some reaction calculations and they can be compared directly to three - body wavefunctions . this paper is organized as follows . in , the theoretical formulation of our model is developed . outlines some technical details and numerical procedures employed . results for @xmath0 are discussed in . a summary can be found in along with an outlook of possible future developments . the model here presented has as its primary goal the applicability to reactions involving light two - neutron halo nuclei and thus needs to capture few - body long - distance features of these systems . a two - neutron halo nucleus will be described by an antisymmetrized product of a microscopic core and a valence part consisting of two neutrons , or schematically . the terms core " and valence " refer to distinct pieces of the wavefunction prior to the action of the core - valence antisymmetrizer @xmath6 upon which nucleons from the two parts of the wavefunction become indistinguishable . at large distances , our wavefunction decouples into the three - body - like form of a desired shape most naturally expressed in jacobi it was therefore our choice from the early start to express the entire wavefunction in jacobi coordinates . as an additional benefit of using these coordinates , translational invariance is guaranteed . all details of the model can be found in @xcite . more precisely , the microscopic core with a fixed total angular momentum and parity @xmath7 , and isospin @xmath8 with projection @xmath9 is described by an antisymmetrized wavefunction @xmath10 corresponding to the core s ground state . if desired , excited states of the core can be included in the future . the valence terms are drawn from a suitable hyper - spherical basis . each valence basis function @xmath11 carries total angular momentum @xmath12 , and isospin @xmath13 with projection @xmath14 , which coupled to the core s quantum numbers give @xmath15 , @xmath16 , and @xmath17 for the full system . for neutron - rich two - neutron halo nuclei , such as @xmath0 and @xmath4 , @xmath18 and @xmath19 , and the core - valence isospin coupling is trivial . it is understood that parities @xmath20 carry the same subscripts as their corresponding @xmath21 and that @xmath22 . then , the total wavefunction is written as : @xmath23_{j^{\pi } m_j \ , t \ , m_t}.\ ] ] by acting on valence particles only , the operator antisymmetrizes the valence part , whereas the core - valence antisymmetrizer permutes valence particles with those inside the core . in these operators , @xmath24 are permutation operators , @xmath25 are permutation parities , and @xmath26 is the mass number of the halo nucleus . the meaning of @xmath27 is explained in . the wavefunction in eq . ( [ eq mich wavefunction ] ) is constructed in two steps . first , the wavefunction of the core as a free nucleus is built within svm , as briefly described in . unlike many microscopic cluster models employing a simple @xmath28-harmonic oscillator approximation to @xmath1 , we use the best possible wavefunction for the core obtainable within svm . by doing so , we hope to attenuate the problem of underbinding relative to the three - body threshold @xcite . once the core wavefunction is optimized , its parameters remain unchanged in the subsequent minimization procedure . this implies that , as in other microscopic cluster models , distortion of the core due to valence neutrons is not accounted for explicitly , although some distortion is delivered through the core - valence antisymmetrization . in the second step , the valence part is constructed by drawing terms @xmath29 from a hyper - spherical basis described in . valence basis functions contain discrete as well as non - linear continuous parameters that are optimized variationally , along with the linear coefficients @xmath30 in eq . ( [ eq mich wavefunction ] ) , by minimizing the expectation value of the hamiltonian @xmath31 : @xmath32 as outlined in . to feed reaction calculations involving nuclei of interest , two - neutron overlap functions are extracted from the resulting optimized wavefunctions , as presented in . our choice of the microscopic model for the core has been motivated by the following factors . the model should provide accurate structure for the core ; it needs to be extendable to cores heavier than @xmath1 ; and it must handle central and non - central nucleon - nucleon forces . unlike for the valence part , there is no need to impose special asymptotic requirements on the core . finally , as mentioned earlier in this section , the wavefunction of the core should be expressed in jacobi coordinates . it is for these reasons that the svm model seemed to be the most appropriate . in this section , basic ingredients of svm applied to @xmath1 are outlined ; more details on the method can be found in @xcite . for a four - particle ( @xmath33 ) system : k - like ( left ) and h - like ( right ) . each jacobi coordinate connects centers of masses of subgroups of particles.,scaledwidth=50.0% ] in svm , the core wavefunction is written as a linear combination of basis functions @xmath34 : @xmath35 with the projection of @xmath36 suppressed . the operator @xmath37 antisymmetrizes particles inside the core . basis states @xmath34 take the form of correlated gaussians expressed in jacobi coordinates . for a four - particle system , there exist two different k- and h - like sets of jacobi coordinates @xmath38 as shown in . to accelerate the optimization of the core s wavefunction , both jacobi sets enter eq . ( [ eq svm wavefunction ] ) as they invoke different inter - particle correlations . for @xmath1 , a basis term in either jacobi basis is given by : @xmath39_{j_{core}^{\pi } } \tau_{t_{12 } t_{123 } t_{core } m_{t_{core}}}.\ ] ] the function @xmath40 is taken as a vector - coupled product of solid harmonics @xcite : @xmath41_{l_{12 } } \otimes \mathcal{y}_{l_3 } ( \vec{x}_3 ) \right]_l.\ ] ] the spin @xmath42 and isospin @xmath43 parts consist of successively coupled single - particle spins and isospins : @xmath44_{s_{12 } } \otimes \chi_{p_3 } \right]_{s_{123 } } \otimes \chi_{p_4 } \right]_s , \\ \tau_{t_{12}t_{123}t_{core}m_{t_{core } } } & = & \left [ \left [ \left [ \tau_{p_1 } \otimes \tau_{p_2 } \right]_{t_{12 } } \otimes \tau_{p_3 } \right]_{t_{123 } } \otimes \tau_{p_4 } \right]_{t_{core}m_{t_{core}}}.\end{aligned}\ ] ] for @xmath1 , the gaussian part in eq . ( [ eq svm single gaussian ] ) contains a -dimensional positive - definite , symmetric matrix @xmath45 , specific to each basis term . the quadratic form involves scalar products of jacobi vectors : @xmath46 due to the symmetry requirement , the matrix @xmath45 has only 6 independent elements and they are considered non - linear continuous variational parameters . note that , although the gaussian in eq . ( [ eq svm single gaussian ] ) as a whole is a spherically symmetric object , it still carries angular information due to cross terms @xmath47 when the matrix @xmath45 is non - diagonal as considered here . in such a case , numbers @xmath48 in eq . ( [ eq core orbital part ] ) loose their meaning of orbital momentum quantum numbers and can be treated as discrete variational parameters , instead . the composite index @xmath49 comprises other ( quantum ) numbers , elements of the matrix @xmath45 , and the jacobi channel identifier k or h , i.e. @xmath50\{@xmath51 , @xmath52 , @xmath53 , @xmath54 , @xmath55 , @xmath56 , k / h , * a*}. the sum in eq . ( [ eq svm wavefunction ] ) was left without a summation index because in svm the wavefunction @xmath57 is constructed term by term by minimizing the expectation value of energy with all components of @xmath49 optimized stochastically . linear expansion coefficients @xmath30 in eq . ( [ eq svm wavefunction ] ) are determined via energy matrix diagonalization . to inspire the form of valence functions in eq . ( [ eq mich wavefunction ] ) , let us first consider a few - body approach in which a two - neutron halo nucleus is treated as a three - body core+n@xmath58+n@xmath59 system . from among all three - body models , we adopt the formalism of @xcite and ( * ? ? ? 278 ) . for all necessary details , see @xcite . for a three - body core+n@xmath58+n@xmath59 system , one can define two different y- and t - like sets of jacobi coordinates @xmath60 as shown in . these coordinates are further transformed into hyper - spherical coordinates \{@xmath61 } where @xmath62 are spherical angles corresponding to vectors @xmath63 . the hyper - radius @xmath64 is related to the overall size of the system , while the hyper - angle @xmath65 contains information about relative magnitudes of @xmath66 and @xmath67 . in three - body models , the use of hyper - spherical coordinates is motivated by the facts that : i ) the three - body schrdinger equation reduces to a one - dimensional hyper - radial differential equation , and ii ) more importantly for us , in the absence of long - range forces the three - body wavefunction of borromean systems decays asymptotically for @xmath68 as : @xmath69 where @xmath70 is the binding energy relative to the three - body core+n+n threshold , and @xmath71 is the mass of a nucleon . in hyper - spherical coordinates , the volume element becomes : @xmath72 where @xmath73 , @xmath74 are dimensionless reduced mass factors corresponding to vectors @xmath75 . + n@xmath59 system : y - like ( left ) and t - like ( right ) . each jacobi coordinate connects centers of masses of subgroups of objects . these coordinates are not to be confused with those for the core in .,scaledwidth=55.0% ] upon the hyper - spherical transformation , the spatial part of the three - body wavefunction can be separated into its hyper - radial , hyper - angular , and spherical parts , each of which can be conveniently expanded . the spherical part is written as a coupled product of spherical harmonics @xmath76 and @xmath77 . the hyper - angular part is expanded over eigenstates @xmath78 of the grand - angular operator appearing in the hyper - radial schrdinger equation . these functions are explicitly defined in terms of jacobi polynomials @xmath79 of the order @xmath80 with hyper - momentum @xmath81 : @xmath82 the normalization constant @xmath83 is chosen to make these functions orthonormal with respect to a weight from the hyper - spherical volume element in eq . ( [ eq hyper - volume ] ) : @xmath84 having in mind the importance of the asymptotic behavior of the wavefunction in eq . ( [ eq 3body asymptotics ] ) , the three - body model in @xcite employs a hyper - radial basis from @xcite with the desired exponential trend built into it : @xmath85 here , @xmath86 are associated laguerre polynomials of the order @xmath87 . note that this basis is just a suitable mathematical basis whose elements can not be interpreted as hyper - radial eigenfunctions of the physical three - body system . basis functions @xmath88 are orthonormal with respect to the weight factor @xmath89 from the volume element in eq . ( [ eq hyper - volume ] ) : @xmath90 the completeness of the hyper - radial basis for any @xmath91 allows one to use any value of @xmath92 , and yet reconstruct the asymptotically correct form of eq . ( [ eq 3body asymptotics ] ) determined by an a priori unknown three - body binding energy . the index @xmath87 is independent from quantum numbers attached to spherical and hyper - angular parts of the wavefunction . based on these three - body arguments , we have chosen for each valence term in eq . ( [ eq mich wavefunction ] ) the following form : @xmath93 where @xmath94 is a generalized hyper - harmonic function in ls - coupling : @xmath95_l \otimes \left [ \chi_{n_1 } \otimes \chi_{n_2 } \right]_s \bigg]_{j_{val}^{\pi } } \left [ \tau_{n_1 } \otimes \tau_{n_2 } \right]_{1 - 1}.\ ] ] here , @xmath96 and @xmath97 are spinors and isospinors of valence neutrons . the composite index @xmath98 comprises all other numbers as well as the jacobi channel identifier y or t , i.e. @xmath99 . by construction , hyper - harmonics @xmath94 corresponding to a given jacobi basis are orthonormal in all components of @xmath98 . in the t jacobi basis , the pauli principle between valence neutrons is satisfied by requiring ; in the y jacobi basis , the exclusion principle is satisfied upon the action of @xmath100 in eq . ( [ eq mich wavefunction ] ) . valence angular momenta @xmath101 , @xmath102 , @xmath53 , and @xmath103 are not to be confused with those for the core in . in this work , the nuclear hamiltonian includes kinetic energies @xmath104 of all nucleons and two - body nucleon - nucleon potentials @xmath105 : @xmath106 no correction is needed for the kinetic energy of the total center of mass as the wavefunction in eq . ( [ eq mich wavefunction ] ) is expressed in relative jacobi coordinates . in microscopic calculations , the choice of the effective nucleon - nucleon interaction is of crucial importance unless realistic forces are used . if a model is to have anything to do with the real physical problem , one must make sure that the inter - nucleon force is appropriate for all subsystems appearing in the model . in @xmath0 , valence neutrons are mostly in spin - singlet configurations @xcite . the spin - singlet di - neutron state is unbound ; however , many effective nucleon - nucleon interactions , such as the volkov force @xcite , do not distinguish this state from a bound spin - triplet neutron - proton state , the deuteron . in this work we use the semi - realistic minnesota interaction @xcite . this force reproduces the most important low - energy nucleon - nucleon scattering data and therefore it does not bind the di - neutron . the force renormalizes effects of the tensor force into its central component and binds the deuteron by the right amount assuming a proton and a neutron in a relative s - wave . it also gives realistic results for the bulk properties of nuclei in the lowest s - shell . besides the central component , the potential contains spin , isospin , and spin - isospin exchange terms . the potential contains the mixture parameter @xmath107 which can be tuned slightly to adjust the interaction strength . when supplemented by a spin - orbit force @xcite , the minnesota interaction reproduces low - energy @xmath108-nucleon scattering data . in @xcite , it is advised to employ a short - range spin - orbit force ; on that merit we use the spin - orbit parameter set iv from table 1 in @xcite . in , we mostly show results obtained for the minnesota plus the spin - orbit interaction ( mn - so ) . for comparative purposes , some calculations were performed in the absence of the spin - orbit force ( mn ) . to reproduce the two - neutron separation energy 0.97 mev of @xmath0 , the mixture parameter @xmath107 is changed from its default value 1.0 to 1.015 ( mn - so ) and 1.15 ( mn ) . by doing so , we expect to obtain a realistic description of the halo even though the absolute binding of @xmath1 and @xmath0 is not reproduced . essentially , the interaction mixture parameter is the only free parameter in our model . the coulomb interaction is neglected as it should not affect the long - range behaviour of the neutron halo . once the full wavefunction in eq . ( [ eq mich wavefunction ] ) is optimized as outlined in , one can use it to calculate various observables . although most of the observables we calculate are standard , two - nucleon overlap functions deserve special attention . for general considerations regarding these functions , see ( * ? ? ? 16.4.2 ) and @xcite . the overlap integral between a two - neutron halo nucleus ( @xmath109 ) described microscopically by eq . ( [ eq mich wavefunction ] ) and its own core in the ground state ( @xmath57 ) is defined as : @xmath110 the binomial factor accounts for the number of combinations to pick two out of @xmath111 nucleons . the integration is done over all degrees of freedom in the core , and so the overlap integral @xmath112 depends only on the degrees of freedom of two valence ) where valence and core particles become indistinguishable upon the full antisymmetrization . ] neutrons remaining outside the core . for neutron - rich two - neutron halo nuclei , the integral has a well defined isospin and its projection , 1 and -1 , respectively , but it does not have a good angular momentum . the integral can , however , be expanded in a complete set of hyper - harmonics of angular momentum @xmath12 introduced in eq . ( [ eq hyper - harmonic ] ) : @xmath113 where @xmath114 are clebsch - gordan coefficients . the expansion is carried out in the t jacobi basis where the hyper - harmonics @xmath94 satisfy the pauli principle between valence neutrons by construction ( see ) . the hyper - radial part in eq . ( [ eq overlap integral expansion ] ) is not expanded in the laguerre basis from eq . ( [ eq hyper - radial basis ] ) because the basis functions @xmath88 do not have physical significance . instead , the overlap functions @xmath115 are computed directly from : @xmath116_{j^{\pi } m_j \ , t \ , m_t } \left| \frac { \delta(\rho - \rho ' ) } { \rho^5 } \right| \psi_{j^{\pi } m_j \ , t \ , m_t } \right\rangle\ ] ] with the integration carried over degrees of freedom of all nucleons . a meaningful calculation of overlap functions @xmath115 requires both wavefunctions @xmath57 and @xmath109 to be normalized . using overlap functions , a three - body - like component of the wavefunction @xmath109 can be written as : @xmath117_{j^{\pi } m_j \ , t \ , m_t}\ ] ] in a form analogous to that of the three - body wavefunction @xcite . note , however , that the core @xmath57 in eq . ( [ eq three - body decomposition ] ) is fully microscopic , whereas the three - body wavefunction would contain an inert macroscopic core . generally , the expansion in eq . ( [ eq three - body decomposition ] ) would be a part of the fractional - parentage expansion of @xmath109 . overlap functions @xmath115 satisfy a three - body - like schrdinger equation with a source term @xcite . therefore , at least in the asymptotic region , wavefunctions from three - body models and our microscopically founded overlap function @xmath118 should behave similarly . on this merit , we can compare our results with those from three - body models at the level of wavefunctions rather than integrated observables . an overlap term characterized by @xmath98 is referred to as an overlap or a three - body channel . for practical considerations , it is convenient to introduce modified overlap functions : @xmath119 with a regular behavior near the origin . again , in the absence of long - range forces in borromean - like nuclei , overlap functions decay as : @xmath120 just like three - body wavefunctions in eq . ( [ eq 3body asymptotics ] ) . formally , the decay constant @xmath121 in eq . ( [ eq overlap asymptotics ] ) is defined as in eq . ( [ eq 3body asymptotics ] ) , the main difference is that for overlap functions the three - body binding energy needs to be computed microscopically . for the borromean nucleus @xmath0 : @xmath122 where @xmath123 and @xmath124 are binding energies corresponding to @xmath109 and @xmath57 , respectively . given the orthonormality of hyper - harmonics in , the norms spectroscopic factors of overlap channels are given by : @xmath125 in three - body models , spectroscopic factors give the probability of finding the system in a given channel @xmath98 . in , we show the convergence of the binding energy of @xmath1 with the number of gaussians included in the svm wavefunction described in . in converged mn and mn - so ( see ) states containing 20 and 75 basis states , @xmath1 is bound by -30.85 mev and -30.93 mev , respectively . in both cases , all channels with @xmath126 in eq . ( [ eq core orbital part ] ) were present in the model space . : mn ( left ) and mn - so ( right).,title="fig:",scaledwidth=40.0% ] : mn ( left ) and mn - so ( right).,title="fig:",scaledwidth=40.0% ] the expectation value of energy in eq . ( [ eq expectation value of energy ] ) involves multidimensional integrals which must be evaluated efficiently to perform a meaningful variational calculation . for example in svm employing the basis of correlated gaussians , many matrix elements can be evaluated analytically in closed form , and consequently svm can afford a random trial - and - error variational search . in our model , however , the core is combined with a functionally very different valence part , and these two become completely entangled upon antisymmetrization . in addition , there does not seem to be an analytical way of computing matrix elements involving different core - valence permuted pieces of the wavefunction in eq . ( [ eq mich wavefunction ] ) . for these reasons , we are left with numerical evaluation of all matrix elements . for @xmath0 , the integrals in eq . ( [ eq expectation value of energy ] ) involve @xmath127 spatial and @xmath128 spin - isospin dimensions . we use techniques of variational monte carlo ( vmc ) @xcite to perform the variational search . the integration space is sampled statistically to find the regions most relevant for a given physical problem . in nuclear physics , vmc has been successfully applied to the problem of light nuclei @xcite ; however , the work here presented is novel in the choice of the trial wavefunction and therefore has its own challenges . in this section , we briefly mention the most important aspects of the optimization procedure , and advise the expert reader to @xcite for a complete description . the greatest challenge we faced was to develop a robust and yet efficient method to navigate the parameter space . this aspect falls beyond the framework of vmc . we found that optimization methods used in atomic and molecular physics @xcite did not work adequately for the problem at hand , most probably due to the specific nature of non - central , state - dependent nuclear interactions . in addition , due to the closeness to the three - body break - up threshold , an arbitrary starting wavefunction is inevitably three - body unbound . as a consequence , since the trial wavefunction drives the scanning of the integration space , all optimization methods tend to break the nucleus into the core and two individual neutrons as long as the system is three - body unbound . the easiest way to solve this pathological problem is to constrain the radius of the nucleus while performing the energy minimization above the three - body threshold . in our model , this is achieved most readily by taking the same non - linear parameters @xmath92 in eq . ( [ eq valence term ] ) for all valence terms . once the system becomes three - body bound , @xmath92 , still being the same in all valence terms , is adjusted to minimize the binding energy . different optimization procedures were used for the two interaction cases , mn and , of @xmath0 . in the absence of the spin - orbit force , the mn wavefunction contains only spin - singlet valence terms . it is then possible to simply add hyper - angular and hyper - radial valence terms of increasing orders while re - adjusting @xmath92 until convergence in the binding energy is reached , see . starting with a converged wavefunction , different values of @xmath92 are tested using correlated sampling to finally locate the energy minimum , as shown in . the converged mn wavefunction contains all spin - singlet valence terms with @xmath129 and @xmath130 in both y and t jacobi bases . for the mn - so case , the spin - singlet and spin - triplet valence states are mixed , and after antisymmetrization with the core , many components become almost orthogonal which creates numerical noise . it was for this reason that a more advanced optimization technique comparative optimization on two independent random walks needed to be developed . the energy minimum with @xmath92 is first found for an auxiliary wavefunction as shown in , and the mn - so wavefunction is then tailored to the optimum value of @xmath92 . only valence terms lowering the energy the most are admitted to the mn - so wavefunction at any optimization stage , or in other words @xmath87 and all components of @xmath98 in eq . ( [ eq valence term ] ) are treated as discrete variational parameters . it takes ninety carefully selected valence terms with @xmath131 and @xmath130 to reach energy convergence in the case ( see ) . due to a more selective optimization process , the mn - so wavefunction contains fewer valence terms than its mn counterpart . in both cases , the energy minimum in is located at . in either case , the optimization begins with a valence term having low hyper - momentum @xmath132 and @xmath133 . because of their diverging local kinetic energies near the origin , valence terms with @xmath134 are avoided until preliminary convergence with @xmath132 has been reached . to avoid high partial waves in the valence part , both y and t jacobi configurations are mixed . the linear coefficients @xmath30 in eq . ( [ eq mich wavefunction ] ) are determined via energy matrix diagonalization . to report final results , values of observables obtained on several independent random walks are averaged for precision . due to the monte carlo integrations , all results come with statistical errors . ) . in case mn , @xmath92 is adjusted along the optimization route ; in case , the results are for fixed @xmath135 fm . error bars were not computed for all points , and even when present they may be smaller than actual symbols.,scaledwidth=45.0% ] on the non - linear parameter @xmath92 . both curves are constructed in correlated sampling on walks produced for reference values of @xmath92 depicted by stars . in the mn - so case , error bars could be computed for the reference @xmath92 only.,scaledwidth=45.0% ] for comparison purposes , we repeat the inert - core three - body calculations for @xmath0 published in @xcite . here , we make use of the valence basis introduced in . the phenomenological interaction , vanishing for f- and higher - order partial waves , combines a spherically symmetric central woods - saxon and a spin - orbit woods - saxon - derivative parts with parameters taken from @xcite . as for the interaction between valence neutrons , a realistic gogny force is used @xcite . although the core - n and n - n two - body interactions reproduce @xmath108-n and n - n phase shifts satisfactorily , @xmath0 bound by these interactions would miss more than half of its experimental three - body binding energy of -0.97 mev . as in @xcite , the problem of underbinding @xcite commonly encountered in three - body models based on inert cores is cured with an effective three - body force presumably simulating the effects of the closed channel . three - body calculations were performed using the computer code efadd @xcite . the pauli principle was satisfied by writing the wavefunction in the t jacobi basis ( see ) and by projecting the forbidden core - n states before diagonalization @xcite . upon fitting the strength of the ambiguous three - body force , the converged three - body binding energy is -0.98 mev for the wavefunction containing all valence terms with and . more three - body results are shown in alongside with those from our microscopic model developed in this work . in three - body models , the size of the core does not enter the actual calculations and some ad hoc assumption about the radius of the core is needed to estimate the size of the whole system . then , the rms point proton @xmath136 and rms point nucleon matter @xmath137 radii of the three - body system ( mass number @xmath111 ) are related to those of the core ( mass number @xmath138 ) through : @xmath139 } , \ ] ] where @xmath140 is the distance between the core s center of mass and the center of mass of the whole nucleus , and @xmath141 denote expectation values . with the fully optimized wavefunctions , we calculate binding energies , rms radii , and density distributions , study core - valence antisymmetrization effects , and extract two - neutron overlap functions for @xmath0 . binding energies and radii for @xmath1 and @xmath0 from our calculations are shown in along with experimental values and results obtained in the three - body model described in , svm @xcite representing microscopic cluster models , and ab - initio gfmc calculations @xcite . in , experimental rms point proton radii @xmath136 were computed from accurately measured charge radii @xmath142 by using the relationship @xcite : @xmath143 where @xmath144 fm @xcite is the rms charge radius of the proton , @xmath145 @xmath146 @xcite is the mean - square charge radius of the neutron , and @xmath147 and @xmath148 are the neutron and proton numbers , respectively . based on arguments in and , there is a qualitative difference between three - body and microscopic results in . strictly speaking , the three - body results should be taken with caution because the three - body binding energy was fitted using an auxiliary three - body force of an arbitrary strength , while for radii we arbitrarily assumed the size of the core in eq . ( [ eq three - body radii ] ) to be equal to that of our mn - so @xmath1 . the later assumption was made to ensure the best comparison between three - body and our results . similar arbitrary assumptions are made for densities from the three - body model in . in our model , a free @xmath1 in is overbound and smaller relative to experimental data which sets a wrong scale for absolute binding of @xmath0 . we are , however , mostly interested in three - body - like features of @xmath0 which should depend more on three - body rather than the absolute binding energy . as mentioned in , the mixture parameter in the minnesota force was adjusted so that the three - body binding energy is about right . upon this adjustment , there is no additional freedom in computing other observables . .absolute ( @xmath149 ) and three - body ( @xmath70 ) binding energies in [ mev ] , and rms point nucleon ( proton @xmath25 , neutron @xmath150 , matter @xmath71 ) radii in [ fm ] of @xmath1 and @xmath0 . mn and mn - so are results of this work ; for other models , see text . experimental proton radii were computed from eq . ( [ eq charge versus proton radius ] ) using charge radii from references cited in the table . experimental rms neutron radii were computed from experimental values of proton and matter radii using @xmath151 $ ] . the thickness of the neutron halo is defined as . [ cols="^ , < , > , < , > , < , > , < , > , < , > , < , > , < " , ] most likely due to the smaller @xmath1 core , the mn and mn - so proton radii of @xmath0 are smaller than they should be . on the other hand , matter radii are comparable with those deduced from experiments . to assess how strongly the radii of @xmath0 depend on the size of the core , we turn to the three - body model . in a naive three - body picture based on eq . ( [ eq three - body radii ] ) , the larger size of @xmath0 in the three - body model ( when compared to mn and mn - so ) is solely due to the valence neutrons living on average slightly farther from the core . if the radius of the core is increased to its experimental value 1.46 fm , the three - body radii of @xmath0 become , , , and , and the experimental proton radius of @xmath0 would be seemingly reproduced . a similar shift towards larger radii could be expected for our results if larger cores were involved . perhaps due to a stronger three - body binding , mn - so @xmath0 is slightly smaller than its mn counterpart , but the thickness of the neutron halo remains about the same . within svm , @xmath0 has been studied in the past repeatedly , e.g. @xcite . in , svm results obtained in @xcite for central and spin - orbit minnesota and coulomb interactions are quoted . in that reference , several different cluster compositions were considered to study break - up of the core in @xmath0 . for @xmath1 , we quote the results for the @xmath152 model where the @xmath1 wavefunction is a superposition of three 0s - harmonic oscillator slater determinants with common oscillator parameters . due to this simple picture , svm @xmath1 is bound significantly less compared to mn and mn - so cases . the svm results for @xmath0 in are those from model ( b ) in @xcite . in that model , @xmath0 was described as a combination of @xmath1+n+n and with tritons again built from simple 0s - harmonic oscillators . the triton channel was introduced to overcome the insufficient three - body binding of @xmath0 . svm radii and three - body binding energies of @xmath0 are comparable with ours , especially with the mn model . for the sake of completeness , we also show ab - initio gfmc results in . these were obtained using realistic two - body av18 and three - body il2 interactions . the three - body binding energy for this case was computed by using @xmath153 mev @xcite . gfmc results are shown to point out that , by using modern realistic potentials in microscopic calculations , absolute binding energies and proton radii of @xmath1 and @xmath0 can indeed be reproduced . however , as we argued in and unless proven otherwise , questions may arise about how well ab - initio models treat asymptotic regions so important for borromean halo nuclei . also , two - neutron overlap functions are yet to be extracted from ab - initio models of two - neutron halo nuclei . point nucleon density distributions in @xmath0 for the more realistic mn - so case are plotted in along with those obtained from other models listed in . to compute densities from the three - body model , some assumption is needed about the internal structure of the core ; quite arbitrarily , we computed these densities using an auxiliary microscopic wavefunction of the type of eq . ( [ eq mich wavefunction ] ) constructed by combining the mn - so @xmath1 core and the three - body wavefunction serving as the valence part . once again , this arbitrary construction reveals deficiencies in the few - body models . from models in ( except mn ) . svm and gfmc densities are from @xcite and @xcite , respectively ; for densities from the three - body model , see text . for comparison , the proton ( = neutron ) density of the mn - so @xmath1 is shown . all proton ( neutron ) distributions are normalized to the number of protons ( neutrons ) . here , @xmath154 is the distance from the center of mass.,scaledwidth=50.0% ] the densities in from different models are close to one another with small differences reflecting different radii and wavefunction compositions . all models reproduce the most pronounced property , the neutron halo with the neutron distribution extending far beyond that of protons . depleted at short distances , the proton density of @xmath0 stretches farther out than that of @xmath1 . a partial explanation of this effect comes from the three - body model : in @xmath0 , the @xmath108 core does not sit at the center of mass of the entire system , and its motion relative to the center of mass spreads out the proton distribution . due to the same effect , the neutron density in @xmath0 is also expected to be depleted at small distances relative to that of a free @xmath1 , as is also visible in . an important drawback of three - body models is the approximate way in which the pauli principle between the core and the valence particles is taken into account @xcite . the three - body procedure closest to the proper antisymmetrization is feshbach projection where forbidden states are projected before diagonalization @xcite . given the similarities between the three - body basis and our valence basis , the effects of pauli blocking can now be assessed microscopically from our model . this can be done simply by including or not including the core - valence antisymmetrizer @xmath6 in eq . ( [ eq mich wavefunction ] ) . because of the simpler optimization procedure involved , we performed this study for the mn @xmath0 only , but qualitatively the same outcome is also expected for the mn - so case . see @xcite for details . the valence channels with @xmath155 suffer the most from the core - valence pauli blocking . when @xmath6 is not included and the valence channels are present in the wavefunction , @xmath0 is three - body overbound by several tens of mev . when all @xmath156 channels are removed , the nucleus becomes three - body unbound regardless of the inclusion of valence terms with higher hyper - momenta . when @xmath6 is active , converged @xmath0 including valence channels is three - body bound by about -0.9 mev as shown in . upon removal of all channels from the converged wavefunction , the three - body binding reduces to about -0.75 mev . it is evident that the proper antisymmetrization is crucial for the structure of @xmath0 . to produce a meaningful @xmath0 it is not sufficient to simply neglect the most pauli - blocked @xmath156 valence channels rather , all contributing valence channels ought to be included in the model space and carefully antisymmetrized . given this conclusion , we strongly advocate using the best available pauli blocking techniques in three - body models of halo nuclei . all models mentioned in , although different in their nature and predictive power , are in fair agreement on the most commonly computed properties of @xmath0 . to appreciate the amount of microscopic details embedded in different models , it would be better to compare these models at the level of wavefunctions rather than highly integrated observables . in svm , only s - wave overlap functions for @xmath0 have been computed , but not expanded in hyper - spherical coordinates @xcite ; in gfmc , these functions are yet to be computed . here , we compare overlap functions ( ) computed for the mn - so @xmath0 to three - body wavefunctions ( ) . in both cases , the @xmath1 core is in its ground state . the three - body wavefunction is normalized to unity . for a meaningful interpretation of overlap functions , mn - so wavefunctions of @xmath1 and @xmath0 in eq . ( [ eq overlap functions ] ) need to be normalized to unity . for @xmath1 , the normalization is known analytically from svm ; the norm of the @xmath0 wavefunction was determined numerically with accuracy of 0.3% or better by using an auxiliary sampling function @xcite . ordered by spectroscopic factors , the five strongest overlap channels in the mn - so @xmath0 are listed in . these are the only channels that could be resolved , all other potential channels have spectroscopic factors too small and as such are buried in numerical noise . the table also contains three - body results ; these were obtained by summing up all hyper - radial components in a given three - body channel . not only the dominant channels are the same in the two models , but also the order of their spectroscopic strength is preserved . in the three - body model , these five channels account for more than 98% of the wavefunction . therefore , we expect that these channels should also grasp most of the decomposition of the mn - so @xmath0 ground state . ccccccc [email protected] d .. 1.2 & + ( l)1 - 6 ( l)7 - 10 & @xmath132 & @xmath101 & @xmath102 & @xmath53 & @xmath103 & 3body & & + @xmath157 s - waves & 2 & 0 & 0 & 0 & 0 & 0.8089 & 1&1155 ( 0.5% ) & 1.38 + @xmath157 p - waves & 2 & 1 & 1 & 1 & 1 & 0.1103 & 0&1859 ( 0.7% ) & 1.69 + @xmath156 s - waves & 0 & 0 & 0 & 0 & 0 & 0.0417 & 0&0555 ( 2.1% ) & 1.33 + @xmath158 d - waves & 6 & 2 & 2 & 0 & 0 & 0.0164 & 0&0266 ( 3.5% ) & 1.62 + @xmath158 f - waves & 6 & 3 & 3 & 1 & 1 & 0.0078 & 0&0122 ( 3.0% ) & 1.56 + & 0.9851 & 1&3957 & + the hyper - radial dependence of overlap channels from is shown in and . we show these functions as @xmath159 because of their simpler asymptotic fall - off presented in eq . ( [ eq overlap asymptotics ] ) . the three - body and our model agree on the overall shape of their hyper - radial distributions , and the number of nodes in those channels . on the other hand , our model appears to narrow the hyper - radial distributions and shift their peaks towards smaller distances , and to put more weight on smaller hyper - radii . this difference may be related to smaller radii of the mn - so @xmath0 in when compared to those of the three - body model . the most striking difference between our overlap functions and three - body wavefunctions is the normalization of their components quantified in and visible in and . the mn - so spectroscopic factor for the strongest @xmath157 s - waves channel is greater than one due to recoil effects as the @xmath1 core in eq . ( [ eq overlap integral ] ) does not sit at the center of mass of @xmath0 . these recoil effects are present in the three - body model too , but there the wavefunction as a whole is normalized to unity and so all spectroscopic factors are smaller than one . this deficiency of three - body models has been pointed out in @xcite where an upper limit 25/16=1.5625 was derived for an additional renormalization factor to multiply the three - body spectroscopic factor as a mock - up for missing microscopic information . indeed , our microscopic model consistently predicts spectroscopic factors greater by at least 30% than those obtained in the three - body model , but the increase varies between channels in a non - trivial manner as can be seen in . this observation suggests that to account for microscopic effects , it may not be sufficient to simply renormalize the entire three - body wavefunction by a common factor ( such as @xmath160 suggested in @xcite ) . this conclusion is important , for example , for two - neutron transfer reaction theories for @xmath0 in which three - body wavefunctions are traditionally used as structure input . since two - neutron transfer cross sections depend on spectroscopic factors , using microscopically derived overlap functions instead of three - body wavefunctions would certainly have implications for reaction observables . for example , if the transfer reaction is mostly sensitive to the dominant component ( s - waves ) and peripheral , one might not see much change in the normalization , since the asymptotic parts of the overlaps do not significantly change . in the other hand , if the reaction is sensitive to the whole volume of the nucleus , one can expect a renormalization of the cross section consistent with the spectroscopic factors . often , two - neutron transfer reactions are complicated by interference of various mechanisms . preliminary two - nucleon transfer calculations for @xmath161he(p , t)@xmath162he at @xmath163 mev involving the @xmath161he microscopic overlap functions here presented have been performed assuming a 1-step reaction . however , any meaningful comparison with the data requires an extended study of the reaction mechanism which falls beyond the scope of this work . at hyper - radii beyond about 12 fm , the mn - so overlap functions in and become unreliable due to statistical fluctuations . very large hyper - radii would place valence neutrons into regions very distant from the core , and because the monte carlo sampling probability is proportional to the wavefunction squared , such extreme spatial configurations are very unlikely to be visited by a walker during a random walk . moreover , statistical samples in such distant regions may be highly correlated . for example , in extreme configurations of @xmath0 , a hyper - radius of 12 fm corresponds to a di - neutron at a distance of 10.4 fm from the center of the core , or to two neutrons on opposite sides of the core , 17 fm apart . in asymptotic regions , where the two valence neutrons are distant from the core , the core - valence antisymmetrization effects in eq . ( [ eq mich wavefunction ] ) disappear . then , both the three - body wavefunction and the overlap functions should fall off exponentially as shown in eq . ( [ eq 3body asymptotics ] ) and eq . ( [ eq overlap asymptotics ] ) . for @xmath164 mev , we get @xmath165 @xmath166 . at larger hyper - radii in , the three - body wavefunctions have the right slope still influenced by the three - body centrifugal barrier . interestingly , we found in the value @xmath167 fm to be optimal for the valence part of the fully antisymmetrized wavefunction of @xmath0 . by comparing eq . ( [ eq hyper - radial basis ] ) with eq . ( [ eq overlap asymptotics ] ) , the two decay parameters can be related as @xmath168 . then , @xmath135 fm would correspond to @xmath169 @xmath166 , or @xmath170 mev . in other words , all individual valence terms in the mn - so wavefunction decay much faster than the expected asymptotic form of overlap functions . nevertheless , the strongest mn - so overlap functions follow approximately the right long - range trend within the computationally safe region . albeit the efforts in tailoring the basis to the problem , our results still show lingering differences in the asymptotics which could be improved by inclusion of laguerre polynomials of even higher orders and/or by using a value of @xmath171 fm in the mn - so wavefunction . even though not discussed in detail , we also computed overlap functions for the mn case of @xmath0 . due to the absence of the spin - orbit force , valence spin - triplet states are missing in the mn wavefunction and consequently among overlap functions . spectroscopic factors of spin - singlet channels were about the same as those for the mn - so model in . using mn - so . two dominant clustering patterns are shown schematically.,scaledwidth=65.0% ] although @xmath0 has been the object of many experimental studies ( e.g. @xcite ) , there is still an ongoing debate on the neutron clusterization in this nucleus , namely on the dominance of the cigar- or the di - neutron - like configuration . one way of gaining insight into this is by calculating the probability of finding the valence neutrons at definite places within the three - body - like decomposition defined in eq . ( [ eq three - body decomposition ] ) : @xmath172 where @xmath173 and @xmath174 are the distance between valence neutrons and the distance between centers of masses of the core and the valence neutron pair , respectively . the integration in eq . ( [ eq clustering probability ] ) is carried out over all ( unprimed ) degrees of freedom of all nucleons . the probability plot for the mn - so @xmath0 is presented in . the figure indeed exhibits two peaks , as it should for a system dominated by @xmath157 @xmath175waves in the @xmath176basis : a di - neutron - like peak positioned at about @xmath177 fm and @xmath178 fm ( @xmath179 fm ) with the two neutrons close together and far from @xmath1 , and a cigar - like peak at @xmath180 fm and @xmath181 fm ( @xmath182 fm ) with the two neutrons positioned on opposite sides of the core . qualitatively the same clustering picture has been predicted for example within three - body models @xcite and svm @xcite . we looked into the clusterization probability distribution produced within the three - body model and found the peaks to be shifted to slightly larger radii ( the di - neutron - like peak at @xmath183 fm and @xmath184 fm and the cigar - like peak at @xmath185 fm and @xmath186 fm ) due to the three - body wavefunctions peaking at larger hyper - radii in . integrals under the peaks in the microscopic model are larger than those in the three - body model , consistent with the spectroscopic factors presented in table [ tab sfs ] : the integral under the di - neutron peak for the microscopic calculation is @xmath187 , compared to @xmath188 in the three - body model and the integral under the cigar - like peak for the microscopic calculation is @xmath189 , compared to @xmath190 in the three - body model . in both models , it appears that there is coexistence of the two clustering patterns , with @xmath191% di - neutron and @xmath192% cigar - like configurations . in this work , a microscopic structure model for light two - neutron halo nuclei is developed . our goal was to combine advantages of few - body and microscopic methods to create a model capable to deal simultaneously with short- and characteristic long - range halo effects exhibited by these nuclei . to succeed , we combine the stochastic variational method and the hyper - spherical harmonic method into a fully antisymmetrized many - body approach . from the computational point of view , our model is similar to the variational monte carlo method , the novel feature being the form of the basis which incorporates the few - body features of two - neutron halo nuclei , in particular the correct behavior when the two halo neutrons are far from the core . the model is applied to the ground state of @xmath0 bound by an effective minnesota nucleon - nucleon interaction including the spin - orbit force . when comparing three - body binding energies , radii , and densities , our results are comparable with those from other microscopic models using the same interaction . the overall binding energy is not reproduced as that would require more realistic forces , but this is not essential to produce the characteristic halo features determined by the binding relative to the three - body threshold . the halo nature of the nucleus can be seen from its extended neutron density resulting in a large difference between the rms matter and proton radii . we advocate that the standard highly integrated observables , such as binding energies and radii , are not sensitive to details of the halo part of configuration space . to recognize these details , one should compare structure models at the level of wavefunctions or look at observables more sensitive to long - range correlations , such as reaction observables . we properly calculate , to our knowledge for the first time , microscopic two - neutron overlap functions for @xmath0 and compare them with those from the state - of - the - art three - body model . these overlap functions provide a crucial input to reaction calculations involving @xmath0 , in particular to two - neutron transfer reaction models . as our basis was tailored specifically to capture the asymptotic behavior of the wavefunction , it is not surprising that the three - body wavefunction and the overlap functions are indeed similar in this region of coordinate space . in the range of the nuclear interaction , the three - body wavefunction seems to reproduce the properties of the many - body overlap functions only qualitatively . at a quantitative level , however , there are significant differences between overlap functions and three - body wavefunctions . outstanding amongst these , the spectroscopic factor for the dominant microscopic overlap channel is larger by about 40% than its three - body counter - part . this discrepancy reveals the deficiency of three - body models , namely their inert - core approximation . also , our study shows that the underbinding problem in three - body models comes partly from a poor treatment of antisymmetrization . it is often thought that few - body models can be corrected for many - body effects by a simple renormalization of the wavefunction . however , we demonstrate that , in general , such a simple fix may not be sufficient because for the most important three - body - like components in @xmath0 , the ratio between microscopic and three - body spectroscopic factors varies between 1.3 and 1.7 in a nontrivial manner . in this work we have shown that it is possible to obtain the correct long - range properties of halo nuclei within a fully microscopic approach . this opens up the opportunity to many applications . studies of two - neutron transfer reactions involving @xmath0 using our two - neutron overlap functions are underway . given that in our model all matrix elements are calculated numerically , an extension of the method towards more realistic nuclear interactions should be straightforward . it would also be exciting to apply this method to heavier systems such as @xmath4 which would most likely entail more advanced optimization algorithms . on top of that , further code parallelization would be necessary to tackle the factorial growth in dimensionality of the problem . we thank robert wiringa and steven pieper for many discussions on vmc , kalman varga for discussions on svm and for sharing his svm code , and ian thompson for providing details on the previously published three - body results . the calculations were performed at the high performance computing center at michigan state university . this work was supported by the national science foundation through grant phy-0555893 and the u.s . department of energy , office of nuclear physics , under contracts de - fg52 - 08na28552 and de - ac02 - 06ch11357 . jensen et al . , rev . ( 2004 ) 215 . m. zhukov et al . , phys . rep . 231 ( 1993 ) 151 . thompson et al . , phys . c 61 ( 2000 ) 024318 . s. funada , h. kameyama , and y. sakuragi , nucl . phys . a 575 ( 1994 ) 93 . timofeyuk , phys . c 63 ( 2001 ) 054609 . n. vinh mau , nucl . a 592 ( 1995 ) 33 . nunes et al . phys . a 609 ( 1996 ) 43 . i. brida , f.m . nunes , and b.a . brown , nucl . phys . a 775 ( 2006 ) 23 . s.c . pieper and r.b . wiringa , ann . nucl . part . 51 ( 2001 ) 53 . pieper , nucl . phys . a 751 ( 2005 ) 516 . p. navrtil et al . , j. phys . g : nucl . part . 36 ( 2009 ) 083101 . p. navrtil and b.r . barrett , phys . c 57 ( 1998 ) 3119 . p. navrtil et al . 87 ( 2001 ) 172502 . h. feldmeier and j. schnack , rev . 72 ( 2000 ) 655 . n. itagaki , a. kobayakawa , and s. aoyama , phys . c 68 ( 2003 ) 054302 . t. neff , h. feldmeier , and r. roth , nucl . a 752 ( 2005 ) 321 . k. varga and y. suzuki , phys . c 52 ( 1995 ) 2885 . y. suzuki and k. varga , _ stochastic variational approach to quantum - mechanical few - body problems _ , springer - verlag berlin - heidelberg , 1998 . k. arai , y. suzuki , and r.g . lovas , phys . c 59 ( 1999 ) 1432 . k. varga , y. suzuki , and r.g . lovas , phys . c 66 ( 2002 ) 041302 . s. korennov and p. descouvemont , nucl . phys . a 740 ( 2004 ) 249 . armstrong , _ a cluster model of @xmath0 and @xmath193 _ , phd thesis , michigan state university , usa , 2007 . v. vasilevsky et al . , phys . c 63 ( 2001 ) 034607 . nollett , phys . c 63 ( 2001 ) 054002 p. navrtil , phys . c 70 ( 2004 ) 054324 i. brida and f.m . nunes , int . jour . mod . e 17 ( 2008 ) 2374 . thompson , f.m . nunes , and b.v . danilin , comp . comm . 161 ( 2004 ) 87 . thompson and f.m . nunes , _ nuclear reactions for astrophysics _ , cambridge university press , new york , 2009 . g. erens , j.l . visshers , and r. van wageningen , ann . 67 ( 1971 ) 461 . carlson and r.b . wiringa , variational monte carlo techniques in nuclear physics , in _ computational nuclear physics i : nuclear structure _ , edited by k. langanke , j.a . maruhn and s.e . koonin , springer - verlag , 1991 . foulkes et al . , rev . 73 ( 2001 ) 33 . x. lin , h. zhang and a.m. rappe , j. chem . 112 ( 2000 ) 2650 . umrigar et al . , phys . 98 ( 2007 ) 110201 . pieper et al . , phys . c 64 ( 2001 ) 014001 . b . wang et al . , phys . 93 ( 2004 ) 142501 . i. sick , phys . b 576 ( 2003 ) 62 . s. kopecky et al . , phys . 74 ( 1995 ) 2427 . s. kopecky et al . c 56 ( 1997 ) 2229 . i. sick , phys . rev . c 77 ( 2008 ) 041302 . weller and g.m . hale , nucl . a 541 ( 1992 ) 1 . tilley et al . , nucl . a 708 ( 2002 ) 3 . i. tanihata et al . , phys . lett b 206 ( 1988 ) 592 . i. tanihata et al . , phys . lett b 289 ( 1992 ) 261 . k. varga , y. suzuki , and y. ohbayasi , phys . c 50 ( 1994 ) 189 . e. sauvan et al . , 87 ( 2001 ) 042501 . a. chatterjee et al . , phys . ( 2008 ) 032701 .

we investigate the light quarkonium spectrum using a covariant dyson - schwinger - bethe - salpeter - equation approach to qcd . we discuss splittings among as well as orbital angular momentum properties of various states in detail and analyze common features of mass splittings with regard to properties of the effective interaction . in particular , we predict the mass of exotic states , and identify orbital angular momentum content in the excitations of the meson . while the and its second excitation are predominantly-wave , the first excitation is predominantly-wave .

1508.07183

over recent years the covariant dyson - schwinger - bethe - salpeter - equation ( dsbse ) approach to qcd @xcite has matured via numerous individual studies of meson and baryon states and related phenomena . to trace and sketch the relevant literature , one may consult works on : pseudoscalar , vector @xcite , and other mesons @xcite , baryon masses @xcite leptonic , electromagnetic , and hadronic interactions of both mesons @xcite and baryons @xcite , as well as further states @xcite many technical and numerical refinements and developments @xcite have made it possible and desirable to attempt an as - comprehensive - as - possible application of these techniques to hadron phenomenology . for the sake of feasibility , first steps were taken in a rainbow - ladder ( rl ) truncated setup of the dsbse framework using a suitable effective interaction model . anchoring for an effective rl investigation needs to be provided in the heavy - quark domain , first approached in this context in @xcite for bottomonium ground states and extended in our recent studies of heavy quarkonia @xcite , and exotic @xmath5 states @xcite . at the same time , additional model - independent anchoring happens in and towards the chiral limit via the satisfaction of the axial - vector ward - takahashi identity @xcite . for the case of dynamical chiral symmetry breaking , this guarantees a massless pion ground state and vanishing leptonic decay constants for all radial pion excitations in the chiral limit as well as the validity of a generalized gell - mann - oakes - renner relation for any pseudoscalar meson and quark mass @xcite . satisfaction of the the axial - vector and other ward - takahashi identities is one of the strengths of the rl truncated setup in the dsbse approach . on the other hand , these identities can be used to find appropriate constructions of the quark - gluon vertex ( qgv ) and corresponding truncations of the dsbse system . in any truncation , the use of an effective quark - gluon interaction remains , but this interaction will necessarily be different at different levels of truncation . the qgv @xcite can easily be made more complicated in simple interaction models @xcite . also , beyond - rl calculations with more sophisticated kinds of an effective interaction are available @xcite , albeit too tedious for large - scale computations . in addition , one needs to build understanding in more basic truncations first before advancing to more complicated ones ; e.g. , technical details and problems are much more easily worked out this way . furthermore still , the importance of resonant corrections can not be stressed enough . we remark at this point that our calculations provide bound - state solutions from the bse ; hadronic ( and other ) decay widths can be obtained in a semi - perturbative fashion via consistent constructions corresponding to the decay mechanism @xcite . for the present purpose of a comprehensive investigation , we continue along the lines of our fitting strategy first described in @xcite , where we employed it to find an optimal rl - based description of heavy quarkonia . in the application to light mesons it turned out that an overall satisfactory description of light isovector meson spectra seems not possible with our particular model setup and parametrization of our effective one - dressed - gluon interaction . in turn , for the study of exotic mesons in @xcite we investigated several sets of splittings among light - meson masses and were able to find some with a reasonable correlation to exotic spectra in order to warrant using them to obtain predictions for exotic heavy quarkonia . as a starting case , we revisit this problem here in sec . [ sec : exotic ] after sketching the essentials of the setup and strategy in secs . [ sec : setup ] and [ sec : strategy ] . we have also systematically investigated radial , orbital , and other meson mass splittings , whose detailed results are collected in [ sec : radial ] - [ sec : other ] . special attention is given to splittings with regard to the pion ground state due to its particular role in the light isoscalar meson spectrum in sec . [ sec : pion ] . a combined fit is presented in sec . [ sec : combined ] , followed by a discussion of orbital angular momentum of various mesons in sec . [ sec : oamd ] . conclusions and an outlook are presented in sec . [ sec : conclusions ] . our calculations are performed in euclidean - space landau - gauge qcd . for the interested reader we refer to analogous calculations in coulomb - gauge qcd @xcite or minkowski space @xcite . we use the homogeneous @xmath6 bse in rl truncation which reads @xmath7 where @xmath8 and @xmath9 are the quark - antiquark relative and total momenta , respectively , and the ( anti)quark momenta are chosen as @xmath10 . the renormalized dressed quark propagator @xmath11 is obtained from its dse @xmath12 @xmath13 is the quark self - energy , @xmath14 is the current - quark mass given at the renormalization scale @xmath15 , and @xmath16 and @xmath17 are renormalization constants @xcite . @xmath18 represents the free gluon propagator and @xmath19 is the dirac structure of the bare and rl truncated quark - gluon vertex . dirac and flavor indices are omitted for brevity . @xmath20 denotes a translationally invariant regularization of the integral , with the regularization scale @xmath21 @xcite . the effective interaction @xmath22 needs to be specified in order to obtain numerical results . our choice is the well - investigated and phenomenologically successful form of ref . @xcite , which reads @xmath23}.\ ] ] the parameter @xmath24 [ gev ] corresponds to an effective inverse range of the interaction , while @xmath4 [ gev@xmath25 acts like an overall strength of the first term ; they determine the intermediate - momentum part of the interaction , while the second term is relevant for large momenta and produces the correct one - loop perturbative qcd limit . we note that @xmath26)]/s\;,\ ] ] where @xmath27 gev , @xmath28 , @xmath29 , @xmath30 , and @xmath31 , which is unchanged from ref . in a model setup such as ours the use of different parameter sets can lead to similar results . a prime example is the original work of maris and tandy @xcite where a reasonable fit of @xmath32 and @xmath2 properties was obtained for three @xmath33 pairs , which are omitting units for simplicity @xmath34 , @xmath35 , @xmath36 . when simplified to a one - parameter model via the prescription @xmath37 ( the standard - maris - tandy value of the constant is @xmath38 gev@xmath39 ) , one finds that ground - state masses and decay constants are independent of the value of @xmath24 on the domain @xmath40 $ ] gev , whereas any excitations ( radial or orbital ) and their properties depend rather strongly on such a choice of @xmath24 , a picture equally valid for both mesons and baryons @xcite . this is not surprising , since its value in the parametrization of eq . ( [ eq : interaction ] ) defines a scale at which the long - range part of the effective interaction is strongest a characteristic expected to be relevant in excitations rather than ground states . strategies to arrive at results independent of the model - parameter values for excited states can then rely on ratios of calculated masses for selected pairs with a strongly correlated @xmath24 dependence in order to predict isolated quantities , see , e.g. , refs . @xcite . for a more comprehensive approach , however , it is important to have a systematic broad basis of parameter sets to investigate . we work on a grid of @xmath41 gev @xmath42 @xmath43 gev@xmath44 . figure [ fig : couplings ] illustrates the form of the effective coupling defined in eq . ( [ eq : interaction ] ) for all combinations of @xmath24 and @xmath4 used herein . the axes are : @xmath4 , used to group curves with the same @xmath4 value , but different @xmath24 values ( this is shown in the left panel of fig . [ fig : couplings ] ) ; and @xmath45 to show the @xmath46 dependence of @xmath22 , but at the same time highlight the peak of the curve and label it by the corresponding value of @xmath24 ( this can be seen best in the right panel of fig . [ fig : couplings ] ) . obviously , the peak height of each curve rises with increasing @xmath4 for constant @xmath24 . also , for equal @xmath4 , the lowest @xmath47 gev produces the largest peak , while the largest of our @xmath48 gev yields the lowest peak . our quark - mass value of @xmath49 gev ( given at a renormalization point @xmath50 gev ) is fixed such that the experimental value for the pion mass is reproduced , which is universally the case for this current - quark mass value on our @xmath24-@xmath4 grid , and we consider the case of isovector mesons in the isospin - symmetric limit . note at this point that nothing prohibits us from studying isoscalars as well , but this case is a lot more complicated due to flavor mixing with the @xmath51 case . while we could do this and extend our study along these lines , the main difficulty would be a consistent setup of flavor - mixing , which happens on the meson level on the basis of the computed @xmath52 and @xmath51 states , since rl truncation does not contain flavor mixing in the bse kernel . in our strategy the general idea is to plot @xmath53 from the comparison of the experimental splitting under consideration to our calculated values across the @xmath24-@xmath4 grid and analyze and identify trends and regions of better or worse agreement with experiment . in a successful global fitting attempt , we would then find overlapping regions of good agreement with experimental splittings for all relevant cases , such that an overall satisfactory description of the isovector meson masses results . each figure of this kind herein is plotted from the same viewing angle onto the @xmath24-@xmath4 grid , unless the resulting @xmath53 surface can not be seen properly or one of its essential features is hidden . for the cases where our calculations did nt produce a numerically convincing result , `` empty grid points '' may appear in the plot . this is the result of some of the numerical techniques used in order to arrive at our computed results ; details on numerical setup and difficulties as well as solution strategies can be found in @xcite . error bars corresponding to our numerical uncertainties for our results are generated by our calculations automatically . however , for the sake of clarity they are not given in the fitting plots but only where results for a particular set of computed results are plotted versus the data , such as figs . [ fig : isov - spectrum ] or [ fig : isov - spectrum - fit ] . note also that regions of equally low @xmath53 in the plots does not indicate a correlation of the model parameters @xmath24 and @xmath4 , since there is no universal pattern across the various plots herein . we treat these two parameters as completely uncorrelated . a prominent example for a meson - mass splitting is the hyperfine splitting between the @xmath54 and @xmath55 ground states . our comparison is shown in the upper left corner of fig . [ fig : splittingsexotic ] . to properly and clearly identify splittings we introduce the following notation : a splitting is denoted by the quantum numbers @xmath56 together with a subscript @xmath57 for the radial excitation with @xmath58 denoting the ground state , @xmath59 the first radial excitation , etc . for the example of the hyperfine splitting , we thus write @xmath60 $ ] . for a set of splittings fitted simultaneously we denote them together in parentheses . it appears that in order to provide a good fit for the hyperfine splitting in the isovector case one would need to investigate values towards the `` traditional '' values for @xmath24 and @xmath4 from the original work of maris and tandy . the general trend is that the splitting is better reproduced for lower values of both @xmath24 and @xmath4 . to immediately expand this picture we also show the hyperfine splitting of the first radial excitations in the pseudoscalar and vector channels in the center of the first row in fig . [ fig : splittingsexotic ] . there the situation is similar : lower values of @xmath24 lead to a better description of the splitting , but in this case rather independent of the value of @xmath4 . returning to the aforementioned interpretation of @xmath24 as an inverse effective range of the interaction , one would conclude from the hyperfine splittings alone that a larger effective range is favored in the interaction . however , while the hyperfine splitting is iconic , there are many others to consider and the picture is complicated by splittings such as those regarding radial and orbital excitations , which are discussed in detail in the appendix . here , we revisit and deepen the investigation published in ref . @xcite . in short , the problem investigated in @xcite was how to reach a good fit of the @xmath61 spectrum without actually fitting it . to illustrate our region of interest we turn directly to the @xmath61 states and plot a comparison for the radial splitting between the ground and first excited @xmath1 states , @xmath62 $ ] , in the right upper corner of fig . [ fig : splittingsexotic ] on the one hand . on the other hand , we add two more splittings to form a set of three , namely the splitting of each @xmath1 state to the ground - state pseudoscalar meson together with the radial exotic splitting , @xmath63,[1^{-+}_1 - 0^{-+}_0],[1^{-+}_0 - 0^{-+}_0])$ ] and show the result at the left of the center row in fig . [ fig : splittingsexotic ] . while the plot showing merely the exotic vector radial splitting remains inconclusive regarding the interesting parameter domain , the combination clearly shows a region of low @xmath53 in the interior of our grid , which is the necessary prerequisite . the next step was to identify a set of non - exotic splittings that provide similar results for our fitting - attempts , i.e. , a reasonable correlation to the exotic case . it turned out that axial - vector mesons are good indicators . in particular , we studied several sets of splittings . to start , the simple splitting between the @xmath64 and @xmath65 ground states @xmath66 $ ] is shown in the center of fig . [ fig : splittingsexotic ] . while there is a visible trend , this does nt show a clear preference in any small enough region of our grid . while at least the combination high @xmath24 and low @xmath4 are excluded , more information is needed to proceed . as a consequence , we added two more splittings analogous to the case above , namely those of each axial - vector ground state to the pseudoscalar ground state . the result is shown at the right of the center row in fig . [ fig : splittingsexotic ] for the combination @xmath67,[1^{++}_0 - 0^{-+}_0],[1^{+-}_0 - 0^{-+}_0])$ ] , which allows to focus on a well - defined region inside our grid as well . even better , this plot is nicely correlated with the exotic - pseudoscalar splitting set and so our marker set of splittings was found . on a little detour to the isoscalar case , more precisely strangeonium , one can see that additional difficulty is added with the loss of a reliable match for the pseudoscalar ground state due to @xmath68 flavor mixing , which is not represented in an rl bse kernel . thus , an alternative to @xmath67 $ ] , @xmath69,[1^{+-}_0 - 0^{-+}_0])$ ] must be found to predict @xmath1 states in the strangeonium system , i.e. , assuming ideal flavor mixing , along the same lines as for the heavy quarkonia . without too much trouble , one can try two approaches : first , use the pseudoscalar ground state nonetheless by employing some dummy experimental mass for a pure light - light pseudoscalar in the comparison . secondly , one could replace the pseudoscalar ground state by the vector ground state , which is a good choice in strangeonium due to the confirmed ideal mixing of the @xmath70 and @xmath24 mesons . following the second example , we plot @xmath53 for the isovector case and the set @xmath67,[1^{++}_0 - 1^{--}_0],[1^{+-}_0 - 1^{--}_0])$ ] in the left lower corner of fig . [ fig : splittingsexotic ] and also at the center of the lower row in fig . [ fig : splittingsexotic ] from a different angle . apparently this works well by preferring also the large-@xmath24-large-@xmath4 region of our grid , except that there is no clear `` wall '' or `` boundary '' signaling a best value _ inside _ our grid . the same fit on the @xmath51 dataset yields a plot shown in the lower right corner of fig . [ fig : splittingsexotic ] . it is evident , that in this case the `` missing of a boundary '' is even more pronounced , i.e. , our grid is too small to make a definitive statement about whether or not one can find a minimum for @xmath53 in this case . however , one should not forget that with the light and strange quark masses being rather close together , so should be the parameter sets optimal for the description of the data . this comes from the fact that , while we do allow our model parameters to be different for each quark mass as a result of our philosophy that in our phenomenological rl approach effects beyond rl truncation could be absorbed by the effective interaction and its parameters including their values , the actual variation should be moderate and thus small from the light to the strange - quark case . we illustrate the two cases discussed above by plotting the corresponding spectra for the pair @xmath71 gev and @xmath72 gev@xmath44 for the isovector and @xmath48 gev and @xmath72 gev@xmath44 for the strangeonium cases in the left and right panels of fig . [ fig : isov - spectrum ] , respectively . in addition to what has been already provided in ref . @xcite , we thus predict the two lowest - lying @xmath0 states in the @xmath1 channel at @xmath73 and @xmath74 gev , respectively . before proceeding , a short discussion of the basic philosophy regarding our comparison to experimental data is in order : to make things simple and straight - forward , we assume the point of view that each light isovector meson state as given by the pdg @xcite corresponds to a bound state in our calculations . while this is straight - forward indeed , it may be oversimplified , since there is the possibility that @xmath6 states such as used herein are only the basis for the actual content of mesonic states seen in experiment . this is also one reason for us not to continue our study to , e.g. , the isoscalar case , since there many possible admixtures such as glueballs or molecules would have to be taken into account . we also neglect effects of hadronic or other decay mechanisms at this point , since inclusion of any of the effects mentioned here would be beyond the scope of the present study . in summary , our results must be taken with a grain of salt and seen as some sort of @xmath6-core states in a simple matching attempt to the experimental light - meson data set . at the same time , we stress that more informed and detailed comparisons based on matching principles which take into account both non resonant as well as resonant corrections are immediate next steps in a comprehensive study such as the one presented here . in particular , the covariant and well - constrained dsbse formalism provides an ideal candidate for providing reliable @xmath6 core states in such an improved future study . it is also important to note at this point that not all radial excitations used here are well - established experimentally like , e.g. , the @xmath75 in the @xmath64 channel or those for higher meson spin @xmath76 . nonetheless , we plot our basic overview using the values provided by the pdg @xcite . in this section we illustrate all splittings in our study of a particular meson state to the pion ground state . this is particularly interesting , since the pion ground state is fixed to its experimental value and , in addition , its position in our computational framework is extremely robust via the inherent satisfaction of the axial - vector ward - takahashi identity . thus , one has to pay special attention to this set of splittings in a fitting attempt , since varying or readjusting the current - quark mass in our framework has a drastically different effect on pion - related mass splittings on one hand and the rest on the other hand . in short , as immediately illustrated by the gell - mann - oakes - renner relation , the pion mass rises like the square root of the current - quark mass close to the chiral limit , while other meson masses rise linearly plus some small corrections . the set presented here is comprehensive . among these splittings , there are more prominent or dominant ones to observe . first , we mention the hyperfine splitting , which we already used at the beginning of our discussion for a first impression and example of the method and resulting statements . next , the scalar- and axialvector - to - pion splittings are relevant due to the lower - lying masses of these states , after which we complete the picture with the @xmath77 states . precisely , the list of splittings shown in fig . [ fig : pionsplittings ] is : @xmath60 $ ] , @xmath78 $ ] , @xmath79 $ ] , @xmath69 $ ] , @xmath80 $ ] , @xmath81 $ ] , @xmath82 $ ] , @xmath83 $ ] , @xmath84 $ ] . at first sight , the appearance of the plots for the various splittings in this figure is rather diverse ; one observes trends in virtually all possible directions . the hyperfine splitting , e.g. , shows a clear preference for low @xmath24 together with low @xmath4 even beyond our grid . the scalar - pseudoscalar splitting prefers high @xmath4 and a low @xmath24 inside our domain ; both axial - vector- , the exotic - vector- , and the pseudotensor - pseudoscalar splittings have their favored regions toward the center of our grid . the remaining tensor - related splittings for the @xmath85 and @xmath86 tend in opposite directions from each other . despite these different tendencies , the combined fit of all these splittings , whose favored pair of @xmath24-@xmath4 values should also already provide a reasonable match to the experimental spectrum , shows a favored `` valley '' from @xmath87 gev combined with our highest @xmath4 to @xmath88 gev combined with a central value for @xmath4 . this result is presented in the left panel of fig . [ fig : fitpion ] . the center panel of this figure shows a fit of 5 selected splittings of lower mass as an intermediate step , namely @xmath60 $ ] , @xmath79 $ ] , @xmath69 $ ] , @xmath82 $ ] , @xmath83 $ ] . finally , we reduce the fit to the three splittings @xmath60 $ ] , @xmath79 $ ] , @xmath69 $ ] and the corresponding result is shown in the right panel of fig . [ fig : fitpion ] . this refinement does not do much qualitatively except that it favors a value of @xmath87 gev more clearly with our two steps of reducing the numbers of splittings involved in the fit . to provide an comprehensive effort and picture , we present a combination of all splittings presented herein , both above as well as in the appendix below , in the three groups we used throughout each discussion , in fig . [ fig : fitall ] : all splittings combined are shown in the left panel , the combination of the selected splittings from all sections is presented in the center panel , and the combination of the reduced sets from all sections is shown in the right panel of fig . [ fig : fitall ] . the complete set of splittings leads to an inconclusive fit : minimal @xmath53 is found at two edges of our grid , namely for the combinations @xmath88 gev with @xmath89 gev@xmath44 and @xmath47 gev with @xmath72 gev@xmath44 . to check that this is a legitimate result , we need to actually have a look at the comparison , which is shown for the latter data set in the left panel of fig . [ fig : isov - spectrum - fit ] . apparently , the inclusion of too many ( all ) splittings in the fit did not properly capture the essence of the mass spectrum . one of the main reasons is certainly the importance and somewhat exceptional role of the pion ground state and all associated splittings . this peculiarity of the light - meson spectrum , however , is an essential feature and can not be ignored or underrepresented like in this attempt , if it is to be successful . as a comparison the center and right panels of fig . [ fig : fitall ] show a much more conclusive situation ; in fact , they look almost identical : both for the selected and the reduced set ( which contains a total of only eight splittings ) , the optimal set on our grid is @xmath87 gev with @xmath72 gev@xmath44 , for which the corresponding result for light isovector meson masses is plotted in the right panel of fig . [ fig : isov - spectrum - fit ] and provides a reasonable , albeit still unsatisfactory match to the experimental states . while having found the optimal parameter set at the boundary of our grid suggests an enlargement of the grid towards higher values of @xmath4 , the slope of the @xmath53 surface shows signs of a minimum very close to our values quoted here . thus and due to the straight - forward and simple matching strategy employed here , we postpone such a step to future investigations . in fact , an improved description of the data will probably rather be accomplished by exploring more parameters / degrees of freedom in the rl effective interaction or introducing corrections beyond rl truncation , both of which are beyond the scope of the present study . a final remark is in order : the comparison of the two panels in fig . [ fig : isov - spectrum - fit ] shows also , how two points rather close together in our @xmath24-@xmath4 parameter space can yield results of rather different fit quality . on close inspection of the various @xmath53 plots one indeed finds a significant change between the two points @xmath47 gev with @xmath72 gev@xmath44 and @xmath87 gev with @xmath72 gev@xmath44 in their majority . in a covariant dsbse framework the bsa of a state is more complex than a quantum - mechanical wave function . in particular , the defining properties of total angular momentum @xmath76 , parity @xmath9 and , if the meson is its own antiparticle , charge - conjugation parity @xmath90 are not restricted like in a quark - model setup @xcite . there , a @xmath6 meson content permits only certain sets of quantum numbers , which are usually referred to as `` conventional '' , while those unavailable are referred to as `` exotic '' . we list the conventional cases in tab . [ tab : nonexotic ] as they are decomposed with regard to the total @xmath6 spin @xmath46 and orbital angular momentum @xmath91 , together with the usual spectroscopic notation . in correspondence with the results presented herein , we include total angular momentum @xmath76 up to @xmath77 . l@llll type & @xmath56 & @xmath46 & @xmath91 & spectroscopic + peudoscalar & @xmath54 & @xmath92 & @xmath92 & @xmath93 + scalar & @xmath94 & @xmath95 & @xmath95 & @xmath96 + vector & @xmath55 & @xmath95 & @xmath92 & @xmath97 + & & @xmath95 & @xmath98 & @xmath99 + axialvector & @xmath64 & @xmath95 & @xmath95 & @xmath100 + & @xmath65 & @xmath92 & @xmath95 & @xmath101 + tensor & @xmath85 & @xmath95 & @xmath95 & @xmath102 + & & @xmath95 & @xmath103 & @xmath104 + pseudotensor & @xmath105 & @xmath92 & @xmath98 & @xmath106 + & @xmath86 & @xmath95 & @xmath98 & @xmath107 + while @xmath91 and @xmath46 are not observable , they are typical subjects in discussions of meson structure and make sense in our approach in connection with the covariant structures in the meson bsa and their relation to the pauli - lubanski operator , see @xcite and [ sec : oamdapp ] . in short , covariants in the bsa can be put in correspondence with @xmath108 orbital angular momentum in the meson s rest frame . contrary to the restrictions of the quark model , however , one finds @xmath9-wave content in , e.g. , pseudoscalar and vector mesons or @xmath3-wave content in scalar or axial - vector states . the latter also highlights the essential difference in terms of the construction of the relevant amplitude : a nonrelativistic quantum mechanical wave function is constructed via certain limited sets of @xmath46 and @xmath91 to have the correct @xmath56 , while a covariant bsa is defined via @xmath56 and the rest - frame @xmath46 and @xmath91 contents are determined dynamically by the solution of the homogeneous bse . here we present an orbital angular momentum decomposition ( oamd ) , i.e. , an analysis based on contributions of a state s covariants to the canonical norm @xcite of the bethe - salpeter wave function @xmath109 in eq . ( [ eq : bse ] ) . more precisely , in rl truncation the square of the norm is proportional to a derivative of the integral @xmath110 with respect to @xmath111 appearing as arguments of the inverse quark propagators . the conjugate amplitude is defined by @xmath112 where @xmath113 is the charge - conjugation operator in dirac space and the superscript @xmath114 denotes transposition . now we write @xmath109 as a sum of covariants with lorentz - scalar coefficient functions as @xmath115 where @xmath116 if @xmath117 and @xmath118 otherwise , and @xmath119 represents open lorentz indices , where necessary @xcite . having the four vectors @xmath8 , @xmath9 , and @xmath120 at our disposal , we can construct the four standard dirac covariant structures @xmath121\;,\label{eq : sccov}\end{aligned}\ ] ] where @xmath122 represents the unit matrix in dirac space and the transversely projected relative momentum @xmath123 is defined via @xmath124 which is apparently simplified by using the unit momentum in @xmath9 direction instead , @xmath125 . these four covariants constitute a scalar meson ; multiplied by a factor of @xmath126 each , they define a pseudoscalar bsa . covariants for spin-@xmath95 mesons are necessarily transverse with respect to @xmath9 as required of a massive spin-@xmath95 boson and have a bsa with one open lorentz index . one can directly construct the corresponding vector - meson covariants by multiplying both @xmath127 and @xmath128 by each of the four covariants given above in eq . ( [ eq : sccov ] ) , yielding a total of eight structures . axialvector covariants are obtained from these eight by multiplication with a factor of @xmath126 each in analogy to the scalar / pseudoscalar case . the general construction principle for bsas with arbitrary @xmath76 can be found in @xcite . regarding the importance of quark orbital angular momentum in a state , we compute the contributions of each combination of covariants from the sum as defined in eq . ( [ eq : bswf ] ) in the factors @xmath129 and @xmath109 to the norm in eq . ( [ eq : cnorm ] ) . to identify @xmath91 via the @xmath130 , we have computed the eigenvalue of each covariant under the operator @xmath131 as it is defined in [ sec : oamdapp ] . a complete corresponding set of covariants for the cases @xmath132 , @xmath133 , @xmath134 , and @xmath135 is given in this appendix in eqs . ( [ eq : oamd0 + ] ) - ( [ eq : oamd1 + ] ) . simply put , each occurrence of the relative quark - antiquark momentum @xmath8 corresponds to one unit of orbital angular momentum @xmath91 . in this way , among the scalar covariants there are @xmath98 @xmath3-wave and @xmath98 @xmath9-wave . in the pseudoscalar case the situation is the same . for the vector and axialvector cases , there are @xmath98 @xmath3-wave , @xmath136 @xmath9-wave , and @xmath98 @xmath4-wave covariants . with these definitions we are prepared for our analysis : inserting the sum of eq . ( [ eq : bswf ] ) for both @xmath109 and @xmath129 in eq . ( [ eq : cnorm ] ) we obtain @xmath137 terms , each with a specific combination of one covariant from @xmath109 and one from @xmath129 , which we denote via the corresponding pair of numbers assigned to the covariants in [ sec : oamdapp ] for each particular case . squaring the contribution from each combination of covariants and normalizing the sum of all squared contributions to @xmath95 , we have a quantitative as well as illustrative tool at our disposal . the corresponding numbers are plotted in fig . [ fig : oamdall ] for all ground states directly accessible to our solution via the homogeneous bse , i.e. , for @xmath138 , including those with exotic quantum numbers . we note here that mesons with exotic quantum numbers appear naturally in our covariant approach as radial excitations in a certain @xmath139 channel . orbital angular momentum with respect to the @xmath108 relative momentum as investigated here has recently been shown to play no role in distinguishing between exotic and conventional states @xcite , where also an in - depth discussion of issues related to exotic quantum numbers can be found . thus , it is not surprising to see rather strong similarities of , e.g. , the @xmath54 and @xmath140 or the @xmath55 and @xmath1 ground states in fig . [ fig : oamdall ] . four our purpose , we simply regard the states presented here as the ground states in their respective @xmath56 channel for the sakes of demonstration and completeness . we also provide the numbers for the plots of fig . [ fig : oamdall ] for each state in tab . [ tab : orbital ] for easy reference and a thorough basis for the discussion together with the corresponding experimental isovector meson , where available . note that numbers are rounded to one decimal . l@rrrrrrr name & @xmath141 & @xmath3 & @xmath3-@xmath9 & @xmath9 & @xmath9-@xmath4 & @xmath4 & @xmath3-@xmath4 + @xmath32 & @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath146 & @xmath146 + @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath146 & @xmath146 & @xmath146 + @xmath151 & @xmath152 & @xmath153 & @xmath154 & @xmath155 & @xmath146 & @xmath146 & @xmath146 + @xmath146 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath146 & @xmath146 & @xmath146 + @xmath2 & @xmath160 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath145 + @xmath166 & @xmath167 & @xmath168 & @xmath169 & @xmath165 & @xmath145 & @xmath145 & @xmath145 + @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath145 + @xmath177 & @xmath178 & @xmath165 & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath145 + a comment on the model parameters used to obtain the data in figs . [ fig : oamdall ] and [ fig : oamd1 ] is in order . in fig . [ fig : oamdall ] we present results for the best - fit data set with @xmath87 gev and @xmath72 gev@xmath44 . higher states are not accessible to a direct solution via the homogeneous bse due to the fact that quark - propagator singularities can not easily be taken into account numerically in our setup , which results in a maximum meson mass obtainable directly . one can ameliorate this situation and estimate e.g. bound - state masses by extrapolation techniques @xcite , solution of the inhomogeneous bse @xcite , or making _ anstze _ for the @xmath183 in eq . ( [ eq : bswf ] ) together with the quark dressing functions and taking into account the effect of singularities explicitly @xcite . however , in all these cases , the direct connection to qcd via explicit solution of the quark dse and the homogeneous meson bse is lost at least to some extent . in addition , the above techniques for an oamd analysis are based on the normalization of the on - shell bound - state amplitude . our results provide a picture largely consistent with expectations . all ground states are predominantly represented by their expected orbital angular momentum contribution : the pseudoscalar- and vector - meson ground states are @xmath3-wave , while the scalar- and axialvector - meson ground states are @xmath9-wave , including their exotic counterparts . naturally , the next interesting question is the oamd for excitations in a given @xmath56 channel . to achieve this and overcome the limit introduced by singularities in the quark propagators discussed above , we investigate a different set of model parameters , namely @xmath184 gev and @xmath185 gev@xmath44 , where a number of radial excitations is directly accessible for investigation via the homogeneous bse @xcite . as a prominent and instructive example , we investigate the vector meson case , for which the results are plotted in fig . [ fig : oamd1 ] and the corresponding numbers are collected in tab . [ tab : orbital1 ] together with possible experimental assignments . l@lrrrrrr name & @xmath186 & @xmath3 & @xmath3-@xmath9 & @xmath9 & @xmath9-@xmath4 & @xmath4 & @xmath3-@xmath4 + @xmath2 & @xmath160 & @xmath187 & @xmath162 & @xmath188 & @xmath189 & @xmath165 & @xmath145 + @xmath190 & @xmath191 & @xmath188 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath165 + @xmath196 & @xmath197 & @xmath198 & @xmath169 & @xmath165 & @xmath165 & @xmath145 & @xmath145 + @xmath199 & @xmath167 & @xmath200 & @xmath201 & @xmath165 & @xmath165 & @xmath145 & @xmath145 + @xmath202 & @xmath203 & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath145 + @xmath209 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath169 & @xmath165 & @xmath145 + in particular , the first row in fig . [ fig : oamd1 ] shows the ground state as well as the first and second excitations in the @xmath2-meson channel . comparison of the @xmath55 ground - state plots in figs . [ fig : oamdall ] and [ fig : oamd1 ] as well as the corresponding numbers in tabs . [ tab : orbital ] and [ tab : orbital1 ] shows that the oamd is robust under changes of the model parameters @xmath24 and @xmath4 within the reasonable range employed herein such that conclusions from this different set of parameters can be expected to also hold for our best - fit parameter choice . we find the @xmath2 ground state with a pure @xmath3-wave component of roughly @xmath214 % . pure other components are almost negligible and mixes provide the remaining contributions to the canonical norm . for the first excitation in this channel , matched to the experimental @xmath190 , the situation is a bit more complex : the state has a predominant @xmath4-wave component of @xmath215 % . pure @xmath3-wave and @xmath9-wave contributions are small and the remaining contributions come from mixed terms . this is interesting to compare , e.g. , to recent studies in lattice qcd @xcite where the @xmath190 is found to be predominantly @xmath4-wave . while this is in contrast to the traditional quark model , where this meson is expected to be @xmath3-wave , we do not support the interpretation that the @xmath190 is not a radial @xmath2 excitation due to its @xmath4-wave content , since these states belong to the same tower of @xmath55 excitations in our covariant approach in a natural way and the character of an excitation is not simply discernable into _ orbital _ or _ radial _ like in nonrelativistic quantum mechanics . for the second excitation in the @xmath55 channel we find a predominantly @xmath3-wave oamd , much like for the @xmath2 with the pure @xmath3-wave percentage even higher at @xmath216 % . in this case , the experimental assignment in terms of the next excitation in the channel is not clear , however , since the @xmath217 needs confirmation ; the next higher - lying state in this channel is the @xmath218 . we thus leave this correspondence open for the sake of simplicity . the presence of both @xmath3- and @xmath4-wave in this channel suggests more excited states than , e.g. , in the pseudoscalar case . generally speaking , in our approach there is no reason why the excitations of the @xmath2 found in experiment should not be arbitrary mixtures of @xmath3- and @xmath4-waves , a feature appearing naturally also in our rl treatment . other interpretations of the excited states found in experiment in the @xmath55 channel include also hybrid admixtures @xcite . the notion of hybrids makes use of explicit gluonic degrees of freedom , which is content implicit in our approach via the construction and degrees of freedom contained in a covariant quark - bilinear bethe - salpeter amplitude @xcite . thus , there is no contradiction of our results with such interpretations . further information regarding an identification of our states with regard to experimental data will come from a study of hadronic decay widths along the lines of @xcite , which is work in progress . this will also allow a comparison with quark - model results and interpretation , where hadronic partial widths are decisive elements pro or contra hybrid admixtures for excited @xmath2 states @xcite . to complement the vector - meson picture , we present the ground and first excited states for the exotic @xmath1 , as well as the strange ground state , i.e. , the @xmath209 . the oamd for the two lowest - lying @xmath1 is very similar to the @xmath55 case . while the identification of these two @xmath1 states with the experimental @xmath199 and @xmath219 is debatable , e.g. , @xcite , we find a predominantly @xmath3-wave ground state and a @xmath4-wave first excitation . the @xmath209 looks very similar to the @xmath2 as well , which is not surprising , given the context of @xmath68 flavor symmetry in qcd . to complete this section , we present all oamd results combined in a single plot in fig . [ fig : orbitalo ] . a similar analysis for heavy quarkonia will be provided elsewhere . overview of oamd for all states presented in figs . [ fig : oamdall ] and [ fig : oamd1 ] . bars without hatch correspond to the best fit model parameter set used in fig . [ fig : oamdall ] , while hatched bars correspond to the secondary parameter set used in fig . [ fig : oamd1 ] . colors like in fig .. [ fig : oamdall ] . ] we present a covariant study of the light isovector meson spectrum . via various comparisons of our calculated results for mass splittings to experimental data , we investigated the role and importance of the two characteristic features of the effective interaction used herein , which are the strength of the interaction s intermediate - momentum part represented by the parameter @xmath4 on the one hand , and the inverse effective range of the interaction represented by the parameter @xmath24 and easily visible as the peak position of the effective interaction in momentum space . at the beginning of the analysis we revisit the problem of isovector exotic vector states , which was already discussed previously @xcite and add an in - depth discussion of the situation , presenting also predictions for the lowest - lying @xmath0 states in the @xmath1 channel . then , in a step - by - step fashion , we investigate our results for radial , orbital , other , and pion - related isovector - meson mass splittings and their dependence on the two parameters in our model . the following conclusions may be drawn from our results : first , for most individual splittings in our investigation , the dependence on @xmath24 was more pronounced and important for a good match to experimental data than the one on @xmath4 in the sense that for the optimal @xmath24 region the dependence on @xmath4 was very small . second , since our rl - truncated approach somewhat simplifies the problem and ignores both nonresonant and resonant effects expected beyond rl , achieving a good fit to experimental light - meson data would not be a sign of an accurate description of the physics mechanisms behind those states . it is thus also unsurprising that fitting all available splittings does not provide the best match to the data . in order to find a better match , we select sets of splittings from each category in two steps and , in the end , combine them as such to arrive at an overall result . our general impression is that choosing a small but reliable set of splittings provides the best results . third , the pion emerges as a critical state for the fitting process in the sense that its mass is fixed in our study due to its protection via the axial - vector ward - takahashi identity . as a result , pion - related splittings appear to be substantially more important than others for achieving a reasonable match of our calculated results to experimental data . finally , regarding preferred parameter values from our investigations we find that @xmath24 values lower on our grid together with high @xmath4 values provide the best matching result . this means an inverse effective range of the model interaction of @xmath182 gev , which is at the upper end of the domain orginally investigated by maris and tandy @xcite , but with an increased overall strength of the interaction . this is different from our results obtained previously for heavy quarkonia @xcite , where the resulting inverse effective range was @xmath220 gev . this means that for our setup the important features of the effective interaction are shorter range in heavy quarkonia and longer range in the light - quark sector . in our analysis of the orbital angular momentum decomposition of the covariant bethe - salpeter wave functions , the ground states follow intuitive patterns . for excited states in the @xmath55 channel , concretely for excitations of the @xmath2 meson , our results for the @xmath190 show a predominant @xmath4-wave component , which is similar to results from lattice qcd and in contrast to a @xmath108 state in the quark model . quark model interpretations of the @xmath2 excitations in terms of hybrid admixtures are not contradicted , however , since such contributions are implicit in our approach . further insight is expected from upcoming studies of hadronic partial decay widths of these states . as an outlook we note that the possibilities of making the effective interaction more general and flexible have still not been exhausted and present the most promising path for making a study such as ours more successful in the present rl setup . steps beyond the current truncation are promising as well , with a clear emphasis on nonresonant corrections , in particular in the light - quark sector . we acknowledge helpful discussions with m. pak , c. popovici and r. williams . in sec . [ sec : oamd ] we discuss quark orbital angular momentum in the bsa based on the covariants present in each particular state . in this appendix we detail the connection to the pauli - lubanski operator and , in particular , the square of the orbital - angular - momentum operator . such a discussion of particle spin rests on a general , relativistic footing , where the angular momentum of a particle s state is obtained via the pauli - lubanski vector operator @xmath221 the operators @xmath76 and @xmath9 are built from the @xmath222 generators of the poincar group . the square @xmath223 is one of the two casimir operators of the poincar group and , in particular if @xmath9 is taken normalized to one , yields the total angular momentum @xmath224 as the eigenvalue of a relativistic quantum state . in our case , we deal with a covariant quark - bilinear bsa with the main focus on quark orbital angular momentum , which can be obtained by splitting @xmath225 and computing the corresponding eigenvalues @xmath226 of the operator @xmath227 acting on the covariant dirac structures in the bsa . in the meson s rest frame @xmath91 can then be interpreted in the usual way , i.e. , @xmath228 corresponds to an @xmath3-wave , @xmath229 to a @xmath9-wave , @xmath230 to a @xmath4-wave , etc . @xmath231 is given by the expression @xmath232 where @xmath8 and @xmath125 are the quark relative four - momentum and unit total meson momentum , respectively , see , e.g. , @xcite . the dirac part is trivial and we omit it for simplicity . @xmath233 , in turn , is given by @xmath234 where the following definitions and relations are helpful : @xmath235 is the transversely projected relative momentum with respect to @xmath9 , whose square is @xmath236 in addition , analogously , @xmath237 we list the covariants together with their @xmath131 eigenvalues in eqs . ( [ eq : oamd0 + ] ) - ( [ eq : oamd1 + ] ) below . for completeness as well as easy reference , we also provide the eigensign @xmath238 of each covariant under charge conjugation defined in analogy to eq . ( [ eq : cc ] ) , i.e. , @xmath239 where again @xmath113 is the charge - conjugation operator in dirac space and the superscript @xmath114 denotes transposition . applying eq . ( [ eq : lsquared ] ) to the standard four linearly independent scalar dirac covariants as defined in sec . [ sec : oamd ] , one obtains the following correspondence : @xmath240\qquad & 1 & + \end{array}\ ] ] where @xmath122 represents the unit matrix in dirac space . since the dirac part of @xmath131 is trivial , multiplication by @xmath126 does not change any of these correspondences , which yields the pseudoscalar assignments : @xmath241\qquad & 1 & + \end{array}\ ] ] for the vector case , one has eight covariants and finds the following correspondence : @xmath242 & 1 & + \\ 4 & \gamma^{\mu ( t ) } \;[\gamma\cdot q,\gamma\cdot \hat{p}]-2\,q^{\mu ( t)}\;\gamma\cdot \hat{p } & 1 & - \\ 5 & q^{\mu ( t)}\;\mathbf{1 } & 1 & - \\ 6 & q^{\mu ( t)}\;\gamma\cdot \hat{p } & 1 & + \\ 7 & q^{\mu ( t)}\;\gamma\cdot q^{(t)}-\frac{1}{3}q^{(t)2}\;\gamma^{\mu ( t ) } & 2 & - \\ 8 & q^{\mu ( t)}\;[\gamma\cdot q,\gamma\cdot \hat{p } ] -\frac{1}{3}q^{(t)2}\;[\gamma^{\mu ( t)},\gamma\cdot \hat{p}]\quad & 2 & - \end{array}\ ] ] multiplying each covariant again by @xmath126 , one arrives at the axialvector case , where @xmath243 & 1 & + \\ 4 & \gamma_5\;\gamma^{\mu ( t ) } \;[\gamma\cdot q,\gamma\cdot \hat{p}]-2\,q^{\mu ( t)}\;\gamma_5\;\gamma\cdot \hat{p } & 1 & + \\ 5 & q^{\mu ( t)}\;\gamma_5 & 1 & - \\ 6 & q^{\mu ( t)}\;\gamma_5\;\gamma\cdot \hat{p } & 1 & - \\ 7 & q^{\mu ( t)}\;\gamma_5\;\gamma\cdot q^{(t)}-\frac{1}{3}q^{(t)2}\;\gamma_5\;\gamma^{\mu ( t ) } & 2 & + \\ 8 & q^{\mu ( t)}\;\gamma_5\;[\gamma\cdot q,\gamma\cdot \hat{p } ] -\frac{1}{3}q^{(t)2}\;\gamma_5\;[\gamma^{\mu ( t)},\gamma\cdot \hat{p}]\quad & 2 & - \end{array}\ ] ] splittings between radial excitations and ground states of the various quantum numbers in the meson spectrum play a dominant role in a fitting strategy . we investigate the splittings between ground state and first radial excitation individually for all @xmath56 with @xmath244 and add the pseudoscalar second - to - first radial excitation splitting as a testing case for higher excitations ; the resulting plots for @xmath53 as described above are collected in fig . [ fig : radialsplittings ] . from the collection of comparisons in fig . [ fig : radialsplittings ] one sees a few interesting features : first of all , it appears that for radial splittings the dependence on @xmath4 is minor compared to @xmath24 , because in the regions of optimal omega , any @xmath4-dependence is suppressed . investigating which values of @xmath24 favor a good match to the experimental splittings , we arrive at lower @xmath24 for @xmath245 , @xmath94 , @xmath86 , and also @xmath1 ; the latter case is shown above in the right panel of the upper row in fig . [ fig : splittingsexotic ] . higher @xmath24 is favored by @xmath65 , while the situation is inconclusive for @xmath85 , @xmath105 , and the second - to - first @xmath54 radial splitting . @xmath64 as well as @xmath55 provide optimal @xmath24 values inside our domain of study . furthermore , we note at this point that the pseudoscalar channel is special due to the particular role of the pion as the pseudo - goldstone boson in qcd . pion - related splittings are thus discussed in more detail below in sec . [ sec : pion ] . for a more serious and concrete attempt at a good phenomenological description of the meson spectrum with our present setup we select only a few reliable ones from the set of radial splittings together with others provided below . at this point , it is insightful to compare the following fitting results collected in fig . [ fig : fitradial ] : a combination of all eight radial splittings between ground and first radial excited states shown in fig . [ fig : radialsplittings ] together with the same splitting in the exotic vector channel is plotted in the left panel of fig . [ fig : fitradial ] . for the center panel in the same figure we reduced the set and remain only with the radial splittings in the @xmath54 , @xmath55 , @xmath94 , @xmath85 , and @xmath1 channels . the right panel shows a fit with only @xmath54 and @xmath55 radial splittings remaining . the characteristics of the three panels in fig . [ fig : fitradial ] is surprisingly different . for all splittings combined ( left panel ) the region with the lowest @xmath53 seems to lie outside or at the borders of our grid for the most part . reducing the set ( center panel ) , we observe a general trend towards lower values of @xmath24 almost independent of @xmath4 . for the pseudoscalar - vector - only set , however , we see the best fit inside our grid for @xmath87 gev and medium values of @xmath4 . interpreting these results , one may refer to arguments that non resonant corrections to rl truncation are smaller for the pseudoscalar and vector channels than the others @xcite and thus expect the corresponding variant of the fit to be most appropriate regarding a @xmath6-core description . demanding an overall description on the basis of radial splittings alone , however , we would have to conclude that the @xmath24-@xmath4 grid we use is too small . another interesting group of splittings among meson states with various quantum numbers is the set of orbital splittings , i.e. , those corresponding to a change in the internal quantum number corresponding to quark - antiquark orbital angular momentum in the simple quark - model picture of the @xmath6 state . while our covariant approach is more general and the bethe - salpeter amplitudes are more complex than a quantum mechanical wave function , we can still interpret our splittings between different sets of @xmath56 according to their corresponding quark - model deconstructions . investigating the various possibilities , we arrive at a comprehensive set , for which in our notation introduced above the labels are : @xmath80 $ ] , @xmath246 $ ] , @xmath247 $ ] , @xmath248 $ ] , @xmath249 $ ] , @xmath250 $ ] , @xmath83 $ ] , @xmath251 $ ] , @xmath252 $ ] . among the most prominent of these for our purposes is certainly the splitting between the scalar and vector mesons , since experimentally these are , after the pion , the lowest lying meson masses . we therefore include not only the corresponding ground - state splitting , but also the one for the first radial excitations of both the scalar and vector meson channels . interestingly enough , the @xmath190 and the @xmath253 are roughly degenerate , which makes this comparison more than a simple exercise or test of the method for higher excited states . we find for the particular case of the scalar - vector splittings that the ground - state splitting shows an optimal range for @xmath88 or @xmath220 gev and slightly adjusted central values of @xmath4 , while the excited - state case is not particularly well reproduced . in general , again , we observe that for the optimal @xmath24 range or case in each individual plot , it appears that the dependence on @xmath4 is secondary much like for the radial splittings . of similar interest are the splittings of the ground states in the axial vector and tensor @xmath254 channel to the vector channel for reasons of consistency and the still rather low values of their masses . for these we find optimal @xmath24 values inside our grid and thus a good basis for our comparison . from the rest of the set , a convincing case comes also from the pseudoscalar - pseudotensor splitting . as for trends towards certain values of @xmath24 and @xmath4 we observe that lower values of @xmath24 outside our grid are only favored by the pseudoscalar - axialvector splitting , and higher values outside our grid seem to be favored by the ground - state splitting of the @xmath255 channels . inconclusive situations appear only for the radially excited splitting and the tensor - pseudotensor splitting , the rest provides a solid base for our fitting strategy . in the same manner as above for the radial splittings , we present three fitting attempts also for the set of orbital splittings in fig . [ fig : fitorbital ] : all splittings , whose @xmath24-@xmath4 dependence is plotted above in fig . [ fig : orbitalsplittings ] are combined to yield the @xmath53 shown in the left panel of fig . [ fig : fitorbital ] . for the second plot shown in the center panel of fig . [ fig : fitorbital ] we choose the following five splittings : @xmath80 $ ] , @xmath246 $ ] , @xmath248 $ ] , @xmath249 $ ] , @xmath250 $ ] . finally , we reduce the fit to @xmath80 $ ] , @xmath246 $ ] and the result is shown in the right panel of fig . [ fig : fitorbital ] . the result from the combined fit to all orbital splittings depicted in the left panel of fig . [ fig : fitorbital ] shows an optimal @xmath24 of @xmath205 gev with central values of @xmath4 , which is not changed by reducing the number of splittings and plotting the center panel of the same figure . the right panel shows only a slight shift towards @xmath71 gev for the minimal fitting set of splittings and thus a very stable and uniform result . in this section we move towards completing the picture by adding a set of splittings which falls neither in the radial nor the orbital category and was not mentioned above yet ( like , e.g. , the hyperfine splitting ) . the set investigated here and depicted in fig . [ fig : othersplittings ] is @xmath79 $ ] , @xmath256 $ ] , @xmath257 $ ] , @xmath258 $ ] , @xmath259 $ ] , @xmath260 $ ] , @xmath261 $ ] , @xmath262 $ ] . this is led by the pseudoscalar - scalar splitting which again refers to states among the lowest lying in mass . again , we have included the analogous splitting of the corresponding first radial excitations of these states . the result for the ground - state splitting shows a significant @xmath4 dependence for the optimal @xmath24 domain for the first time , namely there is a significant trend towards the low-@xmath24-high-@xmath4 corner . all other plots in the set show a minor @xmath4 dependence in each optimal @xmath24 domain , once again . the next interesting set of splittings concerns those among the typical @xmath229 states , i.e. , the ground states in the @xmath94 , @xmath64 , and @xmath85 channels . here we find a uniform and clear trend towards @xmath24 values outside the lower end of our grid . the combined fitting attempts for the set of splittings presented in this section is shown in fig . [ fig : fitother ] in the usual fashion : the left panel shows the result for all splittings combined , the center panel shows the result for an intermediate set , namely the four splittings : @xmath79 $ ] , @xmath257 $ ] , @xmath258 $ ] , @xmath259 $ ] , and finally , we reduce the fit to @xmath79 $ ] , @xmath258 $ ] and display the result in the right panel of fig . [ fig : fitother ] .

this paper deals with the thermodynamic properties of a phantom field in a flat friedmann - robertson - walker universe . general expressions for the temperature and entropy of a general dark - energy field with equation of state are derived from which we have deduced that , whereas the temperature of a cosmic phantom fluid ( ) is definite negative , its entropy is always positive . we interpret that result in terms of the intrinsic quantum nature of the phantom field and apply it to ( i ) attain a consistent explanation for some recent results concerning the evolution of black holes which , induced by accreting phantom energy , gradually loss their mass to finally vanish exactly at the big rip , and ( ii ) introduce the concept of cosmological information and its relation with life and the anthropic principle . some quantum statistical - thermodynamic properties of the quantum quantum field are also considered that include a generalized wien law and the prediction of some novel phenomena such as the stimulated absorption of phantom energy and the anti - laser effect .

astro-ph0407421

phantom energy remains for the moment a theoretical possibility [ 1 ] with potential application to describe super - accelerated evolution in both the primordial and late universe . but it is not just that . concerning late evolution of the universe , the possibility that the equation of state governing current cosmic evolution would correspond to phantom energy is still not at all excluded [ 2 ] . actually , if the accelerating expansion of the universe turned out to be not due to the existence of a positive cosmological constant , then presently supplied cosmic data appear to favor phantom energy over quintessence models for positive internal energy fields [ 3 ] . from a theoretical point of view , the study of phantom energy is by itself a very interesting subject . it actually bears some resemblance with the study of black holes in the early seventies by its intrinsic interest and in what then black holes were considered to be on the borderline of being or not actual objects . it is therefore not surprising that phantom energy had recently received a great deal of attention [ 4 ] . phantom energy can be defined in all present models for dark energy by simply introducing in such models a spontaneous violation of the dominant energy condition ; that is , if e.g. we consider an equation of state @xmath0 , then the phantom appears when we enforce @xmath2 , which in turn implies that the parameter @xmath1 . thus , in order to check the occurrence of phantom energy in the universe , it suffices to determine what the equation of state of the universe is . actually cosmic phantom energy keeps a fairly large set of rather weird properties , including [ 4 ] the above - mentioned violation of the dominant energy condition , @xmath3 , naive superluminal sound speed , increasing energy density with cosmic time , and ultimately the emergence in a finite time in the future of what has been dubbed big rip [ 5 ] , which is a true curvature singularity where all existing particles are coherently ripped apart . the latter prediction actually was already advanced in 1986 by barrow , galloway and tipler [ 5 ] . in this paper we shall add some extra weirdness concerning the thermodynamic properties of phantom energy ; more precisely , we show that phantom energy is characterized by a negative temperature and that this implies the phantom to be quantized , even though it is the largest possible system . this overall quantum nature of the cosmic phantom fluid comes about as a consequence from preserving the weak energy condition , and can be shown by either taking the phantom stuff to be made up of super - light axions [ 7 ] , or defining the phantom spacetime to be euclidean . thus , the aim of the present paper is at studying the thermodynamic properties of cosmic phantom energy . the influence that the presence of black holes may have on the phantom field is also explored . the main results are : ( i ) we confirm and interpret the result obtained by babichev , dokuchaev and eroshenko [ 8 ] that all black holes in the universe loss their mass to vanish at the big rip ; ( ii ) since the internal phantom energy is negative , whereas the phantom temperature is definite negative , and hence hotter than any other sources in the universe [ 9 ] , its entropy is always positive , even though holding of the second law is not guaranteed by quantum - mechanical reasons ; ( iii ) relative to the observable matter in the universe , the cosmic phantom field can be regarded as a cosmological source of information and negative entropy ; and ( iv ) in the presence of phantom radiation , quantum level systems undergo novel radiative processes that include a stimulated absorption phenomenon by which phantom radiation is attenuated . the paper can be outlined as follows . in sec . ii we consider and extend the process of phantom energy accretion by black holes first studied by babichev , dokuchaev and eroshenko [ 8 ] , interpreting it in terms of a new thermodynamic description of a phantom fluid which is developed and includes a negative temperature and a positive entropy . a discussion of the properties and meaning of the cosmological information induced by the presence of dominating phantom energy in the universe , as well as the implications that it may have on the emergence of life and the anthropic principle , is included in sec . iii . by interpreting the phantom field as a radiation field , we consider some of the quantum statistical - thermodynamic properties of the resulting description , including a generalized wien spectrum ( sec . iv ) and a derivation of the generalized planck law based on introducing novel radiative processes ( sec . v ) . we finally conclude and add some more comments in sec . phantom energy has been criticized by several authors [ 10 - 12 ] . the main difficulty stemming from such criticisms concerns the issue of phantom stability and this could still be circumvented if an axion model is considered for the phantom field [ 7 ] . indeed , the large amount of papers on the phantom subject that have appeared [ 4 - 6 ] before and after references [ 10 - 12 ] reflects the fact that most of the criticisms can actually be regarded as manifesting weird phantom properties that could nevertheless be accommodated into the current and future evolution of the universe without contradicting observations . among such properties , an intriguing recent result is the discovery by babichev , dokuchaev and eroshenko [ 8 ] that all black holes in a universe filled with a fluid violating the dominant energy condition ( i.e. a phantom - energy fluid inducing a big rip singularity in the future ) ) will steadily loss all of their mass to fully disappear , all at once , at the big rip , no matter their initial mass or the moment at which they were formed . in fact , by extending the classical theory of michel [ 13 ] , these authors found that , as a result of dark energy accretion , the mass @xmath4 of a black hole in a universe filled with a general quintessence scalar field describable by means of a fluid with equation of state @xmath0 , varies at a rate given by [ 8 ] @xmath5 where @xmath6 is a dimensionless positive constant and @xmath7 . using then the constant equation of state @xmath0 for the fluid described by the quintessence field , eq . ( 2.1 ) can finally be re - written as @xmath8 thus , since in all conceivable situations @xmath9 should be satisfied , accretion of dark energy onto a black hole leads the mass of this black hole to increase if @xmath10 and , as a consequence from the negative value of the internal energy of the phantom field , to a mass loss in the case that @xmath1 . the stability of the schwarzschild - de sitter universe [ 14 ] to the dark - energy accretion process is ensured by the fact that a positive cosmological constant corresponds to @xmath11 for which the rate @xmath12 . of course , in this case there will still be a continuous loss of black hole mass due to hawking radiation , a process which is not included in the present formalism . for the flat geometry which our universe appears to satisfy , it has been obtained [ 15 ] that the most general expression of the scale factor of a universe filled with such a general quintessence field is given by @xmath13},\ ] ] in which @xmath14 is taken as the initial value of the scale factor at the onset of the accelerating regime . this solution satisfies the friedmann equation for flat geometry and the conservation law for cosmic energy . if we , moreover , re - interpret the constant @xmath15 as the value of the scale factor at the onset of the radiation domination era for @xmath16 , then re - scaling time @xmath17 so that @xmath18 , we in fact obtain from eq . ( 2.3 ) the expression of the scale factor for a decelerating radiation - dominated universe , i.e. @xmath19 . ( 2.3 ) furthermore tells us that for the interval @xmath20 the scale factor steadily increases in an accelerating fashion , with @xmath17 tending to infinity as @xmath21 . from the very onset of the phantom energy regime @xmath1 [ 3 ] assumed to occur at @xmath22 , the scale factor ( 2.3 ) predicts a super - accelerated expansion along that regime which finally reaches the so - called big rip curvature singularity as @xmath17 approaches the finite time value @xmath23 after the big rip , as @xmath24 , the universe would start a period of continuous contraction where its size tended to vanish as @xmath21 . the super - accelerated expanding phase and the contracting phase would however be disconnected to each other because the big rip is a true curvature singularity . if , as it is assumed throughout this paper , we set @xmath25 constant , the integration of the cosmic conservation law for energy , @xmath26 [ 16 ] , in this case leads to an expression for the dark energy density @xmath27 with @xmath28 a constant . thus , the energy density consistently becomes constant for the cosmological constant case @xmath11 , it steadily decreases with the scale factor for @xmath29 , and rather surprisingly increases with @xmath30 for @xmath1 . once the scale factor @xmath31 and the energy density @xmath32 have been obtained , the field theory associated with the dark energy fluid through the definitions of the pressure and energy density , @xmath33 , @xmath34 , can be solved ( i.e. expressions for @xmath35 and @xmath36 can be obtained ) . we in this way get @xmath37 @xmath38 the phantom regime is derived when we let @xmath39 and @xmath1 [ 7 ] . it can be seen that both @xmath40 and @xmath41 diverge at the big rip . inserting now eqs . ( 2.3 ) and ( 2.5 ) into eq . ( 2.2 ) and integrating we finally obtain the black hole mass time - evolution equation [ 17 ] @xmath42 where @xmath43 is the initial mass of the black hole , @xmath44 and @xmath45 . ( 2.6 ) actually shows the same predictions as those which were first deduced in ref . that is : ( i ) for @xmath10 , @xmath4 monotonically increases with time @xmath17 , tending to a maximum value @xmath46 as @xmath21 . this result comes about because quintessence fields with @xmath10 have positive energy which , when accreted onto the black hole , makes the positive mass of this black hole to increase . ( ii ) if @xmath11 , the black hole mass remains constant , i.e. the black hole does not accrete any energy from vacuum simply because there then is anything like a quintessence vacuum field , but a cosmological constant instead . ( iii ) finally , and more importantly , such as it was pointed out by babichev , dokuchaev and eroshenko [ 8 ] , if @xmath1 , then @xmath4 steadily decreases with time @xmath17 and tends to vanish as @xmath17 approaches the big rip singularity at time @xmath47 . in order for accounting for such a highly unconventional behaviour one must resort to the feature that the internal phantom energy is negative definite and therefore , when it is accreted onto black holes , it subtracts rather than adds on their total energy . moreover , near the big rip singularity , we have that the mass of any black hole tends to be @xmath48 which remarkably does not depend on the initial mass of the black hole . according to babichev , dokuchaev and eroshenko [ 8 ] , that result means that all black holes in a universe filled with phantom energy will tend to be equal as the big rip is approached , and that phantom - energy accretion prevails over hawking radiation , at least if quantum - gravity effects are not taken into account . in their original derivation , babichev et al . [ 8 ] obtained exactly the same conclusions by using a simpler expression for the scale factor . one could still wonder , what happens with black holes after the big rip . according to eq . ( 2.6 ) , just after the big rip the mass of the black holes would start increasing and tend to the finite maximum value given by eq . ( 2.7 ) as @xmath21 , such as it happens in the case of dark energy with @xmath10 . the memory of the initial mass would thus be recovered , even though the regions before and after the big rip are mutually disconnected . further generalization of these results can be achieved by considering the case where a positive cosmological constant @xmath49 is added to the dark energy fluid . this situation would describe a schwarzschild - de sitter spacetime embedded in dark energy . if we set a vanishing initial time , @xmath50 , the scale factor @xmath31 then reads [ 15 ] @xmath51}\left(e^{3(1+\omega)\sqrt{\lambda}t/2}- ce^{-3(1+\omega)\sqrt{\lambda}t/2}\right)^{2/[3(1+\omega ) ] } , \ ] ] where @xmath52 with @xmath15 the assumed initial value of the scale factor at the onset of the phantom energy dominance . since we always have @xmath53 , a big rip singularity for the phantom regime where @xmath1 is also predicted in the presence of a cosmological constant . actually , that singularity takes in this case place at a finite time given by @xmath54 which becomes shorter as the value of the cosmological constant is made larger . the field theory can also be solved in this case . following the same procedure as for the scale factor ( 2.3 ) we now attain @xmath55 @xmath56 also in this case the phantom regime can be obtained by introducing the conditions @xmath39 and @xmath1 . once again both the resulting field and its potential tend to blow up as the big rip is approached . using solution ( 2.9 ) we can now obtain from eqs . ( 2.2 ) and ( 2.5 ) that in the presence of a cosmological constant the accretion of dark energy onto a black hole makes its initial mass @xmath43 to vary according to the law @xmath57 where now @xmath58 . clearly , again @xmath4 remains all the time equal to @xmath43 if only a cosmological constant @xmath11 is present . for @xmath10 , @xmath4 again monotonically increases with time @xmath17 towards a maximum @xmath59 , which also occurs at @xmath60 . for @xmath1 , eq . ( 2.12 ) can be cast in the form @xmath61 in this case @xmath4 once again decreases with time @xmath17 , tending to vanish on the neighborhood of @xmath47 . thus , all the qualitative behaviours obtained from solution ( 2.3 ) are matched when one uses solution ( 2.9 ) . the effect induced by the presence of an extra cosmological constant is to make @xmath4 evolve exponentially . actually , it can be shown that in the phantom regime corresponding to all existing dark - energy models , the above result is always obtained , provided that regime will show a big rip singularity in the finite future . in fact , tachyon like models admit [ 18 ] a scale factor solution as given by eq . ( 2.9 ) , and therefore we obtain again for these models result ( 2.13 ) . since generalized chaplygin phantom models do not show any big rip singularities [ 19 ] , we are finally left only with k - essence models with non - canonical kinetic energy [ 20 ] for which there will be a big rip future singularity in the phantom region and this can be defined by [ 21 ] @xmath62 with @xmath63 a constant satisfying @xmath64 , and @xmath65 the hubble parameter defined from the scale factor @xmath66 } , \ ] ] where @xmath67 denotes the arbitrary time at which big rip takes place . inserting the latter two expressions into eq . ( 2.1 ) we finally obtain after integration @xmath68 in which now @xmath69 . note that eq . ( 2.16 ) is formally the same as the expression derived in ref . [ 8 ] for a differently defined @xmath70 , so confirming and generalizing to all phantom models characterized by the existence of a sudden big rip singularity in the finite future the result pointed out in that reference . we finally solve the field theory for this case . once we have wick rotated the field @xmath71 , we get @xmath72 @xmath73 we note that in this case , whereas the phantom field diverges at the big rip , the phantom field potential tends to vanish at that singularity . the results on accretion of dark energy onto black holes first obtained by babichev et al [ 8 ] and generalized above could however get into serious conflict with thermodynamics . first of all , one could always assume the current existence of primordial black holes which , at the onset of dark energy domination , could have reached a very small mass and hence a very large temperature due to a long process of hawking thermal evaporation . this would , in principle , imply that hawking radiation effects would by now prevail over the effects of the accretion of phantom fluid if this accretion was characterized by a lower positive temperature [ 22 ] . on the other hand , one should also expect that if the results derived in subsec . ii b hold , then the generalized entropy [ 23 ] defined for the whole universe , @xmath74 ( where @xmath75 and @xmath76 are the entropy of the phantom fluid and the entropy of the black hole ) , should be expected to decrease with time , so violating the generalized second law [ 23 ] . in what follows we are going to show nevertheless that none of these two apparent difficulties actually matters because , contrary to a recent claim [ 22 ] , the temperature of the phantom fluid is definite negative , and this will allow us to re - interpret results from a different , `` quantum '' standpoint . lima and alcaniz have recently proposed a general thermodynamic theory for dark energy , according to which they argue [ 22 ] that , whereas the temperature for the phantom energy regime is always positive , its entropy is negative definite and that , therefore , a cosmic scenario in which the universe is filled with phantom energy should be ruled out . there exist however rather general thermodynamic arguments which are valid for all available cosmic phantom - energy models , which appear to prevent the holding of that conclusion . thus , if the first law of thermodynamics is assumed to hold in every involved case , then it has been shown [ 24 ] that in a general friedmann - robertson - walker flat universe filled with a dark energy that satisfies the equation of state @xmath0 ( with @xmath25=constant ) , the entropy density @xmath77 per comoving volume stays always constant [ 24 ] , while the temperature of the universe turns out to be given by the expression @xmath78 , with @xmath79 an integration constant . thus , we can generally write for the temperature of a universe equipped with a dominating dark energy and equation of state @xmath0 , @xmath80 where @xmath81 is a positive constant . the interpretation of this result can obviously resort to the most commonly used concept of kinetic temperature which is a measure of the average translational kinetic energy of the system . in the case of a general dark - energy scalar field @xmath71 , the kinetic term @xmath82 can be obtained from the definition of the field itself in terms of the pressure @xmath83 and energy density @xmath84 , the equation of state @xmath0 , and the expression of the energy density derived by integrating the conservation law for cosmic energy , i.e. eq . ( 2.5 ) . this gives @xmath85 . now , an estimate of the isotropic translational kinetic energy is @xmath86 ( with @xmath87 a numerical constant ) , which can in fact be made the same as eq . ( 2.17 ) . along the present paper we shall take the general expression ( 2.17 ) as the temperature that characterizes a universe dominated by dark energy . we next derive from eq . ( 2.17 ) thermodynamic functions of interest . thus , eqs . ( 2.5 ) and ( 2.17 ) will allow us to readily get the following generalized stefan - boltzmann law for a dark energy universe , @xmath88 we notice that , whereas for @xmath11 @xmath84 becomes a simple constant ( which actually corresponds to a cosmological constant ) , for @xmath16 , eq . ( 2.18 ) consistently reduces to the usual law for radiation , and for @xmath89 the dark energy density decreases with the temperature . however , for the phantom regime where @xmath1 , in order to preserve @xmath84 positive , we must necessarily take @xmath90 , which is a condition that really directly stems from eq . ( 2.17 ) for @xmath1 . it follows that @xmath84 will always increase with @xmath91 along the entire phantom regime . the energy of such a regime actually is bounded from above and allows therefore the occurrence of negative temperatures [ 25 ] . since on the phantom regime @xmath84 increases with the scale factor @xmath31 it also follows that @xmath91 increases as the universe expands on that regime . these two characteristics are quite surprising actually and therefore they should be added to the increasing collection of phantom weird properties . finally , a general expression for the entropy of a dark energy universe can also be obtained by using the procedure of ref . [ 22 ] which in the present case leads to @xmath92 where @xmath93 is a positive constant and @xmath94 is the volume of the considered portion within the dark energy fluid . thus , contrary to the claim in ref . [ 22 ] , the entropy of a dark - energy universe is _ always _ positive , even on the phantom regime . actually , by inserting eq . ( 2.17 ) into eq . ( 2.19 ) one in fact attains that @xmath75 becomes a constant when we take @xmath94 to be the volume of the entire universe . for usual radiation corresponding to @xmath16 , entropy and temperature can be linked by means of the general relationship , entropy @xmath95 energy / temperature . according to eq . ( 2.19 ) , however , that relationship should be generalized to read : entropy @xmath95 ( 1+@xmath25)energy / temperature , according to which entropy is positive along the entire interval of conceivable values of parameter @xmath25 . even though it is not very common in physics and could thereby be listed as just another more weird property of the phantom scenario , a negative temperature is not at all unphysical or meaningless [ 26 ] . nuclear spin and other quantum systems with negative temperatures have already been observed in the laboratory and interpreted theoretically . as referred to the case of phantom energy , the existence of a negative temperature would imply that the entropy of a phantom universe monotonically decreased if one would be able to supply energy to that universe . hence , the onset at the coincidence time of the dominance of a phantom energy , @xmath1 , universe would imply the emergence of a necessarily `` hotter '' cosmic evolution regime in such a way that , if two copies of the current universe were taken , one with positive and other with negative temperature , and put them in thermal contact , then heat would always flow from the negative - temperature universe into the positive - energy universe . it could yet be argued that negative temperature is a quantum - statistical mechanics phenomenon and therefore can not be invoked in the classical realm . however , the negative temperature for the universe given by eq . ( 2.17 ) when @xmath1 [ 16 ] can still be heuristically interpreted along a way analogous to how e.g. black hole or de sitter temperature can be interpreted ( and derived ) without using any quantum - statistical mechanics arguments ; that is by simply wick rotating time , @xmath96 , so that the metric becomes positive definite , and checking that in the resulting euclidean framework @xmath97 is periodic with a period which precisely is the inverse of the hawking temperature [ 27 ] . thus , the euclideanized black hole turns out to be somehow `` quantized '' . similarly in the present case , the phantom regime can be obtained by simply wick rotating the classical real scalar field [ 28 ] , @xmath39 , which can be generally seen to be equivalent to rotating time so that @xmath96 , too ( see sec . it is in this sense that the phantom energy universe can be also regarded as a somehow `` quantized '' system and that the emergence of a negative temperature in the phantom regime can be interpreted in a consistent way . since a system with negative temperature is `` hotter '' than any other systems having positive temperature ( i.e. energy will always flow from the former to the latter system ) , even if that positive temperature is @xmath98 [ 9,26 ] , we see that the first of the two thermodynamical difficulties pointed out before becomes fully solved , no matter how high the black hole temperature could be . concerning the second of the above difficulties , let us notice on the other hand that , if the universe is assumed to contain a black hole of mass m , the entropy of the dark energy fluid will be smaller than that is given by ( 2.19 ) for @xmath99 . in this case , the entropy of the universe can be written to be @xmath100 where @xmath101 is the volume occupied by the black hole and @xmath94 is the volume of the entire universe . thus , for @xmath10 , @xmath75 decreases and , for @xmath1 , @xmath75 increases as dark energy is accreted onto the black hole . moreover , after varying @xmath75 with respect to @xmath102 and multiplying by @xmath103 , one can obtain from eq . ( 2.20 ) @xmath104 in which @xmath105 and the upper sign corresponds to @xmath10 and the lower sign corresponds to @xmath1 . on the other hand , if on the black - hole spacetime we assume the time taken by a light signal to travel along the entire hole to be @xmath106 , we can derive from eq . ( 2.1 ) @xmath107 where @xmath76 and @xmath108 are the entropy and temperature of the black hole . let us first analyze the total balance of entropy in the case of dark energy with @xmath10 . thus , from eqs . ( 2.21 ) and ( 2.22 ) we see that the accretion of a given amount of dark energy onto the black hole leads to an increase of dark energy entropy which will exceed the corresponding decrease of black hole entropy , so preserving the second law , provided that @xmath109 , i.e. if @xmath110 , a condition which appears to be of general applicability along the latest accelerating evolution of the universe . it would be in this case expected however that the competion between dark energy accretion and black hole thermal emission , both characterized by positive temperatures , will be governed by usual thermodynamic laws . thus , if initially @xmath111 ( with @xmath112 ) , then thermal emission will prevail over dark energy accretion and the black hole mass decreased initially . that situation could nevertheless be maintained only for a while because , during accelerating expansion , temperature ( 2.17 ) will increase at a much greater rate than black hole temperature did due to thermal evaporation . thus , after a given time , both processes would first tend to balance one another , to inexorably allowing then for dominance of the dark energy accretion process , leading finally to a maximum black hole mass given by eq . ( 2.7 ) . if initially we already had @xmath113 , then dark energy accretion prevailed over thermal radiation along the entire evolution , so that the mass of the black hole steadily increased up to @xmath114 . as to the case of phantom energy for @xmath1 , we ought to recall that according to the carnot equality the coupling between a negative - temperature system and a a positive - temperature system leads to a carnot engine of greater than 100 percent efficiency [ 29 ] . thus , in the present case @xmath115 so that the generalized entropy @xmath74 ought to inexorably decrease along phantom energy accretion , so violating the second law . in fact , eqs . ( 2.21 ) and ( 2.22 ) tell us that in the phantom case the entropy @xmath75 increases by an amount which is smaller than the decrease of @xmath116 provided the reasonable condition @xmath117 is again applied . this violation of the second law was to be expected because , as it was mentioned earlier , negative temperatures are compatible with observationally checked `` quantum '' carnot equalities violating the second law [ 29 ] . in any event , the implication that negative temperatures lead through the quantum carnot engine to violation of the second law can be argued under several presumptions . although discussing these arguments is clearly outside the scope of the present paper , we want to leave open the possibility that they might preserve the second law even in the case being considered . one very remarkable property of phantom energy is worth mentioning here . even though most of its properties might be weird , phantom energy seems to be able to elegantly solve one of the most conspicuous and debated paradoxes of all the physics . in fact , if all existing black holes will simultaneously disappear at the big rip leaving no hawking radiation , then the information initially lost during formation of the black holes should somehow be recovered . actually , the so - called quantum coherence loss paradox , long championed by hawking [ 30 ] and according to which an initial pure state is transformed into a final mixed state , during the whole process of black hole formation and subsequent complete evaporation , is here naturally solved in at least the phantom - energy regime in the sense that neither hawking radiation nor black holes are left in the final singular state where only the big rip singularity plus a naked black hole singularity could take place . thus , even in the classical case , the information paradox appears to be solved . at first sight , one could be tempted to dismiss eq . ( 2.17 ) because a spacetime dominated by a cosmological constant has a nonzero temperature of the order of the hubble constant @xmath65 while , according to eq . ( 2.17 ) the temperature of a fluid with @xmath11 vanishes . if by instance we take for the scale factor the expression ( 2.3 ) , then as one approaches the value @xmath11 that expression can be reduced to @xmath118 , and hence it becomes @xmath119 at the special case where @xmath11 . in such a case we have then @xmath120 , while for @xmath121 the hubble parameter is time - dependent and given by the expression @xmath122 . thus , whereas in the first case we have a cosmological horizon of radius @xmath123 which is characterized by a surface gravity @xmath124 , no constant horizon can be defined for any @xmath121 in which cases every point is subject nevertheless to an isotropic acceleration which can be given by @xmath125 . each of these two cases has therefore a distinct expression for temperature : if @xmath11 the surface gravity implies a temperature @xmath126 [ 27 ] ; if however @xmath121 then every accelerating point will be bathed by a thermal radiation at the quantum unruh temperature [ 31 ] @xmath127 , which is precisely the expression given by eq . it follows that , even though the gibbons - hawking temperature for event horizons and the unruh temperature for accelerating systems are conceptually equivalent as the surface gravity is nothing but a measure of how hard you have to accelerate to stay a given short distance away from the horizon , de sitter space must be endowed with a temperature @xmath128 but not with temperature @xmath129 , and those spacetimes with @xmath121 have in turn a temperature @xmath130 but not temperature @xmath131 . thus , eqs . ( 2.17)-(2.19 ) are valid only for @xmath121 . the above discussion makes it clear that these equations are not mere definitions devoid of physical significance , but they all possess a precise physical meaning . we can finally easily convince ourselves of the relevance of temperature ( 2.17 ) if we take for the parameter of the equation of state the value @xmath16 which obviously corresponds to usual radiation . in fact , if @xmath16 eqs . ( 2.17)-(2.19 ) reduce to @xmath132 , @xmath133 and @xmath134 , which are just the values of the temperature , energy density and entropy for the friedmann - robertson - walker universe dominated by usual radiation . therefore , the equation of state @xmath135 , together with the conservation law for cosmic energy , suffice by themselves to determine the thermodynamic properties of a general relativistic fluid which is assumed to dominate in a friedmann - robertson - walker universe , much as e.g. the gravitational characteristics of black holes or de sitter space are now known to determine well - defined and precise laws of thermodynamics , provided quantum theory is taken into account . as represented by a stuff governed by the conservation law for cosmic energy and satisfying a perfect fluid equation of state the way we have considered so far , dark energy can be looked at as being a generalized radiation field whose characteristics are fixed by the value of parameter @xmath25 . thus , if @xmath10 dark energy would account for a quintessence radiation , if @xmath1 it would describe a phantom radiation , and finally if @xmath16 dark energy described a conventional radiation field for which we obtain from eqs . ( 2.3 ) and ( 2.17 ) that @xmath136 ( with the time @xmath17 re - defined so that @xmath18 and @xmath132 . in field theory in curved space usual radiation can be represented by means of a homogeneous massless , scalar field @xmath40 which is conformally coupled to gravity . the action integral of the system formed by gravity plus a scalar field @xmath40 can be written as @xmath137\nonumber\\ & & + \frac{1}{2}\int d^4 x\sqrt{g}\left[(\nabla\phi)^2 - 2v(\phi)\right ] -\frac{1}{8\pi g}\int d^3 x\sqrt{h}\left(1 - 8\pi g\xi\phi^2\right){\rm tr}k , \end{aligned}\ ] ] where @xmath138 accounts for the coupling between gravity and the scalar field , and all other symbols keep their conventional meaning . we shall use the metric @xmath139 where @xmath140 is the conformal time , @xmath141 is the lapse function and @xmath142 is the metric on the unit three - sphere . for conformal coupling @xmath143 and flat geometry , the action integral ( 2.24 ) becomes @xmath144\nonumber\\ & & = -\frac{1}{2}\int d\eta n\left[-\left(\frac{a'}{n}\right)^2 + \left(\frac{\chi'}{n}\right)^2 -\frac{8\pi g}{3}(1-\omega)\rho_0 a^{1 - 3\omega}\right],\end{aligned}\ ] ] in which @xmath145 is the conformal field , and we have used the definitions of the pressure and energy density @xmath146 , @xmath147 and eq . the equation of motion for @xmath148 and the constraint ( @xmath149 ) are given by ( in the gauge @xmath150 ) @xmath151 @xmath152 with the equation of motion for @xmath30 yielding also eq . ( 2.28 ) . from eqs . ( 2.27 ) and ( 2.28 ) we obtain finally @xmath153 where @xmath4 is an arbitrary integration constant . now , if @xmath16 , then @xmath154 , i.e. unless by a time re - scaling , essentially the same result as that we obtained from eq . ( 2.3 ) for @xmath16 by assuming a minimal coupling @xmath155 . in terms of the conformal time @xmath156 the scale factor ( 2.3 ) for @xmath155 is given by @xmath157^{2/(1 + 3\omega ) } , \ ] ] which of course is not a solution of eq . ( 2.29 ) for any value of @xmath25 other than @xmath16 . thus , for @xmath11 we obtain from eq . ( 2.29 ) @xmath158 finally , at the onset of the accelerating regime @xmath159 , @xmath160 if we then assume that each particular value of @xmath25 corresponds to a given radiation field , it follows that all of the resulting fields are associated with a minimal coupling between the scalar field and gravity , except when the state equation parameter takes on a value @xmath16 , for which case the radiation field can be obtained by either conformally coupling or minimally coupling the field to gravity . it is in this sense that we shall consider in what follows that a universe dominated by dark energy is equivalent to a universe dominated by radiation . the concepts of negative temperature and negative entropy are both meaningless within the realm of classical thermodynamics . the mathematical definition of entropy can not in fact accommodate negative values both in statistical thermodynamics and in the shannon theory of information , where the entropy ( respectively denoted by @xmath75 and @xmath131 ) is given by the conventional equivalent formulas @xmath161 with @xmath162 the boltzmann constant , @xmath163 a given constant usually assumed to be unity , @xmath164 the probability of any given microstate , and @xmath141 the probability of the signal given the reference class of possible signals that could have been sent . the fact that @xmath164 and @xmath141 can not exceed unity makes it impossible to have finite negative values for @xmath75 and @xmath131 in the classical theory . however , more recent developments have shown that quantum theory can allow the emergence of both negative temperatures and negative entropies . cerf and adami studied [ 32 ] what happens to the shannon theory when qubits of quantum information are accounted for . in doing that these authors extended classical probabilities to function describing also the quantum characteristics of the system , including both bits and qubits , within a consistent density - matrix description that otherwise parallels classical shannon theory . a rather tantalizing implication from that generalization is that negative entropies crop up in the case of quantum systems having no classical analog . in particular , it was shown [ 33 ] that entropy becomes negative when two particles are quantum - mechanically entangled , even though that does not violate special relativity . on the other hand , the occurrence of negative absolute temperatures has already become a little bit more conventional [ 26 ] . these take also place in quantum systems or phenomena without classical analog , e.g. in quantum nuclear spin systems [ 34 ] , and their existence has already been experimentally checked [ 34 ] . we shall now contend that at least in systems which admit a statistical thermodynamic description , negative temperature implies negative entropy , in the following sense . given the specific properties of the `` log '' function ( which actually relate characteristics of linear evolution with characteristics of nonlinear evolution ) , one can still introduce a tentative statistical definition of the entropy function which encompasses both positive and negative values of temperature and entropy simultaneously . using the formula for carnot efficiency @xmath165 one can actually introduce a new entropy formula @xmath166 we notice that eq . ( 3.3 ) connects the statistical content of entropy with its purely thermodynamic meaning , as it can now be checked by considering a small variation of @xmath167 which , if @xmath168 , by eq . ( 3.3 ) induces a small decrease of entropy , such that @xmath169 . in any event , if we assume that @xmath170 , then the efficiency @xmath171 and the entropy as defined by eq . ( 3.3 ) becomes negative . it follows that a negative temperature would imply a negative entropy , except in the case that the system exchanges negative amounts of internal energy , for which case eq . ( 3.3 ) no longer holds and must be replaced for @xmath172 eq . ( 3.4 ) would in turn predict that a negative temperature @xmath173 must imply a positive entropy . that is precisely the case to which phantom energy belongs . or more precisely , phantom energy must be regarded as a quantum entity with no classical analog , which is characterized by a negative temperature given by eq . ( 2.17 ) and a positive entropy given by eq . ( 2.19 ) . if we would for a moment adhere to the alternate alcaniz - lima view [ 22 ] , then the phantom energy would still be a quantum entity without classical analog , but characterized now by a positive temperature and a negative entropy . the difference with the interpretation provided in ref . [ 22 ] is that in the present scenario having a phantom negative entropy is not necessarily physically meaningless due to the essential quantum nature of the phantom fluid . we are going next to establish in exactly what sense the phantom fluid can be considered as a quantum entity . it is known [ 7 ] that in order to preserve weak energy condition , @xmath9 , the phantom scalar field should be wick rotated ( e.g. @xmath174 ) . as pointed out in sec . ii , we can readily check that such a rotation is equivalent to wick rotating the time @xmath17 itself ( e.g. [ 35 ] @xmath175 ) , while preserving the field unchanged . thus , in the present general dark - energy scenario , the scalar field @xmath71 is expressed in terms of time @xmath17 as [ 7 ] @xmath176 , \end{aligned}\ ] ] where @xmath15 , @xmath177 and @xmath28 are the initial values of the scale factor , the scalar field and the energy density , respectively , with @xmath178 the planck length . in the case that @xmath1 it can be readily seen that eq . ( 3.5 ) can be approximated to @xmath179 along the entire cosmic evolution ending at the big rip . from eq . ( 3.6 ) we can in fact deduce that wick rotating @xmath71 while keeping @xmath17 unchanged is equivalent to wick rotating time @xmath17 while keeping @xmath71 unchanged . now , it is well known that a euclideanized spacetime metric would describe a somehow quantized system [ 27 ] . it is in this sense that the cosmic phantom fluid is a quantized entity ; that is , in a way which would parallel e.g the procedure through which the quantum temperature and entropy of black holes can be derived by simply wick rotating time in the maximally extended kruskal metric [ 27 ] . once we have made a preliminary discussion of some quantum properties of the cosmic phantom field , let us consider the simplest case in which the universe is initially filled only with phantom stuff . if we then add an elementary piece of ordinary matter to this observationally empty universe , according to our discussion before , the entropy would drop down , meaning that at least a bit ( or qubit ) of information necessarily about the added particle has been created and is available to potential observers ( note that essentially no direct information can be made available on the phantom fluid as this is not directly observable by definition ) . this is exactly the opposite to what would occur if such an elementary piece of matter with positive energy is added to a universe filled with dark energy with @xmath10 , in which case the entropy of the universe increased . the creation of negative entropy in the case of a phantom fluid can not be interpreted as being due to the creation of entangled correlations between the added particle and all `` phantom quanta '' as this implied knowledge about phantom being available to potential observers . imagine now the onset of phantom energy domination in our real universe . the above reasoning would lead to the idea that a huge amount of bits and qubits of information about all observable pieces of matter encountered by the phantom field at the start of its domination would then be created and made available to potential observers . this is what can be dubbed as phantom - energy induced cosmological information . it actually calls for the presence of observers if one appeals to a general principle for natural economics . an important point should be mentioned at this point . it is that once phantom energy starts dominating in the universe , as its temperature is negative , and hence `` hotter '' than anything observable , it will gravitationally be ceacelessly accreted onto all kinds of observable matter steadily annihilating it , all the way until the big rip at which point all of the matter in the universe , including black holes , will completely disappear . neither the above available cosmological information nor this process of all matter annihilation would nevertheless be present if instead of phantom energy the universe were filled with a dark energy vacuum component having @xmath10 . another standpoint from which one can look at the question of cosmic phantom energy is the very concept of life and its origin . in his rather controversial book `` what is life ? '' , which collected a series of lectures delivered in dublin , schrdinger posed [ 36 ] a key conclusion that , a way or another , remains still inescapable today : what a living organism feeds upon is negative entropy , and this in its very statistical thermodynamic sense . that is currently interpreted by considering that living beings necessarily produce a positive supply of entropy to the universe during their vital activities which would compensate the needed stream of negative entropy upon themselves . however acceptable this may be for keeping the second law alive , it goes without solving the key question : for an essential element of life to make its first appearance in the universe , the site where it appeared ought to be prepared to provide the necessary and immediate supply of negative entropy for the incipient life element to continue , as in this case `` egg would precede hen '' . thus , since any living organisms are made up of positive energy , while original life elements could not be consolidated if the universe was then filled with any form of dark energy with @xmath180 , it can perfectly do so in a universe filled with phantom energy , as in this case the very act of the organism appearance does imply a supply of negative entropy . on the other hand , as discussed before , an universe filled with phantom energy is also prepared to be observed . it is in these senses that it could be thought that any form of anthropic principle [ 37 ] had to be formulated in terms of the emergence of an epoch when the universe started to be dominated by phantom energy . we shall apply in this and the next sections some aspects of the thermodynamic theory for @xmath181-fluid developed by lima and maia [ 38 ] to the case of a general quintessential model of dark energy and , in particular , to the case of phantom energy . according to the discussion in sec . ii we can assume that the dark - energy field corresponds to a kind of radiation field with a wien - type spectrum given by @xmath182 where @xmath183 is a positive constant , @xmath184 is the frequency , @xmath103 is absolute temperature , and the parameters @xmath185 and @xmath186 will be determined by imposing the two following constraints to be satisfied by eq . ( 4.1 ) @xmath187 @xmath188 which respectively define the energy density and the particle number density . following lima and maia [ 38 ] , let us then define a new variable @xmath189 , so that the constraints ( 4.2 ) and ( 4.3 ) can be re - cast as @xmath190 @xmath191 where we have used eq . ( 4.1 ) . from eqs . ( 4.4 ) and ( 4.5 ) we can finally deduce that @xmath192 and @xmath193 and , therefore , our generalized wien law is given by @xmath194 which , as expected , reduces to the known wien law for blackbody radiation if we set @xmath16 . we note then that the temperature should not appear as simply @xmath103 but always in the combination @xmath195 ( see sec . ii ) , both in the energy density @xmath196 and in the energy density @xmath197 . we can therefore write the wien spectrum ( 4.1 ) for the phantom energy case as @xmath198 and hence using the same procedure as before , we recover the same wien law , but referred to absolute value of temperature , also for a phantom radiation , i.e. for @xmath1 and @xmath199 , @xmath200 this actually gives the most general expression for the wien law and is valid for any positive and negative value of the parameter @xmath25 . let us consider a set of two level systems characterized by the resonant frequency @xmath184 , immersed in phantom radiation at negative temperature @xmath199 . assuming the simple dipole - moment approximation we label the energy levels of the system by @xmath201 and @xmath202 ( see fig . 1 ) , with @xmath203 . following a line of reasoning analogous to that which led einstein to introduce his celebrated absorption and emission coefficients [ 39 ] , we can now assume the occurrence of novel radiative processes if the radiation field is made up of a substance characterized by an equation of state with @xmath1 . in that framework the probability that a system is in the energy level @xmath204 ( @xmath205 ) is given by @xmath206 population of given energy levels in the cases that : ( left ) @xmath10 and hence @xmath207 , and ( right ) @xmath1 and hence @xmath199 . the area of the circles over the energy levels would qualitatively be proportional to the population . ] where @xmath208 is the boltzmann constant and @xmath209 is the temperature - independent statistical weight of the _ i_th state . now , due to the presence of phantom energy which would only exchange discrete amounts of negative internal energy @xmath210 with the level systems , there will be three new einstein - like coefficients , here denoted by @xmath211 , @xmath212 and @xmath213 ( see fig . 2 ) . @xmath213 will correspond to a spontaneous absorption coefficient which would be evaluable from first principles using quantum mechanics and happens in the absence of any phantom radiation , but creates quanta of phantom energy ; @xmath211 is a new induced emission coefficient and @xmath212 is a new stimulated absorption coefficient . the latter two coefficients would only occur in the presence of phantom radiation . these three coefficients will be related with the following rates of probability transition . @xmath214 @xmath215 @xmath216 thus , an equilibrium condition can be introduced in the considered system which reads @xmath217 and if the approximation @xmath218 is applied , then @xmath219 , and hence @xmath220 where we have used @xmath221 . comparing with the generalized wien law ( 4.8 ) , we finally obtain a generalized planck law , with [ 34 ] @xmath222 @xmath223 being the average occupation number , and finally @xmath224 where the case of fermions has been also included . the law ( 5.9 ) is the most general planck law and can be applied to radiation characterized by an equation of state @xmath0 , where @xmath25 can take on any positive or negative value . pictorial representation of the possible radiative processes induced by phantom radiation on a two - level system as expressed in terms of the einstein - like coefficients : @xmath213 for spontaneous absorption , @xmath211 for induced absorption , and @xmath212 for stimulated absorption . no stimulated emission can take place in the presence of phantom energy with negative temperature . ] as a test of consistency we finally calculate e.g. @xmath197 from eq . ( 5.9 ) for the fermion case . we in fact obtain @xmath225 where @xmath226 and @xmath227 are the gamma and zeta functions [ 40 ] , respectively . that is the result what was to be in fact expected . before closing up this section we would like to briefly comment on the effect that negative phantom temperatures may have on laser effect . in fact , it is well known that laser theory is based on the occurrence of stimulated emission which , together with population inversion , leads to the necessary coherent radiation amplification leading to the laser operation . in the case of negative phantom temperature , even though population inversion is guaranteed , there is anything like a stimulated emission and lasing can never be produced . what one could instead suppose is the possibility for a contrary or anti - lasing process . if in our negative temperature system one would restore a larger population in the lower energy level by the use of an appropriate procedure , then stimulated absorption would produce a damping , rather than amplification , of the coherent radiation . the construction of such an anti - laser , or more properly `` lasar '' ( light attenuation by stimulated absorption of radiation ) , device could be of some usefulness in atomic and molecular physics . on the other hand , stimulated absorption processes might provide a clue for explaining the lack of phantom - energy radiative processes in cosmology . there have been two recent important contributions to the question of the thermodynamics of dark energy . on one hand , alcaniz and lima [ 22 ] derived general dark energy thermodynamic expressions that , when extended to the phantom regime , implied positive values for its temperature and negative values for its entropy . by using two independent compelling arguments , here it is nevertheless shown that what becomes negative in the phantom regime is temperature , while entropy is kept definite positive along the entire range of negative state equation parameter . in any event , since classically having either a negative entropy or a negative temperature , or both , makes no physical meaning , the first obvious conclusion seems to be that phantom energy can not exist . this situation is somehow reminiscent to the one which was posed in the mid - seventies when it was first realized that the black hole entropy had to be finite [ 41 ] . hawking himself has many times stressed [ 42 ] that , at that time , it seemed that since black hole can not emit anything classically , such a conclusion was meaningless , too . the solution to this paradoxical situation was , as it is now widely acknowledged , to appeal to the quantum nature of black holes which allowed them to radiate thermal particles . we argue in this paper that the solution to the above apparent phantom thermodynamic paradox is again by appealing to the essential quantum nature of the phantom stuff . in fact , if , in spite of by definition violating the dominant energy condition , i.e. @xmath3 , phantom energy is consistently assumed to satisfy the weak energy condition , @xmath9 , then it can be interpreted that the phantom spacetime becomes euclideanized and this is known to provide the natural framework where quantum temperature and quantum entropy can be consistently defined [ 27 ] ; i.e. the cosmic phantom fluid can be essentially regarded as a quantum system having not any classical analogs . now , negative temperatures ( or even entropies ) can perfectly exist in the quantum realm . it is in this way that the above thermodynamic phantom paradox is solved in the present paper . our conclusion thus is that there can exist a cosmic quantum field characterized by a temperature that is always negative and a positive definite entropy , and that , moreover , if the universe would happen to be currently dominated by such phantom energy , then this would introduce a universal element of cosmological information that would make every piece of observable matter to be really observable and allow the very existence of living organisms in the universe . on the other hand , when considering accretion of phantom energy onto black holes , babichev , dokuchaev and eroshenko have showed [ 8 ] that the mass of the black holes undergo a gradual decrease and tend all to zero at the big rip . in this work we confirm and extend that result , consistently interpreting it in terms of the above conclusion on phantom thermodynamics . this realization is important in at least two respects . it firstly implies that once black holes are formed before or after phantom energy domination , all known evolution processes of existing black holes become inexorably dominated by phantom energy accretion and finally subject to what could be dubbed as a `` democracy before death '' principle by which such black holes are all equalized before disappearing all at once at the big rip , no matter their initial mass or the time when they were formed . the reason of such a domination of phantom energy accretion ( which actually extends over any emission processes of any observable matter system ) essentially resides on the fact that , however small its absolute value may be , a negative temperature is always hotter than any positive temperature , even if this is infinite . secondly , it is also a conclusion of the present paper that , since the phantom - induced annihilation process prevails always over hawking thermal emission , during the whole black hole evolution process leading from formation to final disappearance , quantum coherence is preserved , as in this case neither the black hole nor any thermal radiation emitted from it is left in the final state . this offers a rather comfortable solution to the so - called black - hole information paradox [ 27 ] and becomes still another positive consequence from the existence of phantom energy in the universe . the formulation of some main theoretical basis for establishing the quantum statistic thermodynamics of phantom energy has been also aimed at in this work . by assuming that any form of dark energy can be taken to be a radiation field , we have thus considered a generalized wien law and hence a generalized planck radiation law . these laws turn out to be described by the same general expressions as for the usual dark energy case first derived by lima and maia [ 38 ] , but referred to the absolute value of phantom temperature . this conclusion comes about as a consequence from the necessary introduction of novel einstein coefficients and the different probability law for the case of phantom energy . the extra einstein coefficients correspond to new radiative processes that include a stimulated absorption phenomenon , which would attenuate the intensity of the phantom radiation . it will only be by collecting more cosmological data that we will be able to finally decide on whether or not phantom energy is the form of dark energy that currently operates in the universe to drive its observed accelerating expansion . however , even in the event that other kind of dark energy turned out to be the favoured stuff dominating the current universe , there could still be sufficient room for phantom energy to dominate over all other cosmic stuffs during other epochs along the universal evolution , including the primordial inflationary period , such as has been recently suggested [ 43 ] . anyway , we hope that the contents of the present paper may help to convince cosmologists that , whatever the final conclusion on the above subjects may be , phantom energy is a physical concept with sufficient theoretical interest by itself as to pursue active research on it . we thank professors lima and babichev for useful explanations , discussions and correspondence . this work was supported by dgicyt under research project bmf2002 - 03758 . the authors wish to respectfully dedicate the present work to the memory of all those that were killed , wounded or directly affected in the terrorist attacks of march 11 in madrid r.r . caldwell , phys . b545 , 23 ( 2002 ) j.s . alcaniz , phys . d69 , 083521 ( 2004 ) . baccigalupi , a. balbi , s. matarrase , f. perrotta and n. vittorio , phys . d65 , 063520 ( 2002 ) ; m. melchiorri , l. mersini , c.j . odman and m. tradden , phys . d68 , 043509 ( 2003 ) ; m. doupis , a. riazuelo , y. zolnierowski and a. blanchard , astron . 405 , 409 ( 2003 ) ; l. tonry _ el al . _ , j. 594 , 1 ( 2003 ) . b. mcinnes , jhep 0208 , 029 ( 2002 ) ; g.w . gibbons , hep - th/0302199 ; a.e . schulz and m.j . white , phys . d64 , 043514 ( 2001 ) ; j.g . hao and x. z. li , phys . d67 , 107303 ( 2003 ) ; s. nojiri and s.d . odintsov , phys . lett.b562 , 147 ( 2003 ) ; b565 , 1 ( 2003 ) ; b571 , 1 ( 2003 ) ; p.singh , m. sami and n. dadhich , phys . d68 , 023522 ( 2003 ) ; j.g . hao and x.z . li , phys . d68 , 043501 ; 083514 ( 2003 ) ; x.z . li and j.g . hao , hep - th/0303093 , phys . d ( in press ) ; m.p . dabrowski , t. stachowiak and m. szydlowski , phys . d68 , 067301 ( 2003 ) ; m. szydlowski , w. zaja and a. krawiec , astro - ph/0401293 ; e. elizalde and j. quiroga h. , mod . a19 , 29 ( 2004 ) ; v.b . johri , astro - ph/0311293 ; l.p . chimento and r. lazkoz , phys . 91 , 211301 ( 2003 ) ; h.q . lu , hep - th/0312082 ; m. sami , a. toporensky , gr - qc/0312009 ; r. naboulsi , gr - qc/0303007 , class . quan . grav ( in press ) ; x.h . meng and p. wang , hep - ph/0311070 ; h. stefancic , astro - ph/0310904 , phys . b ( in press ) ; astro - ph/0312484 ; d.j . liu and x.z . li , phys . d68 , 067301 ( 2003 ) ; a. yurov , astro - ph/0305019 ; y.s . piao and e. zhou , d68 , 083515 ( 2003);y.s . piao and y.z zhang , astro - ph/0401231 ; h.q . lu , hep - th/0312082 ; m. szydlowski , w. czaja and a. krawiec , astro - ph/0401293 ; j.m . aguirregabiria , l.p . cimento and r. lazkoz , gr - qc/0403157 ; j. cepa , astro - ph/0403616 , astron . ( in press ) ; v. faraoni , gr - qc/0404078 , phys . d. ( in press ) ; z .- k . guo , y .- s . piao and y .- z . zhang , astro - ph/0404154 ; e. elizalde , s. nojiri and s. odintsov , hep - th/0405034 ; y .- h . wei and y. tian , gr - qc/0405038 ; f. piazza and s. tsujikawa , hep - th/0405054 . caldwell , m. kamionkowski and n.n . weinberg , phys . 91 , 071301 ( 2003 ) ; j.d . barrow , class . 21 , l79 ( 2004 ) . barrow , g.j . galloway and f.j . tipler , mon . not . r. astron . 223 , 835 ( 1986 ) . gonzlez - daz , phys . d69 , 063522 ( 2004 ) . e , babichev , v. dokuchaev and yu . eroshenko , gr - qc/0402089 . l. onsanger , nuovo cim . suppl . 6 , 279 ( 1949 ) ; m.j . klein , phys . 104 , 589 ( 1956 ) ; d. montgomery , phys . a39 , 7 ( 1972 ) ; j.b . taylor , phys . a49 , 1 ( 1972 ) ; c.e . seyler jr , phys . 32 , 515 ( 1974 ) ; r.h . kraichnan and d. montgomery , rep . 43 , 547 ( 1980 ) ; v.l . berdichevsky , i. kunin and f. hussain , phys . a43 , 2050 ( 1991 ) ; j. casas - vazquez and d. jou , rep . 66 , 1937 ( 2003 ) . carroll , m. hoffman and m. trodden , phys . d68 , 023509 ( 2003 ) . frampton , hep - th/0302007 . cline , s.y . jeon and g.d . moore , hep - ph/0311312 , phys . d. ( in press ) . michel , astrophys . 15 , 153 ( 1972 ) . s.w . hawking and g.f.r . ellis , _ the large scale structure of space - time _ ( cambridge university press , cambridge , uk , 1973 ) . gonzlez - daz , phys . d68 , 084016 ( 2003 ) . peebles , _ principles of physical cosmology _ ( princeton university press , princeton , usa , 1976 ) . gonzlez - daz and c.l . sigenza , phys . b589 , 78 ( 2004 ) . gonzlez - daz , _ tachyon phantom scenarios _ , imaff report imaff - rca-04 - 01 ( 2004 ) . gonzlez - daz , phys . rev . d68 , 021303(r ) ( 2003 ) ; m. bouhmadi and j.a . jimnez - madrid , astro - ph/0404540 . c. armendriz - picn , t. damour and v. mukhanov , phys . b458 , 209 ( 1999 ) ; j. garriga and v. mukhanov , phys . b458 , 219 ( 1999 ) ; t. chiba , t. okabe and m. yamaguchi , phys . d62 , 023511 ( 2000 ) ; c. amendriz - picn , v. mukhanov and p.j . steinhardt , phys . 85 , 4438 ( 2000 ) ; c. amendriz - picn , v. mukhanov and p.j . steinhardt , phys . d63 , 103510 ( 2001 ) ; l.p . chimento and a. feinstein , mod . a19 , 761 ( 2004 ) ; m. malquarti , e.j . copeland and a.r . liddle , phys . d68 , 023512 ( 2003 ) . gonzlez - daz , phys . b586 , 1 ( 2004 ) . j.a.s . lima and j.s . alcaniz , astro - ph/0402265 . s.w . hawking , phys . d13 , 191 ( 1976 ) . d. youm , phys . b531 , 276 ( 2002 ) ; e. verlinde , hep - th/0008140 ; j.l . cardy , nucl . b270 , 186 ( 1986 ) . l. vanzo and d. klemm , _ de sitter space and negative temperatures _ , universit di trento , dipartimento di fisica report : www-th.bo.infn.it/activities/bo11/act2003/vanzo.ps ; s .- hong and s .- w . kim , gr - qc/0303059 . purcell and r.v . pound , phys . 81 , 279 ( 1951 ) ; n.f . ramsey , phys . 103 , 20 ( 1956 ) . gibbons and s.w . hawking , phys . d15 , 2752 ( 1977 ) . one can in principle wick rotate the field either clockwise , @xmath39 , or anti - clockwise , @xmath228 ( see ref . it will be seen in sec . iii that @xmath39 will be equivalent to rotating time anti - clockwise and @xmath228 will be equivalent to rotating time clockwise . w.c . reynolds and h.c . perkins , _ engineering thermodynamics _ ( mcgraw hill inc . , usa , 1977 ) . s.w . hawking , phys . d14 , 2469 ( 1976 ) . unruh , phys . d14 , 870 ( 1976 ) . cerf and c. adami , phys . 79 , 5194 ( 1997 ) . cerf and c. adami , physica d120 , 62 ( 1998 ) . c. kittel and h. kroemer , _ thermal physics . 2nd edition _ ( freeman and co. , new york , usa , 1984 ) . linde , _ inflation and quantum cosmology _ ( academic press , boston , usa , 1990 ) . e. schrdinger , _ what is life ? . the physical aspects of the living cell _ ( cambridge university press , cambridge , uk , 1959 ) ; s.a . kauffman , in _ `` what is life ? '' . the next fifty years _ ( cambridge university press , cambridge , uk , 1995 ) ; l. smolin , _ the life of the cosmos _ ( phoenix , london , uk , 1997 ) . barrow and f.j . tipler , _ the anthropic cosmological principle _ ( clarendon press , oxford , uk , 1986 ) . lima and a. maia jr . , phys . d52 , 5628 ( 1995 ) . m. jammer , _ the conceptual development of quantum mechanics _ ( mc graw - hill , new york , usa , 1967 ) . m. abramowitz and i. stegun , _ handbook of mathematical functions _ ( dover , new york , usa , 1964 ) ; i.s . gradshteyn and i.m . ryzhik , _ table of integrals , series , and products _ ( academic press , san diego , usa , 1994 ) . j. bekenstein , phys . d7 , 2333 ( 1973 ) . s.w . hawking , a brief history of time ( bantam press , londres , uk , 1988 ) . y .- s . piao and y .- z . zhang , astro - ph/0401231

we propose the pagerank model of opinion formation and investigate its rich properties on real directed networks of universities of cambridge and oxford , livejournal and twitter . in this model the opinion formation of linked electors is weighted with their pagerank probability . we find that the society elite , corresponding to the top pagerank nodes , can impose its opinion to a significant fraction of the society . however , for a homogeneous distribution of two opinions there exists a bistability range of opinions which depends on a conformist parameter characterizing the opinion formation . we find that livejournal and twitter networks have a stronger tendency to a totalitar opinion formation . we also analyze the sznajd model generalized for scale - free networks with the weighted pagerank vote of electors .

1204.3806

to understand the nature and origins of mass opinion formation is the outstanding challenge of democratic societies @xcite . in a last few years an enormous development of such social networks as livejournal @xcite , facebook @xcite , twitter @xcite and vkontakte @xcite , with up to hundreds of millions of users , demonstrated the growing influence of these networks on social and political life . the small - world scale - free structure of the social networks ( see e.g. @xcite ) , combined with their rapid communication facilities , leads to a very fast information propagation over networks of electors , consumers , citizens making them very active on instantaneous social events . this puts forward a request for new theoretical models which would allow to understand the opinion formation process in the modern society of xxi century . the important steps in the analysis of opinion formation have been done with the development of various voter models described in a great detail in @xcite , @xcite,@xcite,@xcite,@xcite , @xcite,@xcite,@xcite . this research field became known as sociophysics @xcite,@xcite,@xcite . in this work we introduce several new aspects which take into account the generic features of the social networks . at first , we analyze the opinion formation on real directed networks taken from the academic web link database of british universities networks @xcite , livejournal database @xcite and twitter dataset @xcite . this allows us to incorporate the correct scale - free network structure instead of unrealistic regular lattice networks , often considered in voter models @xcite . at second , we assume that the opinion at a given node is formed by the opinions of its linked neighbors weighted with the pagerank probability of these network nodes . we think that this step represents the reality of social networks : all of network nodes are characterized by the pagerank vector which gives a probability to find a random surfer on a given node as described in @xcite . this vector gives a steady - state probability distribution on the network which provides a natural ranking of node importance , or elector or society member importance . in a certain sense the top nodes of pagerank correspond to a political elite of the social network which opinion influences the opinions of other members of the society @xcite . thus the proposed pagerank opinion formation ( prof ) model takes into account the situation in which an opinion of an influential friend from high ranks of the society counts more than an opinion of a friend from lower society level . we argue that the pagerank probability is the most natural form of ranking of society members . indeed , the efficiency of pagerank rating is demonstrated for various types of scale - free networks including the world wide web ( www ) @xcite , _ physical review _ citation network @xcite , scientific journal rating @xcite , ranking of tennis players @xcite , wikipedia articles @xcite , the world trade network @xcite and others . due to the above argument we consider that the prof model captures the reality of social networks and below we present the analysis of its interesting properties . the paper is composed as follows : the prof model is described in sec . ii , the numerical results on its properties are presented in sec.iii for british university networks . in sec . iv we combine the prof model with the sznajd model @xcite and study the properties of the prof - sznajd model . in sec.v we analyze the models discussed in previous sections on an example of large social network of the livejournal @xcite . the results for the twitter dataset @xcite are presented in sec.vi . the discussion of the results is presented in sec.vii . the prof model is defined in the following way . in agreement with the standard pagerank algorithm @xcite we determine the pagerank probability @xmath0 for each node @xmath1 and arrange all @xmath2 nodes in a monotonic decreasing order of the probability . in this way each node @xmath1 has a probability @xmath3 and the pagerank index @xmath4 with the maximal probability is at @xmath5 ( @xmath6 ) . we use the usual damping factor value @xmath7 to compute the pagerank vector of the google matrix of the network ( see e.g. @xcite,@xcite,@xcite ) . in addition to that a network node @xmath1 is characterized by an ising spin variable @xmath8 which can take values @xmath9 or @xmath10 coded also by red or blue color respectively . the sign of a node @xmath1 is determined by its direct neighbors @xmath11 which have the pagerank probabilities @xmath12 . for that we compute the sum @xmath13 over all directly linked neighbors @xmath11 of node @xmath1 : @xmath14 where @xmath15 and @xmath16 denote the pagerank probability @xmath12 of a node @xmath11 pointing to node @xmath1 ( incoming link ) and a node @xmath11 to which node @xmath1 points to ( outgoing link ) respectively . here , the two parameters @xmath17 and @xmath18 are used to tune the importance of incoming and outgoing links with the imposed relation @xmath19 ( @xmath20 ) . the values @xmath21 and @xmath22 correspond to red and blue nodes respectively . the value of spin @xmath8 takes the value @xmath23 or @xmath10 respectively for @xmath24 or @xmath25 . in a certain sense we can say that a large value of parameter @xmath18 corresponds to a conformist society where an elector @xmath1 takes an opinion of other electors to which he points to ( nodes with many incoming links are in average at the top positions of pagerank ) . on the opposite side a large value of @xmath17 corresponds to a tenacious society where an elector @xmath1 takes mainly an opinion of those electors who point to him . + 0.2 cm as a function of number of iterations @xmath26 . the red and black curves ( top and bottom curves at @xmath27 respectively ) show evolution for two different realizations of random distribution of color with the same initial fraction @xmath28 at @xmath29 . the green curve ( middle curve at @xmath27 ) shows dependence @xmath30 for the initial state with @xmath31 all red nodes with top pagerank @xmath32 indexes ( highest @xmath3 values , @xmath33 ) . the evolution is done at @xmath34 and temperature @xmath35 . _ left panel : _ cambridge network with @xmath36 . _ right panel : _ oxford network with @xmath37 . [ fig1],title="fig:",scaledwidth=23.0% ] 0.2 cm as a function of number of iterations @xmath26 . the red and black curves ( top and bottom curves at @xmath27 respectively ) show evolution for two different realizations of random distribution of color with the same initial fraction @xmath28 at @xmath29 . the green curve ( middle curve at @xmath27 ) shows dependence @xmath30 for the initial state with @xmath31 all red nodes with top pagerank @xmath32 indexes ( highest @xmath3 values , @xmath33 ) . the evolution is done at @xmath34 and temperature @xmath35 . _ left panel : _ cambridge network with @xmath36 . _ right panel : _ oxford network with @xmath37 . [ fig1],title="fig:",scaledwidth=23.0% ] -0.2 cm the condition ( [ eq1 ] ) on spin inversion can be written via the effective ising hamiltonian @xmath38 of the whole system of interacting spins : @xmath39 where the spin - spin interaction @xmath40 determines the local magnetic field @xmath41 on a given node @xmath1 @xmath42 which gives the local spin energy @xmath43 . according to ( [ eq2 ] ) , ( [ eq3 ] ) the interaction between a selected spin @xmath1 and its neighbors @xmath11 is given by the pagerank probability : @xmath44 . thus from a physical view point the whole system can be viewed as a disordered ferromagnet @xcite . in this way the condition ( [ eq1 ] ) corresponds to a local energy @xmath45 minimization done at zero temperature . we note that such an analogy with spin systems is well known for opinion formation models on regular lattices @xcite,@xcite,@xcite . however , it should be noted that generally we have asymmetric couplings @xmath46 that is unusual for physical problems ( see discussion in @xcite ) . in view of this analogy it is possible to introduce a finite temperature @xmath47 and then to make a probabilistic metropolis type condition @xcite for the spin @xmath1 inversion determined by a thermal probability @xmath48 , where @xmath49 is the energy difference between on - site energies @xmath45 with spin up and down . during the relaxation process each spin is tested on inversion condition that requires @xmath2 steps and then we do @xmath26 iterations of such @xmath2 steps . we discuss the results of the relaxation process at zero and finite temperatures @xmath47 in next section . here we present results for prof model considered on the networks of cambridge and oxford universities in year 2006 , taken from @xcite . the properties of pagerank distribution @xmath50 for these networks have been analyzed in @xcite . the total number of nodes @xmath2 and links @xmath51 are : @xmath52 , @xmath53 ( cambridge ) ; @xmath54 , @xmath55 ( oxford ) @xcite . both networks are characterized by an algebraic decay of pagerank probability @xmath56 and approximately usual exponent value @xmath57 , additional results on the scale - free properties of these networks are given in @xcite . we discuss usually the fraction of red nodes since by definition all other nodes are blue . : left / right column is for cambridge / oxford network . the initial fraction of red colors is @xmath28 ( top panel ) and @xmath31 nodes have red color for bottom panels with @xmath36 and @xmath58 for cambridge and oxford network respectively . nodes are ordered by the pagerank index @xmath32 and color plot shows only @xmath59 . [ fig2],title="fig:",scaledwidth=23.0% ] : left / right column is for cambridge / oxford network . the initial fraction of red colors is @xmath28 ( top panel ) and @xmath31 nodes have red color for bottom panels with @xmath36 and @xmath58 for cambridge and oxford network respectively . nodes are ordered by the pagerank index @xmath32 and color plot shows only @xmath59 . [ fig2],title="fig:",scaledwidth=23.0% ] + : left / right column is for cambridge / oxford network . the initial fraction of red colors is @xmath28 ( top panel ) and @xmath31 nodes have red color for bottom panels with @xmath36 and @xmath58 for cambridge and oxford network respectively . nodes are ordered by the pagerank index @xmath32 and color plot shows only @xmath59 . [ fig2],title="fig:",scaledwidth=23.0% ] : left / right column is for cambridge / oxford network . the initial fraction of red colors is @xmath28 ( top panel ) and @xmath31 nodes have red color for bottom panels with @xmath36 and @xmath58 for cambridge and oxford network respectively . nodes are ordered by the pagerank index @xmath32 and color plot shows only @xmath59 . [ fig2],title="fig:",scaledwidth=23.0% ] -0.2 cm the typical examples of time evolution of fraction of red nodes @xmath30 with the number of time iterations @xmath26 are shown in fig . we see the presence of bistability in the opinion formation : two random states with the same initial fraction of red nodes @xmath60 evolve to two different final fractions of red nodes @xmath61 . the process gives an impression of convergence to a fixed state approximately after @xmath62 iterations . a special check shows that all node colors become fixed after this time @xmath63 . the convergence time to a fixed state is similar to those found for opinion formation on regular lattices where @xmath64 @xcite . the corresponding time evolution of colors is shown in fig . [ fig2 ] for first 10% of nodes ordered by the pagerank index @xmath32 . the results of fig . [ fig1 ] show that for a random initial distribution of colors we may have different final states with @xmath65 variation compared to the initial @xmath28 . however , if we consider that @xmath31 nodes with the top @xmath32 index values ( from @xmath23 to @xmath31 ) have the same opinion ( e.g. red nodes ) then we find that even a small fraction of the total number of nodes @xmath2 ( e.g. @xmath31 of about 0.5% or 1% of @xmath2 ) can impose its opinion for a significant fraction of nodes of about @xmath66 . this shows that in the frame of prof model the society elite , corresponding to top @xmath32 nodes , can significantly influence the opinion of the whole society under the condition that the elite members have the fixed opinion between themselves . we also considered the case when the red nodes are placed on @xmath67 top nodes of cheirank index @xmath68 . this ranking is characterized by the cheirank probability @xmath69 for a random surfer moving in the inverted direction of links as described in @xcite . in average @xmath69 is proportional to the number of outgoing links . however , in this case the top nodes with a small @xmath70 values are not able to impose their opinion and the final fraction becomes blue . we attribute this to the fact that the opinion condition ( [ eq1 ] ) is determined by the pagerank probability @xmath50 and that the correlations between cheirank and pagerank are not very strong ( see discussion in @xcite ) . to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] + to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] + to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath75 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath77 iterations ; here @xmath35 . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath78 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig3],title="fig:",scaledwidth=23.0% ] -0.2 cm to analyze how the final fraction of red nodes @xmath61 depends on its initial fraction @xmath70 we study the time evolution @xmath30 for a large number @xmath79 of initial random realizations of colors following it up to the convergent time for each realization . we find that the final red nodes are homogeneously distributed in @xmath32 . thus there is no specific preference for top society levels for an initial random distribution . the probability distribution @xmath74 of final fractions @xmath61 is shown in fig . [ fig3 ] as a function of initial fraction @xmath70 at three values of parameter @xmath17 . these results show two main features of the model : a small fraction of red opinion is completely suppressed if @xmath80 and its larger fraction dominates completely for @xmath81 ; there is a bistability phase for the initial opinion range @xmath82 . of course , there is a symmetry in respect to exchange of red and blue colors . for small value @xmath83 we have @xmath84 with @xmath85 while for large value @xmath86 we have @xmath87 , @xmath88 . + 0.7 cm on the tenacious parameter @xmath17 ( or conformist parameter @xmath89 ) for initial red nodes in @xmath36 values of pagerank index ( @xmath33 ) ; black and red(gray ) curves show data for cambridge and oxford networks ; here @xmath35 . [ fig4],scaledwidth=44.0% ] -0.2 cm ( middle panels ) at @xmath90 but with uniform condition for spin flip being independent of pagerank probability ( top panels : @xmath91 in eq . ( [ eq1 ] ) ) and pagerank probability @xmath92 replaced by @xmath93 in eq.([eq1 ] ) ( bottom panels ) ; left and right panels correspond to cambridge and oxford networks ; here @xmath35 and @xmath76 realizations are used . [ fig5],title="fig:",scaledwidth=23.0% ] ( middle panels ) at @xmath90 but with uniform condition for spin flip being independent of pagerank probability ( top panels : @xmath91 in eq . ( [ eq1 ] ) ) and pagerank probability @xmath92 replaced by @xmath93 in eq.([eq1 ] ) ( bottom panels ) ; left and right panels correspond to cambridge and oxford networks ; here @xmath35 and @xmath76 realizations are used . [ fig5],title="fig:",scaledwidth=23.0% ] + ( middle panels ) at @xmath90 but with uniform condition for spin flip being independent of pagerank probability ( top panels : @xmath91 in eq . ( [ eq1 ] ) ) and pagerank probability @xmath92 replaced by @xmath93 in eq.([eq1 ] ) ( bottom panels ) ; left and right panels correspond to cambridge and oxford networks ; here @xmath35 and @xmath76 realizations are used . [ fig5],title="fig:",scaledwidth=23.0% ] ( middle panels ) at @xmath90 but with uniform condition for spin flip being independent of pagerank probability ( top panels : @xmath91 in eq . ( [ eq1 ] ) ) and pagerank probability @xmath92 replaced by @xmath93 in eq.([eq1 ] ) ( bottom panels ) ; left and right panels correspond to cambridge and oxford networks ; here @xmath35 and @xmath76 realizations are used . [ fig5],title="fig:",scaledwidth=23.0% ] -0.2 cm our interpretation of these results is the following . for small values of @xmath94 the opinion of a given society member is determined mainly by the pagerank of neighbors _ to whom he points to _ ( outgoing links ) . the pagerank probability @xmath92 of nodes , on which many nodes point to , is usually high since @xmath92 is proportional to the number of ingoing links @xcite . thus , at @xmath94 a society is composed of members who form their opinion listening an elite opinion . in such a society its elite with one color opinion can impose this opinion to a large fraction of the society . this is illustrated on fig . [ fig4 ] which shows a dependence of final fraction of red @xmath61 nodes on parameter @xmath17 for a small initial fraction of red nodes in the top values of pagerank index ( @xmath36 ) . we see that @xmath95 corresponds to a conformist society which follows in its great majority the opinion of its elite . for @xmath96 this fraction @xmath61 drops significantly showing that this corresponds to a regime of tenacious society . it is somewhat surprising that the tenacious society ( @xmath97 ) has well defined and relatively large fixed opinion phase with a relatively small region of bistability phase . this is in a contrast to the conformist society at @xmath94 when the opinion is strongly influenced by the society elite . we attribute this to the fact that in fig . [ fig3 ] we start with a randomly distributed opinion , due to that the opinion of elite has two fractions of two colors that create a bistable situation since two fractions of society follows opinion of this divided elite that makes the situation bistable on a larger interval of @xmath70 compared to the case of tenacious society at @xmath97 . to stress the important role of pagerank in the dependence of @xmath61 on @xmath70 presented in fig . [ fig3 ] we show in fig . [ fig5 ] the same analysis at @xmath90 but for the case when in eq.([eq1 ] ) for the spin flip we take all @xmath91 ( equal weight for all nodes ) . the data of fig . [ fig5 ] clearly demonstrate that in this case the bistability of opinion disappears . thus the prof model is qualitatively different from the case when only the links without their pagerank weight are counted for the spin flip condition . we also test the sensitivity in respect to pagerank probability replacing @xmath92 by @xmath93 in eq.([eq1 ] ) as it is shown in fig . [ fig5 ] ( bottom panels ) . we see that compared to the case @xmath91 we start to have some signs of bistability but still they remain rather weak compared to the case of fig . [ fig3 ] . ( middle panel ) at @xmath90 but at finite temperature @xmath47 during the relaxation process with @xmath98 ( top panels ) and @xmath99 ( bottom panels ) ; the number of random initial realizations is @xmath100 , the relaxation is done during @xmath101 iterations . left and right columns correspond to cambridge and oxford networks . [ fig6],title="fig:",scaledwidth=23.0% ] ( middle panel ) at @xmath90 but at finite temperature @xmath47 during the relaxation process with @xmath98 ( top panels ) and @xmath99 ( bottom panels ) ; the number of random initial realizations is @xmath100 , the relaxation is done during @xmath101 iterations . left and right columns correspond to cambridge and oxford networks . [ fig6],title="fig:",scaledwidth=23.0% ] + ( middle panel ) at @xmath90 but at finite temperature @xmath47 during the relaxation process with @xmath98 ( top panels ) and @xmath99 ( bottom panels ) ; the number of random initial realizations is @xmath100 , the relaxation is done during @xmath101 iterations . left and right columns correspond to cambridge and oxford networks . [ fig6],title="fig:",scaledwidth=23.0% ] ( middle panel ) at @xmath90 but at finite temperature @xmath47 during the relaxation process with @xmath98 ( top panels ) and @xmath99 ( bottom panels ) ; the number of random initial realizations is @xmath100 , the relaxation is done during @xmath101 iterations . left and right columns correspond to cambridge and oxford networks . [ fig6],title="fig:",scaledwidth=23.0% ] -0.2 cm in fact the spin flip condition ( [ eq1 ] ) can be viewed as a relaxation process in a disordered ferromagnet ( since all @xmath102 in ( [ eq2 ] ) , ( [ eq3 ] ) ) at zero temperature . such type of analysis of voter model relaxation process on regular lattices is analyzed in @xcite . from this view point it is natural to consider the effect of finite temperature @xmath47 on this relaxation . at finite @xmath47 the flip condition is determined by the thermal metropolis probability @xmath103 as described in previous section . we follow this thermodynamic relaxation process at finite temperature up to @xmath101 iterations and in this way obtain the probability distribution of final @xmath61 fraction of red nodes obtained from initial @xmath70 fraction of red nodes randomly distributed over the network at @xmath29 . the results obtained at finite temperatures are shown at fig . they show that a finite temperature @xmath47 allows to have a finite fraction of red nodes @xmath61 when for their small initial fraction @xmath70 all final @xmath61 were equal to zero . also the bistability splitting is reduced and it disappears at larger values of @xmath47 . thus finite @xmath47 introduce a certain smoothing in @xmath74 distribution . over local energies @xmath45 obtained from the relaxation process during @xmath104 time iterations at temperatures @xmath99 ( black curve ) and @xmath105 ( red / gray curve ) ; average is done over @xmath106 random initial realizations . the insets show the distributions on a large scale including all local energies @xmath45 . left and right panels show cambridge and oxford networks . [ fig7],title="fig:",scaledwidth=23.0% ] 0.2 cm over local energies @xmath45 obtained from the relaxation process during @xmath104 time iterations at temperatures @xmath99 ( black curve ) and @xmath105 ( red / gray curve ) ; average is done over @xmath106 random initial realizations . the insets show the distributions on a large scale including all local energies @xmath45 . left and right panels show cambridge and oxford networks . [ fig7],title="fig:",scaledwidth=23.0% ] -0.2 cm however , the relaxation process at finite temperatures does not lead to the thermal boltzmann distribution . indeed , in fig . [ fig7 ] we show the probability distribution @xmath107 as a function of local energies @xmath45 defined in ( [ eq2 ] ) , ( [ eq3 ] ) . the distribution @xmath107 is obtained from the relaxation process with many initial random spin realizations @xmath79 . even if the temperature @xmath47 is comparable with typical values of local energies @xmath45 we still obtain a rather peaked distribution at @xmath108 being very different from the boltzmann distribution . , coded by color from @xmath109 ( black ) to @xmath110 ( blue / dark gray ) , as a function of initial fraction of red nodes @xmath70 and the total initial energy @xmath111 ; each of @xmath112 random realizations is shown by color point ; data are shown after @xmath77 time iterations at @xmath35 . the energy @xmath113 is the modulus of total energy with all spin up ; here @xmath90 . left and right panels show data for cambridge ( @xmath114 ) and oxford ( @xmath115 ) networks ; bars show color attribution to final probability @xmath61 . [ fig8],title="fig:",scaledwidth=23.0% ] , coded by color from @xmath109 ( black ) to @xmath110 ( blue / dark gray ) , as a function of initial fraction of red nodes @xmath70 and the total initial energy @xmath111 ; each of @xmath112 random realizations is shown by color point ; data are shown after @xmath77 time iterations at @xmath35 . the energy @xmath113 is the modulus of total energy with all spin up ; here @xmath90 . left and right panels show data for cambridge ( @xmath114 ) and oxford ( @xmath115 ) networks ; bars show color attribution to final probability @xmath61 . [ fig8],title="fig:",scaledwidth=23.0% ] -0.2 cm we argue that a physical reason of significantly non - boltzmann distribution is related to the local nature of spin flip condition which does not allow to produce a good thermalization on a scale of the whole system . indeed , there are various energetic branches and probably nonlocal thermalization flips of group of spins are required for a better thermalization . however , the voting is a local process that involves only direct neighbors that seems to be not sufficient for emergence of the global thermal distribution . the presence of a few energy branches is well visible from the data of fig . [ fig8 ] obtained at @xmath35 . this figure shows the diagram of final fraction @xmath61 of red nodes in dependence on their initial fraction @xmath70 and the total initial energy @xmath116 of the whole system corresponding to a chosen initial random configuration of spins . most probably , these different branches prevent efficient thermalization of the system with only local spin flip procedure . in addition to the above points the asymmetric form of @xmath40 couplings plays an important role generating more complicated picture compared to the usual image of thermal relaxation ( see e.g. @xcite ) . we also note that the thermalization is absent in voter models on regular lattices @xcite . the sznajd model @xcite nicely incorporates the well - known trade union principle `` united we stand , divided we fall '' into the field of voter modeling and opinion formation on regular networks . the review of various aspects of this model is given in @xcite . here we generalize the sznajd model to include in it the features of prof model and consider it on social networks with their scale - free structure . this gives us the prof - sznajd model which is constructed in the following way . for a given network we determine the pagerank probability @xmath3 and pagerank index @xmath4 for all @xmath1 nodes . after that we introduce the definition of _ group _ of nodes . it is defined by the following rule applied at each time step @xmath117 : * _ i ) _ pick by random a node @xmath1 in the network and consider the polarization of the @xmath118 highest pagerank nodes pointing to it ; * _ ii ) _ if the node @xmath1 and all other @xmath118 nodes have the same color ( same spin polarization ) , then these @xmath119 nodes form a group whose effective pagerank value is the sum of all the member values @xmath120 ; if it is not the case then we leave the nodes unchanged and perform the next time step ; * _ iii ) _ consider all the nodes _ pointing to any member of the group _ ( this corresponds to the model _ option 1 _ ) or consider _ all the nodes pointing to any member of the group and all the nodes pointed by any member of the group _ ( this corresponds to the model _ option 2 _ ) ; then check all these nodes @xmath121 directly linked to the group : if an individual node pagerank value @xmath122 is less than @xmath123 then this node joins the group by taking the same color ( polarization ) as the group nodes ; if it is not the case then a node is left unchanged ; the pagerank values of added nodes are then added to the group pagerank @xmath123 and the group size is increased . the above time step is repeated many times during time @xmath117 , counting the number of steps , by choosing a random node @xmath1 on each next step . this procedure effectively corresponds to the zero temperature case in the prof model . + 0.2 cm in the prof - sznajd model with the initial fraction of red nodes @xmath124 at one random realization . the curves show data for three values of group size @xmath125 ( blue / black ) ; @xmath126 ( green / light gray ) ; @xmath127 ( red / gray ) . full / dashed curves are for cambridge / oxford networks ; left panel is for option 1 ; right panel is for option 2 . [ fig9],title="fig:",scaledwidth=23.0% ] 0.2 cm in the prof - sznajd model with the initial fraction of red nodes @xmath124 at one random realization . the curves show data for three values of group size @xmath125 ( blue / black ) ; @xmath126 ( green / light gray ) ; @xmath127 ( red / gray ) . full / dashed curves are for cambridge / oxford networks ; left panel is for option 1 ; right panel is for option 2 . [ fig9],title="fig:",scaledwidth=23.0% ] -0.2 cm to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] + to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] + to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] to find a final red fraction @xmath61 , shown in @xmath71axis , in dependence on an initial red fraction @xmath70 , shown in @xmath72 axis ; data are shown inside the unit square @xmath73 . the values of @xmath74 are defined as a relative number of realizations found inside each of @xmath128 cells which cover the whole unit square . here @xmath76 realizations of randomly distributed colors are used to obtained @xmath74 values ; for each realization the time evolution is followed up the convergence time with up to @xmath129 steps . _ left column : _ cambridge network ; _ right column : _ oxford network ; here @xmath130 from top to bottom . the probability @xmath74 is proportional to color changing from zero ( blue / black ) to unity ( red / gray ) . [ fig10],title="fig:",scaledwidth=23.0% ] -0.2 cm a typical example of the time evolution of the fraction of red nodes @xmath131 in the prof - sznajd model is shown in fig . it shows that the system converges to a steady - state after a time scale @xmath132 that is comparable with the convergence times for the prof models studied in previous sections . we see that there are still some fluctuations in the steady - state regime which are visibly smaller for the option 2 case . we attribute this to a larger number of direct links in this case . the number of group nodes @xmath119 gives some variation of @xmath61 but these variations remain on a relatively small scale of a few percents . here , we should point on the important difference between prof and prof - sznajd models : for a given initial color realization , in the first case we have convergence to a fixed state after some convergence time while in the second case we have convergence to a steady - state which continue to fluctuate in time , keeping the colors distribution only on average . the dependence of the final fraction of red nodes @xmath61 on its initial value @xmath70 is shown by the density plot of probability @xmath74 in fig . [ fig10 ] ( option 1 of prof - sznajd model ) . the probability @xmath74 is obtained from many initial random realizations in a similar way to the case of fig . we see that there is a significant difference compared to the prof model ( fig . [ fig3 ] ) : now even at small values of @xmath70 we find small but finite values of @xmath61 while in the prof model the red color disappears at @xmath133 . this feature is related to the essence of the sznajd model : here even small groups can resist against totalitar opinion . other features of fig . [ fig10 ] are similar to those found for the prof model : we again observe bistability of opinion formation . the number of nodes @xmath119 , which form the group , does not affect significantly the distribution @xmath74 , we have smaller fluctuations at larger @xmath119 values but the model works in a stable way already at @xmath125 . the results for the option 2 of prof - sznajd model are shown in fig . [ fig11 ] . in this case the opinions with a small initial fraction of red nodes @xmath70 are suppressed in a significantly stronger way compared to the option 1 . we attribute this to the fact that large groups can suppress in a stronger way small groups since the outgoing direct links are taken into account in this option . but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] + but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] + but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] but for prof - sznajd model , option 2 . [ fig11],title="fig:",scaledwidth=23.0% ] -0.2 cm the significant difference between the two options of prof - sznajd model is well seen from the data of fig . [ fig12 ] . here , all @xmath31 nodes are taken in red ( compare with the prof model in fig . [ fig4 ] ) . for the option 1 the society elite succeeds to impose its opinion to a significant fraction of nodes which is increased by a factor 5 - 10 . visibly , this increase is less significant than in the prof model . however , for the option 2 of prof - sznajd model there is practically no increase of the fraction of red nodes . thus , in the option 2 the society members are very independent and the influence of the elite on their opinion is very weak . + 0.2 cm in the prof - sznajd model with the initial red nodes for the top pagerank nodes : @xmath134 ( blue / black ) ; @xmath58 ( green / light gray ) ; @xmath135 ( red / gray ) ; here @xmath136 . full / dashed curves are for cambridge / oxford networks ; left panel is for option 1 ; right panel is for option 2 . color of curves is red , green , blue from top to bottom at maximal @xmath117 on both panels . [ fig12],title="fig:",scaledwidth=23.0% ] 0.2 cm in the prof - sznajd model with the initial red nodes for the top pagerank nodes : @xmath134 ( blue / black ) ; @xmath58 ( green / light gray ) ; @xmath135 ( red / gray ) ; here @xmath136 . full / dashed curves are for cambridge / oxford networks ; left panel is for option 1 ; right panel is for option 2 . color of curves is red , green , blue from top to bottom at maximal @xmath117 on both panels . [ fig12],title="fig:",scaledwidth=23.0% ] -0.2 cm even if one can expect that the properties of university networks are similar to those of the real social networks it is important to analyze the previous prof models in the frame of real social network . for that we use the livejournal network collected , described and presented at @xcite . from this database we obtain the directed network with @xmath137 nodes , @xmath138 links which are mainly directed ( only about 30% of links are symmetric ) . the google matrix of the network is constructed in a usual way @xcite and its pagerank vector is determined by the iteration process at the damping factor @xmath7 . for the time evolution of fraction of red nodes @xmath139 we use time iterations in @xmath26 and @xmath117 defined as in previous sections . + 0.2 cm ( full curve ) , the fitted algebraic dependence is shown by the dashed line @xmath140 ( for @xmath141 ) with the exponent @xmath142 and @xmath143 . _ right panel : _ time evolution of opinion given by a fraction of red nodes @xmath30 as a function of number of iterations @xmath26 ( cf . [ fig1 ] ) at @xmath90 , few random initial realizations with @xmath28 are shown . [ fig13],title="fig:",scaledwidth=23.0% ] 0.2 cm ( full curve ) , the fitted algebraic dependence is shown by the dashed line @xmath140 ( for @xmath141 ) with the exponent @xmath142 and @xmath143 . _ right panel : _ time evolution of opinion given by a fraction of red nodes @xmath30 as a function of number of iterations @xmath26 ( cf . [ fig1 ] ) at @xmath90 , few random initial realizations with @xmath28 are shown . [ fig13],title="fig:",scaledwidth=23.0% ] -0.2 cm the pagerank probability decay @xmath50 is shown in fig . [ fig13 ] . it is well described by an algebraic law @xmath144 with @xmath145 . the convergence of a fraction of red nodes @xmath30 takes place approximately on the same convergence time scale @xmath146 even if the size of the networks is increased almost by a factor 20 . + 0.2 cm on the tenacious parameter @xmath17 ( or conformist parameter @xmath89 ) in the prof model for initial red nodes in @xmath31 values of pagerank index ( @xmath33 , cf . [ fig4 ] ) . here , @xmath36 blue ; @xmath147 green , @xmath148 red curves ( from bottom to top at @xmath90 ) ; @xmath35 . _ right panel : _ same data as in fig . [ fig3 ] at @xmath90 with same parameters but for livejournal network . [ fig14],title="fig:",scaledwidth=23.0% ] on the tenacious parameter @xmath17 ( or conformist parameter @xmath89 ) in the prof model for initial red nodes in @xmath31 values of pagerank index ( @xmath33 , cf . [ fig4 ] ) . here , @xmath36 blue ; @xmath147 green , @xmath148 red curves ( from bottom to top at @xmath90 ) ; @xmath35 . _ right panel : _ same data as in fig . [ fig3 ] at @xmath90 with same parameters but for livejournal network . [ fig14],title="fig:",scaledwidth=23.0% ] -0.2 cm in a way similar to the university networks we find that the homogeneous opinion of society elite presented in a small fraction of @xmath31 nodes influences a large fraction of the whole society especially when the parameter @xmath17 is not very large ( see fig . [ fig14 ] in comparison with fig . [ fig4 ] ) . the influence of the elite at 1% of red nodes is larger in the case of livejournal network . it is possible that this is related to a 30% larger number of links but it is also possible that other structural network parameters also play a role here . in spite of certain similarities with the previous data for university networks discussed before we find that the opinion diagram for the livejournal network ( see fig . [ fig14 ] right panel ) is very different from those obtained for the university networks ( see fig . [ fig3 ] ) : the bistability practically disappeared . we think that this difference originates from a significantly slower decay exponent for pagerank @xmath50 in the case of livejournal . to check this assumption we compare the probability distribution @xmath74 of final opinion @xmath61 for an initial opinion fixed at @xmath149 using the prof model with the usual linear weight @xmath92 in eq.([eq1 ] ) and a quadratic weight proportional to @xmath150 ( see fig . [ fig15 ] ) . for the linear weight we find that only very small values of @xmath151 can be found for initial @xmath149 while for the quadratic weight we obtain a rather broad distribution of @xmath61 values in the main range @xmath152 with a few large values @xmath153 . thus we see that the final opinion is rather sensitive to the weight used in eq.([eq1 ] ) . however , in contrast to the university networks ( see fig . [ fig3],fig . [ fig5 ] ) , where we have narrow one peak or double peak distributions of @xmath61 , for the livejournal network with quadratic weight we find a rather broad distribution of @xmath61 . in the spirit of a renormalization map description considered in @xcite ( see figs.1,2 there ) , it is possible to assume that one or two peaks corresponds to one or two fixed points attractor of the map . we make a conjecture that a broad distribution as in fig . [ fig15 ] ( right panel ) can correspond to a regime of strange chaotic attractor appearing in the renormalization map dynamics . in principle , such a chaotic renormalization dynamics is known to appear in coupled spins lattices when three - spin couplings are present ( see @xcite and refs . therein ) . it is possible that a presence of weight probability associated with the pagerank in a certain power may lead to a chaotic dynamics which would generate a broad distribution of final opinions @xmath61 . + 0.2 cm of final opinion @xmath61 for a fixed initial opinion @xmath149 and @xmath90 in the prof model . _ left panel : _ usual linear weight @xmath50 in eq.([eq1 ] ) . _ right panel : _ a quadratic weight @xmath154 in eq.([eq1 ] ) . histograms are obtained with @xmath155 initial random realizations , the normalization is fixed by the condition that the sum of @xmath74 over all histogram bins is equal to unity . [ fig15],title="fig:",scaledwidth=23.0% ] 0.3 cm of final opinion @xmath61 for a fixed initial opinion @xmath149 and @xmath90 in the prof model . _ left panel : _ usual linear weight @xmath50 in eq.([eq1 ] ) . _ right panel : _ a quadratic weight @xmath154 in eq.([eq1 ] ) . histograms are obtained with @xmath155 initial random realizations , the normalization is fixed by the condition that the sum of @xmath74 over all histogram bins is equal to unity . [ fig15],title="fig:",scaledwidth=23.0% ] -0.2 cm we also made tests for the prof - sznajd model ( option 1 ) for the livejournal database . however , in this case at @xmath149 , @xmath90 we found only small @xmath61 values ( similar of those in fig . [ fig15 ] , left panel ) both for linear and quadratic weights in eq.([eq1 ] ) . it is possible that the sznajd groups are less sensitive to the probability weight . we also analyzed the opinion formation on the twitter dataset with @xmath156 , @xmath157 taken from @xcite . this size is rather large and due to that we present only main features of the prof model for this directed network . + 0.2 cm ( full curve ) , the fitted algebraic dependence is shown by the dashed line @xmath140 ( for @xmath141 ) with the exponent @xmath158 and @xmath159 ( for the range @xmath160 we find @xmath161 ) . _ right panel : _ dependence of the final fraction of red nodes @xmath61 on the tenacious parameter @xmath17 ( or conformist parameter @xmath89 ) in the prof model for initial red nodes in @xmath31 values of pagerank index ( @xmath33 , cf . [ fig4 ] , fig . [ fig14 ] ) . here , @xmath162 ( blue line at @xmath109 ) ; @xmath163 ( red curve with circles ) , @xmath164 ( top gree line ) ; @xmath35 . [ fig16],title="fig:",scaledwidth=26.0% ] 0.1 cm ( full curve ) , the fitted algebraic dependence is shown by the dashed line @xmath140 ( for @xmath141 ) with the exponent @xmath158 and @xmath159 ( for the range @xmath160 we find @xmath161 ) . _ right panel : _ dependence of the final fraction of red nodes @xmath61 on the tenacious parameter @xmath17 ( or conformist parameter @xmath89 ) in the prof model for initial red nodes in @xmath31 values of pagerank index ( @xmath33 , cf . [ fig4 ] , fig . [ fig14 ] ) . here , @xmath162 ( blue line at @xmath109 ) ; @xmath163 ( red curve with circles ) , @xmath164 ( top gree line ) ; @xmath35 . [ fig16],title="fig:",scaledwidth=19.0% ] -0.2 cm the dependence of pagerank @xmath92 on its index @xmath32 is shown in fig . [ fig16 ] ( left panel ) . for the range @xmath165 we find the decay exponent @xmath166 , being similar to those of livejournal network ( see fig . [ fig13 ] ) even if there is a faster drop of @xmath92 at larger @xmath32 values . we note that the value @xmath167 is rather different from the value usually found for the zipf law @xcite and the www @xcite with @xmath168 . it is possible that this is related to a significantly larger average number of links per node which is increased by a factor 3.5 for the twitter network compared to the university networks analyzed in the previous sections . the effect of the homogeneous elite opinion of all red @xmath31 nodes is shown in fig . [ fig16 ] ( right panel ) . we see that on the twitter network a small fraction of elite with the fixed opinion ( @xmath169 ) can impose it practically to the whole community for all values of the conformist parameter @xmath170 . we find that for @xmath171 all @xmath61 values are very close to unity , while for @xmath172 we find @xmath109 as it is seen in fig . [ fig16 ] , right panel . thus , the transition is very sharp . we attribute such a strong influence of elite opinion to the very connected structure of twitter network with a significantly larger average number of links per node comparing to the university and livejournal networks . at @xmath90 , for a fixed fraction of initial opinion @xmath149 , we find that the probability distribution @xmath74 of final opinion @xmath61 is located in the range of small values @xmath173 both for linear @xmath92 and quadratic @xmath150 weight used in eq.([eq1 ] ) ( we do not show these data ) . for the linear weight the situation is rather similar to the case of livejournal ( see fig . [ fig15 ] ) , but for the quadratic weight we find a significant difference between two networks ( see fig . [ fig15 ] ) . the reason of such a significant difference for the quadratic weight case requires a more detailed comparison of network properties . the large size of twitter network makes numerical simulations of the prof - sznajd model rather heavy and due to that we did not study this model for this network . in this work we proposed the pagerank model of opinion formation of social networks and analyzed its properties on example of four different networks . for two university networks we find rather similar properties of opinion formation . it is characterized by the important feature according to which the society elite with a fixed opinion can impose it to a significant fraction of the society members which is much larger than the initial elite fraction . however , when the initial opinions of society members , including the elite , are presented by two options then we find a significant range of opinion fraction within a bistability regime . this range depends on the conformist parameter which characterizes the local aspects of opinion formation of linked society members . the generalization of the sznajd model for the scale - free social networks gives interesting examples of opinion formation where finite small size groups can keep their own opinion being different from the main opinion of the majority . in this way the proposed prof - sznajd model shows that the totalitar opinions can be escaped by small sub - communities . we find that the properties of opinion formation are rather similar for the two university networks of cambridge and oxford . however , the results obtained for networks of livejournal and twitter show that the range of bistability practically disappears for these networks . our data indicate that this is related to a slower algebraic decay of pagerank in these cases compared to the university networks . however , the deep reasons of such a difference require a more detailed analysis . indeed , livejournal and twitter networks demonstrate rather different behavior for the @xmath150-weighted function of opinion formation . the studies performed for regular networks @xcite show existence of stable or bistable fixed points for opinion formation models that have certain similarities with the opinion formation properties found in our studies . at the same time the results obtained in @xcite show that three - body spin coupling can generate a chaotic renormalization dynamics . some our results ( fig . [ fig15 ] , right panel ) give indications on a possible existence of such chaotic phase on the social networks . the enormous development of social networks in a few last years @xcite definitely shows that the analysis of opinion formation on such networks requires further investigations . this research can find also various other applications . one of them can be a neuronal network of brain which represents itself a directed scale - free network @xcite . the applications of network science to brain networks is now under a rapid development ( see e.g. @xcite ) and the google matrix methods can find useful applications in this field @xcite . this work is supported in part by the ec fet open project `` new tools and algorithms for directed network analysis '' ( nadine @xmath174 288956 ) . we thank a.benczr and s.vigna for providing us a friendly access to the livejournal database @xcite and the twitter dataset @xcite . zaller , _ the nature and origins of mass opinion _ , cambridge university press , cambridge uk ( 1999 ) . wikipedia , _ livejournal _ , march 9 , 2012 ` http://en.wikipedia.org/wiki/livejournal ` . wikipedia , _ facebook _ , march 9 , 2012 ` http://en.wikipedia.org/wiki/facebook ` . wikipedia , _ twitter _ , march 9 , 2012 ` http://en.wikipedia.org/wiki/twitter ` . wikipedia , _ vk ( social network ) _ , march 9 , 2012 ` http://en.wikipedia.org/wiki/vk\_(social\_network ) ` . s.n . dorogovtsev and j.f.f . mendes , _ evolution of networks _ , oxford univ . press ( 2003 ) . g. caldarelli , _ scale - free networks _ , oxford univ . press ( 2007 ) . s. galam , j. math . psychology * 30 * , 426 ( 1986 ) . liggett , _ stochastic interacting systems : contact , voter and exclusion processes _ , springer , berlin ( 1999 ) . s. galam , europhys . lett . * 70 * , 705 ( 2005 ) . watts and p.s . dodds , j. consumer research * 34(4 ) * , 441 ( 2007 ) . s. galam , int . c * 19 * , 409 ( 2008 ) . c. castellano , s. fortunato , and v. loreto , rev . 81 * , 591 ( 2009 ) p. l. krapivsky , s. redner and e. ben - naim , _ a kinetic view of statistical physics _ , cambridge university press , cambridge uk ( 2010 ) . b. schmittmann and a. mukhopadhyay , phys . e * 82 * , 066104 ( 2010 ) academic web link database project ` http://cybermetrics.wlv.ac.uk/database/ ` m. kurucz , a.a . benczur , a. pereszlenyi , _ large - scale principal component analysis on livejournal friends network _ , . _ workshop on social network mining and analysis _ held in conjunction with _ 13th acm sigkdd international conference on knowledge discovery and data mining ( kdd 2008 ) _ , las vegas nv , august 24 - 27 ( 2008 ) ; ` http://dms.sztaki.hu/en/letoltes/livejournal-data ` h. kwak , c.lee , h. park and s. moon , _ what is twitter , a social network or a news media ? www2010 , p.591 , acm , new york , n.y . ( 2010 ) ; the web data are downloaded from the web site maintained by s.vigna http://law.dsi.unimi.it/webdata/twitter-2010/ s.brin and l.page , _ computer networks and isdn systems _ * 30 * , 107 ( 1998 ) a.m. langville and c.d . meyer c d 2006 _ google s pagerank and beyond : the science of search engine rankings _ , princeton university press , princeton ( 2006 ) s. redner , phys . today * 58(6 ) * , 49 ( 2005 ) . f. radicchi , s. fortunato , b. markines , and a. vespignani , phys . e * 80 * , 056103 ( 2009 ) . west , t.c . bergstrom , and c.t . bergstrom , coll . * 71 * , 236 ( 2010 ) ; ` http://www.eigenfactor.org/ ` f. radicchi , plos one * 6 * , e17249 ( 2011 ) . zhirov , o.v . zhirov and d.l . shepelyansky , eur . j. b * 77 * , 523 ( 2010 ) l. ermann and d.l . shepelyansky , acta phys . polonica a * 120(6a ) * , a158 ( 2011 ) ; `` k. sznajd - weron and j. sznajd , int . j. mod c * 11 * , 1157 ( 2000 ) . frahm , b. georgeot and d.l . shepelyansky , j. phys , a : math . theor . * 44 * , 465101 ( 2011 ) l.ermann , a.d.chepelianskii and d.l.shepelyansky , _ towards two - dimensional search engines _ , arxiv:1106.6215[cs.ir ] ( 2011 ) ; ` http://www.quantware.ups-tlse.fr/qwlib/dvvadi/ ` s. galam and b. walliser , physica a * 389 * , 481 ( 2010 ) . n. metropolis , a.w . rosenbluth , m.n . rosenbluth , a.h.teller , and e. teller , j. chem . * 21 * , 1087 ( 1953 ) . v. sood and s. redner , phys . lett . * 94 * , 178701 ( 2005 ) . ananikian and s.k . dallakian , physica d * 107 * , 75 ( 1997 ) . zipf , _ human behavior and the principle of least effort _ , addison - wesley , boston ( 1949 ) . eguiluz , d.r . chialvo , g.a . cecchi , m. baliki , and a.v . apkarian , phys . lett . * 94 * , 018102 ( 2005 ) . x .- zuo , r. ehmke , m. mennes , d. imperati , f.x . castellanos , o. sporns and m.p . milham , cereb . cortex doi:10.1093/cercor / bhr269 ( 2011 ) . shepelyansky and o.v . zhirov , phys . a * 374 * , 3206 ( 2010 ) .

supernovae ( sne ) , the luminous explosions of stars , were observed since antiquity , with typical peak luminosity not exceedingergs ( absolute magnitude mag ) . it is only in the last dozen years that numerous examples of sne that are substantially super - luminous (ergs ; mag absolute ) were well - documented . reviewing the accumulated evidence , we define three broad classes of super - luminous sn events ( slsne ) . hydrogen - rich events ( slsn - ii ) radiate photons diffusing out from thick hydrogen layers where they have been deposited by strong shocks , and often show signs of interaction with circumstellar material . slsn - r , a rare class of hydrogen - poor events , are powered by very large amounts of radioactiveni and arguably result from explosions of very massive stars due to the pair instability . a third , distinct group of hydrogen - poor events emits photons from rapidly - expanding hydrogen - poor material distributed over large radii , and are not powered by radioactivity ( slsn - i ) . these may be the hydrogen - poor analogs of slsn - ii .

1208.3217

supernova explosions play important roles in many aspects of astrophysics , being sources of heavy elements , ionizing radiation and energetic particles ; driving gas outflows and shock waves that shape star and galaxy formation ; and leaving behind compact neutron star and black hole remnants . the study of supernovae in general has thus been actively pursued for many decades . the discovery of super - luminous supernova ( slsn ; figure [ lcfig ] ) events in the past decade is now focusing attention on these extreme explosions . the study of slsne is motivated , among other things , by their likely association with the deaths of the most massive stars ; their potential contribution to the chemical evolution of the universe and , at early times , to its reionization ; and since they may be manifestations of physical explosion mechanisms that differ from those of their more common and less luminous cousins . with extreme luminosities extending over tens of days ( fig . [ lcfig ] ) and , in some cases , copious ultra - violet ( uv ) flux , these events may become useful cosmic beacons to study distant star - forming galaxies and their gaseous environments ; their long duration , coupled with the lack of long - lasting environmental effects , as well as the fact that they eventually disappear and allow their hosts to be studied without interference , offer some advantages over other probes of the distant universe such as short - lived gamma - ray burst ( grb ) afterglows , and luminous high - redshift quasars . accumulated observations suggest that slsne can be naturally grouped into three distinct subclasses . in analogy to lower - luminosity explosions , we split slsne into hydrogen - rich events ( slsn - ii ) and events lacking spectroscopic signatures of hydrogen ( slsn - i ) . the latter group is further divided into a minority of events whose luminosity is dominated by radioactivity ( slsn - r ) and a distinct majority that require some other source of luminosity , which we simply refer to as slsn - i . a detailed discussion of these three classes follows in section @xmath6 below . supernovae have been traditionally classified mainly according to their spectroscopic properties ( see filippenko 1997 for a review ) , while the luminosity of sne does not play a role in the currently used scheme . in principle , almost all luminous sne belong to one of two spectroscopic classes : type iin , hydrogen - rich events with narrow emission lines which are usually interpreted as signs of interaction with material lost by the star prior to the explosion ; or type ic , events lacking hydrogen , helium , and strong silicon and sulphur lines around maximum , presumably associated with massive star explosions . as will be discussed below , the physical properties implied by the huge luminosities of slsne suggest they arise , in many cases , from progenitor stars that are very different from those of their much more common and less luminous analogs ; we will propose a provisional extension of the classification scheme that can be applied to super - luminous events . we consider below sne with reported peak magnitudes m@xmath4mag _ in any band _ as being superluminous ( fig . [ lcfig ] ) . see supplementary online material ( som ) section @xmath7 for considerations related to determining this threshold . as will be shown below , modern studies based on large sn samples and homogeneous , ccd - based luminosity measurements , show that slsne are very rare in nearby , luminous and metal - rich host galaxies ( with the possible exception of luminous sne iin ) . one would therefore expect that older sn surveys , mainly conducted by monitoring nearby galaxies drawn from galaxy catalogs ( dominated by luminous , metal - rich objects , we will refer to these as `` targeted surveys '' for short ) would find very few luminous events . indeed , richardson et al . ( 2002 ) find no solid detections of super - luminous events resulting from the older targeted surveys ( som @xmath8 ) . a handful of events are determined to be `` genuinely overluminous '' ( based on full , modern studies ) ; richardson et al . list sn 1997cy and sn 1999as ; we will add sn 1999bd for which we present here new data . these events , found by the first generation of wide - field non - targeted surveys , are indeed the harbingers of the classes of slsne we can finally define today . we note that even the brightest sne associated with cosmological gamma - ray bursts ( grbs , e.g. , the nickel - rich sn 1998bw ) fall well below our slsn threshold . the structure of this review is as follows : we begin by a brief summary of the observational work leading to the discovery of superluminous sne in section 2 , and we synthesize the accumulated information and define classes of superluminous sne in 3 . we review open questions and controversies in 4 , and conclude with a summary in 5 . ergs@xmath1 ( absolute magnitude @xmath2 ) . the new class of super - luminous sn ( slsn ) reach luminosities @xmath9 times higher . the prototypical events of the three slsn classes ( slsn - i ptf09cnd , quimby et al . 2011 ; slsn - ii sn 2006gy , smith et al . 2007 , ofek et al . 2007 , agnoletto et al . 2009 ; and sn 2007bi , gal - yam et al . 2009 ) are compared with a normal type ia sn ( nugent template ) , type iin sn 2005cl ( kiewe et al . 2011 ) , the average type ib / c light curve from drout et al . ( 2012 ) , the type iib sn 2011dh ( arcavi et al . 2011 ) and the prototypical type ii - p sn 1999em ( leonard et al . all data are in the observed @xmath10 band . see som for additional details.,scaledwidth=100.0% ] we now know that superluminous sne are intrinsically rare . their discovery therefore requires surveys that cover a large volume . the first generation of surveys covering large volumes were designed to find numerous distant type ia sne for cosmological use . these observed relatively small fields of view to a great depth , placing most of the effective survey volume at high redshift . possible early discoveries of slsne by such surveys and their modern counterparts are discussed in the som ( section @xmath8 ) . an alternative method to survey a large volume is to use wide - field facilities to cover a large sky area with relatively shallow imaging . with most of the survey volume at low redshift one can conduct an efficient untargeted survey for nearby sne . such surveys provided the first well - observed examples of slsne . prominent initial discoveries were made by the cosmology group at lawrence berkeley labs ( regnault et al . 2001 ) , e.g. , sn 1999as ( knop et al . 1999 ) , which turned out to be the first example of the class of extremely @xmath5ni - rich type i sne ( gal - yam et al . 2009 ) and sn 1999bd ( nugent et al . 1999 ; fig . [ sn1999bd - spec ] ) , which is probably the first well - documented case of a superluminous sn iin . the study of the peculiar sn 1997cy by germany et al . ( 2000 ) is perhaps the first modern study of a supernova which is substantially more luminous than the norm , though the absolute peak magnitude of this event ( -20.1 mag ) as well as those of similar objects discovered since ( e.g. , sn 2002ic , hamuy et al . 2003 ) fall below our fiducial limit defined above , and we do not discuss them any further . perhaps the most significant breakthrough in our observational knowledge about the most luminous sn explosions resulted from the operation of the texas supernova survey ( tss ) by quimby and collaborators ( quimby 2006 ) . on 2005 march 3 this survey discovered a hostless transient at mag 18.13 , its redshift was @xmath11 , indicating an absolute magnitude at peak around @xmath12mag , marking it as the most luminous sn discovered until that time ( quimby et al . 2007 ) . sn 2005ap is the first example of the class we will define below as slsn - i . on 2006 september 18 tss discovered a bright transient located at the nuclear region of the nearby galaxy ngc 1260 ( sn 2006gy ; smith et al . sn 2006gy suffered substantial host extinction ; correcting for this effect , the measured peak magnitude was extreme ( @xmath13mag ; ofek et al . 2007 ; smith et al . 2007 ) . spectroscopy of sn 2006gy clearly showed hydrogen emission lines with both narrow and intermediate - width components , leading to a spectroscopic classification of sn iin ; this is the prototype and best - studied example of the slsn - ii class . tss went on to discover additional luminous sne , as described in the som . during the last few years , several untargeted surveys have been operating in parallel , and now discover and report most sn events ( gal - yam & mazzali 2010 ) . the large volume probed by these surveys and their coverage of a multitude of low - luminosity dwarf galaxies have led , as expected ( young et al . 2008 ) to the discovery of numerous unusual sne not seen before in targeted surveys of luminous hosts ; indeed , it was shown that the sn population in dwarf galaxies is different than that observed in giant hosts ( arcavi et al . 2010 ) . as will be discussed below , luminous sne generally tend to prefer dwarf hosts ( e.g. , neill et al . 2011 ) and indeed numerous such events are being discovered by this new generation of untargeted surveys . the catalina real - time transient survey ( crts ) reported numerous slsn events including sn 2008fz , an extremely luminous member of the snls - ii family ( drake et al . results from the palomar transient factory ( ptf ; law et al . 2009 , rau et al . 2009 ) include the discovery of sn 2007bi ( gal - yam et al . 2009 ) , the first extensively - observed example of the radioactively - powered superluminous sn variety ( slsn - r ) during a `` dry - run '' initial phase of the project in 2007 , while the more recent ptf discovery of several members of the class of slsn - i allowed quimby et al . ( 2011 ) to define this class and derive its main properties . the panoramic survey telescope and rapid response system 1 ( pan starrs 1 ; hereafter ps1 , kaiser et al . 2010 ) appears to be an efficient program to detect numerous slsne at high redshifts ( out to @xmath14 and beyond , e.g. , chomiuk et al . 2011 ; pastorello et al . 2010 ) . more details about these surveys and their results are provided in the som . emission lines ) . a prominent emission bump around @xmath15(black tick marks ) is also a common feature . at a redshift of z=0.1512 , the absolute magnitude at discovery of sn 1999bd was -21.6 . telluric bands are marked and sections of the spectrum of sn 1999bd affected have been excised ; the telluric a band strongly absorbs the red wing of the h@xmath16 line in this spectrum . , scaledwidth=100.0% ] a total of 18 slsne have been discussed in the literature ( table 1 ) . these objects can be grouped into three classes that share observational and physical attributes . we now describe each of these classes in detail . of all classes of super - luminous sne , this seems to be the best understood . slsn - r events are powered by large amounts ( several m@xmath17 ) of radioactive @xmath5ni ( hence the suffix `` r '' ) , produced during the explosion of a very massive star . the radioactive decay chain @xmath5ni@xmath18co@xmath19fe deposits energy via @xmath20-ray and positron emission , that is thermalized and converted to optical radiation by the expanding massive ejecta . the luminosity of the peak is broadly proportional to the amount of radioactive @xmath5ni , while the late - time decay ( which in the most luminous cases begins immediately after the optical peak ) follows the theoretical @xmath5co decay rate ( @xmath21magday@xmath1 ) . the luminosity of this `` cobalt radioactive tail '' can be used to infer an independent estimate of the initial @xmath5ni mass . the first well - observed example of this group was sn 2007bi , discovered by the ptf `` dry run '' experiment . an extensive investigation of this object and its physical nature is presented in gal - yam et al . the most prominent physical characteristic of this group , the large @xmath5ni mass , is well - measured in this case using both the peak luminosity ( @xmath22mag ) and the cobalt decay tail , followed for @xmath23 days . estimates derived from the observations , as well as via comparison to other well - studied events ( sn 1987a and sn 1998bw ) converge on a value of m@xmath24m@xmath17 . the large amount of radioactive material powers a long - lasting phase of nebular emission , during which the optically thin ejecta are energized by the decaying radio nucleides . analysis of late - time spectra obtained during this phase ( gal - yam et al . 2009 ) provides independent confirmation of the large initial @xmath5ni mass via detection of strong nebular emission from the large mass of resulting @xmath5fe , as well as the integrated emission from all elements , powered by the remaining @xmath5co . estimation of other physical parameters of the event , in particular the total ejected mass ( which provides a lower limit on the progenitor star mass ) , its composition , and the kinetic energy it carries , is more complicated . there are no observed signatures of hydrogen in this event ( either in the ejecta or traces of csm interaction ) so the ejecta mass directly constrains the mass of the exploding helium core , which is likely dominated by oxygen and heavier elements . gal - yam et al . ( 2009 ) use scaling relations based on the work of arnett ( 1982 ) , as well as comparison of the data to custom light - curve models , and derive an ejecta mass of m@xmath25m@xmath17 . analysis of the nebular spectra provides an independent lower limit on the mass of m@xmath26m@xmath17 , with a composition similar to that expected from theoretical models of massive cores exploding via the pair - instability process . moriya et al . ( 2010 ) postulate a lower ejecta mass ( m@xmath27m@xmath17 ) ; this difference becomes crucial to the controversy about the explosion mechanism of these giant cores ( see below ) . in any case there is no doubt these explosions are produced by extremely massive stars , with the most massive exploding heavy - element cores we know . the same scaling relations used by gal - yam et al . ( 2009 ) also indicate extreme values of ejecta kinetic energy ( approaching e@xmath28erg ) . finally , the integrated radiated energy of this event over its very long lifetime is high ( @xmath29erg ) . sn 1999as was one of the first genuine slsne discovered ( knop et al . 1999 , see above ) and was initially analyzed by hatano et al . ( 2001 ) . as shown in gal - yam et al . ( 2009 ; see also fig . [ pisnfig ] ) this object was similar to sn 2007bi during its photospheric phase ( reaching @xmath30mag absolute at peak ) , while the analysis of hatano et al . ( 2001 ) suggests physical attributes ( @xmath5ni mass , kinetic energy , and ejected mass ) that are close to , but somewhat lower than , those of sn 2007bi . unfortunately , no late - time data have been collected for this object , so it is impossible to conduct the same analysis carried for sn 2007bi , but the similarities suggest this was another member of the slsn - r class . recently , the lick observatory supernova survey ( loss ; filippenko et al . 2001 ) using the 0.75 m katzman automatic imaging telescope ( kait ) discovered the luminous type ic sn 2010hy ( kodros et al . 2010 ; vinko et al . 2010 ) ; following the discovery by kait this event was also recovered in ptf data ( and designated ptf10vwg ) . it is interesting to note that while the loss survey is operating in a targeted mode looking at a list of known galaxies , by performing image subtraction on the entire kait field of view it is effectively running in parallel also an untargeted survey of the background galaxy population ( as noted by gal - yam et al . 2008 and li et al . it is during this parallel survey that kait detected this interesting rare sn , residing in an anonymous dwarf host . while final photometry is not yet available for this event , preliminary kait and ptf data indicate a peak magnitude of @xmath31mag or brighter , and it is spectroscopically similar to other slsn - r ( fig . [ pisnfig ] ) suggesting it is also likely a member of this class . objects of this sub - class are exceedingly rare ( this is observationally the rarest class among the slsn classes reviewed here ; i discuss relative volumetric rates below ) , and thus additional examples are scarce . during the last two years , the ptf survey has discovered another likely member ( ptf10nmn ; gal - yam 2011 ; fig . [ pisnfig ] ) with similar properties to sn 2007bi , while ps1 may have discovered another similar object at a higher redshift ( smartt 2011 ) . assembling a reasonable sample of such events may thus be a time - consuming process . young et al . ( 2010 ) present additional observations of sn 2007bi during the decline phase , as well as a detailed study of its host galaxy . they find the host is a dwarf galaxy ( with luminosity similar to that of the smc ) , with relatively low metallicity ( @xmath32 ) - somewhere between those of the lmc and smc . so , while the progenitor star of this explosion probably had sub - solar metal content , there is no evidence that it had very low metallicity . the host galaxy of sn 1999as is more luminous ( and thus likely more metal - rich ) than that of sn 2007bi , but still fainter than typical giant galaxies like the milky way ( neill et al . 2011 ) , while the host of ptf10nmn seems to be as faint or fainter than that of sn 2007bi . it thus seems this class of objects typically explode in dwarf galaxies . considering all available data , it seems there is agreement about the observational properties of this class and their basic interpretation : very massive star explosions that produce large quantities of radioactive @xmath5ni . a controversy still exists about the underlying explosion mechanism that leads to this result , either very massive oxygen cores ( m@xmath26m@xmath17 ) become unstable to electron - positron pair production and collapse ( gal - yam et al . 2009 ) , or else slightly less massive cores ( m@xmath33m@xmath17 ) evolve all the way till the common iron - core - collapse process occurs ( moriya et al . we will return to this question below . assuming , for the sake of the current discussion , that these explosions do arise from the pair instability , a clear prediction of the relevant theoretical models ( e.g. , heger & woosley 2002 , waldman 2008 ) is that for each luminous , @xmath5ni - rich explosion ( from a core around @xmath34m@xmath17 ) there would be numerous less luminous events with smaller @xmath5ni masses but large ejecta masses ( m@xmath26m@xmath17 ) . these should manifest as events with very slow light curves ( long rise and decay times ) and yet moderate or even low peak luminosities ( som ; fig . [ pisn - stats ] ) . this is probably the most commonly observed class of slsn . while some examples have been identified early on ( e.g. , sn 1999bd discussed above ; fig . [ sn1999bd - spec ] ) , these objects only became a focus of attention with the discovery of sn 2006gy ( ofek et al . 2007 ; smith et al . 2007 ) . since then , several additional examples have been studied in some detail . slsn - ii show strong hydrogen features in their spectra ; these explosions therefore typically occur within thick hydrogen envelopes , making investigations of their nature more complicated , as all information carried by electromagnetic radiation from the exploding core is reprocessed by the outer envelope . for this reason , our knowledge about the energy source ( or sources ) of these explosions is still mostly speculative . on the other hand , the physics responsible for converting the explosion energy into the observed radiation is better understood . two main physical processes have been invoked to explain the conversion of explosion energy to emitted radiation in slsn - ii . the first process assumes the explosion launches a powerful shock wave expanding outwards from the center of the star ; this shock heats the material it traverses until it eventually escapes from the _ effective _ outer edge of the star , where this effective edge is the radius around which the material in no longer optically thick to radiation . note that this effective edge does not necessarily coincide with the physical edge of the star , which we can define here as the radius inside which material is gravitationally bound to the star . the energy deposited by the shock is then slowly re - emitted by the hydrogen - rich material as photons diffuse out , in analogy to the more common and much less luminous type ii - p sne where this process occurs within the envelope of a red supergiant star . in order to achieve the much higher luminosities observed for slsn - ii , the radius of the effective edge of the star must be substantially larger compared to radii of even the largest red supergiants ; deposition of shock energy into more compact stars is radiatively inefficient as the deposited energy is quickly drained by adiabatic expansion . the observed luminosities probably require also an energetic explosion shock . the shape of the light curve is determined mostly by the density structure and composition of the material into which the energy was deposited . in order to explain the large effective radii ( @xmath35 cm ) required for this mechanism to work , several options have been suggested . these include very large ( bloated ) stars ( e.g. , gezari et al . 2009 ) , energy deposited into massive ( unbound ) optically - thick shells ejected by previous eruptions of the exploding star that have expanded to the required radius ( e.g. , smith & mccray 2007 ; miller et al . 2009 ) or energy deposited into an optically - thick massive stellar wind ( e.g. , ofek et al . 2010 ; chevalier & irwin 2011 ; moriya & tominaga 2011 ) extending out to the required radius . the second mechanism invoked in order to convert large explosion energies into optical emission is strong interaction between the expanding ejecta and massive circum - stellar material ( csm ) , previously lost from the progenitor star . this mechanism converts the kinetic energy carried by the expanding ejecta into radiation via strong shocks , and is commonly invoked for type iin sne . note that there is no conflict between this process ( converting kinetic energy to radiation ) and the previous one ( converting shock energy stored as internal heat energy into radiation ) and both can contribute in any given object ; there are some debates in the literature which is dominant for particular cases . since csm envelopes can be extremely extended , this process can in principle remain active for many years , and is thus more attractive to explain very long - lived events ( e.g. , sn 2006tf , smith et al . 2008 ; sn 2003ma , rest et al . on the other hand , conversion of kinetic energy into radiation should manifest in an observed decline in expansion velocities ; this process is therefore disfavored for events showing high expansion velocities that do not decrease significantly with time ( e.g. , sn 2008es , gezari et al . possible mechanisms invoked to eject large quantities of mass from the star prior to explosions include lbv - like activity ( e.g. , smith et al . 2007 ; 2008 ) and the pulsational pair instability ( e.g. woosley et al . 2007 ) . regardless of the conversion mechanism , the total emitted energy in several recently observed objects ( @xmath29erg ) is challenging to produce in regular iron - core - collapse explosions ( where @xmath36 of the initial explosion energy , @xmath37erg , is carried away by neutrinos ) . this led several authors to speculate about additional energy sources contributing to these powerful explosions . pair instability explosions can provide large kinetic energies and synthesize large amounts of radioactive @xmath5ni ; however , slsn - ii studied at late time did not follow the expected @xmath5co radioactive decay rate , in contrast to slsn - r ; the derived limits on the amount of initial radioactive nickel generally argue against slsn - ii resulting from energetic pair - instability explosions . spin - down of nascent magnetars ( rapidly spinning neutron stars with strong magnetic fields ) has been proposed as an alternative energy source ( woosley 2010 , kasen & bildsten 2010 ) , this process may be relevant at least for some slsn - ii ( e.g. , sn 2008es ) . one can also consider the so - called `` collapsar '' scenario where energy is extracted from material rapidly accreting onto a newly - formed black hole ; this process may be driving cosmological @xmath20-ray bursts ( macfadyen & woosley 1999 ) . when occurring within a massive star with a thick hydrogen envelope , this process may deposit the energy in the expanding envelope , where it may be thermalized and reemitted in the optical ( young et al . 2005 ; quimby et al . 2007 ) . unfortunately , the fact that any energy injected by such processes at the deep layers of the exploding stars is then reprocessed by the outer , optically thick hydrogen layers , makes investigations of such exotic processes in slsn - ii difficult , and mostly speculative so far . in any case , it is clear that slsn - ii are explosions of massive stars that retained their hydrogen envelopes until they exploded . for some objects spectroscopy indicates these stars have lost substantial amounts of mass prior to explosion ( e.g. , sn 2006gy , smith et al . 2007 ; sn 2006tf , smith et al . 2008 ) , suggesting perhaps the progenitor stars are similar to massive luminous blue variables ( lbvs ) which are known to undergo episodic eruptions involving extreme mass loss ( e.g. , humphreys & davidson 1994 ) ; a similar relation has been established in the case of a lower - luminosity sn iin ( sn 2005gl , gal - yam et al . 2007a ; gal - yam & leonard 2009 ) . the observational characteristics of slsn - ii are quite diverse . the peak luminosities of the brightest events are impressive , reaching well above @xmath38mag absolute ( e.g. , sn 2008fz , drake et al . 2010 ) . however , these seem to be the top of a broad distribution , with examples of luminous type iin events reported with peak magnitudes smoothly extending from these extreme values down to luminosities typical of the general sn population ( e.g. , sn 2010jl , stoll et al . 2011 , smith et al . 2011 ; see kiewe et al . 2011 for a review of older events ) . the light curve shapes are quite diverse , with some slsn - ii showing a rapid rise and decline ( e.g. , sn 2008es , gezari et al . 2009 ) , some showing light curves with a slow rise ( @xmath26 days ) to a broad peak ( e.g. , sn 2006gy , smith et al . 2007 , ofek et al . 2007 ; sn 2008fz , drake et al . 2010 ) , and some with rapid rise and very slow decline ( e.g. , sn 2003ma , rest et al . 2011 ; sn 2008am , chatzopoulos et al . 2011 ; and probably also sn 2006tf , smith et al . spectra ( fig . [ sn1999bd - spec ] ) also show diversity , with most objects showing narrow hydrogen balmer lines , and sn 2008es uniquely not . narrow balmer lines arise from a slow wind blown by the progenitor star prior to its explosion , assumed to have been photoionized by the explosion and to then recombine . the lack of such narrow lines in the spectra of sn 2008es suggests the progenitor star was not blowing massive winds for an extended period prior to its explosion ; any substantial mass loss must have been episodic ( e.g. , miller et al . 2009 ) or otherwise time - variable ( moriya et al . 2011 ) . the environments of slsn - ii are also quite diverse . this is the only slsn subclass that has been detected also in luminous , milky - way - like galaxies ( e.g. , sn 2006gy , ofek et al . 2007 , smith et al . 2007 ; css100217:102913 + 404220 , drake et al . 2011 ) . still , like other slsne , most slsn - ii events reside in dwarf star - forming hosts ( neill et al . 2011 ) . it is interesting to note that published slsn - ii events in giant host galaxies seem to have been discovered very close to their host nuclei , suggesting that perhaps specific conditions that are unique to this environment ( e.g. , in circumnuclear star - forming rings ) somehow mimic the conditions in star - forming dwarf galaxies . a review of slsn - ii events from the ptf survey confirms that five additional unpublished events reside either in dwarf hosts or in nuclei of giant hosts , supporting this interesting trend . since several slsn - ii have exploded in the high - metallicity cores of luminous galaxies , it seems that these events do not require low - metallcity progenitors . this final group of slsne was initially the most difficult to understand , with the two first reported events , sn 2005ap ( quimby et al . 2007 ) and scp 06f6 ( barbary et al . 2009 ) rather sparsly observed . it was only following the discovery of additional members of this class by the ptf survey , bridging the redshift gap between the relatively nearby sn 2005ap ( @xmath11 ) and the high - redshift scp 06f6 ( @xmath39 ) that a comprehensive view of this class of objects could be formed ( quimby et al . 2011 ; including spectroscopic redshifts based on mgii absorption lines , and the first correct identification of the redshift of scp 06f6 ; fig . [ slsn - i - spec ] ) . a major reason for the initial difficulties was that early spectra of these objects are quite featureless and the absorption lines that do appear are mostly of high - excitation low - mass elements ( fig . [ slsn - i - spec ] ) , while the elements commonly observed in most sn classes ( neutral oxygen , magnesium , iron and the ubiquitous ionized calcium ) appear only much later ( pastorello et al . 2010 ; som fig . [ sn2005ap - synow ] ) . observationally , these events are characterized by extreme peak luminosities ( often brighter than @xmath38mag absolute ) , very blue spectra with significant ultra - violet ( uv ) flux persisting for many weeks , and ( compared to other classes of slsne ) relatively fast - evolving light curves with rise times below 50 days and post - peak slopes that decline substantially faster than radioactive cobalt decay rates ( pastorello et al . 2010 ; quimby et al . contrary to early reports ( e.g. , quimby et al . 2007 , miller et al . 2009 , gezari et al . 2009 ) , these events do not show hydrogen in their spectra ( quimby et al . 2011 ; pastorello et al . 2010 ; som fig . [ sn2005ap - synow ] ) , and thus do not belong to the spectroscopic class of type ii sne . technically , these events should be classified as type ic sne ( as they also do not show strong he features in spectra taken around peak ) ; since the class of type ic sne is not positively defined and a physical connection between the events we consider here and more common sn ic events has not been firmly established ( see below ) , we refer to this class as slsn - i . quimby et al . ( 2011 ) provide extensive discussion about the physical nature of these objects based on the accumulated observational data . they show that the observed luminosity requires the deposition of significant amount of internal energy , taking place at large radii ( @xmath40 cm , approximately ten times the size of the largest red supergiants ) , into material expanding at high velocities ( @xmath41kms@xmath1 ) . the data can rule out traditional mechanisms discussed above ( radioactivity ; photon diffusion , interaction with massive hydrogen - rich csm ) . viable options for the energy conversion mechanism include interaction with expanding shells of hydrogen - free material ( chevalier & irwin 2011 ) , perhaps ejected by the pulsational pair - instability ( e.g. , woosley et al . 2007 ) ; or reemission of energy injected by an internal engine , such as magnetar spin - down ( woosley 2010 ; kasen & bildsten 2010 ) or a `` collapsar''-like accreting black hole ( e.g. , quimby et al . 2007 ) ; further discussed by pastorello et al . ( 2010 ) , chomiuk et al . ( 2011 ) and leloudas et al . ( 2012 ) . frustratingly , at this time , and even though these events are not enshrouded by massive , opaque hydrogen shells , the physical nature of the energy source remains speculative , and the energy conversion mechanism is also not clearly understood . the host galaxies of these events are again typically dwarf galaxies , though at higher redshifts luminosity upper limits on undetected hosts are less constraining ( neill et al . 2011 , quimby et al . the most natural explanation for these objects not occurring in more luminous galaxies is that a lower metallicity is required to form the progenitor stars of these events , but other explanations are also possible ( e.g. , different star formation modes or a top - heavy imf in dwarf galaxies ) . the extreme intrinsic luminosity and plentiful uv flux of these sources make them ideal probes of dwarf galaxies at high redshifts ; prospects for detecting numerous such events by deep surveys such as the ps1 medium - deep survey ( chomiuk et al . 2011 ) seem bright . .slsn properties . we list the reported redshifts , absolute peak magnitudes and total radiated energies as taken from the literature . the published post - peak magnitude of sn 2006tf ( smith et al . 2008 ; m@xmath42mag ) is below our fiducial cutoff . however , smith et al . argue it was probably brighter at peak ; this is confirmed by unpublished tss data that do cover the peak which is around @xmath31mag absolute ( r. quimby , private communication ) . [ cols= " < , < , < , < , < " , ] while the observations of slsn - r strongly indicate these events are powered by massive - star explosions that synthesize several solar masses of radioactive @xmath5ni , the physical nature of the explosion is a matter of some controversy . theoretical works suggest two options . the first is an extreme version of the iron - core - collapse model that is generally assumed to take place in explosions of massive stars that manifest as common type ii sne ( umeda & nomoto 2008 ; moriya et al . the second is the pair - instability mechanism ( e.g. , rakavy & shaviv 1967 ; barkat , rakavy & sack 1967 ; bond et al . 1984 ; heger & woosley 2002 ; scannapieco et al . 2005 ; waldman 2008 ) . the pair instability occurs during the evolution of very massive stars that develop oxygen cores above a critical mass threshold ( @xmath43m@xmath17 ) ; these cores achieve high temperatures at relatively low densities ; significant amounts of electron - positron pairs are created prior to oxygen ignition ; loss of pressure support , rapid contraction , and explosive oxygen ignition follow , leading to a powerful explosion that disrupts the star . extensive theoretical work indicates this result is unavoidable for massive oxygen cores ; when the core mass in question is large enough ( @xmath44m@xmath17 , as inferred for sn 2007bi ) many solar masses of radioactive nickel are naturally produced . umeda et al . ( 2008 ) and moriya et al . ( 2010 ) show that if one considers a carbon - oxygen core with a mass of @xmath45m@xmath17 ( just below the pair - instability threshold ) , which explodes with an ad - hoc large explosion energy ( @xmath46erg ) , one can produce the required large amounts of nickel ( umeda & nomoto 2008 ) , as well as recover the light curve shape of the slsn - r prototype , sn 2007bi ( moriya et al . 2010 ) . since both the pair - instability model and the massive core - collapse model fit the light curve shape of sn 2007bi equally well ; and progenitors of pair - instability explosions have larger cores and thus larger initial stellar masses , which are , assuming a declining initial mass function , intrinsically more rare , yoshida & umeda ( 2011 ) favor the core - collapse model . the two models ( pair instability vs. core - collapse ) agree about the nickel mass , but strongly differ in their predictions about the _ total _ ejected mass . total heavy - element masses above the @xmath47m@xmath17 threshold would indicate a core that is bound to become pair unstable , and will rule out the core - collapse model . gal - yam et al . ( 2009 ) estimated the total ejected heavy - element mass of sn 2007bi in several ways , including modelling of the nebular spectra of this event . the core - collapse model of moriya et al . 2010 does not fit these data ( gal - yam 2011 ) ; this model assumes a similar amount of radioactive @xmath5ni and lower total ejected mass ( to avoid the pair - instability ) leading to very strong nebular emission lines that are not consistent with the data . thus this model is not viable for this prototypical slsn - r object , supporting instead a pair instability explosion as originally claimed . it remains to be seen whether the massive core - collapse model does manifest in nature ( the prediction would be for slsne showing large amounts of radioactive nickel but relatively small amounts of total ejecta ) . as a final note , it should be stressed that while the stellar evolution models considered by yoshida & umeda ( 2011 ) require stars with exceedingly large initial masses ( @xmath48m@xmath17 ) to form pair - unstable cores at the moderate metallicity indicated for sn 2007bi ( young et al . 2010 ) , alternative models ( langer et al . 2007 ) predict that stars with much lower initial masses ( @xmath49m@xmath17 ) explode as pair - instability sne at smc- or lmc - like metallicities , though they may have to be tweaked to explain the lack of hydrogen in observed slsn - r spectra . the only measurement of the rate of slsne is a rough estimate provided by quimby et al . ( 2011 ) based on tss statistics , which , normalizing the rate of slsn - i at @xmath50 relative to that of sne ia , yields a rate of @xmath51mpc@xmath52y@xmath1 . this rate is substantially lower than the rates of core - collapse sne ( @xmath53mpc@xmath52y@xmath1 ) , and is also well below those of rare sub - classes like broad - line sne ic ( `` hypernovae '' ; @xmath54mpc@xmath52y@xmath1 ) or long gamma - ray bursts ( @xmath55mpc@xmath52y@xmath1 ; podsiadlowski et al . 2004 ; guetta & della valle 2007 ) . both the reported discovery statistics as well as unpublished ptf counts suggest that the rate of slsn - ii is comparable of larger than that of slsn - i , while slsn - r are rarer by about a factor of five , correcting for their slightly lower peak luminosities . i believe this suggests that slsn - r are indeed the rarest type of explosions studied so far , and quite possibly arise from stars that are at the very top of the imf . pastorello et al . ( 2010 ) emphasize the similarity between late - time spectra of sn 2010gx ( slsn - i ) and spectra of broad - line type ic sne ( see also quimby et al . 2011 ; som fig . [ sn2005ap - synow ] ) . they suggest this may be indicative of a physical connection between all slsn - i and sne ic . however , there are several physical differences between slsn - i and sne ic . detailed modelling of sne ic ( e.g. , sauer et al . 2006 ; mazzali et al . 2007 ; 2010 ) indicate the luminosity is dominated by radioactive @xmath5ni decay , and these objects show a correlation between the peak luminosity and the synthesized @xmath5ni mass ( e.g. , perets et al . 2010 ; drout et al . the same is true for slsn - r ( gal - yam et al . 2009 ) , but not for slsn - i ( quimby et al . 2011 ) in the sense that the nickel mass required to power the observed luminous peaks is in conflict with the later evolution of the light curve ( e.g. , quimby et al 2011 ; pastorello et al . 2010 ) . the luminosity of slsn - i must therefore come from a different source . it is interesting to note in this context a small group of very luminous broad - line sne ic ( sn 2007d , drout et al . 2011 ; sn 2010ay , sanders et al . 2011 ) with peak luminosities approaching those of slsn - r ( @xmath56mag and @xmath57mag absolute for sn 2007d and sn2010ay , respectively ) , but with significantly less @xmath5ni ( @xmath58m@xmath17 ) . other processes may be contributing to the large observed peak luminosity of these events ( e.g. , an internal `` engine '' ; sanders et al . perhaps these are intermediate events between the class powered purely by radioactivity ( normal sne ic and slsn - r ) and slsn - i for which the contribution from @xmath5ni is negligible , and which must be powered by some other process , as discussed above . another important physical distinction between slsn - i and sne ic is the size of the emitting region . as shown by quimby et al . ( 2011 ) , the energy radiated by slsn - i must have been deposited at large initial radii , @xmath59 cm . on the other hand , early observations of sne ic indicate the progenitor stars had an initial radius an order of magnitude smaller , ( @xmath60 cm ; corsi et al . 2011 ; sauer et al . 2006 ) and both the explosion shock energy and and radioactivity must therefore be contained within a much smaller initial radius . it thus seems that the observed spectroscopic similarity between slsn - i ( at late time ) and broad - line sne ic suggests similar ejecta composition and a large kinetic energy ( evident as significant amount of mass at high velocities ) , but other physical properties ( energy source , physical size ) are substantially different and argue for a different physical mechanism powering these two classes of objects . as a final curiosity , some photospheric spectra of slsn - r ( gal - yam et al . 2009 ; young et al . 2010 ; gal - yam 2011 ; fig . [ pisnfig ] ) show forbidden line emission , notably [ ca ii ] and probably also [ mg ii ] . such lines are usually only observed during the nebular phase of sne , when the ejecta are optically thin - this is clearly not the case here . the superposition of this nebular - like emission on an underlying photospheric spectrum may hint at a complex geometry of the emitting region ( motivating spectropolarimetric studies ) , but no explanations for this phenomenon have been put forth so far . during the last dozen years , numerous super - luminous sn events have been discovered and studied . the accumulated data suggest these can be grouped into three distinct subclasses according to their observational and physical attributes . radioactively - powered slsn - r seem to be the best understood ( and rarest ) class , while hydrogen - rich slsn - ii and the most luminous hydrogen - poor slsn - i are more common , but the physical origins of the extreme luminosity they emit is not clear at this time . with several ongoing surveys efficiently detecting additional examples , the amount of information about these objects , their rates and diversity , is likely to increase substantially in the coming few years . i would like to thank ofer yaron ; cccp members emilio enriquez , alicia soderberg , s. b. cenko , douglad leonard , derek fox , dae - sik moon , and david sand ; m. phillips and p. nugent ; e. chatzopoulos ; l. chomiuk ; s. b. cenko and robert quimby for use of data presented here , and e. nakar , p. mazzali , d. xu , s. b. cenko , a. soderberg , r. waldman , a. pastorello , i. arcavi , a. howell , r. quimby , e. o. ofek , a. drake , s. smartt , c. wheller and a. miller for useful advice . members of the ptf collaboration and in particular j. s. bloom , s. b. cenko , m. m. kasliwal , s. r. kulkarni , n. m. law , p. e. nugent , e. o. ofek , r. m. quimby , and d. poznanski are thanked for use of unpublished ptf material . anonymous referees are thanked for useful and constructive suggestions and comments . this work was supported by grants from the israeli science foundation , the us - israel binatioal science foundation , the german - israeli foundation , the minerva foundation , an arches award , and the lord sieff of brimpton fund . 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 . richardson et al . ( 2002 ) conducted a thorough study on the luminosity of supernovae ( sne ) discovered until mid-2001 . the most luminous among the common types of sne are type ia events ; richardson et al . therefore identified a `` sn ridgeline '' at absolute blue magnitude m@xmath61mag ( f@xmath62ergs@xmath1 ; equal to the peak luminosity of a normal sn ia ) as the location where the peak luminosities of cataloged events cluster , and defined `` overluminous sne '' as those being significantly more luminous than this value . to avoid confusion due to possible errors in measured peak luminosities ( arising , e.g. , from distance uncertainties to nearby events ) , and due to the high - luminosity tail of the scatter around the ridgeline value , it is useful to set the arbitrary threshold defining truly super - luminous supernovae ( slsne ) well away from the ridgeline ; examining figure 2 of richardson et al . ( 2002 ) the value of m@xmath63mag ( f@xmath64ergs@xmath1 ) seems appropriate . since the objects discussed below were discovered over a large redshift range and peak luminosities are reported in a variety of observed bands , we consider below sne with reported peak magnitudes m@xmath4mag _ in any band _ as being superluminous . this is justified since the typical bolometric or cross - filter corrections for sne around peak are below 1mag , and our fiducial threshold is well above the sn ridgeline . it would probably be better to use the time - integrated luminosity ( the total radiated energy ) rather than the peak value as an observational manifestation of unusually powerful explosions ; however , the total energy is more difficult to measure and is unknown for some objects . samples of events coming from future surveys may be better observed and allow the use of this alternative criterion . slsn events were not , so far , detected as luminous sources in wavelengths other than the optical - uv ; for example , see non - detection reports by gezari et al . ( 2009 ) or chomiuk et al . this is not surprising in view of the distance to most slsne discovered so far . the nearby slsn - ii event sn 2006gy is an interesting exception as its proximity makes detection of , e.g. , radio or x - ray emission feasible . ofek et al . ( 2007 ) report non - detection of this object both in radio and in x - rays . in contrast , smith et al . ( 2007 ) claimed a weak x - ray detection of this event . in my opinion , the analysis of ofek et al . ( 2007 ) as well as recent theoretical considerations ( chevalier & irwin 2012 ; svirski et al . 2012 ) suggest this event was not detected in x - rays . we now know that superluminous sne are intrinsically rare . their discovery therefore requires surveys that cover a large volume . such surveys have been conducted in the past two decades . the first generation of surveys covering large volumes were designed to find numerous distant type ia sne for cosmological use . they therefore observed relatively small fields of view to a great depth , placing most of the effective survey volume at high redshift . these cosmological surveys indeed found mostly high - redshift sne ia ( the most luminous type of common sne ) but they also discovered some `` contaminants '' - other types of luminous explosions at high redshifts ( e.g. , sn 1995av , deustua et al . 1995 ; sn 2000ei , schmidt et al . see also neill et al . no published investigations of these objects beyond the initial reports exist , but they may well have been examples of the super - luminous events we discuss below . a more recent similar narrow - deep cosmological survey using the hubble space telescope ( _ hst _ ) indeed uncovered one of the first bona - fide super - luminous sne ( slsn - i ; scp 06f6 , barbary et al . another recent survey , the canada - france - hawaii supernova legacy survey ( cfh snls ) , also uncovered unusually luminous events . among those was the first example of the `` super - chandra '' type ia events ( snls-03d3bb ; howell et al . 2006 ) , a class of sne ia that are substantially more luminous than the sn ia ridgeline , though not super - luminous according to the criteria defined above . in addition , the snls also detected events that , in retrospect , were very high - redshift ( @xmath65 ) members of the slsn - i class ( howell et al . 2010 ) . finally , re - inspection of data collected as part of the sdss - ii cosmological sn survey uncovered another fine example of an slsn - i ( leloudas et al . 2012 ) ; sdss photometry captures the rising light curve of this object and reveals an initial short plateau that may reflect the breaking of the initial explosion shock from a dense envelope . an alternative method to survey a large volume is to use wide - field facilities to cover a large sky area with relatively shallow imaging . in this way , most of the survey volume is at low redshift , and one can conduct an efficient untargeted survey for nearby sne . such surveys were initially more difficult to conduct since wide - field ccd cameras were challenging and expensive to construct ( while the narrow - deep surveys could be carried out with smaller focal - plane arrays ) ; analyzing the large data volumes from shallow - wide surveys was also a computational challenge . however , shallow - wide surveys were eventually undertaken and provided the first well - observed examples of super - luminous sne . the cosmology group at lawrence berkeley labs pioneered large - scale shallow - wide surveys , initially during the `` 1999 spring campaign '' ( regnault et al . 2001 ) and later on running the `` sn factory '' project ( copin et al . 2006 ) , which used the palomar 48 schmidt telescope . these surveys were designed to study sne ia , but the fact that the discovered targets were relatively bright allowed high s / n spectroscopy and other follow - up data to be collected , leading to the discovery of the first relatively well - studied superluminous events . the first notable event was sn 1999as ( knop et al . 1999 ) , which turned out to be the first example of the class of extremely @xmath5ni - rich slsn - r ( gal - yam et al . 2009 ; fig . [ pisnfig ] ) . during the same campaign , the extremely luminous sn 1999bd was discovered ( nugent et al . 1999 ; fig . [ sn1999bd - spec ] ) , which is probably the first well - documented case of a superluminous sn iin . interestingly , the mount stromlo abell cluster supernova survey ( msacss ; reiss et al . 1998 ) a similar wide - field shallow survey carried out at the southern hemisphere around the same time discovered sn 1997cy ( germany et al . 2000 ) the other luminous event considered by richardson et al . ( 2002 ) to be a bona - fide superluminous event . this event was technically a type iin event , but arguments about its real nature persist ( e.g. , hamuy et al . 2003 ; benetti et al . regardless , the absolute peak magnitude of this event ( -20.1 mag ) as well as those of similar objects discovered since ( e.g. , sn 2002ic , hamuy et al . 2003 ) fall below our fiducial limit defined above , and we do not discuss them any further . perhaps the most significant breakthrough in our observational knowledge about the most luminous sn explosions resulted from the operation of the texas supernova survey ( tss ) by quimby and collaborators ( quimby 2006 ) . this survey was unique in that the survey telescope , the 0.45 m rotse iiib instrument at mcdonald observatory , had a small aperture , and thus a shallow survey depth , typically 18.5 mag , while the large aperture of the main follow - up telescope , the 9.2 m hobby - eberley telescope ( het ) at the same observatory , enabled high s / n spectroscopic follow - up . while the number of rotse discoveries was modest , it had a high spectroscopic completeness , and in particular did not select against either `` hostless '' events ( transients without a visible host galaxy ) or events located at galactic nuclei . previous surveys , which typically suffered from a shortage in spectroscopic resources , preferred transients clearly offset from well - resolved galaxies to minimize the chances to observe variable - star or active galactic nuclei ( agn ) contaminants . this search strategy led to several important new discoveries . on 2005 march 3 this survey discovered a hostless transient at mag 18.13 ; follow - up spectroscopy with het and the 10 m keck telescope at mauna kea revealed its redshift was @xmath11 , indicating an absolute magnitude at peak around @xmath12mag and making it the most luminous sn discovered until that time ( quimby et al . 2007 ) . the interpretation of broad sn features detected was initially difficult ( since this was the first event of its kind ) and was limited by the availability of only two spectra covering a short temporal baseline , with limited coverage in the red . retrospective analysis informed by additional events ( quimby et al . 2011 ) , as well as additional spectroscopic data we present here ( fig . [ sn2005ap - synow ] ) show that early spectra can be explained almost exclusively by high - excitation lines of intermediate - mass elements ( mostly oii ; quimby et al . 2011 ) , while later spectra evolve to broadly resemble type ic supernova spectra of the broad - line subclass ( fig . [ sn2005ap - synow ] ; see also pastorello et al . 2010 ) ; thus sn 2005ap is the first published example of a slsn - i . on 2006 september 18 tss discovered a bright transient located at the nuclear region of the nearby galaxy ngc 1260 ( sn 2006gy ; smith et al . the close proximity of this event to the galactic nucleus ( 300pc ; e.g. , ofek et al . 2007 ) initially led to suspicion that it was an outburst related to the active galactic nucleus , but high angular resolution imaging confirming the slight offset from the galaxy center eventually indicated its supernova nature . events of this nature would have probably been missed by most surveys as likely related to agn ; the tss inclusive approach , as well as the fortunate proximity of this event , allowed its recognition as a stellar explosion . sn 2006gy suffered substantial host extinction ; correcting for this effect , the measured peak magnitude was extreme ( @xmath13mag ; ofek et al . 2007 ; smith et al . 2007 ) . spectroscopy of sn 2006gy clearly showed hydrogen emission lines with both narrow and intermediate - width components , leading to a spectroscopic classification of sn iin . the initial studies by ofek et al . ( 2007 ) and smith et al . ( 2007 ) were followed by numerous additional studies ( e.g. , woosley et al . 2007 , agnoletto et al . 2009 ; kawabata et al . 2009 ; smith et al . 2010 , miller et al . 2010 ) , and this object became the prototype and best - studied example of the slsn - ii class . tss went on to discover additional luminous sne , including the luminous sne iin 2006tf ( smith et al . 2008 ; @xmath66mag at peak ) and sn 2008am ( chatzopoulos et al . 2011 ; @xmath67mag at peak ) and the unique luminous type ii sn 2008es ( gezari et al . 2009 ; miller et al . 2009 ) . during the last few years , several untargeted surveys have been operating in parallel , and now discover and report most sn events ( gal - yam & mazzali 2010 ) . the large volume probed by these surveys and their coverage of a multitude of low - luminosity dwarf galaxies have led , as expected ( young et al . 2008 ) to the discovery of numerous unusual sne not seen before in targeted surveys of luminous hosts ; indeed , it was shown that the sn population in dwarf galaxies is different than that observed in giant hosts ( arcavi et al . luminous sne generally tend to prefer dwarf hosts ( e.g. , neill et al . 2011 ) and indeed numerous such events are being discovered by this new generation of untargeted surveys . the catalina real - time transient survey ( crts ) is employing telescopes in arizona and australia to search for optical transients ( drake et al . 2009 ) . as this survey utilizes a catalog - based search method ( comparing object catalogs from new imaging with reference catalogs ) it is more sensitive to transients that either have no visible host galaxy , or reside in a host that is much fainter than the transient . this makes this survey biased in favor of discovering very luminous sne , and indeed many such discoveries have been announced . prominent results from this survey include the discovery of sn 2008fz , an extremely luminous member of the snls - ii family ( drake et al . 2010 ) , as well as of the type iin sn 2008iy ( miller et al . 2010 ) which did not achieve a bright peak magnitude , but its unprecedented extended light curve implies a large integrated luminosity , suggesting it may be related to the snls - ii group . the palomar transient factory ( ptf ; law et al . 2009 , rau et al . 2009 ) is a wide - field variability survey using the 48 oschin schmidt telescope at palomar observatory and the refurbished cfht 12k camera . the survey employs image subtraction and is thus sensitive to transients regardless of their host luminosity . the ptf has been running since 2009 , and has discovered and spectroscopically confirmed over 1300 sne , including numerous luminous events . the first significant result from ptf was obtained during a `` dry - run '' initial phase of the project in 2007 conducted using the sn factory infrastructure . this was the discovery of sn 2007bi ( gal - yam et al . 2009 ) , the first extensively - observed example of the radioactively - powered superluminous sn variety ( slsn - r ) . the ptf discovery of several members of the class of slsn - i allowed quimby et al . ( 2011 ) to connect the previously disjoint objects sn 2005ap ( quimby et al . 2007 ) and scp 06f6 ( barbary et al . 2009 ) , define the class of slsn - i , and derive its main properties from the accumulated data . the panoramic survey telescope and rapid response system 1 ( pan starrs 1 ; hereafter ps1 , kaiser et al . 2010 ) project employs a new 1.8 m telescope at haleakala observatory to conduct wide - field sky surveys . so far , results have been reported mainly from the medium - deep survey , which covers relatively narrow sky areas to deep limits ( i.e. , most of its survey volume is at high redshift ) , using image subtraction techniques . this appears to be an efficient program to detect numerous slsne at high redshifts ( out to @xmath14 and beyond ) ; initial results include discovery of two high - redshift slsn - i ( chomiuk et al . 2011 ) , as well as analysis emphasizing the potential relation between slsn - i and less luminous sn ic varieties ( pastorello et al . 2010 ) . days after peak magnitude ( peak date : march 10.6 2005 ut ; quimby et al . 2007 ) compared to a synthetic synow fit . the spectral features are well explained by common elements ( oxygen , magnesium , silicon and iron ) ; the flux decrement on the blue probably results from blends of iron - group elements . the spectrum is overall similar to that of a broad - line type ic supernova ( the best fits using the _ superfit _ software ( howell et al . 2005 ) are of sn 1998bw and sn 2002ap ) , as pointed out by pastorello et al . ( 2010 ) for sn2010gx / ptf10cwr . the feature at rest wavelength @xmath68 is well - explained by a blend of si ii and o i lines . a contribution from h@xmath16 previously considered ( quimby et al . 2007 ) is disfavored since it would require extreme expansion velocities ( @xmath69kms@xmath1 ) ; while this was broadly similar to the photospheric velocities seen by quimby et al . ( 2007 ) in early spectra , it is inconsistent with the moderate photospheric velocity we see in this later spectrum ( @xmath70kms@xmath1 ) , and would further require the hydrogen velocity to increase with time , which seems unphysical . the spectrum ( available in digital form from the weizmann institute of science experimental astrophysics spectroscopy system ( wiseass ) at http://www.weizmann.ac.il/astrophysics/wiseass/ ) was obtained using the double - beam spectrograph ( dbsp , oke & gunn 1982 ) mounted on the palomar observatory 5 m hale telescope on april 15 , 2005 ut , as part of the cccp project ( gal - yam et al . , scaledwidth=100.0% ] ni mass and the he - core mass ( which should roughly equal the total ejected mass for type i events ) based on the ( non - rotating and non - magnetic ) models of heger & woosley ( 2002 ) . superposed are the data points for sn 2007bi ( gal - yam et al . 2009 ) and ptf10hgi ( previously unpublished ) . we also mark the approximate upper limit on object detectability ( m@xmath71m@xmath17 ) below which these events are expected to be too faint to be discovered anywhere except for the nearest galaxies , where the expected rate of these events is probably prohibitively low . , scaledwidth=100.0% ] the data presented in figure 1 are take from the following sources . the light curve of slsn - i ptf09cnd is taken from quimby et al . the data are not corrected for host galaxy extinction , and the distance modulus is calculated from the sn redshift @xmath72 , assuming the standard @xmath73cdm cosmology with h@xmath74kms@xmath1mpc@xmath1 . @xmath10-band observations of slsn - ii sn 2006gy are taken from smith et al . ( 2007 ) and agnoletto et al . ( 2009 ) , are corrected for extinction assuming @xmath75mag ( smith et al . 2007 ) and assuming the distance to ngc 1260 from the nasa extragalactic database ( ned ; @xmath76mpc , distance modulus of @xmath77mag ) . @xmath78-band observations of sn 2007bi are taken from gal - yam et al . ( 2009 ) , are not corrected for host galaxy extinction , and use a distance modulus derived from the host galaxy redshift @xmath79 assuming the same cosmology as above . @xmath10-band observations of type iin sn 2005cl are taken from kiewe et al . ( 2011 ) , assuming the host extinction estimated by these authors , and the distance modulus to the host galaxy provided by ned . the type ia sn light curve is a template @xmath10-band light curve of a normal sn ia provided by nugent ( http://supernova.lbl.gov/$\sim$nugent/nugent$_$templates.html ) . the type ib / c light curve is the `` average '' extinction - corrected light curve from drout et al . 2011 . @xmath10-band observations of type iib sn 2011dh are taken from arcavi et al . ( 2011 ) ; these are not corrected for host galaxy extinction ( which these authors argue is negligible ) and assume the distance modulus to m51 reported there . the light curve of the prototypical sn 1999em is taken from leonard et al . 2002 , adjusted to match the absolute plateau luminosity reported by these authors ( m@xmath80mag ) .

we list all the possible fundamental groups of the complements of real conic - line arrangements with two conics which are tangent to each other at two points , with up to two additional lines . for the computations we use the topological local braid monodromies and the techniques of moishezon - teicher and van - kampen . we also include some conjectures concerning the connection between the presentation of the fundamental group of the complements and the geometry of an interesting family of conic line arrangements .

math0305418

in this paper we compute and list the fundamental groups of complements of all real conic - line arrangements in @xmath0 with two conics , which are tangent to each other at two points , with up to two additional lines in any position . algorithmically , this paper uses the local computations ( local braid monodromies and their induced relations ) , the braid monodromy techniques of moishezon - teicher ( see @xcite , @xcite , @xcite , @xcite , @xcite and @xcite ) , the enriques - van kampen theorem ( see @xcite ) and some group calculations for studying the fundamental groups . see @xcite for detailed exposition of these techniques . these arrangements may appear in a branch curve of a generic projection to @xmath0 of a surface of general type ( see for example @xcite ) . the main results of this paper are as follows ( where @xmath1 is the unit element in the group and @xmath2 is the free group with two generators ) : [ th1 - 2 ] let @xmath3 be a curve in @xmath0 composed of two tangent conics . then : @xmath4 [ th1 ] there are three possible fundamental groups for a conic - line arrangement which consists of two tangent conics and an additional line : 1 . 2 . @xmath6 . @xmath2 [ th2 ] there are five possible fundamental groups for a conic - line arrangement which consists of two tangent conics and two additional lines : 1 . @xmath7=[x_2 , x_3 x_1 x_3^{-1}]=e \end{array } \right\rangle\ ] ] 2 . @xmath8 3 . @xmath9 4 . @xmath10 5 . @xmath11 the proof of proposition [ th1 - 2 ] is presented completely in section [ sec:1 ] . otherwise , we skip all the braid monodromy computations , and give only the presentation of the fundamental group obtained by the van kampen theorem . these computations and the proofs of propositions [ th1 ] and [ th2 ] appear in sections [ sec:2 ] and [ sec:3 ] , respectively . a group is called _ big _ if it contains a subgroup which is free ( generated by two or more generators ) . by the above results , we have the following corollary : [ bigness ] all the possibilities for fundamental groups of conic - line arrangements consist of two tangent conics and up to two additional lines are big . the proof of this corollary appears in section [ sec_big ] . we rule out some possibilities for conic - line arrangements by the following remark . [ remark ] the only possibility for a line to be tangent to both conics is that this line will pass through one of the tangency points of the two conics ( fourth case in section [ 2conics1line_other ] ) . the reason is the following . if a line was tangented to both conics , it would be a common point of the two duals to the two conics . now , since the dual curve of two tangented conics is two tangented conics too , there are only two common points to the two dual conics . these two common points correspond to the two tangency points between the original two conics . hence , a different line which tangents to both conics is impossible . the paper is organized as follows . section [ local ] presents the local computations related to the singular points appearing in the conic - line arrangements which we deal with . section [ sec:1 ] deals with the proof of proposition [ th1 - 2 ] . in section [ sec:2 ] we compute the different cases of conic - line arrangements which consist of two tangent conics and one additional line . section [ sec:3 ] deals with the different cases of conic - line arrangements which consist of two tangent conics and two additional lines . in section [ sec_big ] we show the simple proof of corollary [ bigness ] . section [ general ] deals with some conjectures concerning the connection between the presentation of the fundamental group of the complements and the geometry of an interesting family of conic - line arrangements . in the appendix , we list all the possibilities for two tangent conics and two additional lines with the corresponding fundamental groups of their complements . in this section , we present the local computations related to the singular points appeared in the conic - line arrangements which we deal with . in this section , we compute the local braid monodromy of a tangency point of two conics . the tangency point between two conics can be presented locally by the equation : @xmath12 ( see figure [ 2tancon ] ) . = 4.5 cm the singular point is @xmath13 , and the points of the curve in the fiber over @xmath14 are @xmath15 and @xmath16 . now , let us take a loop in @xmath17-axis based at @xmath14 and circumscribing @xmath18 : @xmath19 , where @xmath20 . lifting it to the curve , we get two paths on the curve : @xmath21 and @xmath22 where @xmath20 . projecting the paths to the @xmath23-axis , we get @xmath24 and @xmath25 where @xmath20 . by substituting @xmath26 , the local braid monodromy of this point is two counterclockwise full - twists . hence , we have : [ thm21 ] the local braid monodromy of the tangency point of two conics is two counterclockwise full - twists ( similar to the tangency point of a conic and a line ) . hence , we have the same induced relations : @xmath27 where @xmath28 are the generators of the standard g - base . now , we compute the corresponding lefschetz diffeomorphism : the lefschetz diffeomorphism of a tangency point between two tangent conics is a counterclockwise full - twist ( similar to the usual tangency point ) . the reason for this corollary is since the lefschetz diffeomorphism is obtained by going on the loop mentioned above from @xmath29 to @xmath26 . along this interval , the two points in the fiber make one counterclockwise full - twist , and hence the lefschetz diffeomorphism of this singular point is a full - twist . in this section , we compute the local braid monodromy of four types of singular points consist of three components ( where at most two of them are conics which are tangent ) . then , we compute the relations induced from these singular points by the classical van kampen theorem ( see @xcite ) . the local equation of the singularity of the first type is @xmath30 ( see figure [ type_b ] ) . = 2.5 cm for computing the braid monodromy , we take a loop around @xmath31 and look at what happens to the points of the curve @xmath30 in the fibers above this loop . let @xmath32 where @xmath20 . for @xmath33 we have that @xmath14 and the points of the curve in the fiber over @xmath14 are @xmath34 and @xmath35 . for @xmath29 , we have @xmath36 , and the points of the curve in the fiber over @xmath36 are @xmath37 and @xmath38 . the point @xmath15 in the fiber @xmath36 corresponds to the point @xmath15 in the fiber @xmath14 , and similarly the point @xmath16 in the fiber @xmath36 corresponds to the point @xmath16 in the fiber @xmath14 . the point @xmath38 in the fiber @xmath36 corresponds to the point @xmath35 in the fiber @xmath14 . hence , we get that from @xmath33 to @xmath29 , the points @xmath15 and @xmath16 do a counterclockwise full - twist and the point @xmath35 does a counterclockwise half - twist around the points @xmath34 . if we continue to @xmath26 , the points @xmath15 and @xmath16 do two counterclockwise full - twists and the point @xmath35 does a counterclockwise full - twist around the points @xmath34 together . now , we want to compute the induced relations from the braid monodromy of the singular point . according to van kampen s theorem , one should compute the g - base obtained by applying the action induced by the local braid monodromy of the singular point on the standard g - base . this is shown in figure [ gbase_type_b ] : in the first step , we perform two counterclockwise full - twists of points @xmath39 and @xmath40 which represent @xmath16 and @xmath15 respectively . in the second step , we perform a counterclockwise full - twist of point @xmath41 ( which represents the point @xmath35 ) around the points @xmath39 and @xmath40 . now , by van kampen s theorem , we get the following induced relations from the new g - base ( where @xmath42 are the generators of the standard g - base ) : 1 . @xmath43 2 . @xmath44 3 . @xmath45 from relation ( 1 ) , we get that @xmath46 . relation ( 2 ) becomes @xmath47 . now , we show that relation ( 3 ) is not needed . simplifying relation ( 3 ) yields @xmath48 . by relation ( 1 ) , this relation is equal to relation ( 2 ) , and hence relation ( 3 ) is redundant . to summarize , we proved the following : [ thm23 ] the local braid monodromy of the singularity presented locally by the equation @xmath49 is : two points ( corresponding to @xmath15 and @xmath16 ) do two counterclockwise full - twists and the third point ( corresponds to @xmath35 ) does a counterclockwise full - twist around them . the induced relations of this point are : @xmath50 where @xmath42 are the generators of the standard g - base . since the lefschetz diffeomorphism of the singular point is obtained by computing the action only on half of the unit circle ( from @xmath29 to @xmath26 ) , we have the following corollary : the lefschetz diffeomorphism of the singular point presented locally by @xmath30 is a counterclockwise full - twist of the points corresponding to @xmath15 and @xmath16 , and a half - twist of the point corresponding to @xmath38 ( which becomes @xmath35 ) around them . one can easily see that the singular point presented locally by the equation @xmath51 ( see figure [ type_g2 ] ) has the same braid monodromy and the same induced relations as the singular point which we have dealt with in this section , since these two singular points are locally the same ( from the topological point of view ) . = 2.5 cm the local equation of the singularity of the second type is @xmath52 ( see figure [ type_c ] ) . = 2.5 cm applying almost the same computations we have performed for the first type , we get the following result : the local braid monodromy of the singularity presented locally by the equation @xmath53 is : two points ( corresponding to @xmath15 and @xmath16 ) do two counterclockwise full - twists and the third point ( corresponding to @xmath38 ) do a counterclockwise full - twist around them . the corresponding induced relations are : @xmath54 where @xmath42 are the generators of the standard g - base . the lefschetz diffeomorphism of the singular point is a counterclockwise full - twist of the points corresponding to @xmath15 and @xmath16 , and a half - twist of the point corresponding to @xmath35 ( which becomes @xmath38 ) around them . as in the previous section , one can easily see that the singular point presented locally by the equation @xmath55 ( see figure [ type_g ] ) has the same braid monodromy and the same induced relations as the singular point which we have dealt with in this section , since these two singular points are locally the same . = 2.5 cm the local equation of the singularity of the third type is @xmath56 ( see figure [ type_h ] ) . one should notice that there is a major difference between this type of singularity and the two previous types : in this singularity there are two `` hidden '' branch points . that is , at any fiber one has three real points and two complex points ( i.e. complex level @xmath39 ) , and in each side of the singularity , the complex points belong to a different conic ( by the singularity , two real points become complex and two complex points become real ) . = 2.5 cm for computing the braid monodromy , we take a loop around @xmath31 and look what happens to the points of the curve @xmath56 in the fibers over this loop . let @xmath32 where @xmath20 . for @xmath33 we have that @xmath14 and the points of the curve in the fiber over @xmath14 are @xmath57 and @xmath58 . for @xmath29 , we have @xmath36 , and we get again that the points of the curve in the fiber over @xmath36 are @xmath57 and @xmath58 . by a careful checking , one can see that the point @xmath15 in the fiber @xmath36 corresponds to the point @xmath59 in the fiber @xmath14 , and similarly all the points except for @xmath60 in the fiber @xmath36 made a @xmath61 rotation counterclockwise around @xmath62 from their corresponding points in the fiber @xmath14 . the point @xmath60 remains fixed . now , when we continue to @xmath26 , the points of the fiber continue to move counterclockwise around @xmath60 . when we reach @xmath26 and reach back the fiber @xmath14 , the points in the fiber made a @xmath63 rotation counterclockwise around @xmath62 from their corresponding points in the initial fiber @xmath14 . hence , we have : the action of local braid monodromy of the point presented locally by the equation : @xmath64 is a @xmath63 rotation counterclockwise of the four points ( corresponding to @xmath65 ) around the point corresponding to @xmath60 , as shown schematically in figure [ bm_type_h ] . = 2.5 cm now , we want to compute the induced relations of this singular point . by van kampen s theorem , one should compute the g - base obtained by applying the action induced by the local braid monodromy of the singular point on the standard g - base . since we have two complex points in the fiber before the action of the braid monodromy and after it , we have to start by rotating the two rightmost points by @xmath61 counterclockwise , for representing the two complex points ( see step ( 1 ) in figure [ gbase_type_h ] ) . then , we move the two complex points to be @xmath66 and @xmath59 ( see step ( 2 ) there ) . now , we apply the action of the local braid monodromy ( step ( 3 ) ) , and then we return the two new complex points to the right side , and return them to the real axis by rotating them by @xmath61 clockwise ( step ( 4 ) ) . [ rem_clockwise ] one should notice here that when we return the points back to the real axis in step ( 4 ) , we rotate the points clockwise . the reason is that we want to compute the pure action of the braid on the g - base , and the clockwise rotation cancels the counterclockwise rotation which we have performed in step ( 1 ) . in the global computations , we indeed rotate always counterclockwise , since we work there in a model ( for simplifying the computations ) , and we can define the model as we want . we get the same results in the global computations even if we rotate clockwise the complex points into real points . now , by van kampen s theorem , we get the following induced relations from the new g - base ( where @xmath67 are the generators of the standard g - base ) : 1 . @xmath68 2 . @xmath69 3 . @xmath70 4 . @xmath71 5 . @xmath72 relation ( 2 ) is equivalent to @xmath73 . substituting @xmath74 by @xmath75 in relation ( 3 ) yields @xmath76 , which is equivalent to @xmath77 . by relation ( 2 ) , it can also be written as @xmath78 . we show that relation ( 5 ) is redundant . first , we can cancel @xmath79 , so we get : @xmath80 . since @xmath81 , we can cancel another @xmath79 , and by substituting @xmath74 by @xmath75 and some simplifications we get @xmath82 , which is equal to relation ( 3 ) . hence , relation ( 5 ) is redundant . therefore , we have the following corollary : the induced relations for the singular point presented locally by the equation @xmath64 are : 1 . @xmath73 2 . @xmath82 3 . @xmath68 4 . @xmath71 where @xmath67 are the generators of the standard g - base . now , we compute the corresponding lefschetz diffeomorphism : the lefschetz diffeomorphism of the singular point presented locally by @xmath64 is a @xmath61 rotation counterclockwise of the four points around the fixed point @xmath62 . this type is slightly different from the previous three types . in the previous types , the line intersects the tangency point , but was not tangent to it . in this type , the additional line and the tangency point have a common tangent . the local equation of the singularity of this type is @xmath83 ( see figure [ type_d ] ) . = 2.5 cm by similar arguments as in theorem [ thm21 ] , we get the local braid monodromy of the singularity presented locally by the equation @xmath84 is : the three points do two counterclockwise generalized full - twists . the induced relations of this point are : @xmath85 where @xmath42 are the generators of the standard g - base . the lefschetz diffeomorphism is a full - twist of all the three points . this result can be easily generalized to an arbitrary number of tangented components at the same point with a common tangent : let @xmath86 be a singular point which consists of @xmath87 smooth branches with a common tangent at @xmath86 . let @xmath88 be the generator which corresponds to the @xmath66th branch . then , the local braid monodromy is two generalized full - twists of the whole segment @xmath89 $ ] . moreover , the relations induced by this singular point are : @xmath90 notice that if the tangented components have a higher order of tangency ( i.e. the higher derivatives of the components are also equal up to the @xmath91th derivative , for a given @xmath91 ) , then the braid monodromy will consist of a higher power of a full - twist , and a higher exponent will appear in the relations . in this section , we compute the local braid monodromy of five cases of singular points consist of four components , where at most two of them are conics . then , we compute the relations induced from these singular points by the classical van kampen theorem ( see @xcite ) . a singular point consists of four components in a conic - line arrangements with two tangent conics and up to two additional lines can be of the following two types : take the tangency point of the two conics , and add two lines in the following two ways : one way is to add one line which will be tangent to both conics , and the second line will intersect both conics ( and the line ) at the singular point . the second way will be to add two intersecting lines at the tangency point of the two conics . as before , there are three possibilities for the intersecting line : it can be locally presented as the line @xmath92 , @xmath93 or @xmath31 . in spite of the fact that one can move from the first type to the second type by rotating the line , so locally the singularities are equivalent , from the global point of view these singularities induce different relations in the global fundamental group . [ [ 4comps_type1 ] ] * first type * + + + + + + + + + + + + the local equation of the singularity of the first type is @xmath94 ( see figure [ type_e ] ) . = 2.5 cm by a similar argument as in theorem [ thm23 ] , we have for computing the braid monodromy , we take a loop around @xmath31 and look what happens to the points of the curve @xmath94 in the fibers above this loop . let @xmath32 where @xmath20 . for @xmath33 we have that @xmath14 and the points of the curve in the fiber over @xmath14 are @xmath95 and @xmath35 . for @xmath29 , we have @xmath36 , and the points of the curve in the fiber over @xmath36 are @xmath95 and @xmath38 . the point @xmath15 in the fiber @xmath36 corresponds to the point @xmath15 in the fiber @xmath14 , and similarly the point @xmath16 in the fiber @xmath36 corresponds to the point @xmath16 in the fiber @xmath14 . the point @xmath60 is fixed , and the point @xmath38 in the fiber @xmath36 corresponds to the point @xmath35 in the fiber @xmath14 . hence , we get that from @xmath33 to @xmath29 , the points @xmath96 and @xmath15 do a counterclockwise generalized full - twist and the point @xmath35 does a counterclockwise half - twist around these three points . if we continue to @xmath26 , the three points @xmath97 perform two counterclockwise generalized full - twists and the point @xmath35 does a counterclockwise full - twist around the other three points . for computing the induced relations , we use the van kampen theorem . here , one should compute the g - base obtained by applying the action induced by the local braid monodromy of the singular point on the standard g - base . this is shown in figure [ gbase_type_e ] : in the first step , we perform two counterclockwise generalized full - twists of points @xmath98 and @xmath99 which represent @xmath96 and @xmath15 respectively . in the second step , we perform a counterclockwise full - twist of point @xmath41 ( which represents the point @xmath35 ) around the other three points . by van kampen s theorem , we get the following induced relations ( where @xmath100 are the generators of the standard g - base ) : 1 . @xmath101 2 . @xmath102 3 . @xmath103 4 . @xmath104 from relation ( 1 ) , we get that @xmath105 . relations ( 2 ) and ( 3 ) become @xmath106 and @xmath107 respectively . by relation ( 2 ) , relation ( 3 ) can be written as : @xmath108 and hence we get from relations ( 2 ) and ( 3 ) : @xmath109 we show now that relation ( 4 ) is redundant . from relation ( 4 ) , one can get : @xmath110 using relation ( 1 ) , this relation is equal to relation ( 3 ) , and hence it is redundant . hence , we have : the local braid monodromy of the singularity presented locally by the equation @xmath111 is : three points ( correspond to @xmath112 and @xmath16 ) do two counterclockwise generalized full - twists and the fourth point ( corresponds to @xmath35 ) do a counterclockwise full - twist around them . the induced relations from this singular point are : @xmath113 where @xmath100 are the generators of the standard g - base . the lefschetz diffeomorphism of the singular point is a counterclockwise generalized full - twist of the points correspond to @xmath112 and @xmath16 , and a half - twist of the point corresponds to @xmath38 ( which becomes @xmath35 ) . if we delete the generator @xmath74 which corresponds to the line that intersects the three tangent components , we get the following set of relations : @xmath114 as expected ( see section [ 3comps_type4 ] ) . similarly , if we delete the generator @xmath115 which corresponds to the tangent line , we get the following set of relations : @xmath116 again as expected ( see section [ 3comps_type1 ] ) . [ [ 4comps_type2 ] ] * second type * + + + + + + + + + + + + + the local equation of the singularity of the second type is @xmath117 ( see figure [ type_f ] ) . = 2.5 cm applying almost the same computations we have performed for the previous type , we get : the local braid monodromy of the singularity presented locally by the equation @xmath118 is : the points @xmath95 do two counterclockwise generalized full - twists , and the point @xmath38 does a counterclockwise full - twist around the points @xmath95 together . the corresponding induced relations are : @xmath119 where @xmath100 are the generators of the standard g - base . the lefschetz diffeomorphism of the singular point presented locally by @xmath117 is a counterclockwise generalized full - twist of the points correspond to @xmath112 and @xmath16 , and a half - twist of the point corresponds to @xmath35 ( which becomes @xmath38 ) . [ [ 4comps_type3 ] ] * third type * + + + + + + + + + + + + the local equation of the singularity of the third type is @xmath120 ( see figure [ type_j ] ) . = 2.5 cm for computing the braid monodromy in this case , we use the following trick : the three components in the middle @xmath121 can be thought for a moment as a `` thick '' line which is perpendicular to the other line . after this observation , we have two intersecting lines , whose local braid monodromy is a counterclockwise full - twist ( even though we have here a thick line @xmath31 , which is vertical , one can rotate it a bit getting two `` usual '' intersecting lines , without changing the braid monodromy ) . the local braid monodromy induced by the `` thick '' line itself ( which consists of the curve @xmath121 ) has already been computed ( section [ 3comps_type3 ] ) : we got there that the local braid monodromy is a @xmath63 counterclockwise rotation of four points around one fixed point at the origin . hence , the action of the local braid monodromy of the whole singular point can be summarized as follows : the local braid monodromy of the singular point presented locally by the equation @xmath122 is : first perform a @xmath63 counterclockwise rotation of the points correspond to @xmath65 around @xmath60 , and then the point corresponds to @xmath38 performs a full - twist with a block which consists of all the other points ( see figure [ bm_type_j ] ) . = 2.8 cm now , we want to compute the induced relations of this singular point . according to van kampen s theorem , one should compute the g - base obtained by applying the action induced by the local braid monodromy of the singular point on the standard g - base . since we have two complex points in the fiber before the action of the braid monodromy and after it , we have to start by rotating the two rightmost points by @xmath61 counterclockwise , for representing the two complex points ( see step ( 1 ) in figure [ gbase_type_j ] ) . then , we move the two complex points to be over the second to the left real point ( see step ( 2 ) there ) . now , we apply the action of the local braid monodromy ( steps ( 3 ) and ( 4 ) ) , and then we return the two new complex points to the right side , and return them to the real axis by rotating them clockwise by @xmath61 ( step ( 5 ) , see remark [ rem_clockwise ] ) . now , by van kampen s theorem , we get the following induced relations from the new g - base ( where @xmath123 are the generators of the standard g - base ) : 1 . @xmath124 2 . @xmath125 3 . @xmath126 4 . @xmath127 5 . @xmath128 6 . @xmath129 relation ( 2 ) can be written as : @xmath130 . using relation ( 1 ) , relation ( 3 ) becomes : @xmath131 . by relation ( 1 ) and some cancellations , relation ( 4 ) becomes : @xmath132 by relation ( 3 ) , we get : @xmath133 and hence we have : @xmath134 . now , we show that relation ( 6 ) is redundant . first , this relation can be written as : @xmath135 . by relation ( 1 ) and some cancellations , we have : @xmath136 . by relation ( 3 ) , we have : @xmath137 . by relation ( 2 ) , this relation is equal to relation ( 4 ) , and hence relation ( 6 ) is redundant . therefore , we get the following result for the set of relations for the singular point : the singular point presented locally by @xmath120 has the following set of induced relations : 1 . @xmath130 2 . @xmath138 3 . @xmath139 4 . @xmath140 where @xmath123 are the generators of the standard g - base . since the lefschetz diffeomorphism of the singular point is obtained by computing the action only on half of the unit circle ( from @xmath29 to @xmath26 ) , we have the following corollary : the lefschetz diffeomorphism of the singular point presented locally by @xmath120 is a @xmath61 counterclockwise rotation of the points correspond to @xmath65 around @xmath60 , and then the point corresponds to @xmath38 do a half - twist with a block which consists of all the other points . although the following two types are almost a rotation of each other , we need them both , since the second type includes also `` hidden '' branch points inside the singularity , as we had in section [ 3comps_type3 ] . [ [ 4comps_type4 ] ] * first type * + + + + + + + + + + + + the local equation of the singularity of the first type is @xmath141 ( see figure [ type_k ] ) . = 2.5 cm for computing the braid monodromy , we take a loop around @xmath31 and look what happens to the points of the curve @xmath141 in the fibers over this loop . let @xmath32 where @xmath20 . for @xmath33 we have that @xmath14 and the points of the curve in the fiber over @xmath14 are @xmath142 and @xmath35 . for @xmath29 , we have @xmath36 , and the points of the curve in the fiber over @xmath36 are @xmath143 and @xmath38 . the point @xmath15 in the fiber @xmath36 corresponds to the point @xmath15 in the fiber @xmath14 , and similarly the point @xmath16 in the fiber @xmath36 corresponds to the point @xmath16 in the fiber @xmath14 . the point @xmath38 in the fiber @xmath36 corresponds to the point @xmath35 in the fiber @xmath14 , and similarly the point @xmath35 in the fiber @xmath36 corresponds to the point @xmath38 in the fiber @xmath14 . hence , we get that from @xmath33 to @xmath29 , the points @xmath15 and @xmath16 do a counterclockwise full - twist and the points @xmath38 and @xmath35 do a counterclockwise half - twist around the points @xmath34 . if we continue to @xmath26 , the points @xmath15 and @xmath16 do two counterclockwise full - twists and the points @xmath38 and @xmath35 do a counterclockwise full - twist around the points @xmath34 together . hence we have the following result : the local braid monodromy of the singularity presented locally by the equation @xmath141 is : two points ( correspond to @xmath15 and @xmath16 ) do two counterclockwise full - twists and the two other points ( corresponds to @xmath38 and @xmath35 ) do a counterclockwise full - twist around them . the singular point has the following induced relations : @xmath144 where @xmath100 are the generators of the standard g - base . the lefschetz diffeomorphism of the singular point is a counterclockwise full - twist of the points @xmath15 and @xmath16 , and a counterclockwise half - twist of the points @xmath38 and @xmath35 around the points @xmath34 . [ [ 4comps_type5 ] ] * second type * + + + + + + + + + + + + + the local equation of the singularity of the second type is @xmath145 ( see figure [ type_i ] ) . = 2.5 cm there is a major difference between this type of singularity and the previous one : in this singularity there are two `` hidden '' branch points too ( as in section [ 3comps_type3 ] ) . that is , at any fiber , one has four real points and two complex points ( i.e. complex level @xmath39 ) , and in each side of the singularity , the complex points belong to a different conic ( since by the singularity , two real points become complex and two complex points become real ) . for computing the braid monodromy in this case , we use the same trick we have already used in section [ 4comps_type3 ] : the two lines in the middle , @xmath146 and @xmath60 , can be thought for a moment as a `` thick '' line which is perpendicular to the tangent direction of the two tangent conics . we have already computed this case ( section [ 3comps_type3 ] ) : we got there that the braid monodromy is a @xmath63 counterclockwise rotation of four points around one fixed point at the origin . after observing this , we should add into account that the `` thick '' line stands for two lines . hence , the fixed point at the origin is now decomposed into two close points , which are doing a counterclockwise full - twist ( like a usual node ) . to summarize , we have : the local braid monodromy of the singularity presented locally by the equation @xmath147 is : the four points correspond to @xmath148 and @xmath58 do a @xmath63 counterclockwise rotation , and the two points in the center do a counterclockwise full - twist ( see figure [ bm_type_i ] ) . = 2.8 cm now , we want to compute the induced relations of this singular point . by van kampen s theorem , we compute the g - base obtained by applying the action induced by the local braid monodromy of the singular point on the standard g - base . since we have two complex points in the fiber before the action of the braid monodromy and after it , we have to start by rotating the two rightmost points by @xmath61 counterclockwise , for representing the two complex points ( see step ( 1 ) in figure [ gbase_type_i ] ) . then , we move the two complex points to be over the two middle real points ( see step ( 2 ) there ) . now , we apply the action of the local braid monodromy ( steps ( 3 ) and ( 4 ) ) , and then we return the two new complex points to the right side , and return them to the real axis by rotating them by @xmath61 clockwise ( step ( 5 ) , see remark [ rem_clockwise ] ) . now , by van kampen s theorem , we get the following induced relations from the new g - base ( where @xmath123 are the generators of the standard g - base ) : 1 . @xmath149 2 . @xmath125 3 . @xmath150 4 . @xmath151 5 . @xmath128 6 . @xmath152 from relation ( 2 ) , we get @xmath130 . using this relation , relation ( 3 ) becomes : @xmath153 , and hence @xmath154 . using relation ( 2 ) again , we get @xmath155 . by relation ( 1 ) , relation ( 4 ) can be written : @xmath156 now , we show that relation ( 6 ) is redundant . since @xmath128 , we can simplify relation ( 6 ) to the following form : @xmath157 . now , by relation ( 1 ) , we get : @xmath158 . by some simplifications and relation ( 3 ) , we get relation ( 4 ) , and hence relation ( 6 ) is redundant . therefore , we get the following result : the singular point presented locally by the equation @xmath147 has the following set of relations : 1 . @xmath159 2 . @xmath156 3 . @xmath160 4 . @xmath140 where @xmath123 are the generators of the standard g - base . if we delete the generators @xmath161 or @xmath115 which correspond to the lines @xmath60 or @xmath146 respectively , we get the set of relations as in section [ 3comps_type3 ] as expected . since the lefschetz diffeomorphism is a half of the action of the braid monodromy , we have that : the lefschetz diffeomorphism of the singular point presented locally by @xmath147 is a counterclockwise @xmath61 rotation of the four points around the center ( which consists of two points ) , and then a counterclockwise half - twist of the two points in the center . in this section , we prove proposition [ th1 - 2 ] which states that if @xmath3 is a curve in @xmath0 which composed of two tangent conics , then : @xmath4 figure [ fig ] shows a curve composed of two tangent conics . for computing the braid monodromy of this curve , we first have to compute the lefschetz pairs of the singular points and to identify their types for applying the moishezon - teicher algorithm correctly ( see @xcite ) . in the following table , we summarize this data . [ cols="^,^,^,^",options="header " , ] by the moishezon - teicher algorithm , we get the skeletons related to the braid monodromy as shown in figure [ thm1-bm ] . by the van kampen theorem , we get the following presentation for the group @xmath162 : generators : @xmath100 . + relations : 1 . @xmath163 ( projective relation ) 2 . @xmath164 3 . @xmath165 4 . @xmath166 5 . @xmath167 6 . @xmath168 7 . @xmath164 it remains to show that this presentation is equivalent to the presentation of the group in the formulation of the proposition . by relation ( 2 ) ( which is equal to relation ( 7 ) ) , relation ( 6 ) is equal to relation ( 3 ) and relation ( 5 ) is equal to relation ( 4 ) . hence , relations ( 5),(6 ) and ( 7 ) are redundant . on the other hand , the first relation becomes @xmath169 . by relation ( 5 ) , we have @xmath170 , which yields that : @xmath171 . we also get that @xmath74 is redundant , and therefore we get the requested presentation . a _ simple tangency point _ can be presented locally as a tangency point between two smooth branches of the curve . we split our computations into two type of arrangements : arrangements with only simple tangency points and arrangements with other singular points . in the first subsection , we focus on arrangements with only simple tangency points , and then we deal with the other arrangement ( second subsection ) . in this section we prove that if @xmath3 is a curve in @xmath0 which composed of two tangent conics and one additional line which tangents to one of the conics and intersects the other ( see figure [ thm2-fig ] ) , then : @xmath172 this configuration is the unique configuration of two tangent conics with an additional tangent line , which tangents to the conics in a simple tangency point ( due to remark [ remark ] ) . by the braid monodromy techniques and the van kampen theorem , we get the following presentation for @xmath173 : generators : @xmath67 . + relations : 1 . @xmath174 ( projective relation ) 2 . @xmath175 3 . @xmath176 4 . @xmath177 5 . @xmath178 6 . @xmath179 7 . @xmath180 8 . @xmath181 9 . @xmath182 10 . @xmath183 by relation ( 6 ) , relation ( 5 ) has the following form : @xmath184 . this is equal to @xmath176 , which is already known by relation ( 3 ) . hence , relation ( 5 ) is redundant . by relation ( 3 ) , relation ( 7 ) is equal to @xmath185 , which is equivalent to relation ( 6 ) by relation ( 2 ) . hence , relation ( 7 ) is redundant too . by relation ( 3 ) again , relation ( 8) gets the form @xmath186 , and relation ( 10 ) is reduced to : @xmath187 . since @xmath81 , we have @xmath188 . hence , relation ( 8) is redundant . relation ( 4 ) and relation ( 9 ) are equal , since @xmath81 . therefore , we have the following equivalent presentation : generators : @xmath189 . + relations : 1 . @xmath190 2 . @xmath176 3 . @xmath177 4 . @xmath179 5 . @xmath191 substituting @xmath192 for @xmath161 in relation ( 1 ) and relation ( 2 ) , yields the relations @xmath193 and @xmath194 respectively . hence , we get the following presentation : generators : @xmath195 . + relations : 1 . @xmath193 2 . @xmath194 3 . @xmath177 4 . @xmath191 by the first relation , @xmath196 . substituting it for @xmath74 in relations ( 2 ) and ( 4 ) yields the relation @xmath197 twice , which is already known by relation ( 3 ) . hence , relations ( 2 ) and ( 4 ) are redundant . so the final presentation is @xmath198 which appears in proposition [ th1 ] as needed . this fundamental group was computed independently by degtyarev too ( see ( * ? ? ? * section 3.3.5 ) ) . apart from the previous case , we have four more possibilities to locate a line into a configuration of two tangent conics : 1 . the line intersects the conic transversally . the line passes through one of the tangency points between the two conics , but it is not tangent to the conics ( see figure [ case1_2 ] ) . the line passes through one of the tangency points between the two conics , and it is tangent to the conics ( see figure [ case1_3 ] ) . the line passes through the two tangency points ( see figure [ case1_1 ] ) . + based on proposition [ th1 - 2 ] and @xcite , the fundamental group of the complement of the first case is @xmath199 where @xmath1 is the identity element of the group . in the following subsections , we compute the third and the fourth cases , whence the result of the second case turns out to be isomorphic to the third case . using braid monodromy techniques and the van kampen theorem , we get the following presentation for the fundamental group of the complement of the curve which appear in figure [ case1_3 ] : generators : @xmath67 . + relations : 1 . @xmath174 ( projective relation ) 2 . @xmath175 3 . @xmath200 4 . @xmath201 5 . @xmath166 6 . @xmath165 7 . @xmath202 by relations ( 2 ) and ( 5 ) , relation ( 4 ) becomes @xmath203 , and hence @xmath204 by relation ( 2 ) , we get from relation ( 6 ) that @xmath205 . using relations ( 2 ) and ( 5 ) again , the first relation becomes : @xmath206 . by some cancellations and relation ( 6 ) , we get : @xmath207 . hence , we get the following equivalent presentation : generators : @xmath208 . + relations : 1 . @xmath207 2 . @xmath209 3 . @xmath210 4 . @xmath165 5 . @xmath211 by relation ( 1 ) and ( 4 ) , relations ( 3),(5 ) and the right equation of relation ( 2 ) become trivial . for the left equation of relation ( 2 ) , be relation ( 1 ) we substitute @xmath115 by @xmath212 , to get : @xmath213 by cancellations , we get : @xmath214 which becomes trivial by relation ( 4 ) and hence it is redundant . therefore , we get the following final presentation : @xmath215 using braid monodromy techniques and the van kampen theorem , we get the following presentation for the fundamental group of the complement of the curve which appear in figure [ case1_1 ] : generators : @xmath67 . + relations : 1 . @xmath174 ( projective relation ) 2 . @xmath71 3 . @xmath68 4 . @xmath216 5 . @xmath78 6 . @xmath217 7 . @xmath218 8 . @xmath219 9 . @xmath220 by relations ( 2 ) and ( 3 ) we replace @xmath79 and @xmath74 by @xmath221 and @xmath75 respectively . by these replacements , relations ( 4 ) and ( 8) become equal . moreover , relations ( 5 ) and ( 9 ) are the same ( using relation ( 4 ) ) . so we get the following presentation : generators : @xmath67 . + relations : 1 . @xmath222 2 . @xmath216 3 . @xmath78 4 . @xmath223 5 . @xmath224 relation ( 1 ) can be simplified ( using relation ( 2 ) ) to the following form : @xmath225 . hence , we have : @xmath226 . using this , all the other relations become trivial ( by simple computations ) . hence , the resulting group is the free group with two generators . the above two results were computed independently by degtyarev too ( see @xcite ) . as in the previous section , we divide our treatment into two cases : we first focus on arrangements with only simple tangency points ( first subsection ) , and then we deal with the other arrangement ( second subsection ) . in this section , we present the fundamental groups of the complement of the two different possibilities for curves with two tangent conics and two additional lines which are tangent in only simple tangency points : the case where each line is tangent a different conic , and the case where the two lines are tangent to the same conic . let @xmath3 be a curve in @xmath0 composed of two tangent conics and two additional lines , where each line is tangent to a different conic ( see figure [ thm3-fig ] ) , then : @xmath227 let @xmath3 be a curve in @xmath0 composed of two tangent conics and two additional lines , which are both tangent to the same conic ( see figure [ thm4-fig ] ) , then : @xmath228=[x_2 , x_3 x_1 x_3^{-1}]=e \end{array } \right\rangle\ ] ] the proofs of the above theorems are similar to the ones in section [ 2conics_1line ] . apart from the two previous cases of arrangements with only simple tangency points , we have @xmath229 more possibilities to locate two lines into a configuration of two tangent conics . in the following subsections , we compute only the cases which turn out to have nonisomorphic fundamental groups of the complements . in the appendix , we list all the 16 possibilities with the corresponding fundamental groups of their complements . in this subsection , we compute the fundamental group of the complement of the curve @xmath3 presented in figure [ case2_16 ] . based on proposition [ th1 - 2 ] and @xcite , the fundamental group of the complement of this case is @xmath230 where @xmath1 is the identity element of the group . in this subsection , we compute the fundamental group of the complement of the curve @xmath3 presented in figure [ case2_1 ] . using braid monodromy techniques and the van kampen theorem , we get the following presentation for the fundamental group of the complement : generators : @xmath123 . + relations : 1 . @xmath231 ( projective relation ) 2 . @xmath232 3 . @xmath233 4 . @xmath234 5 . @xmath235 6 . @xmath236 7 . @xmath237 8 . @xmath71 in the first step , we use relation ( 8) to substitute @xmath221 for @xmath79 everywhere . moreover , since relations ( 5 ) and ( 6 ) are both equal to @xmath74 , we can write the following equality : @xmath238 using relation ( 5 ) again , we have that the projective relation gets the following form : @xmath239 hence , the generator @xmath74 is redundant . so we get the following equivalent presentation : generators : @xmath240 . + relations : 1 . @xmath241 2 . @xmath242 3 . @xmath243 4 . @xmath244 5 . @xmath245 6 . @xmath237 from the first relation , we have that : @xmath246 , and hence : @xmath247 now , we replace @xmath115 by @xmath248 , in any place it appears . relation ( 2 ) gets the following form : @xmath249 by some reorderings , we get that : @xmath250 relation ( 3 ) can now be written as : @xmath251 by some cancellations , we get that : @xmath252 . by relation ( 2 ) , this is equal to : @xmath253 , which is known ( relation ( 4 ) ) , and hence relation ( 3 ) is redundant . relation ( 5 ) can be written : @xmath254 by some cancellations and reorderings , this is equal to : @xmath255 which is known ( from relation ( 2 ) ) , and hence this relation is redundant too . by similar computations , relation ( 6 ) can be written as : @xmath256 which is known ( relations ( 2 ) and ( 4 ) ) , and hence relation ( 6 ) is redundant too . the right part of relation ( 4 ) can be written as : @xmath257 which is known by relation ( 2 ) , and therefore it is redundant too . hence , we get the following equivalent presentation : @xmath258 as needed . in this subsection , we compute the fundamental group of the complement of the curve @xmath3 presented in figure [ case2_2 ] . using braid monodromy techniques and the van kampen theorem , we get the following presentation for the fundamental group of the complement : generators : @xmath123 . + relations : 1 . @xmath231 ( projective relation ) 2 . @xmath71 3 . @xmath176 4 . @xmath259 5 . @xmath260 6 . @xmath261 7 . @xmath262 8 . @xmath179 9 . @xmath263 10 . @xmath264 11 . @xmath265 using relations ( 2 ) and ( 8) , we can cancel generators @xmath161 and @xmath79 , and replace them by @xmath75 and @xmath221 respectively . so we get the following simplified presentation : generators : @xmath266 . + relations : 1 . @xmath267 2 . @xmath268 3 . @xmath269 4 . @xmath270 5 . @xmath261 6 . @xmath262 7 . @xmath271 8 . @xmath272 9 . @xmath273 by relation ( 5 ) , relation ( 2 ) is reduced to @xmath274 . by this relation and relation ( 5 ) again , relation ( 3 ) can be simplified to the trivial relation , and hence it is redundant . moreover , relation ( 4 ) gets the following form : @xmath275 . by relation ( 9 ) , we can simplify relations ( 7 ) and ( 8) : they get the following forms respectively : @xmath276 and @xmath277 . hence , we get the following simplified presentation : generators : @xmath266 . + relations : 1 . @xmath267 2 . @xmath194 3 . @xmath275 4 . @xmath261 5 . @xmath262 6 . @xmath276 7 . @xmath277 8 . @xmath278 since @xmath279 commutes with all the other generators , relation ( 7 ) is reduced to @xmath280 the right equation is known ( relation ( 3 ) ) , so we have to simplify only the left equation . by some simplifications , we get @xmath281 which is known too , and hence this relation is redundant . now , by relation ( 1 ) , we have that @xmath282 . so we can replace @xmath279 by @xmath283 in any place it appears . relation ( 5 ) gets the form : @xmath284 since @xmath74 commutes with @xmath115 and @xmath221 , this relation becomes trivial . relations ( 6 ) and ( 8) get the following forms respectively : @xmath285 and @xmath286 . again , by using relations ( 2 ) , ( 3 ) and ( 4 ) , these relations become trivial . hence , we get the following simplified presentation : generators : @xmath287 . + relations : 1 . @xmath194 2 . @xmath275 3 . @xmath261 this group is obviously isomorphic to : @xmath288 as needed . in this subsection , we compute the fundamental group of the complement of the curve @xmath3 presented in figure [ case2_7 ] . using braid monodromy techniques and the van kampen theorem , we get the following presentation for the fundamental group of the complement : generators : @xmath123 . + relations : 1 . @xmath231 ( projective relation ) 2 . @xmath71 3 . @xmath68 4 . @xmath73 5 . @xmath78 6 . @xmath289 7 . @xmath290 8 . @xmath291 9 . @xmath292 10 . @xmath293 11 . @xmath294 12 . @xmath295 13 . @xmath296 using relations ( 2 ) and ( 3 ) , we can cancel generators @xmath74 and @xmath79 , and replace them by @xmath75 and @xmath221 respectively . by a trivial simplification , relation ( 8) is equal to relation ( 4 ) , so it is redundant . so we get the following simplified presentation : generators : @xmath240 . + relations : 1 . @xmath297 2 . @xmath73 3 . @xmath78 4 . @xmath289 5 . @xmath298 6 . @xmath292 7 . @xmath299 8 . @xmath300 9 . @xmath295 10 . @xmath301 from relation ( 5 ) we have @xmath302 which is equal to relation ( 3 ) by relation ( 2 ) . hence , relation ( 5 ) is redundant . also , relation ( 8) follows immediately from relation ( 10 ) , and hence it is redudant too . since the right side of relation ( 1 ) is @xmath1 , we can move @xmath75 to the left , in order to get : @xmath303 . by relation ( 10 ) and a cancellation , we have @xmath304 . hence , we get the following presentation : generators : @xmath240 . + relations : 1 . @xmath304 2 . @xmath73 3 . @xmath78 4 . @xmath289 5 . @xmath292 6 . @xmath299 7 . @xmath295 8 . @xmath301 by the first relation @xmath305 , so we can replace @xmath161 by @xmath306 in any place it appears . we start with relation ( 7 ) . we have : @xmath307 . by some simplifications , we have : @xmath308 by relations ( 4 ) and ( 8) , we have : @xmath309 . relation ( 2 ) gets the following form : @xmath310 . by some simplifications , we have : @xmath311 . by the relation @xmath309 , we finally have that : @xmath278 . therefore , relation ( 4 ) is now redundant . moreover , relation ( 8) can be reduced to @xmath312 , which is known , and hence relation ( 8) is redundant too . since @xmath115 and @xmath221 commute with @xmath279 , relations ( 5 ) and ( 6 ) get the following forms respectively : @xmath313 and @xmath314 . if we continue the simplifications , we get that both relations become trivial , and hence they are redundant . finally , we simplify relation ( 3 ) . by relation ( 1 ) , it gets the following form : @xmath315 . by some simplifications , we get that this relation becomes trivial , and hence it is redundant too . therefore , we get the following simplified presentation : generators : @xmath316 . + relations : 1 . @xmath276 2 . @xmath278 this group is obviously the group : @xmath317 as needed . a group is called _ big _ if it contains a subgroup which is free ( generated by two or more generators ) . in this section we show that the fundamental groups which appear in this paper are big . we start with a simple observation . if @xmath318 is a group which has a big quotient , then @xmath318 is big itself : let @xmath319 be a normal subgroup of @xmath318 such that @xmath320 is big . let @xmath321 be the two generators of the free subgroup of @xmath320 , then the group generated by @xmath322 in @xmath318 is big . otherwise , there is a relation @xmath323 in @xmath318 , and therefore we have a corresponding relation @xmath324 , which is a contradiction to the freeness of the subgroup @xmath325 . using the observation , we show the following result : the group @xmath326 is big . first , we change the presentation : let us take new generators @xmath327 , then the relation @xmath328 becomes : @xmath329 , which is equal to : @xmath330 . hence , we have : @xmath331 now , the quotient of @xmath318 by the subgroup generated by @xmath332 is @xmath333 where @xmath334 is the free product . since @xmath335 is known to be big , by the observation we have that @xmath318 is big as needed . now we can prove corollary [ bigness ] . let @xmath336 be a conic - line arrangement with two tangent conics and up to two additional lines . since the two tangent conics is a sub - arrangement of @xmath336 , it is known that the fundamental group of the two tangent conics is a quotient of the fundamental group of @xmath336 ( by sending the generators which correspond to the additional lines to @xmath1 ) . since , we have shown in the previous proposition that the fundamental group of the two tangent conics is big , then we have that the fundamental group of @xmath336 is big too ( by using again the observation ) . hence , we are done . motivated by the presentations we achieved in the cases of two tangent conics , with up to two tangent lines with only simple tangnecy points , we want to propose some conjectures about the simplified presentation of the fundamental group of two tangent conics , with an arbitrary number of tangented lines , which are tangent in simple tangency points . let @xmath3 be a conic - line arrangement with two tangent conics , and @xmath87 lines which are tangent to these conics . assume that the tangency points of @xmath3 are simple . let @xmath337 be the conic with the maximal number @xmath338 of lines which are tangent to it . then the following holds in the simplified presentation of @xmath339 : 1 . we have @xmath338 relations of the type @xmath340 where @xmath341 is the generator corresponding to the conic @xmath337 and @xmath88 are some generators of the group . 2 . let @xmath23 and @xmath342 be two generators of @xmath339 which are different from @xmath341 . we have @xmath343 in some of the computed presentations , we do not have all the relations of the type mentioned in the second part of the conjecture . this happens since some of these relations are simplified to commutative relations between generators by other relations . for a specific case , we should get a decomposition : let @xmath3 be a conic - line arrangement with two tangent conics and @xmath87 lines , where @xmath344 lines are tangent to one conic , and one line @xmath345 is tangent to the other one . assume that all the tangency points are simple . let @xmath17 be the generator that corresponds to @xmath345 . then : @xmath346 in the appendix , we list all the possibilities for two tangency points and two additional lines , and the corresponding fundmanetal groups of the complement ( we assume that the line at infinity is transversal to our curve , i.e. there are no singularities on the line at infinity ) . @xmath347 @xmath348 = 2.5 cm @xmath349 = 1.5 cm @xmath350 we wish to thank alex degtyarev for fruitful discussions . the first and second authors wish to thank the institute of mathematics , erlangen ( germany ) , institut fourier , grenoble ( france ) , hershel farkas , ron livne and einstein institute of mathematics for hosting their stays . 99 amram , m. and teicher , m. , _ braid monodromy of special curves _ , j. knot theory & ramifications * 10*(2 ) , 171212 ( 2001 ) . degtyarev , a. i. , _ quintics in @xmath351 with non - abelian fundamental group _ , algebra i analiz * 11*(5 ) , 130151 ( 1999 ) ; translation in : st . petersburg math . j. * 11*(5 ) , 809826 ( 2000 ) . garber , d. , _ on the connection between affine and projective fundamental groups of line arrangements and curves _ , singularits franco - japonaises ( j .- p . brasselet and t.suwa , eds . ) , sminaires & congrs * 10 * , 6170 ( 2005 ) . garber , d. and teicher , m. , _ the fundamental group s structure of the complement of some configurations of real line arrangements _ , complex analysis and algebraic geometry , edited by t. peternell and f .- o . schreyer , de gruyter , 173223 ( 2000 ) . hirzebruch , f. , _ some examples of algebraic surfaces _ , contemp . math . * 9 * , 5571 ( 1982 ) . moishezon , b. , _ stable branch curves and braid monodromies _ , notes in math . * 862 * , 107192 ( 1981 ) . moishezon , b. and teicher , m. , _ braid group techniques in complex geometry i , line arrangements in @xmath352 _ , contemp . math . * 78 * , 425555 ( 1988 ) . moishezon , b. and teicher , m. , _ braid group techniques in complex geometry ii , from arrangements of lines and conics to cuspidal curves _ , algebraic geometry , lect . notes in math . * 1479 * , 131180 ( 1990 ) . moishezon , b. and teicher , m. , _ braid group technique in complex geometry iii , projective degeneration of @xmath353 _ , contemp . math . * 162 * , 313 - 332 ( 1994 ) . moishezon , b. and teicher , m. , _ braid group technique in complex geometry iv , braid monodromy of the branch curve @xmath354 of @xmath355 and application to @xmath356 _ , contemp . math . * 162 * , 332 - 358 ( 1994 ) . van kampen , e. r. , _ on the fundamental group of an algebraic curve _ , amer . j. math . * 55 * , 255260 ( 1933 ) .

we present an analytic calculation of the recoil and radiative recoil corrections to energy levels of positronium states and their hyperfine splitting . a complete analytic formula valid to is given for the spectrum of states . technical aspects of the calculation are discussed in detail . theoretical predictions are given for various energy intervals and compared with experimental results . _ pacs numbers : 36.10.dr , 06.20.jr , 12.20.ds , 31.30.jv_ # 1

hep-ph9901394

spectroscopy of positronium ( ps ) provides a sensitive test of bound state theory based on the quantum electrodynamics ( qed ) . because of the small mass of electron and positron , the effects of strong and weak interactions are negligible compared with the accuracy of present experiments . for this reason positronium represents a unique system which can , in principle , be described with very high precision by means of the qed only . tests of the qed predictions are made possible by the very high experimental accuracy of positronium spectroscopy @xcite . the gross spectrum of positronium is well described by the schrdinger equation with the coulomb potential . energy levels are @xmath3 where @xmath4 is the principal quantum number . for the purpose of interpreting modern experiments the precision of eq . ( [ elo ] ) is insufficient . corrections to the energy levels can in part be described by the quantum mechanics ; however , for a complete description one has to resort to the quantum field theory ( qft ) . unfortunately , an application of the qft to the bound states is difficult and special methods have to be devised @xcite . various approaches to bound state calculations have been reviewed e.g. in @xcite . here we focus on a method close to the so called non - relativistic quantum electrodynamics ( nrqed ) @xcite , which is an effective field theory based on the qed , for small energies and momenta . ( [ elo ] ) implies that the characteristic velocity of the electron and positron in positronium is of the order of the fine structure constant @xmath5 . it is appropriate to apply a non - relativistic approximation to this system . recently much progress has been achieved in the framework of non - relativistic effective theories , mainly by employing dimensional regularization . it has been shown @xcite that this regularization procedure permits an exact separation of effects arising at various characteristic energy scales . using that method , which we will call dimensionally regularized nrqed ( nrqed@xmath6 ) the complete energy spectrum of ps has been reproduced to order @xmath7 @xcite . more recently , we have computed @xmath8 corrections to the hyperfine splitting ( hfs ) of the ps ground state @xcite , confirming one of previously obtained numerical results @xcite . in the present paper we generalize that result to all @xmath2 states , confirming @xcite , and compute also their spin independent shift at @xmath0 ( obtained numerically in @xcite ) . it is convenient to describe the energy of an @xmath1 state of ps by dividing it up into the spin averaged part and a part dependent on the total ps spin ( hyperfine splitting ) : @xmath9 where @xmath10 is the total spin value of the ps and @xmath11 are the spins of the electron and positron , respectively . one finds : @xmath12 both the spin averaged energy and the hyperfine splitting can be represented by series in powers and logarithms of the fine structure constant . in the lowest order @xmath13 is given by eq . ( [ elo ] ) , and @xmath14 . to order @xmath7 the results for @xmath15 and @xmath16 were found in @xcite . those corrections have several sources : electron and positron charge radii and anomalous magnetic moments , vacuum polarization , two - photon exchange , two - photon annihilation and one - loop correction to the single - photon annihilation . current accuracy of high precision experiments requires a complete calculation of the @xmath0 corrections @xmath17 and @xmath18 . the most precisely measured property of positronium is the ground state hfs , i.e. the energy difference between the two lowest states with total spin @xmath19 and @xmath20 . two best experimental values are @xmath21 found in @xcite and @xmath22 obtained in @xcite . another quantity of the experimental interest is the energy difference of @xmath23 and @xmath24 states @xcite : @xmath25 the absolute accuracy of this measurement is clearly less impressive than that of the hyperfine splitting . however , since @xmath26 , a complete calculation of the energy levels at this order is warranted . at order @xmath8 both @xmath17 and @xmath18 can be written as @xmath27 the logarithmic contributions at this order , @xmath28 , present in the annihilation @xmath29 and recoil @xmath30 corrections , were found first @xcite . @xmath31 arises from the radiative corrections to the breit potential at @xmath32 @xcite . the three , two , and one - photon annihilation contributions giving @xmath29 were found in @xcite , @xcite , and @xcite , respectively . the non - annihilation radiative recoil contributions @xmath33 were calculated in @xcite , while pure recoil corrections @xmath30 were obtained in @xcite for the hfs and in @xcite for @xmath15 . in this paper we present an analytic calculation of the recoil and radiative recoil corrections , @xmath31 and @xmath34 , to energy levels of arbitrary @xmath1 positronium states . the rest of this paper is organized as follows : in section [ sec : frame ] we discuss our method in general terms . section [ sec : hfs ] is devoted to the calculation of the hfs . many technical details of this calculation are discussed there . in section [ sec : lev ] we present a calculation of the average energy @xmath35 . it is very similar to hfs , except that some additional operators contribute . also the @xmath0 radiative recoil corrections are discussed . our results are summarized in section [ sec : sum ] , where also an overview of the theoretical and experimental situation is given and a complete analytic formula for the @xmath1 energy levels to order @xmath8 is presented . before getting into details , let us describe the general framework of our calculation of the @xmath0 corrections to energy levels . first , we calculate an on shell scattering amplitude for non - relativistic @xmath36 particles to the needed order ( the fact that @xmath37 in ps serves as a counting rule for contributions of various operators ) . in addition to the leading , single coulomb exchange , this includes the relative @xmath38 breit corrections and also higher order @xmath39 terms . this non - relativistic amplitude is gauge invariant , and taken with a minus sign provides the potential for non relativistic particles . next , we use the ordinary quantum mechanical perturbation theory to find the corrections due to that potential ; as unperturbed states we use the solutions of the schrdinger equation with the coulomb potential . we get the @xmath0 correction to energy levels as the sum of the first order correction due to @xmath40 perturbation and of the second order correction due to the breit hamiltonian . previously , this scheme was used for the calculation of the @xmath41 corrections to the levels of @xmath2-states @xcite and of the @xmath42 corrections to the levels of @xmath43-states @xcite . in the present calculation the result of the non - relativistic calculation is divergent . this is because also the short - distance ( `` hard '' ) corrections contribute . they arise from virtual momenta regions of the order of electron mass and can not be obtained from the non relativistic expansion . our calculation is performed in the spirit of nrqed . we apply dimensional regularization , which offers technical advantages over more common techniques , based on the introduction of an intermediate cut - off to separate the relativistic and non relativistic momentum regions . dimensional regularization makes the matching of the low - scale effective theory and the complete qed extremely simple . we find that in the sum of the short and long distance contributions the singularities in the parameter and @xmath44 . ] @xmath45 disappear and one arrives at a finite result . the spinor algebra in dimensional regularization requires some comments . in order to obtain the energy shift due to an operator @xmath46 one has to calculate the trace of the form @xmath47 $ ] , where @xmath48 is an appropriate wave function . the spinor parts of the relevant wave functions are @xmath49 for para and orthopositronium states , respectively . in the latter case , @xmath50 is the polarization vector ( we average over its directions ) . the traces are calculated in a standard way in the @xmath51-dimensional space . one encounters only even numbers of @xmath52 matrices , and we treat them as anticommuting . since the matrix elements involve the positronium wave function , it is easiest to calculate for the ground state ( @xmath53 ) . however , once the corrections to the ground state have been found , there is a convenient way of finding them for excited states , with an arbitrary value of the principal quantum number @xmath4 . only the non - relativistic contributions have a non - trivial dependence on @xmath4 . their computation in dimensional regularization would be difficult . however , this task is simplified using other regularizations . finally , we eliminate the cut - off dependence by requiring that for @xmath53 the result matches the formula we found for @xmath53 . the freedom of choosing the regularization scheme simplifies considerably this part of the calculation . in this section we present a calculation of the recoil corrections to the ps ground state , @xmath54 . it is given as a sum of soft ( non - relativistic ) eq . ( [ hfsnonrel ] ) and hard eq . ( [ hard ] ) scales : @xmath55 those two groups of contributions are computed , respectively , in sections [ sec : soft ] and [ sec : hard ] . further , in section [ sec : nhfs ] , we find a generalization of this result for radially excited states ( arbitrary @xmath4 ) : @xmath56 , \label{ehfsrecn}\ ] ] where @xmath57 is the logarithmic derivative of the @xmath58-function and @xmath59 is the euler constant . the @xmath4 dependence of this result and its numerical value at @xmath53 are in agreement with @xcite . we divide up the non - relativistic contributions to hfs into 6 parts : tree level coulomb and magnetic photon exchanges , retardation , one - loop operators , and second iteration of breit hamiltonian which includes intermediate s and d wave states : @xmath60 these partial results , given in eqs . ( [ deltac ] , [ deltam ] , [ deltaret ] , [ delta1 ] , [ swavehfs ] , [ deltad2 ] ) add up to @xcite @xmath61 in the remainder of this section we discuss in detail how these contributions are calculated . according to standard procedure @xcite we identify the on shell scattering amplitude , taken with the minus sign , with the matrix element of an interaction operator in the momentum representation . the soft scale contributions are calculated using the time - independent `` old - fashioned '' perturbation theory and the coulomb gauge . since this technique is not very common , let us recall its basic ingredients . exchange of a coulomb or magnetic photon is described , respectively , by @xmath62 or @xmath63 . in the latter case , the denominator @xmath64 arises from the magnetic photon s phase space element . an intermediate state introduces the factor @xmath65 , where @xmath66 is the energy of the intermediate state and @xmath67 is the total energy of the process . dirac spinors are @xmath68 where @xmath69 denotes the four - spinor of a particle at rest ; projectors on the positive and negative electron energy states are given by @xmath70 in an expression for the potential the projector @xmath71 contributes an additional minus sign . we begin with the contributions of the tree level effective operators , describing an exchange of the coulomb or magnetic quanta . the tree level operators , relevant for the @xmath0 calculation of the hfs , arise as @xmath38 corrections to the breit potential . for the hfs we need the spin - dependent part of the @xmath72 correction to the coulomb exchange ( see eq . ( [ eqa : coul ] ) ) : @xmath73[{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$}},{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$ } } ' ] } { { \mbox{$\vec{q}$}}^2 } .\ ] ] to calculate the spin part of the matrix element , we take the trace with @xmath74-dimensional sigma matrices and find ( the factor @xmath75 in eq . ( [ 14 ] ) arises from the average over directions of the o - ps polarization vector ) @xmath76 \sigma_i [ { \mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$}}',{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$ } } ] \right ) & = & 4\frac { d-4 } { d } \left [ { \mbox{$\vec{p}$}}'^2 { \mbox{$\vec{p}$}}^2 - ( { \mbox{$\vec{p}$}}'{\mbox{$\vec{p}$}})^2 \right ] , \label{14 } \\ \frac{1 } { 2 } { \rm tr } \left ( [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{p}$}},{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{p}$ } } ' ] [ { \mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$}}',{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$ } } ] \right ) & = & 4 \left [ { \mbox{$\vec{p}$}}'^2 { \mbox{$\vec{p}$}}^2 - ( { \mbox{$\vec{p}$}}'{\mbox{$\vec{p}$}})^2 \right],\end{aligned}\ ] ] respectively for ortho and parapositronium . using @xmath77 and noting that the average value of @xmath78 in an @xmath2state vanishes , we obtain the contribution of @xmath79 to the ground state hfs : @xmath80 in eq . ( [ int ] ) the matrix element is to be calculated over the ground state wave function in @xmath74 dimensions : @xmath81 let us briefly explain how the integral in eq . ( [ int ] ) is calculated . although the integrand does not look complicated , the difficulty is that the exact form of the wave function @xmath82 in @xmath74 dimensions is not known . fortunately , it turns out to be unnecessary . there are two alternative ways to calculate this integral . one is to transform it to the coordinate space . a divergence arises at @xmath83 and in the final result is proportional to the @xmath74-dimensional @xmath84 ; the remaining , finite part can be easily calculated in @xmath85 . at @xmath83 , can appear in the divergent part of the integral . however , the schrdinger equation relates it to @xmath84 . ] in the alternative approach we use the fact that the wave function in eq . ( [ int ] ) satisfies the @xmath74-dimensional schrdinger equation , which in the momentum space reads @xmath86 using this equation we rewrite the integral in eq . ( [ int ] ) as @xmath87 where the integration over @xmath88 , @xmath89 , as well as @xmath90 , @xmath91 , in the last expression is understood . the integral over @xmath88 and @xmath89 receives a divergent contribution only from the region where @xmath88 and @xmath89 simultaneously become infinite . therefore , a single subtraction is sufficient to make this integral finite . it is convenient to subtract from ( [ int1 ] ) the following expression : @xmath92 after the subtraction is done , two nice features emerge . in eq . ( [ int2 ] ) the integration over @xmath93 factorizes and leads to @xmath94 times a two - loop integral , which can be easily calculated for arbitrary @xmath74 . on the other hand , the difference between the last integral in eq . ( [ int1 ] ) and the integral in eq . ( [ int2 ] ) is finite and can be calculated for @xmath85 using the explicit form of the wave function , @xmath95 we note that the counterterm ( [ int2 ] ) is constructed in such a way that the above mentioned difference vanishes for the ground state . this can be easily seen by integrating over @xmath93 in eq . ( [ int2 ] ) and using the fact that the @xmath96-dependent terms in the denominator of eq . ( [ int2 ] ) coincide ( up to a normalization factor ) with the three - dimensional ground state wave functions in the momentum representation . both methods described above lead to the same result . for @xmath97 we obtain : and @xmath98 which do not contribute to the final , finite result . ] @xmath99 where @xmath84 is the value of the @xmath74-dimensional ground state wave function at the origin . we now consider the correction caused by the tree level exchange of a magnetic photon , fig . [ fig : soft](b ) . we neglect the energy dependence in the photon propagator ; it will be restored in the following section , where we discuss retardation effects . the relevant potential is obtained from eq . ( [ eqa : magn ] ) : @xmath100 } { { \mbox{$\vec{q}$}}^2 } \left\ { \left [ { \mbox{$\vec{\sigma}$}}\frac { { \mbox{$\vec{p}$}}'+{\mbox{$\vec{p}$ } } } { 2 } , \sigma^i \right ] \left({\mbox{$\vec{p}$}}'^2-{\mbox{$\vec{p}$}}^2\right ) + [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^i]\left({\mbox{$\vec{p}$}}^2+{\mbox{$\vec{p}$}}'^2\right ) \right\ } + ( { \mbox{$\vec{\sigma}$}}\leftrightarrow { \mbox{$\vec{\sigma}$}}').\ ] ] contribution of this interaction to the ground state hfs is @xmath101 in @xmath85 this matrix element is linearly divergent . to demonstrate how we treat linear divergences let us consider the @xmath102 term on the rhs of the above equation : @xmath103 shifting the integration variable @xmath104 we find that the @xmath88-integral in the last term is scale - less . in dimensional regularization such integrals vanish . the first term in eq . ( [ tadpole ] ) is finite in three dimensions . we obtain @xmath105 applying a similar procedure to the last term in eq . ( [ em ] ) we find the contribution of @xmath106 to the ground state hfs : @xmath107 = \frac { \pi\alpha^3 } { 3 m^2 } \psi^2(0 ) \left ( \frac { 1 } { \epsilon } - 4 \ln(m\alpha ) + \frac{5}{3 } \right ) . \label{deltam}\ ] ] let us now consider the retardation effects , which mean that the magnetic photon emitted by the electron propagates for a finite amount of time before being absorbed by the positron . during this time , the electron and positron can interact by several coulomb exchanges ( fig . [ fig : soft](c , d , e ) ) . to calculate the influence of these effects on the hfs , it is sufficient to take the spin - dependent parts of the current @xmath108 in the leading nonrelativistic approximation : @xmath109 } { 4 m } .\ ] ] the scattering operator describing the retardation effects is nonlocal both in space and time : @xmath110 } { 4 m } \frac { 4\pi } { 2k } \frac { \delta_{ij } - \frac{k_i k_j}{{\mbox{$\vec{k}$}}^2 } } { k+h - e } \frac { [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{k}$}},\sigma_j ] } { 4 m } \exp \left ( i{\mbox{$\vec{k}$}}{\mbox{$\vec{r}$}}_e \right ) + { \rm h.c . } \label{retar}\ ] ] here we assume that the magnetic photon with the momentum @xmath90 is emitted by the electron at a point @xmath111 and absorbed by the positron at a point @xmath112 . between those moments , the evolution of the system `` positronium + photon '' is governed by the propagator @xmath113 , @xmath114 being the hamiltonian of the nonrelativistic positronium slowly moving due to recoil . in the region of interest @xmath115 and one can expand the amplitude ( [ aret ] ) over the powers of @xmath116 . the zeroth term of this expansion is the spin dependent part of the breit potential , @xmath117[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^i ] } { { \mbox{$\vec{q}$}}^2 } .\ ] ] we need the second order term : @xmath118 } { 4 m } [ h,\exp \left ( -i{\mbox{$\vec{k}$}}{\mbox{$\vec{r}$}}_p \right ) ] [ h,\exp \left ( i{\mbox{$\vec{k}$}}{\mbox{$\vec{r}$}}_e \right)]\frac { [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{k}$}},\sigma^i ] } { 4 m } + { \rm h.c.}\ ] ] only kinetic part of the hamiltonian , @xmath119 has to be retained in the commutators . we find @xmath120 } { 8m^2 } ( { \mbox{$\vec{k}$}}^2 + 2{\mbox{$\vec{k}$}}{\mbox{$\vec{p}$}}_p)\exp \left ( i{\mbox{$\vec{k}$}}({\mbox{$\vec{r}$}}_e-{\mbox{$\vec{r}$}}_p ) \right ) ( { \mbox{$\vec{k}$}}^2 + 2{\mbox{$\vec{k}$}}{\mbox{$\vec{p}$}}_e)\frac { [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{k}$}},\sigma^i ] } { 8m^2 } + { \rm h.c.}\ ] ] transforming back to the relative coordinate @xmath121 and the relative momentum @xmath122 , we get for the ground state hfs : @xmath123 now we turn to the operators generated by one - loop diagrams . for the hfs the only contribution comes from the graph in fig . [ fig : soft](f ) , which describes the mixed coulomb - magnetic exchange with a transition of one of the particles to a negative energy state . in other words , this corresponds to a creation of an additional electron - positron pair by the electric or magnetic field of the electron or positron . using feynman rules for the time - independent perturbation theory , given at the beginning of this section , we derive the corresponding potential : @xmath124 } { ( { \mbox{$\vec{q}$}}-{\mbox{$\vec{k}$}})^2 } \frac { [ { \mbox{$\vec{\sigma}$}}'{\mbox{$\vec{k}$}},\sigma'^i ] } { { \mbox{$\vec{k}$}}^2 } .\ ] ] it induces the following correction to the ground state hfs ( @xmath74-dimensional integration over @xmath90 is implicitly assumed below ) @xmath125 to complete the calculation of the soft scale contributions to the hfs we have to consider the second iteration of the breit hamiltonian . it is obtained by including the effects of tree level coulomb and magnetic photon exchanges , as well as a correction to the kinetic energy . using eqs . ( [ eqa : coul ] ) and ( [ eqa : magn ] ) we find @xmath126[{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'^i]}{{\mbox{$\vec{q}$}}^2}.\ ] ] in the position representation this hamiltonian becomes @xmath127[{\mbox{$\vec{\sigma}$}}'\nabla,\sigma'^i ] , c(r)\right],\ ] ] where @xmath128 is the @xmath74dimensional coulomb potential . we consider first the contribution of the intermediate @xmath2states . the @xmath2wave part of the breit hamiltonian ( [ hbr ] ) reads @xmath129[\sigma'_i,\sigma'_j ] \delta({\mbox{$\vec{r}$}}).\ ] ] it is convenient to divide up the calculation of the @xmath130 contribution to the hfs into two parts and consider the first and the last two terms in eq . ( [ bs ] ) separately . we begin with the latter , which we denote by @xmath131 : @xmath132 in three dimensions the last sum is ill - defined due to ultraviolet divergences in the zeroth and first terms of its expansion in @xmath133 . we denote these singular terms by @xmath134 and @xmath135 , respectively , and obtain @xmath136.\ ] ] @xmath134 and @xmath135 are calculated in @xmath74 dimensions , @xmath137 and one finds @xmath138 the contribution of the first two terms in eq . ( [ bs ] ) is calculated in the following way . we first write them as @xmath139,\ ] ] where @xmath140 is the leading order hamiltonian . correction to the hfs induced by eq . ( [ s2 ] ) reads @xmath141 g({\mbox{$\vec{r}$}},{\mbox{$\vec{r}$ } } ' ) \delta({\mbox{$\vec{r}$ } } ' ) + { \rm h.c . } \right\rangle.\ ] ] we introduced here the reduced green function @xmath142 which satisfies the equation @xmath143 using obvious short - hand notations one can rewrite eq . ( [ des2 ] ) as follows : @xmath144 we dropped massless tadpoles and separated the contribution of @xmath145 , which is the only one we have to calculate keeping @xmath146 . we find @xmath147 to obtain the last line we used the following equation : @xmath148 where @xmath149 . > from eq . ( [ 57 ] ) we now find @xmath150 the sum of @xmath151 and @xmath152 gives the final result for the correction to the ground state hfs induced by the second iteration of the @xmath2-wave breit hamiltonian : @xmath153 because of the last term in eq . ( [ hbp ] ) breit hamiltonian has non - vanishing matrix elements with @xmath154 . in our case this causes virtual transitions from the triplet @xmath2state into @xmath51states ( transitions from the singlet state are forbidden by the total angular momentum conservation ) . again , power counting shows that only the zeroth and the first order terms in the green function expansion in @xmath133 diverge in three dimensions . we first compute the remaining , higher order terms , which are finite for @xmath85 . the sum of those higher order terms can be written as @xmath155 where @xmath156 is the @xmath154 part of the breit hamiltonian in three dimensions , @xmath157 and @xmath145 are defined in the previous section , and @xmath158 is the coulomb potential . the correction to the ground state wave function , @xmath159 which appears in eq . ( [ d ] ) , satisfies an inhomogeneous schrdinger equation : @xmath160 solving this equation for @xmath161 we obtain @xmath162 where @xmath163 is the @xmath51-wave part of the green function @xmath157 , and the factor @xmath164 arises from @xmath165 ^ 2 \right\rangle = \left\langle 3 + 4\mbox{$\vec{\sigma}\vec{\sigma}'$}+(\mbox{$\vec{\sigma}\vec{\sigma}'$})^2 \right\rangle = 8\delta_{s1}.\ ] ] to calculate the matrix element in eq . ( [ 72 ] ) we note that @xmath166 where @xmath167 is the radial hamiltonian for @xmath168-states . using equations of motion for both the green function and the wave function in eq . ( [ 72 ] ) one finds @xmath169 in agreement with @xcite . to complete the calculation of the @xmath51-wave contribution we have to consider the zeroth and first order terms in the @xmath133 expansion of the green function , @xmath170 the perturbation @xmath171 is extracted from breit hamiltonian , eq . ( [ hbp ] ) , and reads @xmath172[\sigma'_i,\sigma'_j ] } { d } - \frac{[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma_i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'_i]}{{\mbox{$\vec{q}$}}^2 } \right).\ ] ] the average is taken over the @xmath74dimensional wave function . calculating the trace using the triplet wave function we obtain @xmath173[\sigma'_i,\sigma'_j ] } { d } - \frac{[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}}',\sigma^i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}}',\sigma'^i]}{{\mbox{$\vec{q}$}}'^2 } \right ) \left ( \frac { [ \sigma_k,\sigma_l][\sigma'_k,\sigma'_l ] } { d } - \frac{[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^k][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'^k]}{{\mbox{$\vec{q}$}}^2 } \right ) \right\rangle \nonumber \\ & & \qquad\qquad\qquad = \frac{4(d-2)^2}{d}\left\langle b_{ij}({\mbox{$\vec{q}$ } } ' ) b_{ij}({\mbox{$\vec{q}$ } } ) \right\rangle , \qquad b_{ij}({\mbox{$\vec{q}$ } } ) \equiv 4\pi \left ( \frac { q_i q_j } { { \mbox{$\vec{q}$}}^2 } - \frac{\delta_{ij } } { d } \right).\end{aligned}\ ] ] therefore @xmath174 where @xmath175 and @xmath74dimensional integrations over @xmath90 in ( [ d02 ] ) and over @xmath93 in ( [ d12 ] ) are understood . some details of the integrations in eqs . ( [ d02],[d12 ] ) are given in appendix [ app : int ] . adding the higher - order effects found in eq . ( [ deltad ] ) we obtain the complete @xmath51-wave contributions to hfs @xmath176 another contribution to the hfs arises from virtual momenta scales of the order of the electron mass . it can be calculated by considering the on shell @xmath177 scattering amplitude with an exchange of three photons in the @xmath178-channel ( see fig . [ fig1 ] ) exactly at the threshold , i.e. for zero relative velocity of the incoming electron and positron , in dimensional regularization . the use of the dimensional regularization brings in essential simplifications , since almost any other regularization would bring in power - like divergences and hence require additional subtractions . this so - called hard scale contribution gives rise to four - fermion operators in the low - scale lagrangian or , equivalently , to the @xmath179 terms in the effective quantum mechanical hamiltonian . technically , this calculation is similar to the derivation of the matching coefficient of the vector quark - antiquark current in qcd and its nrqcd counterpart , described e.g. in @xcite . here we outline the main steps of this calculation . an arbitrary feynman integral which contributes to the hard scale part of the calculation can be written as @xmath180 where @xmath181 and @xmath182 are integers . in practice we encounter diagrams with only at most 6 different propagators , so that at least 3 exponents @xmath183 are zero . applying the integration by parts technique @xcite to an integral @xmath184 , one obtains a set of relations among integrals with various values of indices @xmath185 . using these relations one can express any @xmath184 in terms of a few master integrals . this is most easily done using symbolic manipulation programs . the result for the hard scale recoil corrections ( fig . [ fig1 ] ) to the hfs reads @xcite @xmath186 the result for the hfs of the ground state can be used to obtain the hfs for an arbitrary excited state . the non trivial dependence on the principal quantum number @xmath4 arises only from the soft scale contributions . therefore , one has to repeat the quantum mechanical calculation of the non - relativistic part using any convenient regularization ( we use a cut - off at @xmath187 ) and compare the result with the known formula for @xmath53 , eq . ( [ ehfsrecground ] ) . one finds @xmath188 -\frac { 1}{6 } \left ( \ln \frac { \alpha}{n } + \psi ( n ) + \gamma_e \right ) + \frac { 7}{12 n } - \frac { 1}{2n^2 } \right ] , \ ] ] the quantity @xmath189 $ ] in the above equation stands for the unknown and @xmath4-independent constant , easily determined by requiring that for @xmath4=1 eq . ( [ ehfsrecground ] ) is reproduced . we then obtain the final result for the recoil corrections to the hfs splitting for an arbitrary @xmath1 state , eq . ( [ ehfsrecn ] ) . to obtain @xmath0 corrections to the triplet and singlet energy levels separately , we have to calculate @xmath190 ( cf . ( [ param ] ) ) . an appropriate formula for this calculation is @xmath191 it is known @xcite that the recoil corrections @xmath15 do not contain @xmath192 at the order @xmath8 . in dimensional regularization this means that the hard - scale and soft - scale contributions are separately finite . conceptually , determination of @xmath15 is very similar to the calculation of the hfs discussed above in detail . the only difference is that several new operators appear , which contribute to @xmath15 but not to the hfs . we begin with the correction to @xmath15 induced by the relativistic corrections to the dispersion law , @xmath193 ( fig . [ fig : soft](g ) ) . expanding @xmath194 in @xmath195 , we obtain : @xmath196 the last term induces a correction of the appropriate order : @xmath197 the @xmath72 spin - independent part of the tree level coulomb exchange amplitude ( cf . ( [ ac ] ) and fig . [ fig : soft](a ) ; we neglect terms odd in @xmath88 , whose average vanishes in an @xmath2-state ) , @xmath198 gives rise to the following correction : @xmath199 virtual transitions to negative energy states induced by the coulomb exchanges , fig . [ fig : soft](h ) , generate an effective spin - independent operator @xmath200 ^ 2.\ ] ] this operator describes the energy shift due to a creation of an additional @xmath201 pair by the coulomb field of either electron or positron . the resulting energy shift is @xmath202 the spin - independent part of the tree level magnetic exchange , fig . [ fig : soft](b ) , induces the following shift in the energy levels : @xmath203 to account for the retardation in the magnetic photon propagation , fig . [ fig : soft](c , d , e ) , we use the approach described in the hfs case . our starting point is similar to eq . ( [ retar ] ) , except that now the full expression for the currents must be used , rather than just their spin - dependent part . we obtain @xmath204 the next contribution comes from the exchange of two magnetic photons with creation of an additional @xmath201 pair in the intermediate state , fig . [ fig : soft](i ) . we find @xmath205 we proceed further with the correction to @xmath15 , induced by the second iteration of the @xmath2-wave breit hamiltonian . the calculation closely follows the hfs case . we arrive at the following result : @xmath206 the iteration of the @xmath51-wave part of the breit hamiltonian only influences the energy levels of the triplet state because of the total angular momentum ( @xmath207 ) conservation . for this reason , to obtain the required correction to @xmath15 it is sufficient to multiply eq . ( [ deltad2 ] ) by the factor @xmath208 . we find @xmath209 it is easy to see that in 3 dimensions the spin - dependent operators do not contribute to @xmath15 . however , since we work with divergent integrals and use dimensional regularization , this is no longer valid for @xmath210 . in this case an `` anomalous '' situation arises : spin - dependent operators provide contributions of the form @xmath211 to @xmath15 , which are finite as @xmath212 . part of these contributions has already been accounted for in the corrections induced by the breit hamiltonian . the remaining contributions give @xmath213 the hard scale contribution , fig . [ fig1 ] , is calculated in the same way as for the hfs . one finds : @xmath214 the sum of all contributions presented above provides the @xmath0 pure recoil correction to the ground state energy : @xmath215 in very good agreement with the numerical result of eq . ( 20 ) in @xcite , @xmath216 . to generalize the result eq . ( [ n=1shift ] ) for arbitrary @xmath4 , we proceed according to the program outlined in section [ sec : nhfs ] . we repeat the calculation of the soft - scale contributions to @xmath35 for arbitrary @xmath4 using a different regularization scheme . namely , we set @xmath85 and cut off the divergent integrals over @xmath217 from below at some @xmath218 . the transition to three dimensions simplifies the calculation . we find @xmath219 + \frac { 69}{64n^3 } -\frac { 8}{3n^2}+\frac { 2}{n } \right ) . \label{eprel}\ ] ] the @xmath4-independent term in the above equation is regularization dependent . it can not be determined by considering the soft - scale contributions alone . nevertheless , by matching eq . ( [ eprel ] ) to the shift for the ground state eq . ( [ n=1shift ] ) , the `` value '' of the divergent constant @xmath189 $ ] is completely determined . we obtain @xmath220 this is our main result for the recoil corrections to the energy levels of positronium . it agrees with the partially numerical result derived in @xcite . so far in this paper we have been considering pure recoil effects . another class of the @xmath0 corrections to positronium energy levels and their hfs are the so - called radiative recoil corrections , where one of the three exchanged photons is created and absorbed by the same particle ( see fig . [ fig2 ] ) . our technique is very convenient for the calculation of these corrections . the key point is that at @xmath221 the radiative recoil corrections do not receive any contribution from the non - relativistic scales . thus it is sufficient to calculate the diagrams shown in fig . [ fig2 ] ( supplemented by the electric charge , electron wave function and mass renormalization ) exactly at the threshold . for the same reason , the @xmath4-dependence of the radiative recoil corrections comes only from the @xmath222 behavior of the @xmath1-wave function at the origin . some details of this calculation are described in section [ sec : hard ] in the context of the hfs . we obtain @xmath223 , \label{radrechfs } \\ \delta_{\rm rad\ ; rec } e_{\rm aver } & = & \frac { m\alpha^6}{n^3 } \left [ \frac { 9\zeta ( 3)}{8\pi^2 } + \frac { 97}{144}-\frac { 1025}{432\pi^2 } \right ] , \label{radreclevels}\end{aligned}\ ] ] respectively for corrections to the hfs and to the average energy , in full agreement with the analytic results of ref . @xcite . for completeness , we give here separately the contributions of electron vacuum polarization effects to radiative recoil corrections @xcite ( they are included in ( [ radrechfs],[radreclevels ] ) ) : @xmath224 the main new results of the present paper are the analytic formulas ( [ ehfsrecn ] ) and ( [ eaverfin ] ) for the pure recoil @xmath42 corrections to the hfs and spin averaged energy levels of positronium @xmath1 states . these recoil effects provide the last pieces needed to present complete analytical formulas for the total corrections to @xmath15 and @xmath16 . we use the parameterization introduced in eq . ( [ param ] ) , @xmath225 and find @xmath226 \nonumber \\ & & + \frac{m\alpha^6}{32n^3 } \left [ -\ln\frac{\alpha}{n } - \psi(n ) - \gamma_e + \frac{141}{4}\frac{\zeta(3)}{\pi^2 } + \left ( \frac{137}{6 } - \frac{68}{\pi^2 } \right ) \ln 2 + \frac{1421}{27\pi^2 } \right . \nonumber \\ & & \qquad \qquad \left . - \frac{2435}{432 } - \frac{7}{n } + \frac{17}{12n^2 } - \frac{69}{16n^3 } \right ] , \label{aver : final}\end{aligned}\ ] ] and @xmath227 . \label{hfs : final}\end{aligned}\ ] ] we have also recalculated the radiative recoil corrections , eqs . ( [ radrechfs],[radreclevels ] ) , confirming recent result of ref . let us make a technical remark . in dimensional regularization , used in this paper , the calculation of the radiative recoil corrections is particularly simple . since there are no low - scale contributions to the radiative recoil corrections , it suffices to calculate corresponding feynman graphs exactly at the threshold . no matching or subtractions are required . formulas ( [ aver : final],[hfs : final ] ) , together with @xmath43 state energy levels given in appendix [ app : p ] , can be used to compute quantities which can be directly confronted with experimental data . we use the following values for the rydberg @xcite and fine structure @xcite constants : @xmath228 in addition to the full corrections @xmath0 we include the leading logarithmic terms @xmath229 found in @xcite for hfs , and in @xcite for the spin - averaged energy levels : @xmath230 for the most precisely measured quantity , the ps ground state hfs , we find @xmath231 the @xmath0 recoil corrections to this observable have been subject of some debate . in the literature three different results have been reported @xcite . our result for this correction , eq . ( [ ehfsrecground ] ) , evaluates numerically to @xmath232 . this is in excellent agreement with ref . @xcite , where for the non - logarithmic part of the correction a number 0.3767(17 ) was obtained . the framework of our calculation is similar to ref . @xcite . however , in that study a different regularization method was used . the agreement of the results gives us confidence in their correctness . comparing eq . ( [ hfsth ] ) with the experimental results , eqs . ( [ mills],[hughes ] ) , we observe a significant deviation of the order of @xmath233 experimental errors . it is not very likely that the uncalculated higher order effects alone can account for this discrepancy . the size of the @xmath0 corrections gives no indication of bad behavior of the perturbative expansion . on the other hand , the leading logarithmic term @xmath234 is sizable . a calculation of the subleading terms at this order remains an important theoretical challenge . for another experimentally interesting quantity , the energy interval of the @xmath235 transition , we get @xmath236 in fair agreement with the experimental result , eq . ( [ fee ] ) . other quantities , for which high precision measurements have been made or are being planned , have recently been reviewed in @xcite . in table [ tab : one ] we update the theory predictions for those observables . our predictions are in good agreement with @xcite . we have been able to decrease the error bars by including the analytical results ( [ aver : final],[hfs : final ] ) and the value of the leading quadratic logarithms ( [ eq : llogs ] ) . finally we would like to comment on our error estimates . the errors due to uncertainties in the fine structure constant and the electron mass are well below @xmath237 mhz level . the dominant theoretical error source is the uncalculated remainder of the perturbation expansion . although formally @xmath238 , the leading @xmath239 terms contribute @xmath240 to the hfs @xcite . it remains very important to calculate the remaining , non - leading terms in @xmath241 . for the present analysis we assume that the leading logs @xmath239 dominate the higher order contributions and take half their size as the theoretical error estimate . the spectrum of the @xmath1 and @xmath242 positronium energy levels is now known analytically , including effects @xmath0 . our calculation for the @xmath1 levels was made possible by new theoretical tools which have their roots in the recent perturbative calculations in high - energy physics . we hope that these methods will find further applications . the agreement between theoretical predictions and experimental results in ps spectroscopy is impressive with a few exceptions . one can only hope to find something new and unexpected by trying to put these exceptions in line with the overall picture . we look forward to future improved measurements of positronium energy levels and their confrontation with qed . we are grateful to a. burichenko for informing us about his results prior to publication . we thank s. karshenboim , k. pachucki and e. remiddi for reading the manuscript and helpful comments . k.m . would like to thank the high energy theory group at brookhaven national laboratory for hospitality during the final stage of this project . this research was supported in part by the u.s . department of energy under grant number de - ac02 - 98ch10886 , by bmbf under grant number bmbf-057ka92p , by graduiertenkolleg `` teilchenphysik '' at the university of karlsruhe and by the russian foundation for basic research under grant number 99 - 02 - 17135 . we present here formulas for the potential arising from a single coulomb or magnetic photon exchange between an electron and a positron , valid to @xmath72 . the virtual annihilation is not taken into account here . we also drop those terms which annihilate the @xmath2state wave function . these formulas , valid in @xmath74-dimensions , are useful in the derivations of hfs and spin averaged energy levels . for a single coulomb exchange between two particles of opposite charges , fig . [ fig : soft](a ) , the minus on shell scattering amplitude is @xmath243 where @xmath88 and @xmath244 are spatial momenta of the incoming and outgoing electron ; @xmath245 ; and the charge density is @xmath246 . in momentum representation a single coulomb photon exchange gives rise to the potential @xmath247[{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$}}',{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$ } } ] \over 64m^4}\right\}. \nonumber \\ \label{eqa : coul } \ ] ] in the leading nonrelativistic approximation eq . ( [ eqa : coul ] ) gives the coulomb potential . next we consider a magnetic photon exchange , fig . [ fig : soft](b ) . we neglect retardation effects . the scattering amplitude is @xmath248 where @xmath249 is the matrix element of the current . the resulting potential is @xmath250 -{1\over 4}[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'^i]\right ) \nonumber \\ & & -{\pi\alpha\over 2m^4{\mbox{$\vec{q}$}}^2}\left\ { ( { \mbox{$\vec{p}$}}^2+{\mbox{$\vec{p}$}}'^2 ) \left ( { 4\over { \mbox{$\vec{q}$}}^2}\left [ ( { \mbox{$\vec{p}$}}{\mbox{$\vec{p}$}}')^2 - { \mbox{$\vec{p}$}}^2{\mbox{$\vec{p}$}}'^2\right ] -{1\over 4}[{\mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'^i]\right ) \right . \nonumber \\ & & \left . \qquad\qquad + { { \mbox{$\vec{p}$}}^2-{\mbox{$\vec{p}$}}'^2\over 16}\left ( [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{q}$}},\sigma^i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{p}$}},\sigma'^i ] + [ { \mbox{$\vec{\sigma}$}}{\mbox{$\vec{p}$}},\sigma^i][{\mbox{$\vec{\sigma}$}}'{\mbox{$\vec{q}$}},\sigma'^i]\right ) \right\}. \label{eqa : magn}\end{aligned}\ ] ] these formulas are valid in the center of mass frame . we use @xmath88 and @xmath89 to denote incoming and outgoing electron momenta , and @xmath251 , @xmath252 . the primed @xmath253-matrices act on the positron spinor . in this appendix we present various integrals which were useful in the calculations presented in this paper . the following formulas have been used throughout the paper , especially for the tree level diagrams : @xmath254 in the remainder of this appendix we describe some details of the @xmath51-wave contribution to the second iteration of the breit hamiltonian . first , we rewrite eq . ( [ d02 ] ) in the following way : @xmath255 here and below we use the schrdinger equation in the form @xmath256 similarly , by rearranging terms in ( [ d12 ] ) we get @xmath257 using the symmetry with respect to @xmath258 , we rewrite the first term in eq . ( [ 0mod ] ) , @xmath259 in the same way , the second term in eq . ( [ 0mod ] ) is transformed to @xmath260 considering the divergent part of this expression we find @xmath261.\ ] ] the sum of ( [ d02 ] ) and ( [ d12 ] ) reads @xmath262 - \frac{8\pi\gamma^2}{3 } + \frac{4\pi({\mbox{$\vec{p}$}}'{\mbox{$\vec{q}$}})({\mbox{$\vec{q}$}}{\mbox{$\vec{p}$}})}{{\mbox{$\vec{q}$}}^2 } \right . \nonumber \\ & & + 2 p'_i\frac{4\pi}{({\mbox{$\vec{p}$}}'-{\mbox{$\vec{k}$}})^2 } \left ( 2k_j - p'_j \right ) g(k)\frac{4\pi(k_i - p_i)}{({\mbox{$\vec{k}$}}-{\mbox{$\vec{p}$}})^2 } p_j \nonumber \\ & & -2 b_{ij}({\mbox{$\vec{p}$}}'-{\mbox{$\vec{k}$}}')g(k ' ) \frac{4\pi}{({\mbox{$\vec{k}$}}'-{\mbox{$\vec{k}$}})^2}k_i g(k)\frac{4\pi}{({\mbox{$\vec{k}$}}-{\mbox{$\vec{p}$}})^2}p_j \nonumber \\ & & \left . + b_{ij}({\mbox{$\vec{p}$}}'-{\mbox{$\vec{k}$}}')g(k ' ) \frac{4\pi}{({\mbox{$\vec{k}$}}'-{\mbox{$\vec{k}$}})^2}g(k)\frac{4\pi}{({\mbox{$\vec{k}$}}-{\mbox{$\vec{p}$}})^2}p_i p_j \right\rangle.\end{aligned}\ ] ] only two terms here contain the logarithmic divergence : @xmath263 all other terms are finite and we compute them in three dimensions . here we list some useful integrals ( @xmath264 and @xmath265 ) : @xmath266 , \nonumber \\ & & \int { { \rm d}^3 p\over ( 2\pi)^3 } { { \rm d}^3 k'\over ( 2\pi)^3 } \frac{g(k')b_{ij}({\mbox{$\vec{k}$}}'-{\mbox{$\vec{p}$ } } ) } { ( { \mbox{$\vec{k}$}}-{\mbox{$\vec{k}$}}')^2 } \phi(p ) = \frac { \gamma\psi(0 ) } { k } \left ( \frac{k_ik_j}{{\mbox{$\vec{k}$}}^2 } - \frac{\delta_{ij}}{3 } \right ) \left [ \frac{a(x)}{2 } \left ( 1 + \frac{3}{x^2 } \right ) - \frac{3}{2x } \right ] . \nonumber \\\end{aligned}\ ] ] using these above formulas in eq . ( [ d2 ] ) we find the final result for @xmath267 . in this appendix we present formulas for the energy levels of @xmath43 states , to order @xmath0 . correcting some minor misprints in ref . @xcite one finds @xmath268 for the numerical evaluations we use the following values of bethe logarithms @xmath269 $ ] @xcite @xmath270 & = & \phantom{- } 2.984\;128\;555\;765\;498 , \nonumber \\ \ln\left [ k_0(2,0)/r_\infty \right ] & = & \phantom{- } 2.811\;769\;893\;120\;563 , \nonumber \\ \ln\left [ k_0(2,1)/r_\infty \right ] & = & -0.030\;016\;708\;630\;213.\end{aligned}\ ] ] ld transition & theory [ mhz ] + @xmath271 & 1233607222.18(58 ) + @xmath272 & 203392.01(46 ) + @xmath273 & 18498.25(8 ) + @xmath274 & 13012.41(8 ) + @xmath275 & 8626.71(8 ) + @xmath276 & 11185.37(8 ) + @xmath277 & 25424.67(6 ) + @xmath278 ( a ) & \hspace*{11mm}(b ) & \hspace*{10 mm } ( c ) \\[9 mm ] \psfig{figure = soft4.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = soft5.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = soft6.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } \\[7 mm ] ( d ) & \hspace*{11mm}(e ) & \hspace*{10 mm } ( f ) \\[9 mm ] \psfig{figure = soft7.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = soft8.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = soft9.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } \\[7 mm ] ( g ) & \hspace*{11mm}(h ) & \hspace*{10 mm } ( i ) \\[2 mm ] \end{tabular } } \ ] ] @xmath279 \psfig{figure = hfs1c.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = hfs1d.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } \end{tabular } } \ ] ] @xmath280 \psfig{figure = hfs2c.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } & \hspace*{10 mm } \psfig{figure = hfs2d.ps , width=30mm , bbllx=72pt , bblly=291pt , bburx=544pt , bbury=530pt } \end{tabular } } \ ] ]

we calculate low - energy proton deuteron scattering in the framework of pionless effective field theory . in the quartet channel , we calculate the elastic scattering phase shift up to next - to - next - to - leading order in the power counting . in the doublet channel , we perform a next - to - leading order calculation . we obtain good agreement with the available phase shift analyses down to the scattering threshold . the phase shifts in the region of non - perturbative coulomb interactions are calculated by using an optimised integration mesh . moreover , the coulomb contribution to theheh binding energy difference is evaluated in first order perturbation theory . we comment on the implications of our results for the power counting of subleading three - body forces .

1101.5939

although quantum chromodynamics ( qcd ) is widely accepted as the underlying theory of strong interactions , _ ab initio _ calculations of nuclear properties in lattice qcd remain a large theoretical challenge @xcite . in nuclear physics , the relevant degrees of freedom are pions and nucleons , and much of the computational effort in such a calculation would be required for generating the correct degrees of freedom from quarks and gluons rather than their interactions . traditionally , nucleon nucleon interactions are described via phenomenological nuclear forces fitted to scattering data . effective field theory ( eft ) provides a powerful method to construct nuclear forces with a direct connection to qcd in a systematic , model - independent way @xcite . for very low energies and momenta @xmath1 , the non - analyticities from pion exchange can not be resolved and one can hence use an eft including only short - range contact interactions between nucleons @xcite . this theory is constructed to reproduce the effective range expansion @xcite in the two - body system and recovers efimov s universal approach to the three - nucleon problem @xcite . an advantage of the eft formulation is that it can be extended to higher - body systems and external currents in a straightforward way . the extension of this eft to include the long - range coulomb interaction was first discussed by kong and ravndal for the proton proton channel @xcite . in ref . @xcite , this analysis was extended to next - to - next - to - leading order . a renormalisation group analysis of proton proton scattering in a distorted wave basis was carried out in refs . moreover , this theory was applied to proton proton fusion in refs . an extension of this formalism to three charged particles would be important for the possible interpretation of the hoyle state in @xmath2c as an efimov state of @xmath3 particles @xcite and the cluster eft for halo nuclei @xcite . in this work , we are interested in the simpler problem of a three - body system with two charged particles . close to threshold , the coulomb interaction is strong . its long - range nature requires special care in the non - perturbative treatment using momentum space integral equations . at higher energies , the coulomb interaction becomes perturbative . rupak and kong have formulated a power counting for the coulomb contributions in the quartet channel of proton deuteron ( @xmath4@xmath5 ) scattering . they calculated the phase shifts to next - to - next - to - leading order ( ) in the pionless eft and included coulomb effects to next - to - leading order ( ) @xcite . however , they were not able to extend their calculation to the threshold region below center - of - mass momenta of 20 @xmath6 . they did not consider the doublet channel and the @xmath0he bound state . a leading order calculation of the @xmath0he nucleus including non - perturbative coulomb interactions was recently presented by ando and birse @xcite . including isospin breaking effects in the nucleon nucleon scattering lengths , they obtain a surprisingly accurate description of the @xmath0he@xmath0h binding energy difference but they did not consider scattering observables . a similar study to in the pionless eft was carried out using the resonating group method @xcite . their results do not include isospin breaking and are consistent with other determinations of the @xmath0he@xmath0h binding energy difference . in this paper , we focus on @xmath4@xmath5 scattering observables in the quartet and doublet channels . we extend the power counting by rupak and kong for the coulomb contribution to the doublet channel . by using a special integration mesh , we are able to calculate the phase shifts in both channels down to momenta of order 3 @xmath6 . we also provide a perturbative evaluation of the coulomb contribution to the @xmath0he@xmath0h binding energy difference . in this section , we briefly summarise the formalism required for calculating @xmath4@xmath5 scattering in the pionless theory . more technical details can , _ e.g. _ , be found in refs . @xcite . we use the effective lagrangian @xmath7d^i -t^{a\dagger}\left[{\sigma_t}+\left({\mathrm{i}}d_0+\frac{{\boldsymbol{d}}^2}{4{m_n}}\right)\right]t^a \\[0.2 cm ] + { y_d}\left[d^{i\dagger}\left(n^t p^i_d n\right)+{\mathrm{h.c.}}\right ] + { y_t}\left[t^{a\dagger}\left(n^t p^a_t n\right)+{\mathrm{h.c.}}\right ] + \mathcal{l}_\mathrm{photon}+\mathcal{l}_3 \ , , \label{eq : l - nd}\end{gathered}\ ] ] with the nucleon field @xmath8 and two dibaryon fields @xmath9 ( with spin 1 and isospin 0 ) and @xmath10 ( with spin 0 and isospin 1 ) , corresponding to the deuteron and the spin - singlet virtual bound state in s - wave nucleon nucleon scattering . both dibaryon fields are formally ghosts since their kinetic terms have a negative sign . this is required to avoid the wigner bound and reproduce the positive value of the effective range with short - range interactions @xcite . spin and isospin degrees of freedom are included by treating the field @xmath8 as a doublet in both spaces , but for notational convenience we have suppressed the spin and isospin indices of @xmath8 . the projection operators , @xmath11 with @xmath12 ( @xmath13 ) operating in spin ( isospin ) space , project out the @xmath14 and @xmath15 nucleon nucleon partial waves , respectively . the covariant derivative @xmath16 where @xmath17 is the charge operator , includes the coupling to the electromagnetic field . furthermore , we have the kinetic and gauge fixing terms for the photons , @xmath18 of which we only keep contributions from coulomb photons . these correspond to a static coulomb potential between charged particles , but for convenience we introduce feynman rules for a coulomb photon propagator , @xmath19 which we draw as a wavy line , and factors @xmath20 for the vertices . following @xcite , we have introduced a small photon mass @xmath21 to regulate the singularity of the propagator at zero momentum transfer . as we will discuss later on , this regulator will be removed by numerical extrapolation back to vanishing photon mass . in the doublet - channel of the three - nucleon system , a three - body contact interaction is required for renormalisation at leading order @xcite . it can be written as @xmath22 \bigg ) \ , , \label{eq : l-3}\end{gathered}\ ] ] where @xmath23 is a momentum cutoff applied in the three - body equations discussed below and @xmath24 a known log - periodic function of the cutoff that depends on a three - body parameter @xmath25 . in the strong sector , we adopt the standard power counting for large s - wave scattering length @xcite . a nucleon bubble together with a bare dibaryon propagator scales as @xmath26 . the bare dibaryon propagators therefore are dressed by nucleon bubbles to all orders . the resulting geometric series for the full propagators are shown in fig . [ fig : dibaryonprop ] . state ( _ i.e. _ the deuteron ) and ( b ) the @xmath15 state . ] for convenience , we also resum the effective range corrections . if desired , the perturbative expressions can always be obtained by re - expanding the propagators . we do not go into the details of the calculations here and simply quote the results for the renormalised propagators , which we obtain by demanding that the effective range expansions @xmath27 around the deuteron pole , and @xmath28 for the singlet channel are reproduced . in writing eq . ( [ eq : er - t ] ) we have used that @xmath29 to the order we are working . this yields the expressions @xmath30 and analogously @xmath31 for the spin - singlet dibaryon . these expressions are valid to . at leading order , effective range corrections are not included and the dibaryon kinetic terms do not contribute . the corresponding propagators are obtained by setting @xmath32 and @xmath33 in eqs . ( [ eq : prop - d - high ] ) and ( [ eq : prop - t - high ] ) . the deuteron wave function renormalisation constant is given as the residue at the bound state pole : @xmath34 the coulomb interaction breaks the isospin symmetry that is implicit in the dibaryon propagators from the previous subsection . for the @xmath35-part of the singlet dibaryon we can also have coulomb photon exchanges inside the nucleon bubble . these can be resummed to all orders , yielding a dressed nucleon bubble @xcite , which is then used to calculate the full singlet dibaryon propagator in the @xmath35-channel , as shown in fig . [ fig : dressedbubble ] . -channel . ] the result for the leading order propagator is @xcite @xmath36 with @xmath37 and @xmath38 where @xmath39 denotes the logarithmic derivative of the @xmath40-function . effective range corrections can be included in the same way as described above . the power counting of pionless effective field theory has been extensively discussed in the literature ( see the reviews @xcite and references therein ) . we will thus be rather brief on this subject here . we will , however , elaborate a bit on the power counting for the coulomb sector of the theory , as it was introduced in @xcite . the low - energy scale @xmath41 of the theory is set by the deuteron binding momentum @xmath42 . we can formally count the external momenta @xmath43 to be of the same order . since we are working in a theory without explicit pions , the natural ultraviolet cutoff of our theory is of the order of the pion mass , @xmath44 . which cutoff is best to use in practice depends on whether one discusses the quartet - channel system ( where short - range effects are suppressed by the pauli principle ) , or the doublet - channel system . in the first case one finds that already @xmath45 is sufficient for an accurate description , whereas in the latter case we have to set the cutoff to a few hundred @xmath6 to reach convergence . the combination of the two scales yields the expansion parameter @xmath46 . a further relevant scale in our system is the nucleon mass @xmath47 . it appears explicitly in kinetic energies , which scale as @xmath48 . as a consequence , the nucleon propagator scales as @xmath49 and the loop integration measure @xmath50 scales as @xmath51 . we assume @xmath52 for the nucleon dibaryon coupling constants and @xmath53 for the bare dibaryon propagator constants . from the form of the coulomb potential in momentum space , @xmath54 it is clear that coulomb contributions are dominant for small momentum transfers . as noted in @xcite , they enter @xmath55 . this behaviour is not captured by the power counting for the strong sector . hence , when one wants to perform calculations including coulomb effects for small external momenta , one can no longer assume that all momenta scale with @xmath56 . instead , one has to keep track of the new scale introduced by the external momenta separately . we generically denote this scale by @xmath4 and assume @xmath57 for the power counting . as noted in @xcite , this means that we make a simultaneous expansion in _ two _ small parameters @xmath58 and @xmath59 . for @xmath60 , the coulomb contributions are small and the results in both schemes agree . with this modified counting , it is not straightforward to deduce the scaling of loops anymore . kinetic energies always scale like @xmath41 , so the scaling of @xmath61 and the nucleon propagator is not modified in the presence of coulomb effects . however , where we could simply assume that all loop momenta scale like @xmath41 before , we now have to check first which contribution is picked up ( or rather enhanced ) after carrying out the @xmath61-integral . in general , we have that 1 . the loop integration measure @xmath62 scales as @xmath63 , and 2 . the photon propagator scales as @xmath64 , where either @xmath65 or @xmath66 . these rules will become more transparent when we apply them to deduce the scaling of the diagrams shown below . @xmath5 scattering involving the exchange of coulomb photons . ] in this subsection , we discuss several diagrams contributing to @xmath4@xmath5 scattering that include coulomb photons ( see fig . [ fig : coulombdiagrams ] ) . for the discussion we always assume @xmath67 . diagram ( a ) simply scales as @xmath68 . the diagram ( b ) is enhanced relative to ( a ) by a factor @xmath69 from the nucleon bubble and hence gives the leading - order coulomb contribution . diagram ( a ) enters only at since the dibaryon kinetic energy operators , which generate the direct coupling of the photons to the dibaryons , enter only as effective range corrections . diagram ( c ) contains two loops , which have to be analysed separately . the upper nucleon bubble does not contain any photon propagators , so there all momenta scale as @xmath41 . in the lower loop , momenta certainly scale @xmath70 due to the two photon propagators that involve the external momentum . including the remaining nucleon propagator , which cancels the contribution from the integration measure , we are left with a total scaling @xmath71 for diagram ( c ) . this means that compared to diagram ( b ) it is suppressed by a factor @xmath72 . for diagrams of the form ( c ) with more than two photons attached to the bubble we simply quote the results from @xcite . the diagram with three photons could contribute with a factor @xmath73 , whereas the diagrams with @xmath74 photons attached to the bubble are even infrared finite and suppressed by factors @xmath75 . following @xcite , we neglect them all and also the logarithmically - scaling diagram with three photons ( which is already small for @xmath76 ) . diagram ( d ) is a little ambiguous since _ a priori _ it is not clear whether the loop momentum should scale @xmath77 or @xmath70 . in @xcite , the first alternative is chosen , yielding that compared to the same diagram without the photon it is suppressed by a factor @xmath78 . a direct numerical calculation shows that it is a seven - percent effect at threshold . the diagram ( e ) obviously is irrelevant for the quartet - channel system ( there are never two protons in the dibaryon ) , but , at least in principle , it can play a role in the doublet - channel system . the power counting , however , yields the same suppression factor as for diagram ( c ) , only in this case the scaling of the loop momentum is not ambiguous . a direct numerical evaluation yields that at threshold it is a 15% contribution ( again compared to the simple nucleon - exchange diagram without the photon ) . we take this value as the _ a priori _ theoretical uncertainty of our doublet - channel calculation . the bottom line of the discussion above is that , as done in @xcite , we iterate the diagrams ( a ) and ( b ) to all orders and do not include any of the other diagrams shown in fig . [ fig : coulombdiagrams ] . the claim is that this procedure is adequate for both the quartet - channel and the doublet - channel system . the coulomb effects are thus included at accuracy in our calculation . we start with a review of @xmath8@xmath5 quartet - channel scattering , where the spin 1 of the deuteron and the spin @xmath79 of the nucleon couple to a total spin of @xmath80 . since this coupling is only possible when the spins of all three nucleons taking part in the reaction are aligned , the pauli principle applies . hence , the system is rather insensitive to short - range physics . furthermore , only the dibaryon field representing the deuteron can appear in the intermediate state . in the quartet channel . ] in fig . [ fig : nd - inteq - q ] we show a diagrammatic representation of the strong ( neutron deuteron ) scattering amplitude @xmath81 , which does not include any coulomb effects . it is projected onto the spin quartet channel by setting @xmath82 and @xmath83 for the in- and outgoing deuteron spin indices , respectively , and @xmath84 to select the neutron . after furthermore projecting onto s - waves , we get @xmath85 where @xmath86 and @xmath4 are the incoming and outgoing momenta of the particles in the center - of - mass frame , and @xmath87 more details on the derivation of this equation and the required projections can be found in appendix [ sec : scatteq - details ] . we have introduced a cutoff @xmath23 to regularise the loop integral , which is particularly convenient for a numerical treatment of the equation . strictly , however , this regulator is only required for the full amplitude including coulomb interactions and is applied here for convenience only . after setting the energy @xmath88 to the total center - of - mass energy , @xmath89 the equation is solved numerically with standard linear algebra routines after discretising the integrals . from the result we then obtain the s - wave scattering phase shift @xmath90 which can be compared to experimental data . in order to simplify the expressions in the following sections , we introduce a short - hand notation for the scattering equations . defining @xmath91 and @xmath92 along with the operation @xmath93 we find that we can write as @xmath94 \,,\ ] ] where we have omitted all arguments . in the quartet channel . ] . ] in order to include coulomb effects and hence discuss proton deuteron scattering we follow @xcite and define a full scattering amplitude @xmath95 ( see fig . [ fig : pd - inteq - q ] ) that includes both strong and coulomb interactions and a pure coulomb scattering amplitude @xmath96 ( see fig . [ fig : pd - inteq - coulomb ] ) . after spin- , isospin- and s - wave projection we find the integral equations @xmath97 \label{eq : pd - inteq - q - full}\ ] ] and @xmath98 \label{eq : pd - inteq - q - c}\ ] ] with @xmath99 after solving the individual equations , we calculate the phase shifts @xmath100 and @xmath101 according to . the final result , which we will compare to experimental data , is the coulomb - subtracted phase shift @xcite , @xmath102 note that in the integral equations above an artificial dependence on the bare coupling constant @xmath103 is kept for notational convenience . in all observables , this dependence drops out . we now go on to the doublet channel , where the spins of the nucleon and the deuteron couple to a total spin of @xmath79 . the spin - singlet dibaryon can now appear in the intermediate state , which leads to two coupled amplitudes that differ in the type of the outgoing dibaryon . in contrast to the quartet channel , the three nucleon spins no longer need to be aligned in the same direction , which means that a non - derivative three - nucleon interaction is no longer prohibited by the pauli principle . this channel is expected to be more sensitive to short - range physics in general , and in fact the three - body interaction ( [ eq : l-3 ] ) is _ needed _ at leading order to ensure correct renormalisation @xcite . in the doublet channel . the diagrams involving the three - body force have been omitted . ] as we did in the quartet channel , we look at the neutron deuteron system first . [ fig : nd - inteq - d ] shows a diagrammatic representation of the coupled - channel integral equation for the scattering amplitude @xmath81 , of which we only needed to consider the upper left part for the quartet - channel system . the contribution of the three - body interaction ( [ eq : l-3 ] ) is omitted here and will be included below . after projecting onto the @xmath104@xmath5 doublet channel with @xmath105 we find @xmath106 + { { { \mathcal{t}}_{\mathrm{s}}}^\mathrm{d , b}}\otimes \left[\frac{3{m_n}{y_d}{y_t}}2 d_t { k_{\mathrm{s}}}\right ] \ \label{eq : nd - inteq - d - b - no - h } { { { \mathcal{t}}_{\mathrm{s}}}^\mathrm{d , b}}&= -\frac{3{m_n}{y_d}{y_t}}2{k_{\mathrm{s}}}+ { { { \mathcal{t}}_{\mathrm{s}}}^\mathrm{d , a}}\otimes \left[\frac{3{m_n}{y_d}{y_t}}2 d_d { k_{\mathrm{s}}}\right ] - { { { \mathcal{t}}_{\mathrm{s}}}^\mathrm{d , b}}\otimes \left[\frac{{m_n}{y_t}^2}2 d_t { k_{\mathrm{s}}}\right ] \,.\end{aligned}\ ] ] [ eq : nd - inteq - d - no - h ] more details on the derivation and projection are again given in appendix [ sec : scatteq - details ] . we still need to include the contribution of the three - nucleon interaction ( [ eq : l-3 ] ) in eqs . ( [ eq : nd - inteq - d - a - no - h ] ) and ( [ eq : nd - inteq - d - b - no - h ] ) . a straightforward calculation shows that this can be achieved with the replacement @xcite @xmath107 it is , however , important to note that for the terms with an additional factor of @xmath108 in front of @xmath109 , it is cancelled in the @xmath24-part by the additional factor @xmath110 in . with this , we arrive at the final version of our integral equations , @xmath111 { { { \mathcal{t}}_{\mathrm{s}}}^\mathrm{d , b}}\end{pmatrix } = \begin{pmatrix } g_{dd}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right)\\[{0.5em } ] -g_{dt}\left(3{k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) \end{pmatrix } \\ + \begin{pmatrix } -g_{dd}d_d\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & g_{dt}d_t\left(3{k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) \\[{0.5em } ] g_{dt}d_d\left(3{k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & g_{tt}d_t\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) \end{pmatrix } \otimes\begin{pmatrix}{{{\mathcal{t}}_{\mathrm{s}}}^\mathrm{d , a}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{s}}}^\mathrm{d , b}}\end{pmatrix } \,,\end{gathered}\ ] ] written in a compact matrix vector notation . we have furthermore introduced the abbreviations @xmath112 again , the dependence on the bare coupling constants @xmath103 and @xmath113 is only kept for notational convenience and drops out in all observables . finally , we have all the ingredients to discuss the proton deuteron system in the doublet channel . in the doublet channel . the diagrams representing the three - nucleon force have been omitted . ] due to the fact that the electromagnetic interaction does not couple to isospin eigenstates we now need two different projections for the amplitude @xmath114 with the outgoing spin - singlet dibaryon : @xmath115 [ eq : proj - d - b1b2 ] the latter corresponds to the amplitude with the outgoing spin - singlet dibaryon in a pure @xmath35-state . for the diagrams that have this component in the intermediate state , we have to insert the propagator into the unprojected equations in appendix [ sec : scatteq - details ] and find @xmath116{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b1}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b2}}\end{pmatrix } = \begin{pmatrix } g_{dd}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right)\\[{0.5em } ] -g_{dt}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{3\lambda^2}\right)\\[{0.5em } ] -g_{dt}\left(2{k_{\mathrm{s}}}+\frac{4h(\lambda)}{3\lambda^2}\right ) \end{pmatrix } + \begin{pmatrix}g_{dd}{{k_{\mathrm{c}}}^{(d)}}\\[{0.5em}]0\\[{0.5em}]0\end{pmatrix } \\ + \begin{pmatrix } -g_{dd}d_d\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & g_{dt}d_t\left(3{k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & 0\\[{0.5em } ] g_{dt}d_d\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{3\lambda^2}\right ) & g_{tt}d_t\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & 0\\[{0.5em } ] g_{dt}d_d\left(2{k_{\mathrm{s}}}+\frac{4h(\lambda)}{3\lambda^2}\right ) & -g_{tt}d_t\left(2{k_{\mathrm{s}}}+\frac{4h(\lambda)}{\lambda^2}\right ) & 0 \end{pmatrix } \otimes \begin{pmatrix } { { { \mathcal{t}}_{\mathrm{full}}}^\mathrm{d , a}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b1}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b2}}\end{pmatrix } \\ + \begin{pmatrix } -g_{dd}d_d{{k_{\mathrm{c}}}^{(d ) } } & 0 & g_{dt}d_t^{pp } \left(3{k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right)\\[{0.5em } ] 0 & -g_{tt}d_t{{k_{\mathrm{c}}}^{(t ) } } & -g_{tt}d_t^{pp } \left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right)\\[{0.5em } ] 0 & 0 & 0 \end{pmatrix } \otimes\begin{pmatrix } { { { \mathcal{t}}_{\mathrm{full}}}^\mathrm{d , a}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b1}}\\[{0.5em}]{{{\mathcal{t}}_{\mathrm{full}}}^\mathrm{d , b2}}\end{pmatrix } \label{eq : pd - inteq - d - full}\end{gathered}\ ] ] with @xmath117 the terms in have been separated in such a way that the sub - channels with coulomb contributions can be easily identified . the equation for the coulomb scattering amplitude @xmath96 is exactly the same as in the quartet channel . the dibaryon propagators with the resummed kinetic energy insertions have an unphysical deep bound state pole at the radius of convergence of the geometric series . in the quartet channel the cutoff can be chosen low enough to avoid that pole . due to the larger cutoff needed in the doublet channel , however , we can not use the resummed propagators here . instead , we use linear and quadratic insertions of the kinetic energy operator in in the kernel of the integral equations in order to include effective range corrections and obtain the next - to - leading order and next - to - next - to - leading order propagators @xmath118 and @xmath119 @xcite . alternatively , we can think of this as re - expanding the renormalised propagators and up to linear and quadratic order in @xmath120 , respectively . this still resums some higher order effective range contributions , but removes the unphysical pole . the question of when higher - order three - body forces enter in the doublet channel is still under discussion . in ref . @xcite , a subleading three - body force was included as required by nave dimensional analysis . a lepage - plot analysis showed that its inclusion reduces the errors in the calculation . this was supported by ref . @xcite , where a corresponding logarithmic divergence at n@xmath121lo , requiring a subleading three - body force , was identified . more recently , platter and phillips , using the subtractive renormalisation scheme , showed that the leading three - body force is sufficient to achieve cutoff independence up to @xcite . a perturbative analysis recently showed that there is a new three - body parameter already at if the scattering length is not fixed @xcite . in this work , the scattering length _ is _ fixed and we will not include a subleading three - body force . assuming the counting of @xcite , our calculation will correspond to in the quartet channel and to in the doublet channel . we will also perform a calculation including only the two - body interactions to in the doublet channel . the integral equations presented in the previous sections have to be solved numerically . we do so by discretising the integrals , using gaussian quadrature , principal value integration to deal with the singularity of the deuteron propagator , and appropriate transformations of the integration domain . the latter are especially important to deal with the numerical difficulties caused by the coulomb photon propagators . even though we have regulated the singularity with the artificial photon mass @xmath21 , the latter has to be kept small , which then yields strongly peaked functions . it turns out that the coulomb peaks in the inhomogeneous parts of eqs . , and are the major numerical problem . we solve it by concentrating the quadrature points around this peak . together with always putting half of the quadrature points into the low - momentum region ( the interval from zero to the peak position ) , we are able to ( linearly ) extrapolate our results for the scattering phase shifts back to the physical value @xmath122 ( screening limit ) . as a typical example , we show the quartet - channel phase shift at @xmath123 as a function of @xmath21 in fig . [ fig : phase - l - q ] . the linear dependence is clearly visible , with deviations only for very small @xmath21 . in the doublet channel , the qualitative behaviour is the same , only the curvature becomes visible already for somewhat larger photon masses . this can be understood by noting that the absolute value of the phase shift is smaller in the doublet channel , especially for low center - of - mass momenta @xmath86 ( _ cf . _ sec . [ sec : scatt - results ] ) . we settled to use the intervals @xmath124 and @xmath125 for the extrapolations in the quartet and doublet channel , respectively . we note that the error introduced by the extrapolation to @xmath122 can generally be neglected compared to the theoretical error from the eft expansion discussed below . @xmath5 quartet channel s - wave scattering phase shift at for center - of - mass momentum @xmath123 and cutoff @xmath126 as a function of the regulating photon mass @xmath21.,scaledwidth=80.0% ] we have used the experimental input parameters shown in tab . [ tab : params ] in the numerical calculation . .parameters used in the numerical calculation , @xmath127 . [ cols="^,^,^,^",options="header " , ] @xmath5 quartet channel s - wave scattering phase shifts as functions of the center - of - mass momentum @xmath86 . error bands generated by cutoff variation from 120 to 160 @xmath6 . experimental @xmath4@xmath5 phase shift data taken from @xcite ( diamonds ) and @xcite ( circles).,scaledwidth=80.0% ] in fig . [ fig : phase - nd - q ] we show the phase shift results for both neutron deuteron and proton deuteron scattering as functions of the center - of - mass momentum @xmath86 . the error bands are generated by varying the cutoff within a range of @xmath128 to @xmath129 . since the cutoff variation is small , we conclude that the calculation is well converged at these cutoffs . for @xmath130 some numerical artifacts show up from integrating over the unphysical second pole in the full deuteron propagator . for the @xmath104@xmath5 curve in fig . [ fig : phase - nd - q ] we have used @xmath126 . the fact that the bands do not overlap is no point of concern since they only give a lower bound on the error of the calculation . from the expansion parameter @xmath131 of the eft , the error can be estimated as 30% , 10% , and 3% at , , and , respectively . thus , at , the 30% error from the expansion parameter clearly dominates . at and , however , the band from the cutoff variation gives a reasonable estimate of the total error in the calculation . the result to the right of the dotted line at @xmath132 agrees nicely with the results presented in @xcite and also with the experimental data included in the plot . at this point we remark , however , that for @xmath133 the coulomb parameter @xmath134 is of order @xmath110 , which means that in this regime the non - perturbative treatment of coulomb effects might not even be necessary . more important are hence the @xmath4@xmath5 results for small momenta ( @xmath135 mev ) to the left of the dotted line , which we could obtain thanks to our optimised numerical procedure . it would of course be good to have some data points in this region to test our prediction . @xmath5 doublet channel s - wave scattering phase shifts as functions of the center - of - mass momentum @xmath86 . error bands generated by cutoff variation from 200 to 600 @xmath6 . experimental @xmath4@xmath5 phase shift data taken from @xcite.,scaledwidth=80.0% ] the doublet - channel results for the @xmath4@xmath5 scattering phase shifts as functions of the center - of - mass momentum @xmath86 are shown in fig . [ fig : phase - nd - d ] . as in the quartet channel , the @xmath4@xmath5 curve lies above the @xmath104@xmath5 curve and agrees quite well with the experimental data . a more quantitative comparison is , unfortunately , not possible since there are no errors given for the data points . the error bands are generated by varying the cutoff within a natural range of @xmath136 to @xmath137 ( _ i.e. _ a few times the pion mass ) . assuming the power counting of @xcite , our calculation is incomplete since the subleading three - body force is not included . a full calculation , however , is beyond the scope of the paper since gauging the subleading three - body force creates new three - body contributions to the photon coupling . our partial result is stable to within about ten percent under the cutoff variation , which is consistent with a 715% error estimate based on the neglected coulomb diagrams ( see sec . [ sec : power - counting ] and @xcite ) . figure [ fig : phase - nd - d ] furthermore shows how the results improve from order to order . the stability of the partial result with respect to variation of the cutoff suggests that the scattering is relatively insensitive to the subleading three - body interaction . at higher energies , however , there is some room for such a contribution as a our partial result consistently lies two to four degrees above the data . we observe that the shift from to is of the same order of magnitude as the shift from to . this behaviour is typical for effective range corrections in the doublet channel @xcite . the smallness of the corrections can be understood as a cancellation between two different contributions to this correction . the two contributions are proportional to @xmath138 and @xmath139 , respectively , where @xmath140 is the typical momentum scale of the process . furthermore , it is known that at observables are often described better than expected from the power counting once the exact pole position of the two - body propagator is reproduced @xcite . as a consequence , the shifts in observables from to can be small and of a size comparable to the corresponding shifts from to . since the power counting we used for the coulomb contributions is not valid for the regime of typical bound state energies , we can not simply use the above equations to calculate @xmath0he . there are various strategies to proceed . we could extend the power counting and include additional coulomb diagrams in our equations . alternatively , we could use an analytic expression for the full off - shell coulomb amplitude as it is done in ref . @xcite for an calculation . here , we choose the much simpler approach of calculating the coulomb energy shift for @xmath0he in first order perturbation theory as the expectation value of the coulomb interaction between proton and deuteron using trinucleon wave functions in the isospin limit . in order to obtain the trinucleon wave functions we need to solve the homogeneous coupled - channel equation @xmath141 with @xmath142 , @xmath143 , and @xmath144 g_{dt}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{3\lambda^2}\right ) & g_{tt}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right ) & -g_{tt}\left({k_{\mathrm{s}}}+\frac{2h(\lambda)}{\lambda^2}\right)\\[{0.5em } ] g_{dt}\left(2{k_{\mathrm{s}}}+\frac{4h(\lambda)}{3\lambda^2}\right ) & -g_{tt}\left(2{k_{\mathrm{s}}}+\frac{4h(\lambda)}{\lambda^2}\right ) & 0 \end{pmatrix } \,.\ ] ] it is obtained by applying the projections to the raw equation without coulomb contributions . the reason to separate the wave function in this way is that the part with the dibaryon leg in the @xmath35-channel does not contribute to a perturbative calculation of the energy shift since in that case the third nucleon in the system necessarily is a neutron . the energy in the equation is set to the experimental triton binding energy , @xmath145 and the existence of solution is ensured by adjusting the three - nucleon force @xmath24 appropriately , as it was already done to renormalise the scattering equations . having obtained the wave functions as solutions of , we still need to normalise them properly . since the eft generates an energy - dependent interaction in the three - body system , this is done by demanding that @xmath146 where @xmath147 with @xmath148 a short derivation of this normalisation condition can be found in appendix [ sec : triton - norm ] . with the normalised trinucleon wave functions we can obtain the coulomb - induced energy shift in first order perturbation theory and find @xmath149 with the s - wave projected coulomb potential @xmath150 in momentum space . our prediction for the @xmath0he binding energy is then given by @xmath151 the results are shown in fig . [ fig : en - wf ] . it is remarkable that the result , which should be accurate to about 10% , agrees very well with the experimental value @xmath152 over a large cutoff range . in our partial calculation , the results are still quite stable against cutoff variations within @xmath136 to about @xmath153 , but our value lies about @xmath154 above the experimental value . this shift again leaves room for a natural - sized contribution of the subleading three - body we have not included . brandenburger , coon and sauer have determined the @xmath0he@xmath0h binding energy difference in a largely model - independent way using experimental charge form factors @xcite . they found that the coulomb contribution is about 10% below the experimental value for the total energy difference . within the expected error of 10% , our result is consistent with this . ando and birse have carried out a non - perturbative calculation to leading order in the pionless eft including the full off - shell @xmath155-matrix for the coulomb interaction and found @xmath156 @xcite . their calculation included isospin breaking effects in the nucleon nucleon scattering lengths . et al_. found the smaller value @xmath157 in an calculation using the resonating group model to solve the pionless eft with a charge - independent value of the spin - singlet scattering length and non - perturbative coulomb interactions @xcite . binding energy difference in dependence of the cutoff @xmath23 . bottom curve : result . middle curve : result . top curve : result.,scaledwidth=80.0% ] the increase in our result for @xmath158 that is seen in fig . [ fig : en - wf ] also occurs at and , but for larger cutoffs . in fig . [ fig : h - en - nlo ] we show the prediction for @xmath159 together with the three - nucleon force obtained from fitting the triton binding energy . it is obvious that a drop in the binding energy prediction occurs whenever the three - nucleon force has gone through a pole . we interpret this as an artifact of the theory which is related to the efimov effect . for cutoffs beyond the position of the first pole in @xmath24 , the triton is not the true ground state of the system anymore ; after each pole transition a new ( unphysical ) deep bound state emerges . he binding energy together with the three - nucleon force ; results . solid curve : @xmath0he binding energy prediction . dashed curve : three - nucleon force . dotted line : experimental value for the @xmath0he binding energy.,scaledwidth=80.0% ] these unphysical deep states lead to additional nodes in the triton wave function at short distances which affect our perturbative results . this suggests that an additional short - distance counterterm is required to cancel these contributions if one wants to go to cutoffs much larger than the pion mass . in this paper , we have investigated s - wave proton deuteron scattering in pionless effective field theory . in the quartet channel , we have calculated the elastic scattering phase shift up to using the power counting for coulomb contributions suggested by rupak and kong @xcite . the coulomb effects are included at accuracy in our calculation . using an optimised integration mesh we were able to extend their calculation into the threshold region were the coulomb interaction becomes highly non - perturbative . we found good agreement both with available phase shift analyses and with the results of rupak and kong at momenta @xmath160 . moreover , we extended the power counting to the doublet channel and performed a complete calculation of the phase shifts to in agreement with the available phase shift data . we also carried out a partial calculation that neglected the contribution of the subleading three - body force entering at this order . the results of this calculation are stable under variations of the cutoff . furthermore , there is good agreement with the phase shift data at low momenta and room for a small contribution of the neglected three - body force at larger momenta . overall , however , the doublet channel phase shifts are only weakly sensitive to the subleading three - body force entering at . although we were mainly interested in @xmath4@xmath5 scattering , we have also calculated the coulomb contribution to the @xmath0he@xmath0h binding energy difference @xmath161 . this observable has previously been calculated in the pionless theory by treating the coulomb interaction non - perturbatively @xcite . here , we treat the coulomb potential between proton and deuteron in first order perturbation theory using trinucleon wave functions . higher order corrections to this quantity are expected to be small . our result is in reasonable agreement with the experimental value and other evaluations . we find @xmath161 to be more sensitive to the subleading three - body force . the partial result is about 10% too large , thus leaving room for a contribution from the omitted three - body force . we also observe steps in the calculated value of @xmath161 as the cutoff is increased beyond its natural range . whenever the leading three - nucleon force has gone through a pole , a drop in the calculated binding energy occurs . we interpret this as an artifact of the theory related to the efimov effect . at higher cutoffs , spurious deep three - body bound states appear and the triton is not the true ground state anymore . it appears that an additional short - distance counterterm is required to cancel these contributions if one wants to go to cutoffs much larger than the pion mass . a further study of this issue would be interesting . in the future , a full calculation including the subleading three - body force and the electromagnetic interaction terms generated from gauging its momentum dependence should be carried out . such an accuracy will , _ e.g. _ , be required for high - precision calculations of low - energy astrophysical processes in pionless effective field theory and the effective field theory for halo nuclei . we thank a. rusetsky , b. metsch , m. hoferichter and p. hagen for discussions and s .- ando , s. coon , d. r. phillips and u. van kolck for comments on the manuscript . this research was supported in part by the dfg through sfb / tr 16 `` subnuclear structure of matter '' and the bmbf under contract no . 06bn9006 . s.k . was supported by the `` studienstiftung des deutschen volkes '' and by the bonn - cologne graduate school of physics and astronomy . using the feynman rules that follow from the lagrangian and inserting appropriate symmetry factors , we find @xmath162 for the @xmath104@xmath5 scattering equation depicted in fig . [ fig : nd - inteq - q ] . in the same way we get @xmath163 \\ + \int{{\!\!\frac{\mathrm{d}^3q}{(2\pi)^3}}}\,\delta_d\left(e-\frac{{\mathbf{q}}^2}{2{m_n}},{\mathbf{q}}\right ) \cdot({\mathrm{i}}{{\mathcal{t}}_{\mathrm{full}}}^{ik})^{\gamma c}_{\alpha a}(e;{\mathbf{k}},{\mathbf{q } } ) \cdot\left\{\frac{{m_n}{y_d}^2}{2}\frac{(\sigma^j\sigma^k)^\beta_\gamma\delta^b_c } { { \mathbf{q}}^2+{\mathbf{q}}\cdot{\mathbf{p}}+{\mathbf{p}}^2-{m_n}e-{\mathrm{i}}{\varepsilon}}\right . \\ + \left.\,\alpha{m_n}^2{y_d}^2\cdot\delta^{kj}\delta^\beta_\gamma \left(\frac{{\mathbf{1}}+\tau_3}{2}\right)^b_c \left[\frac{\mathcal{i}_\mathrm{bubble}(e;{\mathbf{q}},{\mathbf{p}})}{({\mathbf{q}}-{\mathbf{p}})^{2}+\lambda^2 } -\frac{{\rho_d}}{2}\frac{1}{({\mathbf{q}}-{\mathbf{p}})^2+\lambda^2}\right]\right\}\end{gathered}\ ] ] for the full @xmath4@xmath5 scattering equation shown in fig . [ fig : pd - inteq - q ] , and @xmath164 \\ + \int{{\!\!\frac{\mathrm{d}^3q}{(2\pi)^3}}}\,\delta_d\left(e-\frac{{\mathbf{q}}^2}{2{m_n}},{\mathbf{q}}\right ) \cdot({\mathrm{i}}{{\mathcal{t}}_{\mathrm{c}}}^{ik})^{\gamma c}_{\alpha a}(e;{\mathbf{k}},{\mathbf{q } } ) \\ \times\alpha{m_n}^2{y_d}^2\cdot\delta^{jk}\delta^\beta_\gamma \left(\frac{{\mathbf{1}}+\tau_3}{2}\right)^b_c \left[\frac{\mathcal{i}_\mathrm{bubble}(e;{\mathbf{q}},{\mathbf{p } } ) } { ( { \mathbf{q}}-{\mathbf{p}})^{2}+\lambda^2}-\frac{{\rho_d}}{2}\frac{1}{({\mathbf{q}}-{\mathbf{p}})^2+\lambda^2}\right]\end{gathered}\ ] ] for the pure coulomb scattering equation ( fig . [ fig : pd - inteq - coulomb ] ) , where @xmath165 corresponds to the loop integral in the diagram with the photon attached to a nucleon bubble . the expression looks quite complicated , but it can be simplified . the dominant terms of @xmath166 are those with @xmath167 and @xmath168 , respectively , due to the prefactors of @xmath169 and @xmath170 . in the latter case we can furthermore assume that @xmath171 because of the pole at this position in the propagator . furthermore inserting the total center - of - mass energy @xmath172 , we get @xmath173 and @xmath174 [ eq : bubble - approx ] where we have used and the expansion @xmath175 . the same simplifications , which effectively amount to keeping only loop contributions with @xmath66 , are used in @xcite and appear to be well supported by comparing the results with experimental data ( see sec . [ sec : scatt - results ] ) . in the doublet channel we find @xmath176 @xmath177 [ eq : nd - inteq - d - raw ] for the coupled - channel @xmath104@xmath5 equation shown in fig . [ fig : nd - inteq - d ] , and analogously we have @xmath178 \\ + \int{{\!\!\frac{\mathrm{d}^3q}{(2\pi)^3}}}\,\delta_d\left(e-\frac{{\mathbf{q}}^2}{2{m_n}},{\mathbf{q}}\right ) \cdot({\mathrm{i}}{{\mathcal{t}}_{\mathrm{full}}}^{\mathrm{a},ik})^{\gamma c}_{\alpha a}(e;{\mathbf{k}},{\mathbf{q } } ) \cdot\biggl\{\frac{{m_n}{y_d}^2}{2}\frac{(\sigma^j\sigma^k)^\beta_\gamma\delta^b_c } { { \mathbf{q}}^2+{\mathbf{q}}\cdot{\mathbf{p}}+{\mathbf{p}}^2-{m_n}e-{\mathrm{i}}{\varepsilon } } \\ + \alpha{m_n}^2{y_d}^2\cdot\delta^{kj}\delta^\beta_\gamma \left(\frac{{\mathbf{1}}+\tau_3}{2}\right)^b_c \left[\frac{\mathcal{i}_\mathrm{bubble}(e;{\mathbf{q}},{\mathbf{p}})}{({\mathbf{q}}-{\mathbf{p}})^{2}+\lambda^2 } -\frac{{\rho_d}}{2}\frac{1}{({\mathbf{q}}-{\mathbf{p}})^2+\lambda^2}\right]\biggr\ } \\ + \int{{\!\!\frac{\mathrm{d}^3q}{(2\pi)^3}}}\,\delta_t\left(e-\frac{{\mathbf{q}}^2}{2{m_n}},{\mathbf{q}}\right ) \cdot({\mathrm{i}}{{\mathcal{t}}_{\mathrm{full}}}^{\mathrm{b},ic})^{\gamma c}_{\alpha a}(e;{\mathbf{k}},{\mathbf{q } } ) \cdot\frac{{m_n}{y_d}{y_t}}{2}\cdot\frac{(\sigma^j)^\beta_\gamma(\tau^c)^b_c } { { \mathbf{q}}^2+{\mathbf{q}}\cdot{\mathbf{p}}+{\mathbf{p}}^2-{m_n}e-{\mathrm{i}}{\varepsilon } } \label{eq : pd - inteq - d - a - raw}\end{gathered}\ ] ] @xmath179\biggr\ } \label{eq : pd - inteq - d - b - raw}\end{gathered}\ ] ] [ eq : pd - inteq - d - raw ] for the @xmath4@xmath5 equation depicted in fig . [ fig : pd - inteq - d ] . in this section we will give a brief derivation of the normalisation condition for the triton wave functions used in sec . [ sec : he-3 ] . in order to do that we need to introduce a little more formalism . for simplicity , we work with a simplified nucleon deuteron system , where the virtual spin - singlet state is neglected . the discussion could easily be carried out for the full coupled - channel system , but that would only complicate the notation . we start by considering the full two - body nucleon deuteron propagator ( green s function ) @xmath180 , which fulfils the ( inhomogeneous ) bethe salpeter equation in momentum space : @xmath181 @xmath182 is essentially a product of a nucleon propagator @xmath183 and a deuteron propagator @xmath184 . more precisely , we have @xmath185 where @xmath186 is the total four - momentum of the system and @xmath187 . @xmath188 represents the ( doublet - projected ) one - nucleon exchange diagram , @xmath189 as shown , for example , in fig . [ fig : nd - inteq - q ] . we now consider a bound state at rest , _ @xmath192 , and define the _ amputated _ wave function @xmath193^{-1 } \cdot\left[\delta_n\left(-{\eta_ne_b}-p_0,{\mathbf{p}}\right)\right]^{-1 } \ , , \label{eq : bs - amputated}\ ] ] which fulfils the homogeneous equation @xmath194 carrying out the @xmath61-integration picks up the residue from the nucleon propagator pole at @xmath195 . from the resulting right hand side of we then find that @xmath196 fulfils the equation @xmath197 this is essentially the single - channel equivalent of ( before s - wave projection ) , so we have established the connection of our current formalism to the triton wave functions @xmath198 in sec . [ sec : he-3 ] . note furthermore that @xmath199 is a schrdinger wave function . we now write in an operator notation as @xmath200 defining @xmath201 we find @xmath202 where @xmath203 corresponds to the wave function given in , and @xmath204 from this we readily derive the normalisation condition @xmath205 where @xmath206 a straightforward calculation shows that @xmath207^{-1 } \ , , \label{eq : g0m1 - 3dbs}\ ] ] and we also see that @xmath208 . the expression for @xmath209 _ a priori _ looks more complicated , but one finds that in the formal expansion @xmath210{{\widetilde{g}}_0}^{-1 } \label{eq : v-3dbs - exp}\ ] ] everything but the term @xmath211 drops out , and we have @xmath212 the essential ingredient to see this is @xmath213 which , in turn , follows from the fact that the nucleon propagator residues are always picked up in such a way that one deuteron propagator is cancelled by the inverse propagator in @xmath214 , _ cf . _ altogether , we have shown that is just the single - channel version of the normalisation condition stated in sec . [ sec : he-3 ] ( modulo s - wave projection ) , where the functions @xmath215 correspond to @xmath214 . s. r. beane , w. detmold , k. orginos and m. j. savage , prog . * 66 * ( 2011 ) 1 [ arxiv:1004.2935 [ hep - lat ] ] . s. r. beane , p. f. bedaque , w. c. haxton , d. r. phillips and m. j. savage , arxiv : nucl - th/0008064 . p. f. bedaque and u. van kolck , ann . nucl . part . * 52 * ( 2002 ) 339 [ arxiv : nucl - th/0203055 ] . e. epelbaum , h .- w . hammer and u .- g . meiner , rev . * 81 * ( 2009 ) 1773 [ arxiv:0811.1338 [ nucl - th ] ] . d. b. kaplan , m. j. savage and m. b. wise , nucl . b * 534 * , ( 1998 ) 329 [ arxiv : nucl - th/9802075 ] . u. van kolck , nucl . a * 645 * , ( 1999 ) 273 [ arxiv : nucl - th/9808007 ] . h. a. bethe , phys . rev . * 76 * ( 1949 ) 38 . v. efimov , nucl . a * 362 * ( 1981 ) 45 . hammer and l. platter , ann . * 60 * ( 2010 ) 207 [ arxiv:1001.1981 [ nucl - th ] ] . x. kong and f. ravndal , phys . b * 450 * ( 1999 ) 320 [ arxiv : nucl - th/9811076 ] . x. kong and f. ravndal , nucl . phys . a * 665 * ( 2000 ) 137 [ arxiv : hep - ph/9903523 ] . ando , j. w. shin , c. h. hyun and s. w. hong , phys . c * 76 * ( 2007 ) 064001 [ arxiv:0704.2312 [ nucl - th ] ] . t. barford and m. c. birse , phys . c * 67 * ( 2003 ) 064006 [ arxiv : hep - ph/0206146 ] . ando and m. c. birse , phys . c * 78 * ( 2008 ) 024004 [ arxiv:0805.3655 [ nucl - th ] ] . x. kong and f. ravndal , phys . c * 64 * ( 2001 ) 044002 [ arxiv : nucl - th/0004038 ] . ando , j. w. shin , c. h. hyun , s. w. hong and k. kubodera , phys . b * 668 * ( 2008 ) 187 [ arxiv:0801.4330 [ nucl - th ] ] . v. efimov , phys . b * 33 * ( 1970 ) 563 . c. a. bertulani , h .- w . hammer and u. van kolck , nucl . phys . a * 712 * ( 2002 ) 37 [ arxiv : nucl - th/0205063 ] . g. rupak and x. kong , nucl . a * 717 * ( 2003 ) 73 [ arxiv : nucl - th/0108059 ] . s. ando and m. c. birse , j. phys . g : nucl . part . * 37 * ( 2010 ) 105108 [ arxiv:1003.4383 [ nucl - th ] ] . j. kirscher , h. w. griehammer , d. shukla and h. m. hofmann , eur . phys . j. a * 44 * ( 2010 ) 239 [ arxiv:0903.5538 [ nucl - th ] ] . f. bedaque , h .- w . hammer and u. van kolck , nucl . phys . a * 676 * ( 2000 ) 357 [ arxiv : nucl - th/9906032 ] . f. bedaque , g. rupak , h. w. griehammer and h .- w . hammer , nucl . a * 714 * ( 2003 ) 589 [ arxiv : nucl - th/0207034 ] . e. braaten and h .- w . hammer , phys . rept . * 428 * ( 2006 ) 259 [ arxiv : cond - mat/0410417 ] . f. gabbiani , p. f. bedaque and h. w. griehammer , nucl . phys . a * 675 * ( 2000 ) 601 [ arxiv : nucl - th/9911034 ] . d. r. phillips and t. d. cohen , phys . b * 390 * ( 1997 ) 7 [ arxiv : nucl - th/9607048 ] . j. d. jackson and j. m. blatt , rev . * 22 * ( 1950 ) 77 . h. w. griehammer , nucl . a * 760 * ( 2005 ) 110 [ arxiv : nucl - th/0502039 ] . l. platter , d. r. phillips , few body syst . * 40 * ( 2006 ) 35 [ arxiv : cond - mat/0604255 ] . c. ji , d. r. phillips , l. platter , europhys . * 92 * ( 2010 ) 13003 [ arxiv:1005.1990 [ cond-mat.quant-gas ] ] . c. van der leun and c. anderliesten , nucl . a * 380 * ( 1982 ) , 261 . j. j. de swart , c. p. f. terheggen and v. g. j. stoks , arxiv : nucl - th/9509032 . j. r. bergervoet , p. c. van campen , w. a. van der sanden and j. j. de swart , phys . c * 38 * ( 1988 ) 15 . j. arvieux , nucl . phys . a * 221 * ( 1973 ) 253 .

fractional dehn twists give a measure of the difference between the relative isotopy class of a homeomorphism of a bordered surface and the thurston representative of its free isotopy class . we show how to estimate and compute these invariants . we discuss the the relationship of our work to stabilization problems in classical knot theory , general open book decompositions , and contact topology . we include an elementary characterization of overtwistedness for contact structures described by open book decompositions .

1201.5290

given an automorphism @xmath0 of a bordered surface @xmath1 that is the identity map on @xmath2 , the fractional dehn twist coefficient of @xmath0 with respect to a component @xmath3 of @xmath2 is a rational number . it measures the amount of rotation that takes place about @xmath3 when isotoping @xmath0 to its thurston representative . see section [ basics ] for precise definitions of @xmath4 . this concept was first studied by gabai and oertel in @xcite , where it is called the degeneracy slope and was applied to problems related to dehn fillings of manifolds containing essential laminations . it is also used , with different coordinates , by roberts @xcite to describe surgeries on fibred knots in which she can construct taut foliations . most recently it was used by honda , kazez , and mati @xcite to quantify the concept of right - veering that arises in contact topology . section [ basics ] describes techniques for estimating and computing @xmath4 and gives some examples . we prove a general result , theorem [ stabilizedc(h ) ] , that @xmath5 $ ] if @xmath0 is a stabilization , and begin to develop properties of the important case when @xmath6 . in section [ pa ] we construct a class of examples , motivated by an example of gabai s @xcite , for which @xmath6 . the idea is to replace a fibred knot with its @xmath7-cable . this creates a fibred knot with reducible monodromy and fractional dehn twist coefficient @xmath8 . following this with surgery on an unknotted curve in the fibre of the @xmath7-cable can produce examples of fibred knots with pseudo - anosov monodromy that also have fractional dehn twist coefficient 1/2 . our interest in producing examples of fibred knots in @xmath9 with fractional dehn twist coefficient 1/2 arises from conjecture [ main1 ] . this conjecture states that a knot in @xmath9 with fractional dehn twist coefficient 1/2 can not be destabilized . section 4 includes a brief history of this conjecture . the relevance of conjecture [ main1 ] in contact topology is discussed in section [ applications ] . denote by @xmath10 the contact structure compatible with a pair @xmath11 . if true , this conjecture would provide the simplest known counterexamples to a conjecture of honda , kazez , and mati @xcite by producing fibred knots in @xmath9 that are not stabilizations , are right - veering , yet are still overtwisted . the section includes some context for the conjecture of @xcite , references to earlier counterexamples for links bounding planar surfaces by lekili @xcite , lisca @xcite , and ito and kawamuro @xcite , and some remarks on work of colin and honda @xcite . the technique we use for recognizing that a contact structure is overtwisted is completely elementary : we exhibit an overtwisted disk by finding an unknotted , untwisted curve on the seifert surface of the knot . we would like to thank david gabai and jennifer schultens for helpful conversations . we recall thurston s classification of surface automorphisms . [ thurston ] @xcite let @xmath1 be an oriented hyperbolic surface with geodesic boundary , and let @xmath12 . then @xmath0 is freely isotopic to either \(1 ) a pseudo - anosov homeomorphism @xmath13 that preserves a pair of geodesic laminations @xmath14 and @xmath15 , \(2 ) a periodic homeomorphism @xmath13 , in which case there is a hyperbolic metric for which @xmath1 has geodesic boundary and such that @xmath13 is an isometry of @xmath1 , or \(3 ) a reducible homeomorphism @xmath16 that fixes , setwise , a maximal collection of simple closed geodesic curves @xmath17 in @xmath1 . to avoid overlap in the cases , we refer to a map as reducible only if it is not periodic . given a reducible map , splitting @xmath1 along @xmath18 gives a collection of surfaces @xmath19 with geodesic boundary that are permuted by @xmath16 . choose an integer @xmath20 so that @xmath21 maps @xmath22 to itself . maximality of @xmath17 implies that applying thurston s classification theorem to @xmath23 produces either a pseudo - anosov or periodic representative . our primary focus will be in describing homeomorphisms from the point of view of a single boundary component @xmath24 that is fixed pointwise by @xmath0 . if such a map is reducible , let @xmath25 be the subsurface @xmath22 ( as above ) of @xmath1 which contains @xmath3 . then @xmath26 , and we let @xmath27 be the pseudo - anosov or periodic representative of @xmath28 . with this notation , we make the following definition . [ thurstonrep ] a map @xmath29 , freely isotopic to @xmath0 , is called a _ thurston representative _ of @xmath0 , if it is pseudo - anosov , periodic , or reducible with @xmath30 . consider the open book determined by the data @xmath11 and the suspension flow of @xmath13 . this flow , when restricted to the component corresponding to @xmath3 in the complement of the binding , necessarily has periodic orbits . let @xmath31 be one such , and write @xmath32 where @xmath33 , @xmath34 is the meridian , and @xmath35 are relatively prime integers . the _ fractional dehn twist coefficient _ of @xmath0 with respect to a component @xmath3 of @xmath2 @xcite is given by @xmath36 in the context of surgery on knots in @xmath9 , the reciprocal quantity , @xmath37 , is called the _ degeneracy slope _ by gabai @xcite . see also @xcite for motivating work describing the location of cusps from the point of view of a toroidal boundary component in a 3-manifold carrying an essential lamination . fractional dehn twists , though with different coordinates , also play an important role in roberts constructions of taut foliations @xcite . see @xcite for a description of the change in coordinates . an alternate description of the fractional dehn twist coefficient , which is meant to emphasize that @xmath0 looks like a fraction of a dehn twist has been applied to the boundary of @xmath1 is given as follows . first extend @xmath0 by the identity map to a homeomorphism of @xmath38 where the annuli added to the boundary of @xmath1 are glued along @xmath39 . next , freely isotop @xmath0 to its thurston representative , @xmath13 , on @xmath1 , and extend this representative across @xmath40 so that the resulting homeomorphism of @xmath41 is isotopic to @xmath42 relative to @xmath43 . the extension can be chosen to be a shearing map , that is , to be a straight - line homotopy when lifted to the universal cover of @xmath40 thought of as a subset of the plane . the periodic orbit @xmath31 determines a collection of points @xmath44 labelled cyclically and ordered compatibly with the induced orientation on @xmath3 . the shearing homeomorphism that extends @xmath13 across @xmath45 maps @xmath46 to an arc that may spiral more than once around @xmath3 in either direction . lifting to the the universal cover and indexing with integers rather than working mod @xmath47 , it must end at some @xmath48 . it then follows that @xmath49 . we now describe how to compute @xmath4 with respect to @xmath3 by computing the action of @xmath0 on some arcs . let @xmath50 be an oriented , properly embedded arc in @xmath1 starting on @xmath3 . define @xmath51 to measure the number of intersections of @xmath50 and @xmath52 which occur after their common initial point but in an annular neighborhood of @xmath3 . . ] topologically , this means first isotoping the arcs @xmath50 and @xmath52 , relative to their boundaries , so they intersect minimally . then @xmath51 is defined to be a signed count of the number of points , @xmath53 , in the intersection of the interiors of @xmath52 and @xmath50 with the property that the union of the initial segments of these arcs , up to @xmath53 , is contained in an annular neighborhood of @xmath3 . it is also useful to define @xmath51 geometrically by putting the arcs in a canonical position . we can think of arcs as living in @xmath38 a hyperbolic surface with geodesic boundary together with annuli extending the boundary components . denote by @xmath54 the unique geodesic arc in the free isotopy class of @xmath50 that is orthogonal to the initial and terminal boundary components of @xmath1 . now let @xmath55 and @xmath56 be monotonically spiraling arcs in @xmath40 such that @xmath57 is an arc isotopic , relative to its endpoints , to @xmath50 . write @xmath58 for the canonical form of @xmath52 . the geometric description of @xmath51 depends on properties of its thurston representative , @xmath13 . in all cases @xmath59 . if @xmath13 is pseudo - anosov , it fixes no compact geodesic , and hence @xmath60 and @xmath61 intersect minimally . it follows that @xmath51 is the number of intersections of the interiors of @xmath62 and @xmath55 counted with sign ; namely , @xmath63 the same formula for @xmath51 applies when @xmath13 is periodic , provided that @xmath64 . in the case that @xmath65 , @xmath13 is an orientation preserving isometry that agrees with the identity map at a point of @xmath2 , and hence is the identity map . it follows that @xmath0 is the composition of dehn twists along the boundary components of @xmath1 . in the case that @xmath13 is reducible let @xmath66 and @xmath25 be as in definition [ thurstonrep ] . if @xmath67 , then the formulas given above can be applied to compute @xmath68 . otherwise , the geometric description breaks into cases . although an analysis of these cases is straightforward , it seems to yield little insight , and so we do not include this here . @xcite let @xmath50 and @xmath69 be properly embedded oriented arcs in @xmath1 with a common initial point . suppose , without changing notation , that they have been isotoped , while fixing endpoints , to intersect transversely in the minimum possible number of points . we say @xmath50 is _ to the right of _ @xmath69 if the frame consisting of the tangent vector to @xmath50 followed by the tangent vector to @xmath69 agrees with the orientation on @xmath1 . similarly , we say @xmath50 is _ to the left of _ @xmath69 if the frame consisting of the tangent vector to @xmath69 followed by the tangent vector to @xmath50 gives the orientation on @xmath1 . this definition implies that every arc is both to the left and right of itself . see @xcite for several equivalent definitions . @xcite a homeomorphism @xmath0 that restricts to the identity map on @xmath2 is called _ right - veering at @xmath3 _ if for every embedded oriented arc @xmath50 with initial endpoint on @xmath3 , @xmath52 is to the right of @xmath50 . similarly , a homeomorphism @xmath0 that restricts to the identity map on @xmath2 is called _ left - veering at c _ if for every embedded oriented arc @xmath50 with initial endpoint on @xmath3 , @xmath52 is to the left of @xmath50 . the first step in using @xmath70 to compute @xmath4 is the following proposition . @xcite[rv ] let @xmath12 . then @xmath0 is right - veering if and only if @xmath71 for every component of @xmath2 , and similarly @xmath0 is left - veering if and only if @xmath72 for every component of @xmath2 . we now return our focus to a component @xmath3 of @xmath2 . in what follows , @xmath4 denotes the fractional dehn twist coefficient with respect to @xmath3 and properly embedded oriented arcs have initial point on @xmath3 as necessary . [ c(h)<>0]if @xmath50 is a properly embedded oriented arc in @xmath1 , then @xmath52 is to the right of @xmath50 implies @xmath73 , and @xmath52 is to the left of @xmath50 implies @xmath74 . this gives a quick way to show that @xmath75 by producing two arcs , one moved to the right by @xmath0 and the other moved to the left by @xmath0 . see example [ example_8_20 ] . if @xmath75 it is always possible to find such a pair of arcs . note that if @xmath13 is periodic or @xmath13 is reducible with periodic @xmath27 , then we can choose a single arc @xmath50 fixed by @xmath0 ( and hence moved both to the left and to the right by @xmath0 ) . for @xmath0 with @xmath76 with @xmath77 , the next proposition shows how to produce a homeomorphism @xmath78 for which computing @xmath79 is equivalent to showing @xmath76 . [ formula ] if @xmath80 , then @xmath81 . propositions [ formula ] , [ rightveeringi ] , and [ leftveeringi ] make it possible to estimate @xmath4 experimentally . [ example_8_20 ] figure [ 8_20 ] shows the knot @xmath82 , its seifert surface , @xmath1 , and two curves living on @xmath1 . this is a fibred knot with pseudo - anosov monodromy @xmath0 . the curves shown are the cores of left and right hopf bands . it follows that the indicated arcs @xmath83 and @xmath84 in @xmath1 are moved to the right and to the left , respectively . by corollary [ c(h)<>0 ] , @xmath75 . [ rightveeringi ] suppose @xmath0 is right - veering at @xmath3 . then either * @xmath85 and @xmath86 or * @xmath87 and @xmath88 . consider first the case that @xmath1 is an annulus . in this case , necessarily @xmath89 , with @xmath90 and @xmath91 a positive dehn twist about the core of the annulus . also , @xmath92 and so the result holds . so we may assume @xmath1 is not an annulus . let @xmath13 be the thurston representative of @xmath0 . let @xmath93 denote the restriction of @xmath13 to @xmath3 . if @xmath13 is pseudo - anosov or @xmath13 is reducible with @xmath27 pseudo - anosov , then let @xmath94 be any perodic orbit of @xmath93 . we choose @xmath95 to be ordered cyclically about @xmath3 , with order compatible with the induced orientation on @xmath3 . if @xmath13 is periodic , then choose a properly embedded geodesic arc @xmath96 in @xmath1 with both endpoints on @xmath3 . if @xmath13 is reducible with @xmath27 periodic , then since @xmath25 is not an annulus , we may choose a properly embedded geodesic arc @xmath96 in @xmath25 with both endpoints on @xmath3 . in each case , set @xmath97 and let @xmath94 denote the corresponding periodic orbit of @xmath93 . recall when constructing this orbit that @xmath13 and @xmath27 respectively are isometries . again , we assume that @xmath95 is ordered cyclically about @xmath3 , with order compatible with the induced orientation on @xmath3 . notice that @xmath98 for some @xmath99 . also , @xmath100 moreover , if @xmath69 is a properly embedded geodesic arc orthogonal to the boundary with @xmath101 , then @xmath102 . when @xmath13 is pseudo - anosov or reducible with @xmath27 pseudo - anosov , this is immediate since any geodesic @xmath69 intersects the stable and unstable laminations transversally and minimally . when @xmath13 or @xmath27 is periodic , this follows from the choice of @xmath97 . since @xmath103 and all of its images under @xmath13 are geodesics , necessarily @xmath103 will intersect @xmath69 transversally and minimally . now consider the shearing homeomorphism , @xmath104 , that extends @xmath13 across @xmath105 , and , in particular , maps the arc @xmath106 to an arc connecting @xmath107 to @xmath108 . define @xmath109 by @xmath110 , where @xmath104 maps the arc @xmath111 to an arc connecting @xmath112 to @xmath113 . lift @xmath78 to a map @xmath114 , and index the lifts of the points @xmath115 with integers in the natural way . notice that since @xmath116 , @xmath117 where @xmath118 . finally , we compute @xmath51 . let @xmath119 . we may assume that @xmath120 , since we may relabel the indices of the @xmath115 as necessary while preserving their cyclic order . if @xmath121 , then no lift of @xmath122 lies in @xmath123 and it follows that @xmath86 . if @xmath124 , then both @xmath125 and @xmath122 have unique lifts in @xmath126 . label these lifts @xmath127 and @xmath128 respectively . if @xmath127 lies to the right of @xmath128 in @xmath126 , then @xmath129 otherwise , @xmath130 . if @xmath0 is right - veering at @xmath3 , then @xmath131 . [ compute ] if @xmath0 is right - veering at @xmath3 and there exist arcs @xmath50 and @xmath69 with @xmath132 , then @xmath133 . for completeness , we record the left - veering version of these results . [ leftveeringi ] suppose @xmath0 is left - veering at @xmath3 . then either * @xmath85 and @xmath134 or * @xmath87 and @xmath135 . if @xmath0 is left - veering at @xmath3 , @xmath136 . if @xmath0 is neither right nor left - veering at @xmath3 , that is @xmath75 , then for any @xmath50 , @xmath137 . if @xmath138 , then @xmath139 . so we may assume @xmath140 . let @xmath141 where @xmath142 is a right dehn twist about @xmath3 . by proposition [ formula ] , @xmath143 , thus @xmath78 is right - veering and by proposition [ rightveeringi ] , @xmath144 . since @xmath140 , it follows that @xmath145 . next let @xmath146 . then @xmath147 , thus @xmath148 is left - veering and by proposition [ leftveeringi ] , @xmath149 . it follows that @xmath150 . [ stabilize ] let @xmath151 and let @xmath83 be a properly embedded arc in @xmath152 . define @xmath1 to be the union of @xmath152 and a band attached to @xmath153 along neighborhoods of the endpoints of @xmath83 in @xmath152 so that the resulting surface is orientable . this presents @xmath1 as the union of @xmath152 and an annulus that intersect in a regular neighborhood of @xmath83 . let @xmath91 be a positive dehn twist on the annulus . let @xmath154 and @xmath155 be the extensions of @xmath78 and @xmath91 , respectively , to the portion of @xmath1 where they are not already defined , by the identity map . the _ positive stabilization of @xmath78 along @xmath83 _ is the homeomorphism @xmath156 . refer to @xmath83 as the _ plumbing arc _ of the stabilization . [ stabilizedc(h ) ] if @xmath0 is a positive stabilization , and @xmath1 has connected boundary , then @xmath157 . let @xmath158 be the cocore of the stabilizing handle added to @xmath152 as shown in figure [ news ] . consider the arcs @xmath159 and @xmath160 . to show that @xmath161 , it is enough , by proposition [ formula ] , to show that @xmath162 . this in turn will follow , by proposition [ rightveeringi ] , if we show @xmath163 . to see this , we follow the arc @xmath69 from its initial point , keep track of the edges labelled @xmath164 and @xmath165 that it crosses , and argue that if it eventually intersects @xmath166 , the intersection point will not contribute to @xmath167 . the first edge that @xmath69 crosses is @xmath168 . from this point on , it never crosses either @xmath158 or @xmath168 . thus if it intersects @xmath166 , it must do so in the portion @xmath104 of @xmath69 in the square bounded by @xmath164 , and @xmath165 . if after hitting @xmath168 , @xmath69 intersects @xmath104 , the initial segments of @xmath69 and @xmath104 are not boundary parallel since their union misses @xmath169 . in this case , this point of @xmath170 contributes nothing to @xmath167 , subsequent intersections can not contribute either , and we conclude @xmath163 . now consider the case instead that @xmath69 first hits @xmath168 , then @xmath165 , and then @xmath165 again . this forces an intersection of @xmath170 , and for the same reason , we conclude @xmath163 . this leaves the case that @xmath69 intersects @xmath171 , and @xmath169 in order . at this point , either the entire arc @xmath69 misses @xmath104 , in which case there is nothing to prove , or @xmath69 intersects @xmath104 . in the latter case , the union of the initial segments of @xmath69 and @xmath104 are a non - separating curve dual to @xmath169 , and hence not boundary parallel . [ lensexample ] let @xmath172 be an annulus , and let @xmath91 be a positive dehn twist of @xmath172 . let @xmath173 with @xmath174 . next let @xmath175 be an arc that cuts @xmath172 into a disk , and let @xmath176 be the positive stabilization of @xmath16 along @xmath177 . since @xmath1 is a punctured torus , a straightforward homology computation shows that @xmath0 is pseudo - anosov . since @xmath0 is a product of positive dehn twists , it is right - veering , hence @xmath178 . by theorem [ stabilizedc(h ) ] , @xmath179 . since @xmath180 the stable lamination can have only @xmath181 or @xmath182 cusps surrounding @xmath2 . from this it follows that @xmath4 is a multiple of 1/2 . it now follows that @xmath6 . [ prop18 ] let @xmath0 be a homeomorphism of a surface @xmath1 with connected boundary . suppose that @xmath0 is a positive stabilization of a homeomorphism @xmath78 of @xmath183 along a plumbing arc @xmath83 . if @xmath6 , then @xmath184 and @xmath185 . in particular , @xmath186 on each boundary component of @xmath152 . by symmetry , it suffices to show that @xmath184 . let @xmath187 be the properly embedded arcs introduced in the proof of theorem [ stabilizedc(h ) ] and let @xmath188 . consider their geodesic representatives @xmath189 and @xmath190 . notice that @xmath191 if and only if @xmath183 is an annulus and in this case , @xmath6 if and only if @xmath192 for some @xmath174 . so we may restrict attention to the case that @xmath183 is not an annulus and @xmath193 . focus attention on @xmath194 . if this intersection is nonempty , follow the arc @xmath195 from its initial point @xmath196 to its first intersection @xmath197 with @xmath198 . notice that if the geodesic representatives @xmath198 and @xmath199 had been chosen in @xmath183 , rather than in @xmath1 , they would be equal and isotopic to @xmath83 . it follows that @xmath184 if and only if @xmath200 with @xmath201 * \overline{n}[0,t_1]$ ] boundary parallel in @xmath152 . see figure [ twisting>1 ] . we next introduce reference points about the boundary @xmath2 . these are represented in figure [ dynamics ] . as usual , let @xmath13 denote the thurston representative of @xmath0 . consider first the case that @xmath13 is periodic or reducible with @xmath27 periodic . in this case , set @xmath202 and choose distinct points @xmath203 and @xmath204 so that @xmath205 and also the listing @xmath206 is ordered cyclically and compatibly with the induced orientation of @xmath3 , where @xmath207 and @xmath208 are each ( order two ) @xmath13-orbits if @xmath13 is periodic and ( order two ) @xmath27 orbits if @xmath13 is reducible with @xmath27 periodic . consider next the case that @xmath13 is pseudo - anosov or reducible with @xmath27 pseudo - anosov . since @xmath6 , there are an even number , @xmath209 say , of attracting periodic points and an even number , again @xmath209 , of repelling critical points in @xmath3 with respect to @xmath13 or @xmath27 respectively . let @xmath210 be a listing of these , labelled cyclically and ordered compatibly with the induced orientation of @xmath3 . cut @xmath3 open along these periodic points to obtain @xmath211 pairwise disjoint intervals @xmath212 we may further assume that the periodic points are labelled so that @xmath205 . since @xmath6 it follows that @xmath213 . the cyclic ordering in @xmath3 of endpoints of arcs in @xmath1 agrees with the cyclic ordering of their geodesic representatives , provided the arcs have no boundary parallel intersections . using this we see that since the endpoints @xmath214 are cyclically ordered about @xmath3 as @xmath215 , necessarily @xmath216 and @xmath217 . see figure [ dynamics ] . ] since @xmath6 , there is a properly embedded arc @xmath69 so that @xmath218 . in fact , either @xmath219 or we can choose @xmath69 so that @xmath220 . if @xmath219 , then @xmath221 . in the second case , @xmath222 . the goal now is to relate the initial point of a geodesic orthogonal to @xmath2 , such as @xmath223 , to the homeomorphisms @xmath224 and @xmath225 . given a geodesic orthogonal to the boundary , for instance @xmath226 , denote by @xmath227 the initial point of @xmath228 . this notation is not meant to suggest that @xmath229 or @xmath230 are functions that can be applied to arbitrary points of @xmath231 . we now show the locations of @xmath232 or @xmath233 are not possible if @xmath234 . recall that @xmath235 . we begin with @xmath236 . so @xmath237 if @xmath234 we may therefore conclude that @xmath238 and with no boundary parallel intersection . so @xmath239 and with no boundary parallel intersection . but then it follows that @xmath240 , contradicting the conclusion of the preceding paragraph . we may also conclude that @xmath241 , again an impossibility . we conclude that @xmath184 . let @xmath1 be a surface properly embedded in a 3-manifold @xmath242 , and let @xmath3 be an unknot which is properly embedded in @xmath1 . we say that @xmath3 is _ untwisted relative to @xmath1 _ if its regular neighborhood in @xmath1 is an untwisted annulus . this means , if @xmath243 is an embedded disk in @xmath242 bounded by @xmath3 , then @xmath3 is untwisted relative to @xmath1 if and only if @xmath243 can be isotoped relative to its boundary so that @xmath243 and @xmath1 are transverse along @xmath3 . a _ stallings twist _ @xcite is a surgery along any such unknotted untwisted @xmath3 . stallings twists will be applied to @xmath7-cables and exploited in section [ applications ] . as a first step we show in this section that the monodromies associated to the construction are pseudo - anosov and have fractional dehn twist coefficient @xmath8 . consider an open book @xmath244 , with connected binding @xmath245 , such that @xmath0 is isotopic to a pseudo - anosov map @xmath246 and @xmath75 . let @xmath247 denote a regular neighborhood of @xmath245 . abuse notation and let @xmath41 also denote the compact fibre of the surface bundle @xmath248 , and consider two copies of this compact fibre : @xmath249 and @xmath250 . choose a hyperbolic metric with geodesic boundary for @xmath41 and let @xmath251 and @xmath252 inherit this metric . let @xmath253 be a second smaller regular neighborhood of @xmath245 and let @xmath254 denote the @xmath7-cable of @xmath245 , embedded in @xmath255 ( see @xcite , p. 112 ) . letting @xmath256 be an embedding with image @xmath257 , the image under @xmath41 of the @xmath7 torus knot ( shown in figure [ ( 2,1)cable ] ) is what is standardly known as the @xmath7-cable of @xmath245 . -cable of @xmath245 . ] let @xmath258 denote a pair of pants properly embedded in @xmath259 , with boundary components @xmath260 in @xmath261 and a third boundary component lying on @xmath262 , where @xmath263 is a small regular neighborhood of @xmath254 lying in the interior of @xmath247 . see figure [ pants ] . set @xmath264 . extend the hyperbolic metric on @xmath265 over @xmath258 to obtain a hyperbolic metric on @xmath1 with geodesic boundary . notice that @xmath1 can be thought of as the union of two copies of @xmath41 connected by a 1-handle . let @xmath266 denote the geodesic representative orthogonal to @xmath2 of a cocore of this 1-handle . we may choose this hyperbolic metric on @xmath1 so that there is an orientation preserving isometry , @xmath267 say , which interchanges the two components of @xmath268 . lemma [ h_0formula ] gives two slightly different formulas for the monodromy map @xmath1 . depending on the context , it is enough to describe a representative of the free isotopy class , in which case we use @xmath269 and @xmath270 . to compute fractional dehn twist coefficients we need to describe a representative of the relative isotopy class . this is done by adjusting @xmath270 so that it fixes @xmath2 pointwise . to accomplish this , add a collar neighborhood @xmath40 to @xmath1 , and let @xmath271 be the homeomorphism of @xmath272 that extends @xmath270 and is a shear to the right by half a rotation on @xmath40 . for simplicity of notation , we incorporate the collar into @xmath1 and think of @xmath271 as a homeomorphism of @xmath1 . [ h_0formula ] the surface @xmath1 is a fibre of @xmath254 in the fibred 3-manifold @xmath273 . representatives of the free and relative isotopy classes of the monodromy classes are @xmath269 and @xmath274 , respectively , which are given by @xmath275 @xmath276 @xmath277 inherits a product structure from the product structure on @xmath278 . so it suffices to understand a compatible product structure on @xmath279 . a meridional flow line for @xmath280 is shown on the left of figure [ productdisk ] . this splits into two subarcs that are flow lines for the @xmath7-cable of @xmath245 . for each subarc , a parallel flowline is shown as part of the boundary of a disk . the disks in turn are unions of flowlines in @xmath279 . figure [ monodromy ] shows @xmath258 flattened out , and it records the positions of two arcs under the monodromy map which is the identity on the boundary . the monodromy , up to isotopies fixing @xmath2 , is determined by the action on these arcs . direct computation shows that @xmath281 agrees with the monodromy map on these arcs . are shown . ] under the flow . ] choose a simple closed curve @xmath3 in @xmath1 with two properties . first , it intersects @xmath265 in two essential arcs , @xmath282 and @xmath283 . second , @xmath284 is nonseparating in @xmath285 for each @xmath286 . now set @xmath287 . [ h ] let @xmath288 be the map obtained from the above construction applied to a pseudo - anosov homeomorphism @xmath0 of @xmath41 with @xmath75 . the thurston representative @xmath289 of @xmath288 is pseudo - anosov . to simplify the exposition , it helps to pass to a closed hyperbolic surface @xmath290 naturally associated to @xmath1 . topologically , @xmath290 is defined by setting @xmath291 the surface obtained from @xmath1 by capping off @xmath2 with a disk . notice that @xmath292 where @xmath293 is an annulus . the homeomorphism @xmath269 extends to a homeomorphism @xmath294 given by @xmath295 where @xmath296 is a homeomorphism , uniquely determined up to isotopy , extending @xmath267 . now consider the closed hyperbolic surface @xmath297 , obtained by gluing the hyperbolic surfaces @xmath251 and @xmath252 by the isometry @xmath298 given by @xmath299 . choose a diffeomorphism from @xmath290 to @xmath300 , and use this diffeomorphism to pull back the hyperbolic metric on @xmath297 to a hyperbolic metric on @xmath290 . let @xmath172 denote the geodesic simple closed curve which is the pull back to @xmath290 of @xmath301 . notice that @xmath302 set @xmath303 we will use the following characterization of periodic and reducible maps ( @xcite ) . [ first ] a homeomorphism @xmath0 of a surface @xmath1 has a periodic or reducible representative if and only if there exists a possibly immersed multi - curve @xmath304 where each @xmath305 is an essential , embedded , curve in @xmath1 , and , up to isotopy , @xmath31 is fixed by @xmath0 . in the periodic case , each @xmath305 may be assumed to be non - separating , and in the reducible case , @xmath31 may be chosen to be embedded . if the thurston representative @xmath306 of @xmath307 is pseudo - anosov , then the thurston representative @xmath289 of @xmath288 is also pseudo - anosov . by contradiction , suppose that @xmath288 is either periodic or reducible , and let @xmath31 be the multi - curve guaranteed by lemma [ first ] . since the curves @xmath31 , @xmath308 , and @xmath309 and the image of the isotopy between @xmath308 and @xmath309 all lie in @xmath1 and therefore in @xmath290 , it follows that @xmath310 is isotopic to @xmath31 . furthermore , no component of @xmath31 is boundary parallel in @xmath1 . thus applying lemma [ first ] to @xmath311 shows that @xmath307 is either periodic or reducible . theorem [ h ] therefore follows from the following theorem . [ hath ] the thurston representative @xmath312 of @xmath313 is pseudo - anosov . suppose instead that @xmath312 is periodic or reducible . choose @xmath31 as guaranteed by lemma [ first ] so that it is either an embedded curve or is a union of non - separating curves . necessarily @xmath314 is isotopic to @xmath315 . we next establish notation for a sequence of lemmas that will be used to derive a contradiction in corollary [ last ] . let @xmath14 and @xmath15 denote the stable and unstable geodesic laminations of @xmath246 . let @xmath316 , @xmath317 , denote the leaves of @xmath14 bounding the complementary region of @xmath14 containing @xmath43 , ordered cyclically about @xmath43 . similarly , let @xmath318 denote the leaves of @xmath15 bounding the complementary region of @xmath15 containing @xmath43 , ordered cyclically about @xmath43 . the unstable prongs , i.e. , the geodesics perpendicular to @xmath43 found between each consecutive pair @xmath319 , canonically cut each @xmath320 into two open intervals @xmath321 and @xmath322 . similarly , the stable prongs canonically cut each @xmath323 into two open intervals @xmath324 and @xmath325 . since @xmath75 , we may assume that both @xmath246 and @xmath0 fix the leaves @xmath326 , and @xmath327 , for @xmath328 . @xmath320 , the leaves @xmath323 , the leaves @xmath329 , and the leaves @xmath327 , @xmath328 . to simplify the exposition ( and the notation ) , we now assume that @xmath330 , @xmath331 , @xmath172 , @xmath3 , and @xmath31 are each geodesic in @xmath290 . let @xmath332 denote the geodesic representative of @xmath314 , and choose an isotopy representative of @xmath333 so that @xmath309 is a geodesic . let @xmath334 denote the closed complementary region of @xmath335 which contains @xmath172 . since @xmath31 meets @xmath335 efficiently , so does @xmath314 . it follows that we can isotope @xmath314 to be geodesic while preserving this efficient intersection with @xmath335 . the following therefore holds . [ path ] if a component of @xmath336 is a path connecting @xmath337 and @xmath338 for some @xmath339 , @xmath340 , then the corresponding component of @xmath341 is also a path connecting @xmath342 and @xmath343 . [ comp ] let @xmath344 be a geodesic simple closed curve in @xmath290 , and for each integer @xmath345 , let @xmath346 denote the geodesic representative of @xmath347 . if a component of @xmath348 is a path connecting @xmath337 and @xmath338 for some @xmath339 , @xmath340 , then the corresponding component of @xmath349 is also a path connecting @xmath342 and @xmath343 . the leaves @xmath350 which intersect @xmath344 are uniquely determined by the constraint that the intersection @xmath351 is efficient . this efficiency of intersection is unaffected by composition with @xmath352 . symmetric statements hold relative to the unstable lamination @xmath353 . both @xmath354 and @xmath355 . if @xmath356 , then @xmath31 is fixed by @xmath333 and hence , up to isotopy , also by @xmath357 . since @xmath246 is pseudo - anosov , @xmath357 fixes only @xmath172 , up to isotopy , and it follows that @xmath358 . it follows that if @xmath359 , then necessarily @xmath360 . but we know @xmath361 and hence @xmath362 so @xmath363 ( and @xmath360 ) . we now use @xmath3 to get a measure of the amount of twisting of @xmath314 and @xmath309 respectively about @xmath172 . we do this as follows . first consider the components of @xmath364 . call such a complementary region a _ triangular disk region _ if it is a disk with piecewise geodesic boundary consisting of exactly three geodesic subarcs , one contained in each of @xmath31 , @xmath172 and @xmath3 respectively . choose a regular neighborhood @xmath365 of @xmath172 so that 1 . all such triangular disk regions are contained in the interior of @xmath365 , 2 . all intersections of @xmath366 with @xmath3 , @xmath31 , @xmath314 , and @xmath309 are transverse , 3 . @xmath367 consists of two essential embedded arcs : @xmath368 and @xmath369 , 4 . @xmath370 consists of @xmath83 essential embedded arcs : @xmath371 , for some @xmath372 , 5 . @xmath373 consists of @xmath83 essential embedded arcs : @xmath374 , 6 . @xmath375 consists of @xmath83 essential embedded arcs : @xmath376 , and 7 . for @xmath377 and @xmath378 , each of the arcs @xmath379 , @xmath380 , @xmath381 , and @xmath382 has nonempty ( necessarily minimal ) intersection with each of @xmath383 , @xmath384 , @xmath385 , and @xmath386 . in ( 4)(6 ) we are using the fact that @xmath387 , thus @xmath388 choose the indices @xmath389 , @xmath378 , so that @xmath381 is the component corresponding to @xmath390 . orient the arcs @xmath391 , @xmath392 , @xmath393 , and @xmath394 , @xmath395 , so that they run from @xmath396 to @xmath397 . [ small ] for all @xmath398 and @xmath389 , @xmath399 and @xmath400 , @xmath401 . since @xmath402 and @xmath403 each component @xmath394 can meet each of @xmath368 and @xmath369 at most once . for some @xmath404 and @xmath405 , @xmath406 and @xmath407 , @xmath408 . points of intersection @xmath409 lie either in @xmath365 or outside @xmath365 . the condition that all triangular disk regions lie in @xmath365 guarantees that any point of intersection @xmath409 lying outside @xmath365 results in increased geometric intersection of @xmath309 with @xmath172 . and any point of intersection @xmath409 lying inside @xmath365 and with positive intersection number @xmath410 results in increased geometric intersection of @xmath309 with @xmath172 . since @xmath409 is nonempty , it follows that if all intersection numbers @xmath410 are nonnegative , then necessarily @xmath411 . this is impossible . rechoose indices as necessary so that @xmath412 and @xmath413 is the component of @xmath332 corresponding to @xmath414 . [ last ] the geodesic representative of @xmath415 , @xmath332 , and @xmath309 are not equal , and therefore @xmath416 is not isotopic to @xmath31 . focus on @xmath417 and @xmath418 , the latter arc isotoped rel endpoints to be geodesic and oriented so that it runs from @xmath396 to @xmath397 . the arc @xmath419 is a path from @xmath350 to @xmath420 for some @xmath421 . since @xmath357 switches @xmath422 and @xmath423 , and since @xmath75 , it follows from lemma [ comp ] that both @xmath424 and @xmath425 are paths from @xmath426 to @xmath427 . there are two possibilities : either @xmath428 or @xmath429 . consider first the case that @xmath428 . in this case , @xmath430 and hence @xmath431 it follows from lemma [ small ] that @xmath432 for all @xmath433 . next consider the case that @xmath429 . recall that @xmath413 is the portion of the geodesic , corresponding to @xmath417 , that is formed by applying @xmath434 to @xmath31 . the action of @xmath434 on @xmath334 is @xmath435 , thus @xmath418 corresponds to acting on @xmath417 by the map @xmath436 used in the definition of @xmath434 . if @xmath437 , @xmath436 would force @xmath428 . it follows that @xmath438 . moreover , @xmath439 implies @xmath440 . if @xmath441 , the argument of the first case applies . so we may assume @xmath442 . this is possible only if @xmath417 , @xmath443 and @xmath368 are all paths connecting @xmath350 to @xmath427 , pairwise isotopic through paths connecting @xmath350 to @xmath427 . in other words , up to composition by @xmath352 for some integer @xmath345 , they must lie as shown in figure [ intersect].a . the arc @xmath413 is added in figure [ intersect].b . the next step is to show there is a geodesic arc , @xmath69 , properly embedded and essential in @xmath334 , such that @xmath69 is disjoint from @xmath444 . when @xmath445 , choose @xmath69 to be a geodesic path from @xmath446 to @xmath447 for some @xmath448 . when @xmath202 , we argue similarly but work instead with the leaves @xmath449 , @xmath450 . the arcs @xmath417 , @xmath443 and @xmath368 are all paths connecting @xmath451 to @xmath452 , pairwise isotopic through paths connecting @xmath451 to @xmath452 . we then choose @xmath69 to be a geodesic path from @xmath453 to @xmath454 , where @xmath455 . thus in either case , @xmath69 satisfies @xmath456 , whereas @xmath457 for all @xmath458 . so again we may conclude @xmath432 for all @xmath433 . theorem [ hath ] now follows from corollary [ last ] [ h_0 ] with @xmath459 as defined in lemma [ h_0formula ] , and @xmath460 , @xmath461 . let @xmath462 be the arc fixed by the involution @xmath463 . to estimate @xmath464 , we first compute @xmath465 . loosely speaking , it is enough to compute just @xmath271 and @xmath142 on the initial portion of @xmath50 . more precisely , @xmath466 where @xmath467 is an arc ending transversely on @xmath422 , and @xmath468 is an arc which starts on @xmath422 and has an essential first return to @xmath422 . figure [ c(h ) ] shows @xmath467 which is computed by applying @xmath271 and @xmath142 to the initial portion of @xmath50 . only the first intersection of @xmath467 and @xmath50 can contribute to @xmath465 , but this does not yield a boundary parallel union of initial segments . it follows that @xmath469 , and thus by proposition [ rightveeringi ] , @xmath470 . the next step is to estimate @xmath471 by computing @xmath472 and comparing it to @xmath473 . towards this end , write @xmath474 where @xmath417 is essentially truncated at @xmath422 as above . to compute @xmath475 where @xmath476 is essentially truncated at @xmath423 by applying the remaining portion of @xmath477 to @xmath467 . the single intersection of @xmath417 and @xmath467 is enough to conclude that @xmath478 . thus by lemma [ rightveeringi ] we have @xmath479 and it follows by lemma [ formula ] that @xmath480 . we show in the next section , corollary [ hins3 ] , that in @xmath9 , @xmath464 can only be @xmath8 . the first portion of this section is devoted to proving the result , theorem [ onlyvalues ] , that the only values fractional dehn twist coefficients take on in @xmath9 are @xmath481 and @xmath482 where @xmath169 is an integer satisfying @xmath483 . this leads to the statement of conjecture [ main1 ] , or in light of theorem [ onlyvalues ] , the equivalent statement , conjecture [ main2 ] , either of which imply that a fibred knot in @xmath9 with fractional dehn twist coefficient @xmath8 can never be destabilized . this consequence , in the most important case when the monodromy is pseudo - anosov , was first stated , without proof , by gabai in @xcite . part of our interest in this result is that it appears to depend on deep facts about @xmath9 , most likely gabai s thin position strategy , in an essential way . indeed , very simple examples , such as the examples of proposition [ lenspace ] , show the result fails even in lens spaces . we begin by considering more closely the case that the thurston representative @xmath13 of the monodromy @xmath0 is either periodic or reducible with @xmath27 periodic . @xcite let @xmath484 be an open book decomposition which has connected binding @xmath245 . if the thurston representative of @xmath0 is periodic , then @xmath245 is the unknot or a @xmath485-torus knot for some relatively prime integers @xmath486 and @xmath47 satisfying @xmath487 . let @xmath484 be an open book decomposition which has connected binding @xmath245 . if the thurston representative , @xmath13 , of @xmath0 is reducible with @xmath27 periodic , then @xmath245 is a @xmath485-cable knot , for some relatively prime integers @xmath486 and @xmath47 satisfying @xmath487 , and @xmath488 . suppose @xmath13 is reducible with @xmath489 periodic . denote the boundary components of @xmath25 by @xmath490 . let @xmath491 denote the torus in @xmath9 that is the union of flow lines of @xmath27 intersecting @xmath284 . since @xmath27 may permute the @xmath284 s , the @xmath491 s are not necessarily distinct . let @xmath492 denote a solid torus in @xmath9 with @xmath493 . each @xmath491 is essential in @xmath494 . to see this , notice that an infinite cyclic cover of @xmath494 can be created by gluing copies of @xmath494 split along @xmath1 . the inverse image of @xmath491 in this cover is an infinite cylinder which is incompressible since @xmath284 is essential in @xmath1 . it follows that every @xmath492 contains @xmath245 . by connectivity of @xmath25 there can be only one such solid torus , thus @xmath495 and @xmath496 . consider a closed orbit @xmath31 of @xmath27 that lies on the boundary of a regular neighborhood @xmath497 of @xmath245 . if this is a meridian , then @xmath75 . otherwise the orbits of @xmath27 , together with @xmath245 , define a seifert fibre space structure on @xmath498 . a seifert fibre space structure on @xmath498 can have at most one singular fibre . if @xmath245 is a singular fibre , the leaf space of @xmath499 is an annulus . moreover , this annulus is covered by @xmath25 . it follows that @xmath25 is itself an annulus , thereby contradicting the choice of @xmath25 as a reducing surface for @xmath13 . it follows that @xmath245 is a non - singular fibre and hence a @xmath485-cable of the core of @xmath498 . let @xmath500 denote the knot which is the core of @xmath498 , and let @xmath501 denote the complement of @xmath498 in @xmath9 . coordinates @xmath485 are chosen so that a @xmath502 curve is a longitude for @xmath500 , and a @xmath503 curve is a meridian for @xmath500 . let @xmath504 denote the subsurface @xmath505 of @xmath1 . notice that the orientation on @xmath25 induces a @xmath27-invariant orientation on the intersection @xmath506 . equivalently , the oriented complementary surface @xmath504 intersects @xmath507 in @xmath317 parallel consistently oriented simple closed curves : @xmath508 . necessarily these curves are @xmath502 curves and @xmath504 has @xmath169 connected components . moreover , the complement of @xmath504 in @xmath501 is an @xmath509-bundle . it follows that @xmath500 is fibered and @xmath504 consists of @xmath169 copies of a fibre @xmath510 of @xmath500 . it follows also that @xmath511 . moreover , since @xmath25 can not be an annulus , @xmath487 . to compute @xmath4 , consider a periodic orbit @xmath31 near @xmath3 , and express it in ( longitude , meridian ) coordinates with respect to @xmath3 . the orbits of @xmath27 define the seifert fibration on @xmath498 , and since @xmath3 is a non - singular fibre , @xmath31 is also a @xmath485 curve with respect to @xmath500 . it follows that @xmath31 intersects a meridian of @xmath3 once . to compute the number of times @xmath31 intersects @xmath1 , it is enough to compute the intersection between a @xmath485 curve and the @xmath486 @xmath502 curves that are @xmath512 . it follows that @xmath31 is a @xmath513 curve with respect to @xmath3 , thus @xmath488 . [ pqgames ] let @xmath484 be an open book decomposition which has connected binding @xmath245 . suppose that the thurston representative , @xmath13 , of @xmath0 is either periodic or reducible with @xmath27 periodic . then either @xmath245 is the unknot and @xmath75 or @xmath514 , where @xmath483 is the slope of the cabling annulus . since @xmath245 is either a @xmath485-torus knot or a @xmath485-cable knot , it follows that either @xmath245 is the unknot , or the cabling annulus is incompressible and boundary incompressible in @xmath515 . since the cabling annulus has slope @xmath516 , it follows that either @xmath245 is the unknot and @xmath75 or @xmath245 is not the unknot and @xmath517 . [ gatwist]@xcite let @xmath484 be an open book decomposition which has connected binding @xmath245 . suppose that @xmath13 is either pseudo - anosov or reducible with @xmath27 pseudo - anosov . then either @xmath75 or @xmath518 , where @xmath519 . theorem 8.8 of @xcite states that the degeneracy , @xmath520 , of the complement of @xmath245 has one of the forms 1 . @xmath521 , with @xmath522 2 . @xmath523 , with @xmath524 ( see @xcite for a second proof of the lower bound in ( 2 ) . ) as noted in section [ basics ] , when @xmath525 , @xmath4 is the reciprocal of the slope of @xmath520 . it follows that @xmath4 is either @xmath481 or @xmath526 for some integer @xmath270 , @xmath524 . combining corollary [ pqgames ] and theorem [ gatwist ] proves [ onlyvalues ] if @xmath11 is an open book decomposition of @xmath9 with connected binding , then @xmath4 equals @xmath481 or @xmath482 for some integer @xmath169 , @xmath483 . theorem [ gatwist ] and proposition [ h_0 ] immediately imply [ hins3 ] let @xmath459 be the monodromy of a ( 2,1)-cable of a knot in @xmath9 with pseudo - anosov monodromy and fractional dehn twist coefficient @xmath481 . then @xmath460 , as constructed above , satisfies @xmath527 . among all the possible values of the fractional dehn twist coefficient that arise in @xmath9 , it is 1/2 that plays the most mysterious role knot theory . the following was first stated by gabai in @xcite along with a possible proof strategy . [ main1 ] let @xmath484 be an open book decomposition which is stabilized and has connected binding @xmath254 . then @xmath528 . by theorem [ onlyvalues ] , this conjecture can also be stated as follows . [ main2 ] let @xmath244 be an open book decomposition which is stabilized and has connected binding . then either @xmath75 or @xmath514 , where @xmath529 . an appealing aspect of these conjectures is that very simple examples show that they are false outside of @xmath9 . specifically , example [ lensexample ] in section [ basics ] proves [ lenspace ] let @xmath530 and @xmath531 . then @xmath532 has an open book decomposition with connected binding which is stabilized and satisfies @xmath6 . in this section , we give an application of our results in contact topology by associating to each pair @xmath11 a contact structure @xmath10 via the giroux correspondence @xcite . this correspondence is many to one , and giroux shows that the failure of injectivity is generated by positive stabilization ( see definition [ stabilize ] ) . a natural goal is to try to determine properties of @xmath10 from a single pair @xmath11 . some results in this direction are as follows @xcite . if @xmath74 then @xmath10 is overtwisted . if @xmath533 then @xmath10 is tight . contrast this with the uncheckable theorem that @xmath10 is tight if and only if for _ every _ @xmath11 determining @xmath10 , @xmath534 . this leads to the conjecture by honda , kazez , and mati @xcite that if a single representative is not destabilizable and has @xmath534 ( or equivalently is right - veering ) then @xmath10 is tight . lekili @xcite produced counterexamples to this conjecture with @xmath1 a 4 times punctured sphere . lisca @xcite constructed an infinite number of counterexamples for the same surface . ito and kawamuro @xcite have produced an even larger set of counterexamples on the 4 times punctured sphere . this should be contrasted with the work of colin and honda @xcite in which they show that if an open book has connected boundary , pseudo - anosov monodromy , and fractional dehn twist coefficient @xmath535 , then @xmath536 implies the associated contact structure is not only tight , but universally tight , and the universal cover is @xmath537 . our examples show that it is often very easy to recognize an overtwisted contact structure using essentially no technology . it is enough , by proposition [ ot ] , to find a non - separating , untwisted , unknotted curve contained in @xmath1 . let @xmath538 be an open book decomposition and let @xmath3 be a simple closed curve embedded in the interior of @xmath1 which bounds an embedded disk in @xmath242 . let @xmath247 be a regular neighborhood of @xmath3 . note that although there is a unique canonical meridian @xmath539 , there are two natural choices of longitude in this setting : if the orientation for @xmath540 is reversed , then @xmath539 and hence @xmath542 also have their orientations reversed . it follows that the sign of @xmath545 does not depend on the choice of orientation of @xmath243 . in particular , from the point of view of @xmath243 , if the sign of @xmath545 is negative , an annular neighborhood of @xmath3 in @xmath1 twists to the left , when traveling around @xmath547 . thus @xmath545 serves as a topological stand - in for the thurston - bennequin invariant . let @xmath551 denote a regular neighborhood of @xmath243 in @xmath242 . note that @xmath552 with summing sphere @xmath553 . so it suffices to consider the case that @xmath525 . if @xmath554 , simply cut and twist around @xmath243 to realize the homeomorphism between @xmath242 and @xmath549 . computing @xmath555 shows the converse is also true . [ tauqofh ] if @xmath556 and @xmath557 , then @xmath549 has an open book decomposition given by @xmath558 . moreover , the analogues of @xmath559 and @xmath1 exist in @xmath549 as embedded objects , and in @xmath549 it remains the case that @xmath556 . let @xmath172 be an annular neighborhood of @xmath3 in @xmath1 and let @xmath560 $ ] . let @xmath84 be non - separating embedded arc in @xmath172 and let @xmath561 $ ] be a rectangular compressing disk for @xmath247 . to construct the desired homeomorphism from @xmath549 to the manifold built from @xmath558 , start with the identity map from @xmath562 . next extend the homeomorphism to @xmath563 as follows . let @xmath564 . at the quotient level , we have @xmath565 and @xmath566 , thus we define @xmath567 if @xmath568 . next we compute the preimage of @xmath569 to compute the required filling of @xmath261 in the domain . to be explicit about orientations , pick an orientation for @xmath3 . choose the orientation on @xmath542 so that it is isotopic to @xmath3 in @xmath570 . the orientation on @xmath542 determines the orientation on @xmath539 . see figure [ surgery ] . the only portion of @xmath569 not mapped by the identity is the arc @xmath571 . from the formula above , we see the preimage of @xmath571 is an arc @xmath572 in @xmath573 such that @xmath574 , or equivalently , @xmath575 . it follows that the preimage of @xmath569 in @xmath242 is a @xmath576 curve , thus the homeomorphism on @xmath577 extends to a homeomorphism defined on all of @xmath549 . since @xmath3 is non - separating , it is non - isolating . thus by honda s legendrian realization principle @xcite , we may assume @xmath3 is legendrian . the thurston - bennequin invariant of @xmath580 is equal to @xmath545 , thus if @xmath556 , @xmath243 is an overtwisted disk . in the remaining cases , @xmath3 can be isotoped to decrease the thurston - bennequin invariant and thereby produce an overtwisted disk . [ qsurgery ] if @xmath3 bounds an embedded disk @xmath243 in @xmath578 , @xmath581 , and @xmath3 is non - separating in @xmath1 , then the contact structure determined by @xmath582 is overtwisted for any integer @xmath47 . in @xmath242 we have @xmath584 . by proposition [ tauqofh ] we have in @xmath549 , @xmath585 and consequently , @xmath586 . when @xmath556 in @xmath242 , @xmath587 . this formula is satisfied on just the boundary of the regular neighborhood of @xmath3 , thus it holds in @xmath549 as well . [ cable_8_20 ] let @xmath41 be the seifert surface for the knot @xmath82 as described in example [ example_8_20 ] . using the notation of section [ pa ] , form the @xmath7-cable of @xmath41 , and denote its seifert surface by @xmath264 . this gives an open book decomposition @xmath588 of @xmath9 . to produce the curve @xmath3 used in the definition of @xmath589 consider the curves @xmath172 and @xmath551 that live on @xmath41 as shown in figure [ 8_20 ] . pick an arc @xmath590 that runs from a point of @xmath591 to the point where the twisted band is added in the formation of the @xmath7-cable . choose an arc @xmath592 similarly corresponding to @xmath593 . in addition the arcs should be chosen so that @xmath594 , and the projections of @xmath595 to @xmath41 have no interior intersections . define @xmath3 to be the band sum , in @xmath1 , of @xmath591 and @xmath593 along an arc consisting of @xmath596 , and an arc that runs across the twisted band . since @xmath172 and @xmath551 are non - separating in @xmath41 , @xmath3 does not separate when restricted to either @xmath597 or @xmath252 and theorem [ h ] applies . combining theorem [ h ] , proposition [ rv ] , corollary [ hins3 ] , and proposition [ ot ] shows that @xmath288 is pseudo - anosov , right - veering with fractional dehn twist coefficient 1/2 , and the associated contact structure is overtwisted . [ counterexamples ] there exists an infinite collection of fibred knots in @xmath9 such that the monodromy is pseudo - anosov , right - veering with fractional dehn twist coefficient 1/2 , and the associated contact structure is overtwisted . it is enough , by the construction of example [ cable_8_20 ] , to produce knots with pseudo - anosov monodromy whose fibres contain two disjoint hopf bands of opposite signs . one such family of examples are the nontorus fibred 2-bridge knots of genus 2 and higher . the fibres of such knots correspond to words in l and r that use both letters and have even length @xcite . the letters correspond to hopf bands plumbed together in a vertical stack . as long as the surface has genus at least 2 , the word will have length at least 4 . the desired hopf bands correspond to a non - adjacent pair of letters l and r which necessarily exist . if conjecture [ main1 ] is true , then none of the examples of theorem [ counterexamples ] and example [ cable_8_20 ] can be destabilized , and thus they would all provide additional counterexamples to the conjecture of @xcite . e. giroux , _ gomtrie de contact : de la dimension trois vers les dimensions suprieures _ , proceedings of the international congress of mathematicians , vol . ii ( beijing , 2002 ) , 405414 , higher ed . press , beijing , 2002 . j. stallings , _ constructions of fibred knots and links _ , algebraic and geometric topology ( proc . pure math . , stanford univ . , stanford , calif . , 1976 ) , part 2 , pp . 5560 , proc . pure math . , xxxii , amer . math . soc . , providence , r.i . , 1978 .

we present a new library of integrated spectra of 40 galactic globular clusters , obtained with the blanco 4-m telescope and the r - c spectrograph at the _ cerro tololo interamerican observatory_. the spectra cover the range 3350 6430 with 3.1 ( fwhm ) resolution . the spectroscopic observations and data reduction were designed to integrate the full projected area within the cluster core radii in order to properly sample the light from stars in all relevant evolutionary stages . the s / n values of the flux - calibrated spectra range from 50 to 240/ at 4000 and from 125 to 500/ at 5000 . the selected targets span a wide range of cluster parameters , including metallicity , horizontal - branch morphology , galactic coordinates , galactocentric distance , and concentration . the total sample is thus fairly representative of the entire galactic globular cluster population and should be valuable for comparison with similar integrated spectra of unresolved stellar populations in remote systems . for most of the library clusters , our spectra can be coupled with deep color - magnitude diagrams and reliable metal abundances from the literature to enable the calibration of stellar population synthesis models . in this paper we present a detailed account of the observations and data reduction . the spectral library is publicly available in electronic format from the national optical astronomical observatory website .

astro-ph0504313

the determination of the stellar content of unresolved systems from observations of their integrated light is a fundamental challenge to extragalactic astronomers . the problem has been traditionally approached by comparing the integrated properties ( spectral energy distributions , line strengths , colors ) of remote systems with those of better - understood local templates ( e.g. faber 1973 , burstein 1984 , rose 1985 ) . more recently , stellar population synthesis ( sps ) models combined with our knowledge of the stellar evolution of stars and of the behaviour of their emitted light as a function of fundamental stellar parameters have been constructed to predict integrated properties of single and composite stellar populations in integrated light ( e.g. le borgne 2004 , bruzual & charlot 2003 , thomas , maraston & bender 2003 , vazdekis 1999 , worthey 1994 ) . in this context , the galactic globular cluster ( gc ) system plays a crucial role . it consists of a set of increasingly well studied stellar populations which are close enough that their constituent stars can be resolved and studied individually and meaningful color - magnitude diagrams can be constructed , and yet their compact structures enable observations of their integrated light relatively easily . integrated observations of gcs can be used to construct calibrations of integrated observables as a function of the parameters that characterize stellar populations ( mainly age , metallicity , and abundance pattern ) . such calibrations provide useful , fairly model - independent tools for the conversion of integrated observables of distant systems into fundamental stellar population parameters . moreover , comparisons of sps models with integrated observations of galactic gcs has become commonplace ( e.g. proctor , forbes & beasley 2004 , schiavon 2004a , b , maraston 2003 , puzia 2002 , schiavon 2002a , b , beasley , hoyle & sharples 2002 , vazdekis 2001 , gibson 1999 , schiavon & barbuy 1999 , bruzual 1997 , rose 1994 , and references therein ) . such fundamental comparisons aim primarily at verifying the validity of the models , whose predictions must match the integrated properties of gcs for the known set of input stellar population parameters . reliable databases of integrated properties of galactic gcs are therefore precious resources and numerous efforts have been invested towards their collection in the past . notable among these are the works of puzia ( 2002 ) , cohen , blakeslee & ryzhov ( 1998 ) , covino , galletti & pasinetti ( 1995 ) , bica & alloin ( 1986 ) , burstein ( 1984 ) , and zinn & west ( 1984 ) , where integrated spectra of moderate to large - sized samples of galactic gcs were collected . in spite of these and other previous works , there remains considerable room for improvement to the available spectroscopic databases which are deficient in one or more of such crucial features as size , a representative range of metallicities and horizontal branch ( hb ) morphologies , signal - to - noise ratio ( s / n ) , resolution , and homogeneity . these features are all essential both for a robust calibration of stellar population synthesis models and for the direct comparison with observations of distant systems . this work aims at providing this much needed improvement . we present new , high - s / n , moderately high - resolution , integrated spectra of 40 galactic globular clusters obtained with the same instrumental setup and over a single observing run . this is the largest collection of optical integrated spectra of galactic gcs to date with such high s / n and resolution , and should serve as a prime database for comparison with observations of remote stellar systems and the calibration of sps models . this paper focuses on the detailed description of our observations and data reduction procedures , and the presentation of the final data products . a first exploration of this spectral library , focussed on the effects of hb morphology on different balmer lines , was presented in schiavon ( 2004b ) . the paper is organized as follows : we describe the sample selection and observations , in sections [ sec : sselection ] and [ sec : obsdata ] , respectively , and the reductions are discussed in section [ sec : reductions ] . the final spectra are presented in section [ sec : spectra ] and section [ sec : notes ] contains notes on individual clusters . our results are summarized in section [ sec : summary ] . our choice of gc targets was chiefly aimed at producing a spectral library that is representative of the galactic gc system . therefore , gcs spanning a wide range of metallicities , hb morphologies , and galactocentric distances were chosen . another important consideration is that the spectral library will become more useful if gc characteristic parameters such as age , metallicity , and hb morphology are known accurately . therefore , we gave priority to gcs for which [ fe / h ] values , based on abundance analyses of high resolution spectra of member stars ( e.g. carretta & gratton 1997 , kraft & ivans 2003 ) , are available . for the same reason , we also preferred gcs for which high - quality color - magnitude diagrams , preferrably reaching below the turn - off , are available from the literature ( e.g. piotto 2002 ) . the combination of such data with our spectral library optimizes the construction of reliable calibrations of integrated gc properties with fundamental determinations of basic parameters like age , metallicity and hb morphology . the distribution of our gc sample in galactic coordinates is shown in figure [ fig : spdist ] . the main characteristics of our globular cluster sample are summarized in table 1 . we list in columns ( 1 ) and ( 2 ) the cluster ids , in columns ( 3 ) and ( 4 ) the positions in galactic coordinates , and in column ( 5 ) galactocentric distances in kpc . in column ( 6 ) we list core radii , @xmath2 , in arcminutes , in column ( 7 ) the central concentration parameter @xmath3 , where @xmath4 is the tidal radius , and in column ( 8) reddening values , @xmath5 , are listed . we list in column ( 9 ) [ fe / h ] values from high - resolution abundance analyses , when available , and those determined by schiavon ( 2005 , in preparation ) otherwise . in column ( 10 ) the lee , demarque & zinn ( 1994 ) parameter describing the morphology of the hb is listed . finally , in column ( 11 ) we provide a non - exhaustive list of recent references where deep color - magnitude diagrams for the sample clusters can be found . the data in columns ( 3 ) , ( 4 ) , ( 5 ) , and ( 8) were taken primarily from piotto ( 2002 ) , or otherwise from the compilations by djorgovski & meylan ( 1993 ) , djorgovski ( 1993 ) , and peterson ( 1993 ) . the [ fe / h ] values from schiavon ( 2005 , in preparation ) are derived from a new metallicity scale for galactic gcs , based on measurements of iron absorption lines and state - of - the - art high - resolution abundance analyses of cluster members . the hb morphologies listed in column ( 10 ) were taken from borkova & marsakov ( 2000 ) . for gcs whose hb morphologies are not given by borkova & marsakov , we provide qualitative estimates from inspection of the color - magnitude diagrams of piotto ( 2002 ) . core radii and concentration parameters were taken from harris ( 1996 ) ( http://physwww.mcmaster.ca/@xmath6harris/mwgc.dat ) . all observations were obtained with the ritchey - chretien spectrograph mounted on the 4-m blanco telescope at the _ cerro tololo inter - american observatory _ ( ctio ) on the nights of 24 - 27 april 2003 . a loral 3k x 1k ccd detector was used with an arcon controller . the grating kpgl#1 , with 632 grooves / mm , was used in first order with the blue collimator , yielding a dispersion of 1.0 / pix over the wavelength range 33606430 . a spectral resolution of 3.1 fwhm was achieved in the central @xmath01000 of the dispersion curve , deteriorating to @xmath03.6 fwhm at the blue and red ends of the spectrum . a cubic polynomial fit to the fwhm of arc lines as a function of wavelength produced the following curve : @xmath7 where @xmath8 is the wavelength . both fwhm and @xmath8 are given in . the spectra were binned by a factor of two along the slit , yielding a spatial scale of 1/pixel . the gain of the ccd was set to 1.0 e@xmath9/adu , and the read noise was determined to be 7.3 e@xmath10 . for all observations , except those of flux standards , the slit width was set at 225 @xmath11 or 15 . however , as we discuss further in [ sec : anomalies ] , there were problems with the slit width in the second half of the last night of observations , due to a failure in the slit motor and its subsequent reinitialization . the change in slit width also produced a change in spectral resolution for those observations . for the first two nights , the @xmath0 55-long slit was oriented in the north - south direction , and was along the east - west direction for the final two nights . at the beginning of each night a sequence of 25 bias frames and 25 dome flat field frames were acquired . the bias level fluctuated throughout the observing run , but could always be subtracted adequately ( see [ sec : anomalies ] ) . twilight exposures were recorded on the first two nights to provide illumination correction along the slit . the combination of a set of 5 arc lamp exposures taken at the beginning of each night provided a high - s / n master arc . finally , on the last night , we observed the g star hd76151 at several positions along the slit to aid in mapping optical distortions in the spectrograph . specifically , we observed hd76151 at the center of the slit , and repeated the observation with the slit offset by 1 , 2 , and 25 east and west of the central position . due to the spatially extended nature of globular clusters , and our desire to obtain their representative integrated spectra , each globular cluster observation was obtained by drifting the spectrograph slit across the core diameter of the cluster . the cluster core radii are listed in table 1 . the telescope was positioned so as to offset the slit from the cluster center by one core radius . a suitable trail rate would then allow the slit to drift across the cluster core diameter during the typically 15 minute - long exposure . in addition to the trail across the cluster core , a suitable background determination is required . we considered the extraction of a background spectrum from the ends of the @xmath055 slit . however there are several disadvantages to this procedure . for many of the clusters , there would still be significant contamination from the cluster itself even at a radial distance of @xmath023 . use of the extreme ends of the slit for sky subtraction also requires a very secure mapping of the slit illumination and of the geometric distortions in the spectrograph . as well , there is the danger of deteriorating focus at large distances from the slit center . all of these problems are circumvented by taking a separate sky exposure from the cluster exposure , and then directly subtracting the sky frame from the cluster frame . this approach also allows one to carefully select patches of background that appear to be well matched to that of the cluster , which is particularly important in regions towards the galactic bulge , where the background variations are large due to non - uniform extinction and/or diffuse hii emission . we selected background areas in the vicinity of each cluster from film copies of the _ european southern observatory / src _ plates and/or print copies of the _ palomar observatory sky survey_. in crowded regions towards the galactic bulge , areas of typically 5 @xmath12 5 or 10@xmath12 10 were selected for background scans . the criteria used in defining the scan region included visually similar level of faint background stars as for the cluster , avoidance of discrete bright stars , and , if applicable , similar level of hii emission as the cluster ( as best seen on the red sky survey print / film ) . the typical observation sequence included an arc spectrum , a 15 minute trailed cluster observation , and a 15 minute trailed background observation . for clusters with a low star background , the slit was fixed at a particular sky position , i.e. , no trail was made for the background spectrum . in the case of one cluster , ngc 6352 , for which the background spectrum was `` contaminated '' by the presence of several very bright b and a stars that were covered in the trailed background spectrum , we found it safer to use the ends of the slit from the cluster observation to extract the background spectrum . a journal of observations , providing complete information regarding the globular cluster and background spectra , is given in table 2 . we give in columns ( 1 ) , ( 2 ) , and ( 3 ) the cluster ids , the ut date of observation , and the exposure time ( in minutes ) , respectively . the orientation of the spectrograph slit is listed in column ( 4 ) . in column ( 5 ) we specify the angular distance , in arcseconds , over which the spectrograph slit was trailed . a trail of 50indicates that we trailed by @xmath1325on either side of the cluster center . we list in columns ( 6 ) and ( 7 ) the exposure time and angular offset ( from the cluster center ) , in arcminutes , for the sky background exposure . in column ( 8) we give the angular distance , in arcminutes , over which we trailed the spectrograph slit for the sky background exposure . a trail of 10 indicates that we moved the telescope by @xmath135on either side of the sky offset position given in column ( 7 ) . finally , in column ( 9 ) we give the region extracted along the slit for the given cluster . normally this was a symmetrical extraction of @xmath13@xmath2 on either side of the cluster center , but in some instances it was necessary to consider an asymmetrical extraction . in addition to the globular cluster observations , typically 2 or 3 spectrophotometric flux standards were observed per night . for these observations the slit was widened to 10to avoid wavelength - dependent light losses through differential atmospheric refraction . for the cluster observations the slit was trailed across the cluster so that atmospheric dispersion is not an issue . on each night several lick / ids standard stars were also observed to transform our observations into the lick spectral index system ( worthey 1994 ) . to further tie in our observations with population synthesis models , we also observed several stars from two more recent stellar spectral libraries based on the coud feed spectrograph at the kitt peak national observatory . these libraries are both available at the noao website . we note here the more extensive current library , called the indo - us library of coud feed stellar spectra , which is summarized in valdes ( 2004 ) , and available at the url : http://www.noao.edu / cflib/. since the purpose of this database is to provide a homogeneous set of integrated spectra of galactic globular clusters , we summarize here all aspects of the observations that could generate a departure from a strictly homogeneous dataset . we stress that in all cases these departures are minor and can be appropriately corrected or accounted for . from frequent monitoring of bias frames , we determined that the bias was subject to large variations during the night . the mean overscan level fluctuated by up to 80 adu ( hence 80 e@xmath14 ) during the night and could jump by as much as 10 adu between consecutive bias frames . in addition , the shape of the overscan varied from one bias frame to the next . on some bias frames , an overall gradient as large as 6 adu was present across the overscan region , but the trend varied considerably from one frame to the next , and a fifth order polynomial was used to fit the overscan . fortunately , with only one exception , we found that the overscan - subtracted bias frame is stable , thus a useful overscan - subtracted master bias frame could indeed be constructed from a series of bias exposures , and that master bias is well behaved . the sole exception is the 25 bias frames from the first night . hence we used the master bias from the second night for the first night as well . as mentioned in [ sec : setup ] , a glitch in the slit motor occurred approximately midway through the last night of observations . the motor failed while narrowing the slit back to 225 @xmath11 after we had widened it to observe a flux standard . after the problem was fixed and the system rebooted , we encountered a zeropoint problem in the slit width ; the slit looked visibly wider on the tv monitor for the guider camera . for the next observation , of the globular cluster ngc 6362 , the linewidths in the arc spectrum are nearly twice as large as in previous observations . as a result , we narrowed the slit considerably , but ended up with a narrower slit for the rest of that final night than for the first three nights . indeed the lines in the arc spectra are narrower by @xmath00.2 pixels in fwhm than for the earlier part of the night . the clusters affected are ngc 6528 ( also observed on a previous night ) , ngc 6553 , ngc 6569 , ngc 6752 , and 47 tuc . users of the database should be aware that we have slightly smoothed the spectra of these 5 clusters to give them the same spectral resolution as the rest of the observations . the spectrum of ngc 6362 should only be used for low resolution applications . during the first two nights of the observing run , the rotator mirror that switches between the main field viewer and the henear arc lamp was unreliable , and required frequent rebooting and reinitialization of the system . after reinitialization , the wavelength zeropoint could shift by several pixels , which is not a problem since an arc lamp exposure was taken along with the object exposure . however , we caution the reader that we do not consider the radial velocity data extracted from our integrated spectra to be reliable . to circumvent the rotator mirror problem , for the third and fourth nights the penray henear lamp from the hydra multi - fiber spectrograph was used as the wavelength calibration source . this lamp does not provide as uniform and extensive coverage as the henear source for the r - c spectrograph , particularly at the bluer wavelengths . [ sec : reductions ] for a description of how we handled this problem . although the weather was mostly good during the four - night observing run , non - photometric conditions were experienced at times during the first and third nights , i.e. , 2003 april 24 - 25 and 2003 april 26 - 27 . during the first night , thin clouds developed during the second half of the night , and appeared to persist until dawn . thus the exposures of ngc 6121 , ngc 6266 , ngc 6284 , ngc 6342 , and ngc 6626 are all marginally affected by thin clouds on that night . on the third night , however , the evening started with thick cirrus , which eventually developed to the point where we were required to shut down for approximately an hour . thereafter , the sky cleared rapidly , and the conditions were good for the second half of the night . exposures of five clusters , ngc 1904 , ngc 2298 , ngc 2808 , ngc 5286 , and ngc 5904 were taken under quite heavy clouds , with those of ngc 1904 , ngc 2808 , and ngc 5286 taken under approximately a magnitude of extra extinction . note that we have at least one additional exposure for each of these five clusters on other nights . analysis of the spectra was carried out using the iraf software package . we generally followed the procedures described in the tutorials `` a user s guide to ccd reductions with _ iraf _ '' by p. massey and `` a user s guide to reducing slit spectra with _ iraf _ '' by p. massey , f. valdes , and j. barnes , both available at the iraf website . the first step involved fitting and removing the overscan region and then subtracting off a master bias frame . an overscan - subtracted master bias frame was created for each night from a series of 25 bias frames ( taken at the start of the night ) . the 25 frames were averaged , with a _ minmax _ rejection of the highest and lowest values , using the _ zerocombine _ routine in the _ ccdred _ package . the bias - subtracted frames were then flat - field corrected . a master flat was produced from a sequence of 25 overscan and bias - subtracted dome flat field frames . the individual flats were averaged , and cosmic ray rejection was achieved using the _ crreject _ parameter in the _ flatcombine _ task . then the instrumental and flat field lamp response was normalized along the dispersion direction using the _ response _ task and the non - uniform illumination along the slit was measured using a twilight sky exposure according to the _ illumination _ task . all object and arc frames were then overscan and bias subtracted , flat - field divided , and illumination corrected . next we extracted a one - dimensional spectrum from the two - dimensional data frames . this was accomplished using the _ apall _ task in the _ apextract _ package . the extraction aperture was selected by examining the slit profile for the cluster , and generally included the core diameter of the cluster . two important aspects of the extraction are to trace the location of the cluster spectrum as a function of position along the dispersion axis ( the _ aptrace _ routine ) and eliminating cosmic rays ( crs ) by making a variance - weighted , cleaned extraction that fits the aperture profile at each wavelength ( see the _ iraf _ slit spectrum tutorial mentioned above for details ) . for a few clusters ( e.g. , ngc 2298 ) , with particularly large core radii and/or a jagged profile along the slit produced by discrete bright stars , it proved difficult to successfully trace the slit profile at all wavelengths . in these cases we simply used the trace from a bright star taken on the same night , but with the aperture resized and initially centered according to the cluster itself . the _ apall _ procedure produces not only a variance - weighted , cleaned extracted spectrum , but an unweighted , uncleaned extracted spectrum , and a variance spectrum as well . the variance spectrum was calculated with the detector read noise and gain properly accounted for . the variance spectrum for the extracted cluster spectrum was added in quadrature to the variance spectrum for the extracted sky spectrum to produce a total variance for the final sky - subtracted cluster spectrum . the sky - subtracted cluster spectrum was then divided by the variance spectrum to produce a spectrum that gives the s / n ratio at each pixel . wavelength calibration was carried out using an arc spectrum taken at the same pointing position as the cluster spectrum , either immediately before or after the cluster and sky background spectra were acquired . at the start of each night a series of 5 arc spectra were taken , and averaged to form a master arc spectrum for that night . the s / n ratio of the master arc is high enough that many faint lines can be fit , thus stabilizing the wavelength fitting , especially at the blue end of the spectrum . a fourth order legendre polynomial was fit to the master arc , generally producing rms residuals of @xmath00.12 . then an extraction was made from the arc taken along with a particular cluster spectrum , and the extraction was made with the same trace and aperture parameters as for the cluster . since the s / n ratio of the single arc spectrum is considerably lower than for the master arc , we then calculated a zero point shift between the cluster arc and the master arc , using the _ reidentify _ task . the assumption is that variations in the geometric distortion of the spectrograph will be minimal during the night , with only a zero point shift needing consideration . the wavelength zero point did shift significantly during the night , by up to several pixels . however , we have verified that the zero point has been succesfully monitored , since the wavelength of the [ ] @xmath85577 night sky line in the dispersion corrected spectra is very stable from one cluster to another . for the last two nights , when the penray lamp from the hydra spectrograph was utilized ( cf . , [ sec : arcs ] ) , we used the master arc from the second night as our high s / n reference arc spectrum . after the dispersion correction was carried out , the final step was to flux calibrate the data using the typically 3 flux standards observed during each night . given the low number of flux standards observed , along with the fact that the mean extinction data for ctio were used , and that non - photometric conditions were encountered on some nights ( cf . , [ sec : weather ] ) , we consider the flux calibration to be only approximate . the most useful attribute of our spectral library is the relatively high resolution and high s / n , allowing for numerous line indices , which do not depend on an absolute flux calibration , to be defined and used . note that our observations do not suffer from the usual problem encountered in narrow - slit high - resolution spectroscopy , namely differential atmospheric refraction . due to the highly extended nature of our clusters , coupled with our trailing over the cluster core , atmospheric dispersion is not a problem . for our flux standard observations , we opened the slit to 10width , to ensure that atmospheric dispersion did not harm the flux standards either . we determined radial velocities from cross - correlation of the library spectra with an integrated spectrum of 47 tuc obtained in a previous observing run ( see schiavon et al . 2002a for a description ) , using the _ rv.fxcor _ iraf task . we estimated an internal accuracy of @xmath139 @xmath15 for our radial velocity determinations by measuring the central wavelengths of absorption lines in the blue and red ends of the spectra . finally , these radial velocities were used to bring the 1-d cluster spectra to the restframe . the finalized library of integrated spectra and electronic versions of tables 1 and 2 are available at http://www.noao.edu / ggclib/. for each cluster the final flux - calibrated spectrum is presented in a single fits file . another fits file , in multi - spectrum format , contains auxiliary spectral information which may be useful to some investigators . the multi - spectrum file contains four spectral bands , each one containing wavelength - calibrated , but _ not _ flux - calibrated , data . the first band contains the variance - weighted cr - cleaned , sky - subtracted spectrum , while the second band contains the unweighted , uncleaned extraction , both produced by the iraf _ apall _ task . the third band contains the _ apall _ extraction of the sky from the separate sky frame , except for the few instances where we extracted the sky from the ends of the slit . those exceptions are mentioned in the notes on individual clusters in 6 . the fourth band contains the s / n ratio at each pixel in the spectrum . we call the reader s attention for this slight deviation of the iraf standard , according to which the fourth band of _ apall _ output files usually contains the variance spectrum , instead of the s / n ratio . we decided to adopt the latter in order to provide the users with a quick visual estimate of the s / n ratio of the spectra at each pixel . as mentioned earlier , the s / n ratio per pixel was obtained by dividing the sky - subtracted cluster spectrum by the quadrature sum of the variance spectra in the cluster and in the sky . therefore , the variance spectrum can be recovered by taking the ratio between bands 1 and 4 of the auxiliary files . the readme file at the website gives further details on the format of the spectral data . figures [ fig : spectra1]a - h display our complete library of integrated spectra of galactic globular clusters , ordered by increasing [ fe / h ] . while the spectral library is clearly intended for work with quantitative spectral indices , we provide here a few qualitative illustrations of the use of these spectra . we begin by showing the effect of hb morphology on the integrated spectra of old stellar populations . in figure [ fig : hbm ] we compare the integrated spectra of the two moderately metal - poor gcs ngc 6723 and ngc 6652 . while the two clusters have virtually the same [ fe / h ] ( table 1 ) ngc 6652 is characterized by a red hb morphology , whereas ngc 6723 has a comparable number of blue and red hb stars ( piotto 2002 ) . the impact of the blue hb stars in the integrated spectrum of ngc 6723 can be easily seen by the substantially _ stronger _ balmer lines and _ weaker _ metallic features ( see also schiavon et al . 2004b ) . figure [ fig : hbp ] shows a similar comparison in a more metal - rich regime . ngc 6441 and ngc 6304 are more metal - rich than the above two clusters by @xmath0 0.5 dex , which is readily noticeable due to their much stronger metal lines . in ngc 6304 , the horizontal branch is made solely of red stars , while in the case of ngc 6441 it is dominated by red stars but contains a sizeable blue extension . the effect of hb stars on the integrated spectrum of ngc 6441 is more subtle than for the metal poor clusters discussed above . while balmer lines are clearly stronger in the spectrum of ngc 6441 , metal lines appear to be only weakly sensitive to the influence of the blue hb stars , their strengths being only slightly weaker in the spectrum of ngc 6441 , probably because the horizontal branch of the latter has a dominant red component . it is well established that the best age indicators for stellar populations use the sensitivity of balmer lines to the main sequence turnoff temperature ( worthey 1994 ) . hence one must track any conditions under which anomalous emission lines could lead to an artificial fill - in of the balmer lines in the integrated spectrum of a cluster . specifically , clusters at low galactic latitude can have a foreground weak diffuse emission region superimposed on the cluster . an example of this effect is seen in fig . [ fig : n3201 ] , where the two - dimensional spectrum of ngc 3201 is shown , covering the [ ] @xmath83727 emission line which extends across the entire slit . the presence of [ ] emission implies that there is emission in the balmer lines as well . the latter , if not appropriately subtracted , would fill - in the balmer lines in absorption , making them appear artificially weak and thereby making the cluster appear spuriously old . the identification and removal of any diffuse emission is thus crucial . to further emphasize this point with a second cluster , ngc 6352 , which has foreground [ ] emission , we plot in fig . [ fig : n6352spec ] the background - subtracted spectrum of ngc 6352 ( bottom ) and the `` raw '' spectrum before sky subtraction ( top ) . the [ ] 3727 emission is readily seen in the raw spectrum , and is greatly reduced in the subtracted spectrum . ( note that the [ ] 3727 emission is superposed on a pseudo - continuum peak in the line - rich absorption spectrum of ngc 6352 , making it difficult to assess the degree of emission contamination without subtracting out the underlying absorption spectrum of the cluster ) . for one last example , we take a closer look at the integrated spectrum of the relatively metal - rich globular cluster ngc 6637 (= m69 ) . this cluster contains an emission - line star near the center , as can be seen in fig . [ fig : n6637im ] , where we show the two - dimensional spectrum of ngc 6637 in the wavelength region covering h@xmath16 and h@xmath17 . aside from a few random crs , the localized emission in h@xmath17 and h@xmath16 at one specific location on the slit ( where the slit trailed across the emission - line star ) is clearly evident . this fact has been previously noted by rose & tripicco ( 1986 ) , who tentatively classified the star as a long - period variable ( lpv ) . in order to further investigate the nature of the emission - line star we extracted its individual spectrum and verified that it contains tio bands whose strength are consistent with an m2 - 3 spectral type , so that if the star is a cluster member its classification as a lpv is reliable . moreover , the strengths of the balmer lines are consistent with those of emission - line m giants found in the field ( e.g. pereira , franco & de arajo 2003 ) . however , we tried to match the position of the emission - line star in the 2d spectrum with the positions of known lpv stars in the cluster ( hartwick & sandage 1968 , rosino 1962 ) , without success . we also speculated whether the emission line star could perhaps be a symbiotic star or a foregroud flaring m dwarf . however , the spectra of symbiotic stars are characterized by the presence of strong hei and [ oiii ] lines which are absent in our spectrum ( e.g. luna & costa 2005 , warner 1995 ) . the same applies to the case of flare m dwarfs , whose spectra are characterized by the presence of the caii h and k lines in emission ( e.g. petit 1987 ) , which are also not seen in the spectrum of the emission - line star in ngc 6637 . therefore , we suggest that this emission - line star is _ likely _ to be a lpv member of the cluster . given the high luminosity of stars at the giant branch tip , the strong balmer emission lines due only to this star can significantly affect the integrated spectrum of the cluster . this is shown in figure [ fig : n6637spec ] , where the variance - weighted `` optimum '' extraction of the integrated spectrum and the unweighted extraction are both plotted . note the major discrepancy in the strengths of h@xmath16 and h@xmath17 between the two extractions . the difference is due to the fact that the variance - weighted extraction identifies and removes the bright emission spots in the two - dimensional spectrum , just as it does for crs . while the identification of the emission lines with crs is incorrect , they certainly have that appearance in figure [ fig : n6637im ] , and their removal from the variance - weighted extraction fortunately avoids the otherwise catastrophic influence of the lpv emission on the integrated balmer lines of the cluster . the example above raises two important issues . firstly , under special circumstances , the influence of a single star can clearly affect the integrated light of the cluster , even when the entire core diameter is covered . thus it is important to check for any particularly bright object along the slit . secondly , our exercise identified a case where the variance - weighted extraction procedure rejected the localized emission lines from the lpv as crs , which was fortunate , but we can imagine other less benign situations . essentially , any bright star in the cluster with a very different spectrum from the rest of the cluster is a target for removal by the variance - weighted extraction . for most clusters , this is not an issue , because the light profile along the slit is fairly smooth . however , for a few of the less concentrated clusters , the choice of spectral extraction technique can have an impact on the final results . for this reason , in the data products we have included both variance - weighted and unweighted extractions in the multi - spectrum fits file for each cluster , as was discussed above . thus one can directly search for anomalies created by the particular extraction technique . note that the unweighted extraction has _ not _ been cleaned of crs . in this section we provide information on individual clusters . * ngc 104 = 47 tuc : * 47 tuc is so bright that no separate sky exposure was taken . this is one of the clusters observed during the last night when the slit width was slightly narrower than for the other observations . we have smoothed the final spectrum of this cluster to match the spectral resolution of the other clusters . * ngc 1904 : * it was difficult to obtain a successful aperture trace for this cluster , due to the presence of a very red star that traversed the slit right at the center of the cluster . however , by adjusting the step size of the aperture trace , a successful trace was carried out . * ngc 2298 : * ngc 2298 could not be successfully traced for aperture extraction , due to a large number of discrete bright stars crossing the slit , and creating a jagged slit profile . consequently , we used the aperture trace from a stellar spectrum taken on the same night , but recentered the aperture and resized it . there are two spectra of this cluster , taken on different nights , with very slightly different aperture extractions . * ngc 2808 : * two exposures were taken , heavy clouds compromise the second one . we used slightly different aperture extractions for the two exposures . * ngc 3201 : * ngc 3201 is a cluster with a particularly large core diameter . it is somewhat difficult to determine the cluster center . we obtained spectra for the cluster on both ut 25-apr-2003 and ut 26-apr-2003 . there is strong [ oii]@xmath83727 emission evident across the entire slit in both cluster exposures and the two sky background regions scanned ( the sky background regions scanned were different on the two nights , as is recorded in table 2 . an extremely bright star crossed the slit during the sky background trail for the exposure on ut 25-apr-2003 . consequently , we decreased the extracted aperture on the sky background frame to @xmath1355 pixels ( @xmath18 ) , to avoid the contaminated location on the slit . we then renormalized the sky counts to match the @xmath1385 pixel extraction on the cluster . this problem did not occur for the exposure on ut 26-apr-2003 , for which we scanned a different sky background region . note that the [ oii]@xmath83727 emission subtraction appears to come out cleaner for the ut 26-apr-2003 exposure . * ngc 5286 : * there is no arc for ngc 5286 on the night of ut 2003 - 04 - 27 , so we used the arc for ngc 2808 , and made a zeropoint correction using night sky lines . * ngc 5904 : * ngc 5904 could not be successfully traced at all , maybe a result of too many discrete stars crossing the slit with different sed s . so we used the trace from the star hd102574 , but recentered the aperture and resized it . * ngc 5927 : * ngc 5927 was difficult to trace , so used trace from the star hd76151 , but recentered the aperture and resized it . * ngc 5986 : * the trail across ngc 5986 intersected a very bright star . we did not choose to eliminate the bright star from our aperture extraction . * ngc 6171 : * there is extended [ oii]@xmath83727 emission across this spatially extended cluster . to obtain a successful aperture trace for the cluster it was necessary to center the extracted aperture on the brightest peak on the slit profile rather than on the average center of the light distribution . * ngc 6284 : * ngc 6284 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . * ngc 6342 : * ngc 6342 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . * ngc 6544 : * ngc 6544 has an extremely peaked light profile along the slit that is offset by several arcseconds from the centroid of the fainter light levels . we used the peak in the light profile for the center of the extracted aperture , but used asymmetric limits for the aperture . * ngc 6352 : * ngc 6352 has emission at [ oii]@xmath83727 . sky subtraction in [ oii ] emission came out better when the background was subtracted from the ends of the slit rather than from the separate sky background exposure . the background in the separate frame was further compromised by several bright stars that went through the slit in the 5trail . hence we used the ends of the slit for sky subtraction in this case . * ngc 6362 : * the spectrun for this cluster has degraded resolution , due to the problem of the slit width control described in [ sec : slit ] . * ngc 6441 : * ngc 6441 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . * ngc 6528 : * for the 28-apr-2004 observations of ngc 6528 inaccuracy in the tracking led to the 15 minute trail not covering the anticipated 9 . as a result we took a second 15 minute exposure , which picked up the trail where it ended in the first exposure . the counts in the second exposure are very similar to those int he first exposure . we have maintained these as two separate observations ( ngc6528b*.fits and ngc6528c*.fits ) because they are taken from different parts of the cluster . in addition , ngc 6528 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . finally . we have smoothed the final spectra of this cluster taken on 28-apr-2004 to match the spectral resolution of the other clusters . * ngc 6553 : * this is perhaps the most problematic cluster in the sample . it is in a high background region of the bulge , with variable obscuration and variable [ oii ] emission . in addition , ngc 6553 is not a centrally concentrated cluster . consequently , it was a difficult cluster to trace , so we used the trace from a star , but with redefined apertures . we found that the [ oii ] and balmer emission in the background is stronger in the separate sky exposure than at the ends of the slit for the cluster exposure . consequently , we used the ends of the slit for background subtraction . in addition , there is a bright blue foreground star that traversed the slit during the cluster exposure . this star is bright enough to make an appreciable impact in strengthening the balmer lines in the integrated cluster spectrum . hence we separately extracted the slit position corrsponding tot he bright star and subtracted that off . nevertheless , even when the bright star contributioin is removed , the relative depths of ca ii h and k lines indicates a contribution from hot stars to the integrated spectrum of this cluster . we believe this reflects the contribution of blue straggler stars that are seen in the color - magnitude diagram of the cluster . * ngc 6569 : * ngc 6569 is one of the clusters observed during the last night when the slit width was slightly narrower than for the other observations . we have smoothed the final spectrum of this cluster to match the spectral resolution of the other clusters . * ngc 6624 : * ngc 6624 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . * ngc 6637 : * ngc 6637 has a long period variable in emission in the cluster core . the cleaned , variance - weighted extraction is unaffected by this ( the lpv balmer line emission is confused with cosmic rays and cleaned out ) , but the raw counts extraction has the balmer lines seriously weakened by the emission in the star . [ sec : examples ] for further details . * ngc 6652 : * for this cluster either the discrete digitization of the trail rate or inaccuracy in the tracking led to the 15 minute trail not covering the aniticipated 9 . as a result we took a second 15 minute exposure , which picked up the trail where it ended in the first exposure . the counts in the second exposure are @xmath030% higher than in the first exposure . we have maintained these as two separate observations because they are taken from different parts of the cluster . * ngc 6723 : * we took a 10 minute exposure on the cluster and only a 5 minute sky exposure , thus doubled the sky counts before subtracting them from the cluster . * ngc 6752 : * ngc 6752 is one of the clusters observed during the last night when the slit width was slightly narrower than for the other observations . we have smoothed the final spectrum of this cluster to match the spectral resolution of the other clusters . * ngc 7078 : * ngc 7078 is so centrally concentrated that we defined two apertures . the first aperture contains the fwhm of the slit profile , while the second contains a substantially wider region along the slit . * ngc 7089 : * we used sky from the ends of the slit , rather than from the subsequent sky exposure , because the sky background was starting to brighten considerably with approaching dawn . we have collected a new library of integrated spectra of galactic globular clusters , with high s / n and moderately high resolution . we envisage at least two chief applications for this new dataset . firstly , it should serve as a reliable database for local , well - known stellar populations , against which the observations of remote , unresolved , systems can be contrasted . secondly , this library can be used to verify the accuracy and reliability of the predictions of stellar population synthesis models . for both purposes , it is crucial that the fundamental parameters of the clusters that make up the library be well known , and for that reason we selected our targets amongst those with best color - magnitude and metal - abundance data . further applications aiming at a detailed calibration of stellar population synthesis models are currently under way . the spectral library is publicly available in electronic format from the national optical astronomical observatory website . beasley , m.a . , hoyle , f. & sharples , r.m . 2002 , , 336 , 168 bica , e. & alloin , d. 1986 , , 162 , 21 borkova , t. v. , & marsakov , v. a. 2000 , astron . , 44 , 665 brocato , e. , castellani , v. , raimondo , g. & walker , a.r . 1999 , , 527 , 230 bruzual , g. & charlot , s. 2003 , , 344 , 1000 bruzual , g. , barbuy , b. , ortolani , s. , bica , e. , cuisinier , f. , lejeune , t. & schiavon , r.p . 1997 , , 114 , 1531 burstein , d. , faber , s.m . , gaskell , c.m . & krumm , n. 1984 , , 287 , 586 carretta , e. & gratton , r.g . 1997 , , 121 , 95 cohen , j.g . , blakeslee , j.p . & ryzhov , a. 1998 , , 496 , 808 covino , s. , galletti , s. & pasinetti , l.e . 1995 , , 303 , 79 djorgovski , s. 1993 , asp conference series , 50 , 373 djorgovski , s. & meylan , g. 1993 , asp conference series , 50 , 325 faber , s.m . 1973 , , 179 , 731 feltzing , s. & johnson , r.a . 2002 , , 385 , 67 fullton , l.k . , carney , b.w . , olszewski , e.w . , zinn , r. , demarque , p. , janes , k.a . , da costa , g.s . & seitzer , p. 1995 , , 110 , 652 gibson , b.k . , madgwick , d.s . , jones , l.a . , da costa , g. & norris , j.e . 1999 , , 118 , 1268 hargis , j.r . , sandquist , e.l . & bolte , m. 2004 , , 608 , 243 harris , w.e . 1996 , , 112 , 1487 hartwick , f.d.a . & sandage , a. 1968 , , 153 , 715 howell , j.h . , guhathakurta , p. & gilliland , r.l . 2000 , , 112 , 1200 janes , k.a . & heasley , j.n . 1991 , , 101 , 2097 kraft , r.p . & ivans , i.i . 2003 , , 115 , 143 kanatas , i. , griffiths , w.k . , dickens , r.j . & penny , a.j . 1995 , , 272 , 265 le borgne , d. , rocca - volmerange , b. , prugniel , p. , lanon , a. , fioc , m. & soubiran , c. 2004 , , 425 , 881 lee , y .- w . , demarque , p. & zinn , r. 1994 , , 423 , 248 luna , g.j.m . & costa , r.d.d . 2005 , , in press , astro - ph/0502308 maraston , c. , greggio , l. , renzini , a. , ortolani , s. , saglia , r.p . , puzia , t.h . & kissler - patig , m. 2003 , , 400 , 823 markov , h.s . , spassova , n.m . & baev , p.v . 2001 , , 326 , 102 momany , y. , piotto , g. , recio - blanco , a. , bedin , l.r . , cassisi , s. & bono , g. 2002 , , 576 , l65 momany , y. , ortolani , s. , held , e.v . , barbuy , b. , bica , e. , renzini , a. , bedin , l.r . , rich , r.m . & marconi , g. 2003 , , 402 , 607 ortolani , s. , momany , y. , barbuy , b. , bica , e. , catelan , m. 2000 , , 362 , 953 ortolani , s. , bica , e. & barbuy , b. 2001 , , 374 , 564 pereira , c.b . , franco , c.s . & de arajo , f.x . 2003 , , 397 , 927 peterson , c.j . 1993 , asp conference series , 50 , 337 petit , m. 1987 , variable stars ( new york : john wiley & sons ) piotto , g. et al . 2002 , , 391 , 945 proctor , r.n . , forbes , d.a . & beasley , m.a . 2004 , , 355 , 1327 pulone , l. , de marchi , g. , covino , s. & paresce , f. 2003 , , 399 , 121 puzia , t.h . , saglia , r.p . , kissler - patig , m. , maraston , c. , greggio , l. , renzini , a. & ortolani , s. 2002 , , 395 , 45 rose , j.a . 1985 , , 90 , 1927 rose , j.a . 1994 , , 107 , 206 rose , j.a . & tripicco , m.j . 1986 , , 92 , 610 rosenberg , a. , piotto , g. , saviane , i. & aparicio , a. 2000a , , 144 , 5 rosenberg , a. , aparicio , a. , saviane , i. & piotto , g. 2000b , , 145 , 451 rosino , l. 1962 , mem.soc.astron.it . 33 , 351 samus , n. , ipatov , a. , smirnov , o. , kravtsov , v. , alcaino , g. , liller , w. & alvarado , f. 1995 , , 112 , 439 sandquist , e. , bolte , m. , stetson , p.b . & hesser , j.e . 1996 , , 470 , 910 schiavon , r.p . & barbuy , b. 1999 , , 510 , 934 schiavon , r.p . , faber , s.m . , castilho , b.v . & rose , j.a . 2002a , , 580 , 850 schiavon , r.p . , faber , s.m . , rose , j.a . & castilho , b.v . 2002b , , 580 , 873 schiavon , r.p . , caldwell , n. & rose , j.a . 2004a , , 127 , 1513 schiavon , r.p . , rose , j.a . , courteau , s. , & macarthur , l. 2004b , apj , 608 , l33 thomas , d. , maraston , c. & bender , r. 2003 , , 339 , 897 valdes , f. , gupta , r. , rose , j.a . , singh , h.p . & bell , d.j . 2004 , , 152 , 251 vazdekis , a. 1999 , , 513 , 224 vazdekis , a. , salaris , m. , arimoto , n. & rose , j.a . 2001 , , 549 , 274 von braun , k. & mateo , m. 2001 , , 121 , 1522 von braun , k. , mateo , m. , chiboucas , k. , athey , a. & hurley - keller , d. 2002 , , 124 , 2067 warner b. 1995 , cataclysmic variable stars ( new york : cambridge univ . press ) worthey , g. 1994 , , 95 , 107 worthey , g. , faber , s.m . , gonzlez , j.j . & burstein , d. 1994 , , 94 , 687 zinn , r. & west , m.j . 1984 , , 55 , 45 zoccali , m. , renzini , a. , ortolani , s. , bica , e. & barbuy , b. 2001 , , 121 , 2638 llrrcrrcccl [ tab : mastertable ] 104 & 47 tuc&305.90 & -44.89 & 7.4 & 0.40 & 2.03 & 0.04 & -0.70@xmath19 & -0.99 & 1,2,12 + 1851 & & 244.51 & -35.04 & 16.7 & 0.06 & 2.32 & 0.02 & -1.21@xmath20 & -0.36 & 1,2 + 1904 & m 79 & 227.23 & -29.35 & 18.8 & 0.16 & 1.72 & 0.01 & -1.55@xmath20 & -0.89 & 1,2 + 2298@xmath21 & & 245.63 & -16.01 & 15.1 & 0.34 & 1.28 & 0.15 & -1.97@xmath19 & 0.93 & 2 + 2808 & & 282.19 & -11.25 & 11.0 & 0.26 & 1.77 & 0.23 & -1.29@xmath20 & -0.49 & 1,2 + 3201 & & 277.23 & 8.64 & 9.0 & 1.43 & 1.30 & 0.21 & -1.56@xmath19 & 0.08 & 1,2,14 + 5286@xmath21 & & 311.61 & 10.57 & 7.2 & 0.29 & 1.46 & 0.24 & -1.51@xmath22 & 0.80 & 4 + 5904 & m 5 & 3.86 & 46.80 & 6.2 & 0.42 & 1.83 & 0.03 & -1.26@xmath19 & 0.31 & 1,3,13,21 + 5927 & & 326.60 & 4.86 & 4.5 & 0.42 & 1.60 & 0.45 & -0.64@xmath23 & -1.00 & 1,2 + 5946 & & 327.58 & 4.19 & 7.4 & 0.08 & 2.50 & 0.54 & -1.54@xmath22 & b@xmath24 & 1 + 5986 & & 337.02 & 13.27 & 4.8 & 0.63 & 1.22 & 0.27 & -1.53@xmath20 & 0.97 & 1,2,17 + 6121@xmath21 & m 4 & 350.97 & 15.97 & 6.2 & 0.83 & 1.59 & 0.40 & -1.15@xmath19 & -0.06 & 2,5 + 6171 & m 107 & 3.37 & 23.01 & 3.3 & 0.54 & 1.51 & 0.33 & -1.13@xmath20 & -0.73 & 1,2 + 6218 & m 12 & 15.72 & 26.31 & 4.5 & 0.72 & 1.39 & 0.19 & -1.32@xmath20 & 0.97 & 1,3,15,16 + 6235 & & 358.92 & 13.52 & 2.9 & 0.13 & 1.33 & 0.36 & -1.36@xmath20 & 0.89 & 1 + 6254@xmath21 & m 10 & 15.14 & 23.08 & 4.7 & 0.86 & 1.40 & 0.28 & -1.51@xmath19 & 0.98 & 3,15 + 6266 & m 62 & 353.58 & 7.32 & 1.7 & 0.18 & 1.70 & 0.47 & -1.20@xmath20 & 0.32 & 1,2 + 6284 & & 358.35 & 9.94 & 6.9 & 0.07 & 2.50 & 0.28 & -1.27@xmath22 & b@xmath24 & 1 + 6304 & & 355.83 & 5.38 & 2.1 & 0.21 & 1.80 & 0.52 & -0.66@xmath23 & -1.00 & 1,2 + 6316 & & 357.18 & 5.76 & 3.2 & 0.17 & 1.55 & 0.51 & -0.90@xmath22 & -1.00 & 1 + 6333@xmath21 & m 9 & 5.54 & 10.71 & 1.7 & 0.58 & 1.15 & 0.35 & -1.65@xmath22 & 0.87 & 6 + 6342 & & 4.90 & 9.73 & 1.7 & 0.05 & 2.50 & 0.46 & -1.01@xmath22 & -1.00 & 1 + 6352@xmath21 & & 341.42 & -7.17 & 3.1 & 0.83 & 1.10 & 0.19 & -0.70@xmath23 & -1.00 & 2,7,8 + 6356 & & 6.72 & 10.22 & 7.6 & 0.23 & 1.54 & 0.28 & -0.74@xmath22 & -1.00 & 1 + 6362 & & 325.55 & -17.57 & 5.3 & 1.32 & 1.10 & 0.08 & -1.17@xmath20 & -0.58 & 1,2,18 + 6388 & & 345.56 & -6.74 & 4.4 & 0.12 & 1.70 & 0.40 & -0.68@xmath22 & i@xmath24 & 1 + 6441 & & 353.53 & -5.01 & 3.5 & 0.11 & 1.85 & 0.44 & -0.65@xmath22 & i@xmath24 & 1 + 6522 & & 1.02 & -3.93 & 0.6 & 0.05 & 2.50 & 0.48 & -1.39@xmath20 & 0.71 & 1 + 6528@xmath21 & & 1.14 & -4.17 & 1.5 & 0.09 & 2.29 & 0.62 & -0.10@xmath20 & -1.00 & 9,19 + 6544 & & 5.84 & -2.20 & 5.4 & 0.05 & 1.63 & 0.73 & -1.38@xmath20 & 1.00 & 1,2 + 6553@xmath21 & & 5.25 & -3.02 & 4.6 & 0.55 & 1.17 & 0.80 & -0.20@xmath20 & -1.00 & 10 + 6569 & & 0.48 & -6.68 & 1.2 & 0.37 & 1.27 & 0.56 & -1.08@xmath22 & mr@xmath24 & 1,20 + 6624 & & 2.79 & -7.91 & 1.2 & 0.06 & 2.50 & 0.28 & -0.70@xmath23 & -1.00 & 1 + 6626@xmath21&m 28 & 7.80 & -5.58 & 2.4 & 0.24 & 1.67 & 0.38 & -1.21@xmath20 & 0.90 & 2 + 6637 & m 69 & 1.72 & -10.27 & 1.6 & 0.34 & 1.39 & 0.16 & -0.78@xmath23 & -1.00 & 1,2 + 6638 & & 7.90 & -7.15 & 1.6 & 0.26 & 1.40 & 0.40 & -1.08@xmath20 & -0.30 & 1,2 + 6652 & & 1.53 & -11.38 & 2.4 & 0.07 & 1.80 & 0.09 & -1.10@xmath22 & -1.00 & 1 + 6723 & & 0.07 & -17.30 & 2.6 & 0.94 & 1.05 & 0.05 & -1.14@xmath20 & -0.08 & 1,2 + 6752@xmath21 & & 336.50 & -25.63 & 5.1 & 0.17 & 2.50 & 0.04 & -1.57@xmath19 & 1.00 & 2,11 + 7089 & m 2 & 53.38 & -35.78 & 10.4 & 0.34 & 1.80 & 0.06 & -1.49@xmath22 & 0.96 & 1 + ngc 104 & 2003 - 04 - 28 & 5 & e - w & 48 & & & & @xmath1325 + ngc 1851 & 2003 - 04 - 25 & 5 & n - s & 9 & 5 & 20w & & @xmath139 + ngc 1904 & 2003 - 04 - 26 & 15 & e - w & 20 & 15 & 15e & & @xmath1310 + ngc 1904 & 2003 - 04 - 27 & 15 & e - w & 20 & 15 & 15e & & @xmath1314 + ngc 2298 & 2003 - 04 - 27 & 15 & e - w & 36 & 15 & 15e & & -24 , + 20 + ngc 2298 & 2003 - 04 - 28 & 15 & e - w & 36 & 15 & 15e & & @xmath1320 + ngc 2808 & 2003 - 04 - 25 & 10 & e - w & 30 & 5 & 15e & & @xmath1320 + ngc 2808 & 2003 - 04 - 27 & 10 & e - w & 30 & 10 & 15w & 5 & -17 , + 15 + ngc 3201 & 2003 - 04 - 25 & 15 & n - s & 170 & 15 & 12w & 5 & @xmath1385 + ngc 3201 & 2003 - 04 - 26 & 15 & n - s & 170 & 15 & 12n & 5 & @xmath1385 + ngc 5286 & 2003 - 04 - 25 & 10 & n - s & 36 & 10 & 12n & 5 & @xmath1320 + ngc 5286 & 2003 - 04 - 27 & 10 & e - w & 36 & 10 & 12w & 5 & @xmath1320 + ngc 5286 & 2003 - 04 - 28 & 15 & e - w & 36 & 15 & 12w & 5 & -20 , + 23 + ngc 5904 & 2003 - 04 - 27 & 15 & e - w & 54 & 15 & 20w & 5 & -27 , + 30 + ngc 5904 & 2003 - 04 - 28 & 15 & e - w & 45 & 15 & 20n & 5 & -27 , + 30 + ngc 5927 & 2003 - 04 - 25 & 15 & n - s & 50 & 15 & 20w & 10 & @xmath1329 + ngc 5927 & 2003 - 04 - 25 & 15 & n - s & 50 & 15 & 15e & 10 & @xmath1329 + ngc 5927 & 2003 - 04 - 28 & 15 & e - w & 46 & 15 & 15e & 10 & @xmath1325 + ngc 5946 & 2003 - 04 - 26 & 15 & n - s & 10 & 15 & 10w & 10 & -6 , + 7 + ngc 5986 & 2003 - 04 - 25 & 15 & n - s & 76 & 15 & 12n & 5 & -32 , + 42 + ngc 6121 & 2003 - 04 - 25 & 15 & n - s & 100 & 15 & 14s , 14w & 10 & @xmath1350 + ngc 6171 & 2003 - 04 - 26 & 15 & n - s & 64 & 15 & 11n , 11w & 10 & -32 , + 36 + ngc 6171 & 2003 - 04 - 26 & 15 & n - s & 64 & 15 & 10s & 10 & -32 , + 37 + ngc 6218 & 2003 - 04 - 27 & 15 & e - w & 90 & 15 & 20w & 10 & @xmath1345 + ngc 6235 & 2003 - 04 - 28 & 15 & e - w & 45 & 15 & 20n & 10 & @xmath1325 + ngc 6254 & 2003 - 04 - 27 & 15 & e - w & 108 & 15 & 15n & 10 & @xmath1354 + ngc 6266 & 2003 - 04 - 25 & 15 & n - s & 22 & 15 & 15n & 5 & -12 , + 16 + ngc 6284 & 2003 - 04 - 25 & 15 & n - s & 9 & 15 & 15n & 5 & -6 , + 7 + & & & & & & & & @xmath1314 + ngc 6304 & 2003 - 04 - 26 & 15 & n - s & 26 & 15 & 10n & 5 & @xmath1317 + ngc 6316 & 2003 - 04 - 26 & 15 & n - s & 20 & 15 & 10e & 5 & -10 , + 11 + ngc 6316 & 2003 - 04 - 28 & 15 & e - w & 18 & 15 & 10w & 5 & -14 , + 11 + ngc 6333 & 2003 - 04 - 26 & 15 & n - s & 70 & 15 & 15e & 5 & @xmath1335 + ngc 6342 & 2003 - 04 - 25 & 10 & n - s & 12 & 10 & 10s & 5 & @xmath134 + & & & & & & & & @xmath1310 + ngc 6352 & 2003 - 04 - 27 & 15 & e - w & 99 & & & & @xmath1350 + ngc 6356 & 2003 - 04 - 26 & 15 & n - s & 28 & 15 & 15e & 5 & @xmath1322 + ngc 6362 & 2003 - 04 - 28 & 15 & e - w & 160 & 15 & 15w & 5 & @xmath1380 + ngc 6388 & 2003 - 04 - 26 & 10 & n - s & 18 & 10 & 10e & 5 & @xmath1310 + ngc 6441 & 2003 - 04 - 26 & 10 & n - s & 12 & 10 & 10e & 5 & @xmath1310 + & & & & & & & & @xmath1318 + ngc 6522 & 2003 - 04 - 26 & 15 & n - s & 9 & 15 & 8s & 5 & @xmath139 + ngc 6528 & 2003 - 04 - 27 & 15 & e - w & 9 & 15 & 8n & 5 & @xmath138 + & & & & & & & & @xmath1315 + ngc 6528 & 2003 - 04 - 28 & 15 & e - w & 5 & 15 & 15w & 5 & @xmath137 + & & & & & & & & @xmath2514 + ngc 6544 & 2003 - 04 - 26 & 15 & n - s & 18 & 15 & 15e & 10 & -12 , + 6 + ngc 6553 & 2003 - 04 - 28 & 15 & e - w & 64 & 15 & & & @xmath1335 + ngc 6569 & 2003 - 04 - 28 & 15 & e - w & 45 & 15 & 8w , 8s & 10 & @xmath1325 + ngc 6624 & 2003 - 04 - 27 & 15 & e - w & 9 & 15 & 15s & 5 & -7 , + 5 + & & & & & & & & @xmath1311 + ngc 6626 & 2003 - 04 - 25 & 10 & n - s & 28 & 10 & 15s & 5 & @xmath1316 + ngc 6637 & 2003 - 04 - 27 & 15 & e - w & 36 & 15 & 15n & 10 & @xmath1320 + ngc 6638 & 2003 - 04 - 27 & 15 & e - w & 27 & 15 & 15n & 5 & @xmath1315 + ngc 6652 & 2003 - 04 - 27 & 15 & e - w & 9 & 15 & 12w & 5 & @xmath137 + ngc 6723 & 2003 - 04 - 27 & 10 & e - w & 108 & 5 & 12w & 5 & @xmath1354 + ngc 6752 & 2003 - 04 - 28 & 5 & e - w & 18 & 5 & 12w & & -13 , + 11 + ngc 7078 & 2003 - 04 - 26 & 5 & n - s & 9 & & & & @xmath135 + & & & & & & & & @xmath1320 + ngc 7089 & 2003 - 04 - 28 & 5 & e - w & 36 & & & & @xmath1322 +

in a bernal - stacked graphene bilayer , an electronic state in landau level is described by its guiding - center index ( in the landau gauge ) and by its valley , spin , and orbital indices and when coulomb interaction is taken into account , the chiral two - dimensional electron gas ( c2deg ) in this system can support a variety of quantum hall ferromagnetic ( qhf ) ground states where the spins and/or valley pseudospins and/or orbital pseudospins collectively align in space . in this work , we give a comprehensive account of the phase diagram of the c2deg at integer filling factors $ ] in landau level when an electrical potential difference between the two layers is varied . we consider states with or without layer , spin , or orbital coherence . for each phase , we discuss the behavior of the transport gap as a function of the spectrum of collective excitations and the optical absorption due to orbital pseudospin - wave modes . we also study the effect of an external in - plane electric field on a coherent state that has both valley and spin coherence and show that it is possible , in such a state , to control the spin polarization by varying the strength of the external in - plane electric field .

1301.3094

electrons in a bernal - stacked graphene bilayerbilayerreview , castrojphys behave as a chiral two - dimensional gas of massive dirac fermions ( c2deg)@xcite . the chiral nature of the electrons lead to transport and optical properties that are different from those of conventional semiconductor 2deg s or of the 2deg in monolayer graphene . in particular , in the absence of coulomb interaction , the landau level ( ll ) spectrum is given by @xmath8 where the landau level index @xmath9 and the effective cyclotron frequency @xmath10 where @xmath11 is the magnetic field and @xmath12 the effective mass of the electrons . each landau level is four - time degenerate when counting valley and spin degrees of freedom with the exception of landau level @xmath5 which is eight - time degenerate . indeed , an electronic state in landau level @xmath5 is specified by its guiding - center @xmath1 ( in the landau gauge ) , spin @xmath13 valley @xmath14 and orbital @xmath15 indices@xcite . ( in landau level @xmath16 valley and layer degrees of freedom are equivalent . ) when the coulomb interaction is negligible with respect to the disorder broadening at low temperature and when the small zeeman splitting is neglected , the eight states in @xmath5 are degenerate and the hall conductivity has plateaus at @xmath17 where @xmath18@xcite . in recent transport experimentsfeldman , martin , bao , dean , zhao , velasco , freitag , weitz , it was shown that in sufficiently pure sample , when disorder is low or when the magnetic field is large enough , the coulomb interaction completely lifts the degeneracy of the @xmath5 octet and lead to the formation of seven new plateaus in the hall conductivity i.e. @xmath19 where @xmath20 .$ ] these plateaus were attributed to the formation of broken - symmetry many - body ground states . these states can alternatively be described as quantum hall ferromagnets ( qhf s ) where the spin and/or valley pseudospins and/or orbital pseudospins are spontaneously and collectively aligned in space@xcite . in bilayer graphene , a top - bottom gates voltage imbalance can be applied to create a potential difference @xmath6 ( which we call the `` bias '' hereafter ) between the two layers . in dual - gated bilayer graphene , @xmath21 and the total density of electrons in the bilayer can be controlled independently . this allows the phase diagram of the c2deg to be studied as a function of @xmath7 magnetic field and temperature . such study has been done by weitz et al.@xcite in high - quality bilayer graphene suspended between a top gate electrode and the substrate . the measurements show a series of phase transitions between different qhf states as @xmath21 is increased at given filling factor and magnetic field . special attention has been given to the filling factor @xmath22 where the precise nature of the ground state when @xmath23 near zero bias is still debated@xcite . all experiments were done at relatively small magnetic field @xmath24 t with the exception of the experiments reported in ref . where @xmath11 reached @xmath25 t. various aspects of the qhf states in bilayer graphene ( in particular the nature and the evolution of the ground state of the c2deg near charge neutrality as @xmath23 ) have been studied theoretically by a number of authorsnandkishore , gorbarnu0,gorbarnutot , gorbarconductivity , shizuya2012,shizuyadipole , shizuyacyclotron , shizuyamodes01,sari . in the work of gorbar et al.@xcite and shizuyashizuya2012 , the phase diagram of the c2deg as a function of bias for @xmath26 is presented . gorbar et al . have considered the effect of both static@xcite and dynamical screenings@xcite of the coulomb interaction . the modified gap equation captures the linear scaling of the transport gaps with the magnetic field which is seen in all the experiments at low magnetic field . the transport gaps are also strongly reduced with respect to the unscreened case and become more comparable to those observed experimentally . the dynamical screening was found to reproduce the offset in the behavior of the gap with magnetic field in the spin - polarized qhf state near zero bias which is seen in the experiments@xcite . shizuya points out that the filled levels @xmath27 can not be considered as completely inert . instead , they lead to a correction @xmath28 of the energy of the orbital levels @xmath15 that can change the ordering of these states in a way that depend on their occupation . a negative capacitance effect is also found that suppress rotation of the valley pseudospins . the work that we present in this paper extends our previous study of the phase diagram of the c2deg at zero biasbarlasprl1,barlasprl2,coteorbital , coteorbital2 and complements the study of gorbar et al . and shizuya . we give a comprehensive account of the phase diagram of the c2deg in bilayer graphene at all integer filling factors @xmath29 $ ] in landau level @xmath5 as a function of an applied bias @xmath6 at a fixed magnetic field . our analysis is based on an effective two - band model@xcite which describes the low - energy physics near the valleys @xmath30 . we explicitly take into account as symmetry - breaking terms the zeeman splitting @xmath31 and the bias @xmath32 . in the effective two - band model , the bias @xmath6 lifts the degeneracy between the orbital levels @xmath33 and @xmath34 by a small amount @xmath35 where @xmath36 with @xmath37 the interlayer hopping between carbon atoms that are part of a dimer . we also include in our model the interlayer next nearest - neighbor hopping term @xmath38 between carbons atoms in the same sublattices . this term causes a small asymmetry in the electronic band structure and was neglected in previous studies@xcite . it combines with the correction @xmath39 to give an energy difference @xmath40 between the @xmath34 and @xmath33 orbital levels where @xmath41 is the intralayer hopping energy between nearest - neighbors . this correction is thus finite at zero bias and breaks the orbital degeneracy . our phase diagram is not electron - hole symmetric around @xmath22 and the sequence of phase transitions is different for each filling factor . in our analysis , we treat the electron interaction in the hartree - fock approximation ( hfa ) and compute the collective excitations and electromagnetic absorption of the different phases of the c2deg in the generalized random - phase approximation ( grpa ) . we include in our study both uniform and non - uniform states and allow for the possibility of any type of coherent ( or qhf ) state . by coherent state , we mean a state where the average value @xmath42 for @xmath43 and/or @xmath44 and/or @xmath45 where @xmath46 creates an electron in a state with quantum numbers @xmath47 in our phase diagram which is summarized on fig . 5 , the layer - coherent states occur at very small bias because of the small interlayer distance @xmath48 nm in bilayer graphene . as the bias is increased , we find around a critical bias corresponding to the regions where the hall conductivity ceases to be quantized in the experiments@xcite a state with both layer and spin coherence . the orbital coherent states occur at a much larger bias corresponding to the situation where level @xmath34 gets lower in energy than level @xmath33 in valley @xmath49 ( see fig . figniveaux ) . in - between these coherent states are various incoherent states , some of which have been studied before@xcite . interestingly , we find that the application of an electric field in the plane of the layers can produce a new state where all three coherences ( layer , orbital and spin ) are present . in such a state , it is possible to control the degree of _ spin _ polarization by changing the strength of the external in - plane _ electric _ field . we also present a study the properties of the different ground states in the phase diagram . for all filling factors , we show how the transport gaps evolve with bias . in most cases , this evolution is qualitatively similar to that obtained with screening corrections@xcite . we compute all the intra - ll collective excitations in the various phases showing that all coherent states but the orbital state are characterized by a linearly - dispersing gapless ( in the long - wavelength limit ) goldstone mode . this mode becomes gapped after the transition to an adjacent incoherent state . in the orbital phase , the orbital - pseudospin goldstone mode dispersion is anisotropic and this mode becomes unstable at a finite wave vector indicating a transition to a charge - density - wave statecoteorbital , coteorbital2 . we identify the number of spin - waves and orbital modes in each phase . these later modes are active in optical absorption . the inter - ll and some intra - ll magnetoexcitons have been computed recentlysari , shizuyadipole , shizuyacyclotron , shizuyamodes01 and we comment on the difference with our results and how the presence of the inter - ll magnetoexcitons in the spectrum may complicate the detection of the intra - ll excitations . the main results of our paper are summarized in fig . 5 ( phase diagram ) , fig . 8 ( transport gaps ) and fig . 9,10 ( collective mode dispersions ) . this paper is organized in the following way . section ii introduces the two - band model of bilayer graphene with the resulting ll spectrum in finite magnetic field . section iii summarizes the hartree - fock and generalized random - phase approximations that we use to take into account the coulomb interaction and gives the formalism for the calculation of the electromagnetic absorption . our numerical results for the phase diagram , transport gaps , collective excitations and optical absorption are presented in sec . iv . in sec . v , we show how the application of an in - plane electric field allow to control the spin polarization in some phases . we conclude in sec . vi . the crystal structure of a bernal - stacked graphene bilayer is shown in fig . [ figstructurebicouche ] . each graphene layer is a two - dimensional crystal with a honeycomb lattice structure . the honeycomb lattice can be described as a triangular bravais lattice with a basis of two carbon atoms @xmath50 and @xmath51 where @xmath52 is the layer index . the two basis vectors are given by @xmath53and @xmath54 where @xmath55 @xmath56 is the lattice constant of the underlying triangular bravais lattice and @xmath57 is the separation between two adjacent carbon atoms . the distance between the two graphene layers is @xmath58 . in the bernal stacking arrangement , the upper @xmath59 sublattice is directly on top of the lower @xmath11 sublattice while the upper @xmath11 sublattice is above the center of a hexagonal plaquette of the lower layer . the brillouin zone of the reciprocal bravais lattice is shown in fig . we choose the two nonequivalent valley points to be@xmath60where @xmath61 is the valley index . ] the electronic dispersion is obtained from a tight - binding model with the parameters@xcite : @xmath41 the nearest - neighbor ( nn ) hopping in each layer , @xmath37 the interlayer hopping between carbon atoms that are part of a dimer ( i.e. @xmath62 these sites are called the high - energy sites ) , @xmath63 the interlayer nn hopping term between carbon atoms of different sublattices ( i.e. @xmath64 ) and @xmath38 the interlayer next nn hopping term between carbons atoms in the same sublattice ( i.e. @xmath65 and @xmath66 ) . the energy @xmath67 represents the difference in the crystal field between sites @xmath68 and sites @xmath69 . in this work , we neglect the trigonal warping term @xmath70 , a correct approximation at sufficiently high magnetic field@xcite . if we define the spinor@xmath71where @xmath72 creates an electron on site @xmath73 in layer @xmath74 with wave vector @xmath75 and spin @xmath76 then we can write the second - quantized tight - binding hamiltonian in the basis @xmath77 as @xmath78 with the matrix @xmath79where @xmath80 with @xmath81 is the zeeman energy . we have also included in @xmath82 an external transverse electric field that creates an electrical potential difference ( or bias ) @xmath6 between the layers . the function @xmath83 is defined by@xmath84where the summation is over the vectors connecting a site @xmath85 to its three nearest - neighbors in the same plane i.e. @xmath86 using @xmath87 and setting @xmath88 we find for the the band structure near the points @xmath89 the bands @xmath90where the momentum @xmath91 is measured with respect to @xmath92 and the effective mass is defined by@xmath93the band structure consists of four bands . in the absence of bias , the two middle bands meet at the six valley points . the two high - energy bands are separated by a gap @xmath37 from the two middle , low - energy bands . the bands @xmath94 remain degenerate at @xmath95 when @xmath96 and @xmath97 are finite if @xmath98 this degeneracy is lifted by a finite @xmath6 . for a neutral bilayer , the chemical potential is at the energy @xmath99 the low - energy excitations ( @xmath100 ) of the tight - binding model can be studied using an effective two - band model@xcite . this model gives for each valley@xmath101where we used the basis @xmath102 for @xmath49 and @xmath103 for @xmath104 and defined @xmath105 and @xmath106 in this model , the presence of a quantizing perpendicular magnetic field is accounted for by making the peierls substitution @xmath107 ( with @xmath108 ) , where @xmath109 defining the ladder operators @xmath110 and @xmath111 with the magnetic length @xmath112 we get@xmath113where@xmath114and@xmath115the effective cyclotron frequency is @xmath116 in eq . ( [ s1 ] ) , the ladder operators are defined such that @xmath117 and @xmath118 where @xmath119 with @xmath120 are the eigenfunctions of the one - dimensional harmonic oscillator . for all calculations done in this paper , we choose@xcite for the value of the parameters@xmath121we have checked that the band dispersion obtained with this choice of signs for the hopping terms is consistent with that reported in the literaturepartoens . with the magnetic field in tesla , we have @xmath122while @xmath123 in the calculation of @xmath124 we take @xmath125 for the effective dielectric constant at the position of the graphene layers . at @xmath126 t , @xmath127 when @xmath128 the landau level spectrum of @xmath129 is given by@xmath130where @xmath9 is the landau level index and sgn is the signum function . the corresponding eigenvectors of a given spin are @xmath131for @xmath132 . we use the landau gauge @xmath133 where the eigenstates are @xmath134with @xmath1 the guiding - center index . all landau levels @xmath132 are four - fold degenerate including spin and valley degrees of freedom in addition to the guiding - center degeneracy @xmath135 where @xmath136 is the area of the 2deg . the landau level @xmath5 is an exception because there are two degenerate spinors with zero energy which are given , in the basis @xmath137 for @xmath49 and @xmath138 for @xmath104 , by @xmath139it follows that @xmath5 is eight - fold degenerate . in this paper , we restrict the hilbert space to the landau level @xmath5 and use the index @xmath15 to refer to the two `` orbitals '' @xmath140 and @xmath141 . with finite values of @xmath38 , @xmath142 or @xmath6 , the valley , spin , and orbital degeneracies are lifted and the noninteracting energies become @xmath143the corresponding eigenspinors are still given by eq . ( [ bg2 ] ) . note that the structure of the sublattice spinors in eqs . ( [ bg2 ] ) is such that states from different valleys ( which are localized on different layers ) have no overlap . for @xmath5 , the layer index is thus equivalent to the valley index . [ figcomparaisonmodeles ] shows a comparison between the four - band and two - band models for the electronic dispersion in landau levels @xmath144 and valley @xmath49 using the values of the parameters given previously . the agreement between the two models is excellent for @xmath5 where the difference in energy is of the order of @xmath145 for levels @xmath146 the difference in energy between the two models is much more important . note that the two sub - landau levels of @xmath5 intersect level @xmath147 at @xmath148 ev . for the valley @xmath104 ( not shown in the figure ) the crossing occurs at a smaller bias @xmath149 ev corresponding to an electric field @xmath150 mev / nm between the layers . in our calculation we must keep the bias smaller than @xmath151 ev ( i.e. @xmath152 for @xmath153 ) for our model to be valid . and valley @xmath154 ] [ figniveaux ] shows the ordering of the four levels of a given spin in @xmath5 at finite bias . the correction @xmath155 opens a gap between the two orbital states @xmath33 and @xmath34 which is independent of the bias . the effective two - band model introduces a correction @xmath156 to this gap that has different signs in the two valleys as indicated in the figure . when combined with @xmath155 , the gap in valley @xmath104 is positive at all biases while the gap in valley @xmath49 changes sign ( level @xmath34 gets below level @xmath33 ) when @xmath157 i.e. for @xmath158 ev ( i.e. @xmath159 for @xmath160 ) . at finite bias @xmath6 . ] we now add the coulomb interaction to the noninteracting hamiltonian . hereafter , we use the same basis @xmath102 for both valleys and define the field operators @xmath161 by @xmath162and@xmath163the second - quantized noninteracting part of the hamiltonian is given by@xmath164for the second - quantized coulomb interaction , @xmath165where the coulomb potential @xmath166has the fourier transform@xmath167where @xmath168 is a two - dimensional vector in the plane of the bilayer . the terms that do not conserve the valley index in eq . ( [ vcoulomb ] ) are very small and usually neglected@xcite . in order to describe the different phases of the uniform c2deg , we define the operators@xmath169where @xmath170 creates(destroys ) an electron in a state @xmath171 . the index @xmath172 combines the spin and valley indices and we use @xmath173 and @xmath174 to refer to the specific spin or valley index of @xmath175 . the set of average values @xmath176 gives a complete description of an uniform ground state . they are the order parameters of that state . the diagonal elements @xmath177 are the filling factors of levels @xmath178 while the off - diagonal elements are the `` coherences '' . for nonuniform states , it is necessary to define the order parameters @xmath179 where @xmath180 is a reciprocal lattice vector and @xmath181 the fourier transform of @xmath182 . we refer the reader to refs . where the formalism for this case is discussed in more details . the hartree - fock hamiltonian can be written in terms of these operators by ( we adopt the convention that repeated indices are summed over ) @xmath183 where@xmath184with @xmath185 given by eqs . ( [ s10]-[s11 ] ) . the fock interaction is defined by@xmath186with the form factors@xmath187these form factors capture the character of the two different orbital states . in eq . ( [ capa ] ) , @xmath188 is the number of filled levels in @xmath189 we reserve the symbol @xmath4 $ ] for the filling factor of the c2deg . in deriving eq . ( [ hhf2 ] ) , we have taken into account a neutralizing positive background so that the only contribution from the hartree and background terms is the capacitive energy given by the term in parenthesis in eq . ( [ capa ] ) . in this term , @xmath190 is the total filling factor in valley @xmath191 detailed expressions for the hartree ( see next section ) and fock interactions @xmath192 and @xmath1 are given in appendix a of ref . . the hartree - fock energy per electron is given by@xmath193where @xmath194 is the number of electrons in the 2deg and @xmath195 are the filling factors of the two valleys . at @xmath196 the only nonzero matrix elements of the fock interactions are @xmath197and the corresponding interlayer terms which must be computed numerically . we have defined@xmath198with @xmath199 we define the single - particle matsubara green s function@xmath200where @xmath201 is the imaginary time ordering operator , such that the order parameters are given by@xmath202 the equation of motion for the green s function in the hartree - fock approximation is @xmath203where @xmath204 is the chemical potential , @xmath205 a fermionic matsubara frequency and@xmath206are the self - consistent fock potentials . the self - consistent eq . ( [ ehf ] ) can be put in a @xmath207 matrix form by defining superindices and then solved numerically in an iterative way in order to get the order parameters . the hartree - fock equation of motion for the green s function leads to the sum rule ( at @xmath208 k)@xmath209where @xmath210 is the filling factor of the @xmath211 level . by definition@xmath212 to study the collective excitations , we compute the two - particle green s functions @xmath213 in the generalized random - phase approximation ( grpa ) . in this approximation , @xmath214 is the solution of the equation @xmath215 where @xmath216 is a bosonic matsubura frequency and the hartree interaction@xmath217the two - particle green s functions @xmath218 satisfy the set of equations @xmath219 \chi _ { n_{1},n_{2},n_{3},n_{4}}^{\left ( 0\right ) a , b , c , d}\left ( \mathbf{q}% , \omega _ { n}\right ) \label{grpa2 } \\ & = & \hslash \left\langle \rho _ { n_{1},n_{4}}^{a , d}\right\rangle \delta _ { b , c}\delta _ { n_{2},n_{3}}-\hslash \left\langle \rho _ { n_{3},n_{2}}^{c , b}\right\rangle \delta _ { a , d}\delta _ { n_{1},n_{4 } } \notag \\ & & + u_{m , n_{1}}^{a , e}\chi _ { m , n_{2},n_{3},n_{4}}^{\left ( 0\right ) e , b , c , d}\left ( \mathbf{q},\omega _ { n}\right ) \notag \\ & & -u_{n_{2},m}^{e , b}\chi _ { n_{1},m , n_{3},n_{4}}^{\left ( 0\right ) a , e , c , d}\left ( \mathbf{q},\omega _ { n}\right ) . \notag\end{aligned}\ ] ] eq . ( [ grpa1 ] ) can be represented by a set of bubbles ( hartree terms ) and ladder ( fock terms ) diagrams . the function @xmath220 is the hartree - fock approximations for the two - particle green s functions . it includes the hartree - fock self - energy corrections but not the vertex corrections . note that two - particle green s functions depend only on the order parameters @xmath221 computed in the hfa . ( grpa1,[grpa2 ] ) can be solved numerically by defining superindices and then writing them in a @xmath222 matrix form . the collective excitations are then given by the poles of the retarded green s functions @xmath223 which are obtained by the analytic continuation @xmath224 of the corresponding two - particle green s functions . to derive the dispersion relations , we follow these poles as the wave vector @xmath168 is varied . we showed above that the coherent states of the c2deg can be described by the set of order parameters @xmath225 these states are also quantum hall ferromagnets ( qhf s ) and can also be described by using a pseudospin language where the two valley states @xmath226 ( @xmath227 ) are associated with valley - pseudospin up ( down ) and the two orbital states @xmath33 ( @xmath34 ) with orbital - pseudospin up ( down ) . in this language , the total spin , valley pseudospin , and orbital pseudospin components of the electron gas are given by@xmath228where @xmath229 are the pauli matrices and the total filling factor is@xmath230note that these @xmath231 fields do not provide a complete description of a state . one must also consider the @xmath232 other combinations of indices ( the @xmath233 order parameters are not all independent , however).we will use both @xmath234 and the order parameters @xmath235 to characterize the ground states of the c2deg . the coupling of the c2deg with a uniform external electric field _ in the plane of the layers _ is given by @xmath236where @xmath237 the total density is given by@xmath238fourier transforming eq . ( [ couplage ] ) and using the form factors defined in eqs . ( [ form1]-[form4 ] ) , we can show that in an homogeneous state@xcite , @xmath239 gives the dipolar coupling @xmath240 with the total dipole vector defined by @xmath241 where @xmath242is the dipole moment in valley @xmath173 with spin @xmath243 we have defined here@xmath244it is possible to control the orientation of the orbital pseudospins in the @xmath245 plane with an external electric field . the total current operator in second quantization is given by @xmath246with @xmath247the current operator @xmath248 is derived from the hamiltonian in eq . ( [ hamilp ] ) by making the peierls substitution @xmath249 and then taking the derivative with respect to the external vector potential @xmath250 @xmath251where @xmath252 this gives@[email protected] . the bare gap in valley @xmath255 and@xmath256 , \]]where @xmath257 is the noninteracting hamiltonian . to compute the electromagnetic absorption per unit area , we define the two - particle current - current green s function@xmath258which gives @xmath259 where @xmath260 and @xmath261,@xmath262 using eqs . ( [ cun],[cdeux ] ) , we have for example@xmath263and similarly for the other components . the absorption can only involve these four combinations of orbital indices whatever the polarization of the electric field of the electromagnetic wave . the retarded current - current response function @xmath264 is obtained from the analytic continuation @xmath265 and the electromagnetic absorption for an electromagnetic wave of amplitude @xmath266 linearly polarized in the direction @xmath267 is given by@xmath268 e_{0}^{2}. \label{abso}\]]this formula is valid at finite frequency only since we have neglected the diamagnetic contribution to the current . if there is no coherence in a phase , then @xmath269 in this case , we can solve analytically for the absorption because this restriction leads , from eq . ( [ grpa2 ] ) to@xmath270now , at @xmath271 the only nonzero fock interactions in the grpaequations are given in eqs . ( [ v1]-[v3 ] ) while the only hartree interactions that need to be considered are those of the form @xmath272these interlayer hartree interaction involves combinations of the form @xmath273 that give a finite contribution at @xmath271 ( and also a diverging contribution that is cancelled by the other terms ) . it follows that eq . ( [ grpa1 ] ) gives for the grpa response functions @xmath274 and @xmath275 } , \]]i.e . valley and spin must be conserved in an optically active electronic transition . since @xmath276(and a similar expression with @xmath277 for @xmath278 ) , we have easily@xmath279 /\hslash } \]]and@xmath280 /\hslash } .\]]the functions @xmath281 and @xmath282 are the response to the two circular polarizations of light . the absorption in an incoherent phase is finally given by@xmath283with @xmath284 and @xmath285 given by eq . ( [ xsip ] ) . in the numerical calculation , we introduce a small landau level width in order to get a finite value for the optical absorption . we can follow the same type of reasoning to show that , in a phase with no orbital coherence but with possibly layer and/or spin coherence , the functions @xmath286 and the absorption depends again only on @xmath287 and @xmath288 in this special case , the equation of motion for @xmath289 is@xmath290and a similar expression with @xmath277 for @xmath291 at zero bias , the qhf states follow a set of hund s rules : the spin polarization is maximized first , then the layer polarization is maximized to the greatest extent possible and finally the orbital polarization is maximized to the extent allowed by the first two rules@xcite . in this section , we study the phase transitions that occur when a _ finite _ bias ( or transverse electric field ) is turned on . fig . [ figdiagrammeglobal ] shows our numerical result for the phase diagram of the c2deg as a function of an applied transverse electric field @xmath292 for @xmath293 t and @xmath160 and for all integer filling factors @xmath4 .$ ] we indicate the eight non - interacting levels by horizontal lines and number them according to the scheme indicated in the top inset . note that the lines are only offset vertically for clarity . their position does not reflect the true ordering of the energy levels which changes with bias . we name the phases @xmath294 and @xmath295 according to the type of coherence that is present : incoherent , layer - coherent , orbital - coherent or spin - layer coherent respectively . when there is more than one incoherent phase at a given filling factor , we use the notation @xmath296 for the second phase , @xmath297 for the third phase and so on . the critical electric field for the transition between two phases is indicated by @xmath298 to @xmath299 and is in units of mv / nm . a circle on an energy level represents a fully filled level while an ellipse that connects two levels indicates a coherent superposition of these two states . we list in fig . figdiagrammeglobal some properties of each phase : spin polarization @xmath300 number of goldstone modes ( g ) , number of collective modes gapped at the zeeman energy ( z ) and number of peaks in the optical absorption spectrum ( a ) . the transition between a coherent and an incoherent phase is continuous while a transition between two incoherent phases is discontinuous . the @xmath302 phases have no coherence of any kind and so @xmath303 ( the parallel component is in the plane of the bilayer ) . each level is either full or empty and so @xmath304 and @xmath305 vary from one phase to another . the state corresponding to a specific diagram is easily read from fig . 5 . for @xmath306 , we have for example@xmath307and the order parameters @xmath308 we include the phases @xmath309 and @xmath310 in the phase diagram only to make it more complete . indeed , the bias ca nt produce any more transition after these states . but , the critical bias needed to reach these states is well outside the limits of validity of our two - band model . using eq . ( [ hfenergy ] ) , the hartree - fock energy of two adjacent incoherent phases are readily compared to extract the critical biases . we find , with @xmath311 @xmath312if we ignore the @xmath295 phases at @xmath313 , we find that the transition between the two incoherent phases occurs in the middle of the @xmath314 phase ( see sec . iv.a.4 below ) i.e. at @xmath315note that these last three results are independent of coulomb exchange corrections and so of screening corrections . ( the capacitive term comes from the hartree self - energy and is not screened . ) from eq . ( [ gap0 ] ) , we find a critical electric field @xmath316 $ ] mv / nm when @xmath317 for the transition @xmath318 . this critical field depends linearly on the magnetic field , in agreement with the experiments@xcite . experimentally , however , the slope is @xcite @xmath319 mv nm@xmath320 t@xmath321 or@xcite @xmath322 mv nm@xmath320 t@xmath320 or@xcite @xmath323 mv nm@xmath320 t@xmath320 and thus larger than the hfa value . moreover , experiments measure an offset of @xmath324 mv / nm at @xmath325 t@xcite . this offset ca nt be captured by our hfa which is only valid at sufficiently large magnetic field where landau level mixing can be neglected . apart from the extra @xmath326 corrections , eqs . ( [ gapm1]-[gapp1 ] ) are identical to those given by gorbar et al.@xcite . the second type of phase , @xmath328 , has layer coherence between two states with the same spin and orbital indices and so @xmath329 but @xmath330 the tilt angle of the pseudospin vector @xmath331 varies with bias in this phase but @xmath305 and @xmath332 are constant . an example is phase @xmath333 which is described by @xmath334where the coefficients @xmath172 and @xmath335 depend on the bias and are related by @xmath336 with increasing bias , @xmath337 and @xmath338 continuously . the level populations and the coherence in @xmath333 are given by @xmath339fig . [ figpoplm1 ] shows how these variables depend on the transverse electric field for the similar phase @xmath340 the populations of the coherent levels vary linearly with the bias in all @xmath328 phases with the exception of @xmath341 where the variation is not exactly linear . in @xmath342 for example , @xmath343where the critical bias is , to order @xmath344@xmath345these critical biases all scale with the magnetic field as @xmath346 in phases @xmath347 there is a layer coherence in orbitals @xmath33 and @xmath348 phase @xmath349 for example , is described by the state@xmath350the critical bias @xmath351 for the transitions @xmath352 has a complicated analytical expression that we do not reproduce here but the numerical values of the critical electric field for @xmath293 t and @xmath160 is indicated in fig . [ figdiagrammeglobal ] . phases with layer coherence occur in a small range of bias and at very small bias because the interlayer separation @xmath353 is very small in bilayer graphene and so is the capacitive energy . in semiconductor bilayers , @xmath354 can be of order unity and interlayer coherence can survive to a much higher bias@xcite . ] for @xmath356 the ordering of the energy levels @xmath15 is reversed ( this change is not shown in fig . [ figdiagrammeglobal ] ) . when this happens , the kinetic energy is minimized by filling level @xmath34 before @xmath357 however , this increases the coulomb exchange energy because @xmath358 the c2deg optimizes its energy by creating a coherent superposition of @xmath33 and @xmath34 with the same valley and spin indices . we use the notation @xmath359 for such a phase . an example is phase @xmath360 which is described by @xmath361 in this state , @xmath362 and @xmath363 and @xmath332 are constant . it is now the tilt angle of the pseudospin vector @xmath364 that varies with bias . orbital coherence begins when the bare energy of the state @xmath365 is equal to that of state @xmath366 at @xmath367 this occurs when@xmath368and the critical bias does not depend on the magnetic field , coulomb interaction or on the value of the dielectric constant . we find @xmath369 mev i.e. @xmath370 mv / nm which is in the range of validity of the two - band model . the orbital phase survives until a transition to an incoherent phase occurs at the critical bias@xmath371since @xmath372 is the onset of the orbital phase , we see that the range of existence of the orbital phase scales as @xmath373 . part of this range is outside the limit of validity of our model . [ figpopo3 ] shows how the populations and coherence vary with the transverse electric field in phase @xmath360 . the same behavior is found in phase @xmath374 we remark that in previous work where the spin degree of freedom is frozen@xcite , the orbital coherent phase occurs at @xmath375 and @xmath376 . ] the orbital phase exists in a large range of bias . as we explained in sec . iii.e , a finite orbital coherence implies a finite density of electric dipoles in the plane of the layers . the orientation of these dipoles can be controlled by an electric field in the plane of the layersshizuyadipole . on of us has studied in detail the interesting properties of this state@xcite . for example , the collective mode associated with the orbital coherence is highly anisotropic . this mode softens at a finite wave vector in the direction perpendicular to the dipoles when the bias is increased . this suggests a transition to a charge - density - wave state . in the hartree - fock approximation , it was found that this transition is preempted by a transition to a crystal phase with one electron per site and a skyrmion - like pseudospin texture of the orbital pseudospin at each crystal site . as the bias is increased , the crystal state is followed by a helical state where the orbital pseudospin rotates along one spatial direction . in both phases , the total electronic density is modulated spatially but the local filling factor is not . by further increasing the bias , the crystal state and then the uniform states are recovered . the critical electric fields for the transition to the uniform ( up ) , skyrmion crystal ( skp ) and helical phases ( hp ) are given bycoteorbital2,note2 : @xmath377the phase diagram in @xmath378 and @xmath360 is symmetric with respect to the center of the helical phase . we remark that this sequence of phase transitions is similar to that observed in a thin film of the helical magnet fe@xmath379co@xmath379si when a perpendicular magnetic field is increased@xcite . it has been showncoteorbital2 that the hamiltonian of the c2deg in the orbital phase contains a dzyaloshinskii - moriya ( dm ) interaction@xcite that is responsible for the rotation of the pseudospins . its origin in bilayer graphene is purely coulombic while the dm interaction comes from spin - orbit coupling in fe@xmath379co@xmath379si . we do not find any sign of instability in the collective mode dispersions for the other phases in fig . [ figdiagrammeglobal ] . however , we remark that phases with lower energy than those considered in this figure are possible . in order to establish the phase diagram of the c2deg , we choose a set of possible ground states and compare their energies . this does not ensure however that the true ground state is amongst the states that we have chosen to compare ! for the incoherent states , this is not a problem because there is a finite number of states to compare . but for the nonuniform states , the number of possible ground state is enormous . the fourth type of phase has coherence between two states with the same orbital index but different spin and layer indices . we use for these phases the notation @xmath381 an example is phase @xmath382 where the ground state is @xmath383because the coherence is now between two states with different spin _ and _ layer indices , i.e. @xmath384 and @xmath385 in @xmath386 , we can not describe the change with bias as the tilting of one of the pseudospin @xmath387 or @xmath388 in fact this state has @xmath389 both @xmath363 and @xmath390 vary with bias however . this phase is characterized by the order parameter @xmath391 the variation of @xmath392 and @xmath393 with bias in this phase is similar to that shown in fig . [ figpopo3 ] . the critical biases for the beginning ( @xmath394 ) and end ( @xmath395 ) of the @xmath396 phases are given by @xmath397 where @xmath398 thus , @xmath399 these critical biases scale linearly with the magnetic field . the phase @xmath314 is the ground state in a small range of bias of the order @xmath400 which is approximately @xmath401 mev for @xmath402 a sufficiently large bias is necessary for spin down states to cross the spin up states and produces a @xmath295 phase . the @xmath403 is special because it involves coherence in both @xmath33 and @xmath34 . as for @xmath341 , the exact critical bias in this case has a complicated analytical expression which we do not reproduce here . we indicate for each phase in fig . [ figdiagrammeglobal ] the spin polarization @xmath332 . the polarization is constant in all phases with the exception of the phases @xmath295 where it varies continuously between the two numbers indicated . the biggest change in @xmath332 and @xmath363 occurs at filling factor @xmath22 where the c2deg makes a transition from a fully spin polarized and layer unpolarized gas ( @xmath404 ) at small bias to a spin unpolarized and layer polarized gas ( @xmath405 ) at large bias . for @xmath406 the @xmath295 phase interpolates between @xmath407 and @xmath408 the only jumps in @xmath332 occur at the transitions @xmath409 where the system goes from a spin polarized to a spin unpolarized state . another quantity that is accessible experimentally is the transport gap @xmath410 which is defined by the difference in energy between the first empty state and the last filled state of the hartree - fock hamiltonian . it was shown previously@xcite that the gap at zero bias follow the hierarchy @xmath411 this implies that the first plateau to appear when the magnetic field is turned on has @xmath412 at larger field , the @xmath413 plateaus appear and at still larger field , the @xmath414 plateaus . this is indeed what is seen experimentallyfeldman , martin , zhao . [ figglobalgap ] shows the hartree - fock gaps as a function of the transverse electric field for the different phases of the c2deg . for this figure , we have taken @xmath293 t and @xmath402 in some phases , one or more level crossing occurs that change the behavior of the gap . this is clearly visible for @xmath415 in fig . [ figglobalgap ] ( a ) and for @xmath416 in fig . [ figglobalgap ] ( c ) . with the exceptions of the phases where coherence occurs in two levels ( @xmath417 ) , it is possible to obtain a simple analytical expression for the gap . we list these expressions below . when one or more level crossings occur , we use the notation @xmath418 to denote the different behaviors of the gap and @xmath419 for the values at which the level crossings occur@xmath420 in the incoherent phases with @xmath421 @xmath422 @xmath423 where , for @xmath424@xmath425for the incoherent phases with @xmath426@xmath427with@xmath428while for @xmath429@xmath430with @xmath431for @xmath432 we find : @xmath433with @xmath434the gap changes rapidly in the @xmath396 phase while it is almost independent of the bias in phase @xmath403 . its value in the middle of the @xmath435 phases is given approximately by@xmath436where @xmath437 are defined in eqs . ( [ gapm1],[gapp1 ] ) . the gaps are twice as big in phases @xmath438 than in phases @xmath439 and vary more rapidly with bias in the former than in the latter . the presence of the @xmath295 phase smoothens the jump of the gap in the transition from @xmath440 to @xmath441 the gap is independent of the bias in the orbital phases @xmath378 and @xmath360:@xmath442 for the @xmath328 phases , the gaps at zero bias are given approximately by@xmath443the correction @xmath444 is very small and the gaps at zero bias for @xmath445 and @xmath446 are almost equal . ( the difference comes from the fact that the coherence is not in the same orbital in @xmath447 and @xmath448 . ) these gaps are not shown in fig . figdiagrammeglobal because the corresponding phases occur at very small biases . the gap increases ( decreases ) with bias in phases @xmath449 ( @xmath450)@xmath420 it is almost constant in @xmath341 . the main contribution to all gaps is the coulomb exchange interaction . with @xmath451 our gaps for phases @xmath452 agree with those of gorbar _ et al_.@xcite if screening is neglected in their calculation . for @xmath453 and @xmath454 however , our exchange correction is @xmath455 which is consistent with ref . while gorbar _ et al_. have @xmath456 at @xmath457 we find for @xmath458 the gap @xmath459 mev for @xmath460 t and @xmath461 while , with static screening , gorbar et al . find @xmath462 mev . similarly , for phase @xmath463 the hfa gives @xmath464 mev while the result with static screening is @xmath465 mev . static screening leads to a reduction of the gap by a factor of at least @xmath466 dynamical screening and landau mixing corrections , however , increase the gaps calculated with static screening by a factor of two to three@xcite . as for the behavior of the gap with bias . 5 of gorbar _ et al_.gorbarnutot shows that , with screening , the gap of the phase @xmath467 increases with the electric field even when the correction @xmath468 is neglected . the slope is approximately @xmath401 nm - c for @xmath460 t and @xmath469 . the ( unscreened ) hfa predicts a slope of @xmath470 nm - c . when both screening and @xmath156 are considered , the gaps for @xmath471 will probably increase with bias ( contrary to the behavior illustrated in our fig . [ figglobalgap](b ) ) but the rapid change of the gap in the @xmath396 will still be present . the energy gaps obtained from local compressibility measurements on suspended bilayer graphene by martin _ et al_.@xcite are of size @xmath472 @xmath473 $ ] mev , @xmath474 @xmath473 $ ] mev and @xmath475 @xmath473 $ ] mev ( with less data points in this case ) . a more recent transport experiment by velasco _ et al_.@xcite on suspended bilayer graphene with a higher mobility reports a larger gap @xmath476 @xmath473 $ ] mev . the measured gaps scale linearly with the magnetic field contrary to the hfa prediction . in fact , gorbar _ et al_ .. @xcite have shown that a linear scaling is obtained if static screening is considered . ( one set of experiments at higher magnetic field reported gaps that scaled as @xmath477 however@xcite . ) experiments@xcite show that @xmath478 ceases to be quantized at @xmath479 in the region corresponding to the @xmath295 phase and at @xmath480 and @xmath376 in the region around zero bias . a possible explanation is that the conductance quantization is broken by disorder in the regions corresponding to a minimum of the gap@xcite . however , that argument does not seem to work at @xmath481 where the screened hfa gap increases continuously with the bias and is also not compatible with our unscreened result . the kink in the behavior of @xmath415 are due to level crossings . a value of @xmath482 is assumed for the dielectric constant and @xmath293 t. ] each phase of the c2deg is characterized by a set of collective excitations . the number of dispersive modes when @xmath483 levels are filled is @xmath484 . we have calculated the dispersion relation of these modes using the grpa described in sec . iiii . in the limit @xmath485 , the vertex corrections vanish and the response function @xmath486 where @xmath487 is the response function evaluated in the hfa . thus , in this limit , the collective mode frequencies must correspond to transitions between a filled and an empty eigenstate of the hartree - fock hamiltonian . at finite value of @xmath488 some modes mix together and it becomes difficult to identify their character ( layer , orbital , spin transitions , etc . ) . our numerical results are summarized in fig . [ figmodes ] for the coherent phases and in fig . [ figmodeio ] for the incoherent phases . the coherent phases sustain one gapless ( goldstone ) mode . the number of goldstone modes is indicated for each phase in fig . [ figdiagrammeglobal ] . for example , in phase @xmath489 this mode is due to the fact that the layer pseudospin @xmath490 can rotate freely around the @xmath491 axis . the same situation occurs for the coherent phase @xmath359 where again the orbital pseudospin @xmath364 can rotate freely around the @xmath492 axis . phases @xmath341 support coherence in both @xmath493 and @xmath34 orbitals and we can define a layer pseudospin @xmath494 for @xmath33 and @xmath495 for @xmath496 the goldstone mode in this case correspond to an in - phase rotation of both pseudospins . alternatively , we can see this mode as a fluctuation of the relative phase of the two order parameters @xmath497 and @xmath498 in @xmath499 or @xmath500 and @xmath501 in @xmath502 . the dispersion off all gapless modes ( with the exception of the gapless mode in the orbital phase ) is linear in wave vector at very small wave vector i.e. for @xmath503 phases @xmath341 have the same collective mode spectrum . the dispersion of the goldstone mode in phases where the coherence occurs in the orbital @xmath34 ( i.e. @xmath504 ) has a roton minimum while there is none if the coherence occurs in @xmath33 ( i.e. @xmath505 ) . this is due to the particular form factor for @xmath34 involved in the coulomb matrix elements ( see eqs . ( [ form1]-form4 ) ) . phases @xmath506 contain coherence in both @xmath33 and @xmath507 and a small shoulder appears in the dispersion . for phase @xmath508 the goldstone mode is the famous layer - pseudospin - wave mode which has been extensively studied in semiconductor bilayerlayerpseudospin at filling factor @xmath481 and detected experimentallyspielman . in semiconductor bilayer , this mode becomes soft at a finite wave vector as the separation between the layers is increased ( around @xmath509 ) . in bilayer graphene @xmath510 and the layer - coherent phases are stable . the only instability in the collective modes is seen in the orbital phases @xmath378 and @xmath511 for @xmath451 the goldstone mode of @xmath512and @xmath513 ) has a quadratic dispersion at zero bias and becomes unstable@xcite at finite bias . ( we have checked that these conclusions remain valid if @xmath514 is finite but small . ) a consequence of this instability is that the c2deg is expected to go from a smectic ( non - homogeneous ) phase at low temperature to an isotropic phase at higher temperature . the smectic phase would lead to anisotropic electrical transport . the dispersion of the gapless orbital pseudospin - wave mode was studied in detail for the uniform phase@xcite as well as for the crystal and helical phases@xcite . in the uniform phase , it has a strongly anisotropic dispersion : linear in the direction of the orbital pseudospins and quadratic in the other directions i.e.@xmath515where @xmath516 is the angle between the wave vector and the @xmath517 axis . the goldstone mode softens at a finite wave vector @xmath518 in the direction perpendicular to the orbital pseudospins at a bias @xmath519 mev ( @xmath520 mv / nm ) . this suggests a transition to a charge - density - wave state . as we explained above , this transition is preempted by a transition to a crystal phase at @xmath521 mev . the number of spin - wave modes gapped at @xmath31 is indicated for each phase in fig . [ figdiagrammeglobal ] . all coherent phases with the exception of @xmath522 have two modes gapped at the zeeman energy @xmath31 at @xmath523 their degeneracy is lifted at finite wave vector . because of the occupation of the levels in @xmath378 and @xmath524 only one intralayer spin - flip transition is possible in theses phases . the incoherent phases can have @xmath525 or @xmath526 modes gapped at @xmath527 to be gapped at @xmath528 these modes must involve transitions within the same valley and orbitals . from fig . [ figdiagrammeglobal ] , it is easy to see that no intralayer and intraorbital spin - flip transition is possible for @xmath529 and @xmath530 and that the occupation of the levels permit only one such mode in @xmath531 in @xmath532 two transitions seem possible but they are degenerate and the coupling between them leaves one mode gapped at @xmath31 and the second mode has its frequency renormalized . the same mechanism operates in phases @xmath533 resulting in two modes gapped at @xmath527 in the absence of coulomb interaction , the dynamical conductivity has intra - octet peaks at the bare gap energy @xmath534 in addition to the inter - landau - level peaks which do not appear in our calculation . figabso shows the absorption in different phases when coulomb interaction is considered . the number of absorption peaks is also indicated for each phase in fig . [ figdiagrammeglobal ] and , in figs . [ figmodes],figmodeio , we identify by ( blue ) arrows the modes that lead to optical absorption . in the incoherent phases , optical absorption is possible only for transitions that occur between states with the same valley and spin indices but different orbital indices . the incoherent phases have one or zero absorption peak . the latter case occurs when both levels @xmath15 with the same layer and spin indices are filled ( phases @xmath535 ) . when level @xmath33 is filled , the absorption , as shown in sec . iii.g is exactly at @xmath536 when level @xmath34 is filled , the absorption is at @xmath537 i.e. affected by exchange corrections . the former case applies to most of the incoherent phases in our phase diagram . the latter case applies to phases @xmath538 and @xmath310 which occur at very high bias and are outside the limits of validity of our two - band model . in those phases , level @xmath34 is filled because it is below @xmath33 in energy . in phases @xmath539 , the two allowed transitions have the same energy and the intensity of the absorption peak is doubled . [ figabso ] ( a ) shows the absorption as a function of frequency in phase @xmath540 at finite bias . the absorption is concentrated in one strong peak at the frequency @xmath541 \right ) \times 10^{11}\text { hz } \notag\end{aligned}\]]at @xmath293 t and @xmath402 the absorption frequency varies widely with bias@xmath420 for example , at the onset of the transition from @xmath542 to @xmath543 the frequency @xmath544 ( see eq . ( [ orbit ] ) ) while @xmath545 hz at the onset of the @xmath546 phase . by contrast , if the absorption occurs in layer @xmath547 the minimum frequency is @xmath548 hz since the absorption frequency must increase with bias in this case . in the @xmath522 phases , the goldstone mode has orbital character ( i.e. electric dipole fluctuations ) but does not lead to absorption at _ finite _ frequency . interestingly , the coherent phases @xmath328 and @xmath295 show two absorption peaks ( see fig . [ figabso ] ) . the second peak in @xmath549 is extremely weak however and is absent at zero bias . the two weak peaks in @xmath341 are also extremely weak and disappear at zero bias . in @xmath550 the first absorption peak at zero bias is exactly at @xmath551 while for @xmath552 the frequency is at @xmath553 which is slightly shifted from @xmath554 the layer eigenstates of the hartree - fock hamiltonian at zero bias are the symmetric ( @xmath136 ) and antisymmetric ( @xmath555 ) combinations of @xmath104 and @xmath556 using eqs . ( [ chizero ] ) and ( [ grpa2 ] ) , we can show from the grpa equations that the response functions that enter in eq . ( [ abso ] ) for the absorption in the absence of orbital coherence and at zero bias are of the form @xmath557 or @xmath558 where @xmath173 now stands for the @xmath136 or @xmath555 layer combinations . that is , the layer combination is conserved in the absorption at zero bias and in the absence of orbital coherence . it follows that , in phase @xmath333 , the state @xmath559 is filled and only the transition @xmath560 is optically active . for @xmath349 no transition conserving the valley index ( @xmath136 or @xmath555 ) is possible and for @xmath561 the only allowed transition is between @xmath562 since levels @xmath563 are filled . the same argument applies to the spin down states . at finite but small bias , the @xmath136 and @xmath555 layer combinations are replaced by bonding and anti - bonding combinations i.e. @xmath564 and @xmath565 for example where @xmath172 and @xmath335 depend on the bias and on the orbital index @xmath566 . the strong peak in the absorption corresponds to the transition @xmath567 for @xmath333 and to @xmath568 for @xmath569 by contrast to the zero bias case , the absorption given by eq . ( [ absoabso ] ) contains response functions which are not just of the form @xmath570 with @xmath571 so that transitions that do not conserve the layer combination @xmath572 are weakly optically active@xcite . a second peak appears in the absorption spectrum which corresponds to the transition @xmath573 for @xmath333 and @xmath340 the two weak peaks in @xmath499 come from the transitions @xmath574 and @xmath575 fig . [ figabso ] ( b ) shows the absorption in phases @xmath403 and @xmath576 in @xmath577 we find by analyzing the eigenvectors of the modes involved in the optical absorption that the strong peak corresponds to the transition @xmath578 @xmath579 and the weak peak to the transition @xmath580 @xmath581 where @xmath582 and @xmath583 and @xmath584 depend on the bias and on the orbital index @xmath566 . the layer combination here are between two states with opposite spin orientations . the electromagnetic absorption in the crystal and helical phases are much more complex and was discussed previously@xcite . in the helical phase , for example , the absorption depends on the orientation of the polarization of the electromagnetic wave in the @xmath585 plane . t and @xmath586 in phases ( a ) @xmath333 ( @xmath587 mev)@xmath588 @xmath467 ( @xmath589 mev ) and @xmath499 ( @xmath590 mev ) ; ( b ) @xmath403 ( @xmath591 mev ) and @xmath382 ( @xmath592 mev)@xmath420 ] the collective mode dispersions computed here for the intra - landau level ( @xmath0 ) transitions should be combined with the dispersion of the inter - landau level magnetoexcitons computed by sari and tke@xcite and by shizuya@xcite to get a complete picture of the absorption for the c2deg in bilayer graphene . in this section , we consider the effect of adding a uniform in - plane electric field to the hartree - fock hamiltonian of eq . ( [ hhf2 ] ) . the coupling between the c2deg and this external electric field is given by @xmath593 where @xmath594 is the total dipole moment of the electron gas given by eq . ( [ dipole ] ) . the main effect of @xmath595 is to induce orbital coherence . we consider here the case of @xmath596 but a similar effect occur at other filling factors and will be discussed elsewhere@xcite . we set @xmath597 in the ground state , the electric dipoles are aligned with @xmath598 with finite @xmath599 the @xmath382 state is replaced by a state with spin , orbital , and layer coherences i.e. @xmath600 . this is represented by the inset in the top - left corner of fig . [ multiferro ] . the wave function of the ground state becomes @xmath601with @xmath602we show in fig . [ multiferro ] how the different polarizations @xmath603 and @xmath604 change with bias in @xmath382 when @xmath605 is increased . the other polarizations are zero . the spin(layer ) polarization @xmath606 increases(decreases ) with with @xmath607 until @xmath608 mv / nm where it remains constant . the orbital coherence has not yet saturated at @xmath609 . the ground state above this critical electric field is represented in the inset at the top - right corner of fig . [ multiferro ] . it is interesting that the _ spin _ polarization can be varied in this phase by an external _ electric _ field . the state @xmath610 has one gapless goldstone mode and its dispersion is anisotropic in wave - vector space ( as is the dispersion of the other modes ) . by contrast , if we apply @xmath605 to a phase @xmath359 , the u(1 ) symmetry of the dipoles in the @xmath245 plane is broken . the orbital pseudospin @xmath364 is then forced to align with @xmath595 and the goldstone mode is gappedshizuyadipole . the absorption in @xmath610 shows two peaks as in @xmath611 the first , low - energy , peak is shown in fig . [ absoslo ] for two different orientations of the electromagnetic wave polarization . clearly , the absorption is anisotropic in phase @xmath600 by contrast to all the other uniform states that we studied before . the bias @xmath6 has been taken near the middle of the @xmath612 phase where @xmath613 parameters are @xmath293 t and @xmath614 phase for two different polarizations of the electromagnetic field . ] in this work , we have derived the phase diagram of the c2deg in a bernal - stacked graphene bilayer . for the non - interacting hamiltonian , we used a tight - binding model with the hopping terms @xmath615 and @xmath67 and introduced a potential bias between the two layers given by @xmath616 the coulomb interaction was treated in the hartree - fock approximation . to reduce the complexity of the problem , we used an effective two - band model which described the low - energy behavior of the c2deg and is valid for @xmath617 we also restricted the hilbert space to the @xmath5 landau level only and worked at zero temperature . our method allows us to include both coherent and incoherent phases in the phase diagram . indeed , we found phases with layer coherence at small bias , spin and layer coherence at intermediate bias and orbital coherence at large bias . the application of a parallel electric field , as we showed , can also lead to a state with orbital , layer and spin coherence . we have included in our analysis the hopping parameter @xmath38 which is often neglected in theoretical calculations . in our calculations , we find that the phase diagram in sensitive to the precise value of this parameter . if the value of this parameter is modified , the phases that we have discussed are still present in the phase diagram but they occur at different bias . moreover , other phases may appear . for example , with @xmath618 the orbital - coherent phase becomes possible at @xmath596 and a new phase with valley and orbital coherence appears as discussed in ref . . we have written down the ground - state wave function for each phase in the global phase diagram of the c2deg . we have also calculated for each phase the transport gap , the spin polarization , the collective mode dispersions and the electromagnetic absorption spectrum . the change in these properties from one phase to another should facilitate their experimental identification . strictly speaking , however , our results are only valid within the limits of validity of the approximations listed above . in particular , we have neglected screening corrections which are known to reduce significantly the transport gaps . these corrections were considered for the incoherent phases in ref . . in principle , these screening corrections should be smaller at larger magnetic field . the stability of the different phases that we found should also be studied by considering quantum and thermal fluctuations as well as disorder effects . more subtle corrections specific to graphene have also been considered by shizuya@xcite . the quantum fluctuations of the dirac sea ( the filled landau levels from the valence band ) have been shown to be sizable and to lead to corrections of the energy of the octet of states in @xmath5@xcite . according to shizuya , the orbital degeneracy of the zero - energy levels is lifted by coulombic vacuum fluctuations , leading to an appreciable shift and splitting of the @xmath33 and @xmath34 levels and to a negative capacitance effect that blocks the rotation of the valley pseudospins . the negative capacitance effect appears when the full four bands of the tight - binding model are considered . 99 for a review on bilayer graphene , see edward mccann and mikito koshino , arxiv:1205.6953 ; johan nilsson , a. h. castro neto , f. guinea , and n. m. r. peres , phys . b * 78 * , 045405 ( 2008 ) and ref . below . e. v. gorbar , v. p. gusynin , and v. a. miransky , phys . rev . b * 81 * , 155451 ( 2010 ) ; i d . , jetp letters , * 91 * , 314 ( 2010 ) ; e. v. gorbar , v. p. gusynin , v. a. miransky , and i. a. shovkovy , phys . rev . b * 85 * , 235460 ( 2012 ) . h. a. fertig , phys . b * 40 * , 1087 ( 1989 ) ; a. h. macdonald , p. m. platzman and g. s. boebinger , phys . * 65 * , 775 ( 1990 ) ; r. ct , l. brey , and a.h . macdonald , phys . b * 46 * , 10239 ( 1992 ) ; xiao - gang wen and a. zee , phys . rev . lett . * 69 * , 1811 ( 1992 ) ; j. p. eisenstein and a. h. macdonald , nature ( london ) * 432 * , 691 ( 2004 ) . we remark that the transitions that do not conserve the layer combinations @xmath572 are absent of the spectrum if the small term @xmath622 is artificially set to zero in @xmath623 in eq . ( [ absoabso ] ) .

observations from the rossi x - ray timing explorer have shown the existence of high frequency quasi - periodic oscillations ( hfqpos ) in the x - ray flux from accreting black hole binary systems . in at least two systems , these hfqpos come in pairs with a 2:3 frequency commensurability . we propose a simple `` hot spot '' model to explain the position and amplitude of the hfqpo peaks . using the exact geodesic equations for the kerr metric , we calculate the trajectories of massive test particles , which are treated as isotropic , monochromatic emitters in their rest frames . photons are traced from the accretion disk to a distant observer to produce time- and frequency - dependent images of the orbiting hot spot and background disk . the power spectrum of the x - ray light curve consists of multiple peaks at integral combinations of the black hole coordinate frequencies . in particular , if the radial frequency is one - third of the azimuthal frequency ( as is the case near the inner - most stable circular orbit ) , beat frequencies appear in the power spectrum at two - thirds and four - thirds of the fundamental azimuthal orbital frequency , in agreement with observations . we also model the effects of shearing the hot spot in the disk , producing an arc of emission that also follows a geodesic orbit , as well as the effects of non - planar orbits that experience lens - thirring precession around the black hole axis . by varying the arc length , we are able to explain the different qpo features observed in `` type a '' and `` type b '' x - ray outbursts from xte j1550 - 564 . in the context of this model , the observed power spectra allow us to infer values for the black hole mass and angular momentum , and also constrain the parameters of the model , such as the hot spot size and luminosity .

astro-ph0309458

in the past decade , observations of x - ray emission from accreting neutron stars and black holes have introduced new possibilities for astrophysical tests of fundamental physics . recent discoveries made by satellites such as asca , rxte , bepposax , chandra , and xmm - newton provide direct evidence for strong - field gravitational effects in compact binary systems and active galactic nuclei ( agn ) . these results include doppler - broadened iron k@xmath0 fluorescent emission from microquasars @xcite and millisecond variability of the x - ray flux from black holes in low - mass x - ray binaries @xcite . these measurements give the exciting prospect for determining a black hole s mass and spin , as well as tests of general relativity in the strong - field regime . the strong gravitational fields near a black hole ( bh ) introduce significant deviations from newtonian physics , including the existence of an inner - most stable circular orbit ( isco ) , a feature absent in the classical kepler problem . since accreting gas can efficiently lose energy and angular momentum only outside of the isco , the hydrodynamic and radiative behavior of the inner accretion disk should be strongly dependent on the structure of the space - time metric near the isco . the famous `` no hair '' theorem states that the only observable features of an electrically neutral black hole are functions of its mass @xmath1 and specific angular momentum @xmath2 . by understanding the behavior of matter near the isco , we can determine the mass and angular momentum , and thus completely describe the black hole . many authors have approached the problem of accretion in compact binaries with a variety of different methods , including early analytic models by @xcite and @xcite . some have simplified the hydrodynamics in favor of a flat , thin , steady - state disk and a more detailed treatment of general relativistic effects @xcite . to include dynamic effects , essential for modeling qpos , others have included magnetohydrodynamics ( mhd ) in a pseudo - newtonian model @xcite or with smoothed particle hydrodynamics @xcite . a family of perturbative models has given rise to the field of diskoseismology @xcite , where different global modes in the disk oscillate with different frequencies . @xcite propose a magnetic coupling between the rotating black hole and the accretion disk as a means of producing high - frequency qpos , analogous to the blandford - znajek process @xcite used to describe agn . recent observations of commensurate relationships in the high - frequency qpos of black hole accretion disks @xcite , as well as the longstanding puzzles of the frequency variability of qpo peaks and their correlations with x - ray flux and energy @xcite , motivate more detailed study of the qpo phenomenon as a means to determining the black hole parameters . we have developed a model that is a combination of many of the above approaches in which additional physics ingredients can be added incrementally to a framework grounded in general relativity . the model does not currently include scattering , radiation pressure , magnetic fields , or hydrodynamic forces , instead treating the disk as a collection of cold test particles radiating isotropically in their respective rest frames . the trajectories of emitted photons are integrated through the metric to a distant observer to construct time - dependent images and spectra of the disk . the dynamic model uses the geodesic trajectory of a massive particle as a guiding center for a small region of excess emission , a `` hot spot , '' that creates a time - varying x - ray signal , in addition to the steady - state background flux from the disk . this hot spot model is motivated by the similarity between the qpo frequencies and the black hole ( or neutron star ) coordinate frequencies near the isco @xcite as well as the suggestion of a resonance leading to integer commensurabilities between these coordinate frequencies @xcite . @xcite investigated primarily the qpo frequency pairs found in lmxbs with a neutron star ( ns ) accretor , but their basic methods can be applied to black hole systems as well . @xcite have presented a thorough analysis of this hot spot model for a collection of ns binaries for which pairs of qpos have been observed . based on a number of experimental and theoretical arguments , they conclude that the geodesic hot spot model is not a physically viable explanation for the observed ns qpos . for low to moderate eccentricity orbits , the coordinate frequencies simply do not agree with the qpo data . for highly eccentric geodesics , they argue that the relative power in the different frequency modes are qualitatively at odds with the observations . furthermore , they show that hydrodynamical considerations place strong constraints on the possible size , luminosity , coherency , and trajectories of the hot spots . many of these points are addressed in our version of the hot spot model . also , by including full 3-dimensional ( 3d ) ray - tracing , we can quantitatively predict how much qpo power will be produced by a hot spot of a given size and emissivity moving along a geodesic orbit near the isco . along with the special relativistic beaming of the emitted radiation , we find that strong gravitational lensing can cause high - amplitude modulations in the light curves , even for relatively small hot spots . the issues of differential rotation and shearing of the emission region are addressed below when we consider the generalization of the hot spot model to include arcs and non - planar geometries . perhaps the most powerful feature of this hot spot model is the facility with which it can be developed and extended to more general accretion disk geometries . in addition to providing a possible explanation for the commensurate hfqpos in at least two systems ( xte j1550 - 564 and gro j1655 - 40 ) , the hot spot model with full general relativistic ray - tracing is a useful building block toward any other viable model of a dynamic 3d accretion disk . within the computational framework of the kerr metric , we can investigate many different emission models and compare their predicted x - ray spectra and light curves with observations . for example , our ray - tracing code could be used in conjunction with a 3d mhd calculation of the accretion disk to simulate the time - dependent x - ray flux and spectrum from such a disk . it is with this motivation that we present the initial results of the model . section 2 describes the relativistic ray - tracing methods and discusses numerical techniques . in section 3 , we develop the basic model of a steady - state disk with the potential application to broad iron k@xmath0 emission . section 4 introduces the basic hot spot model for circular geodesics and shows the effect of binary inclination and black hole spin on the qpo power spectrum . in section 5 the model is extended to trajectories with non - zero eccentricity and inclination , as well as elongated hot spots or arcs , giving a set of model parameters that best fit the qpo data from a large number of outbursts from xte j1550 - 564 . in section 6 we present our conclusions and a discussion of future work . we begin by dividing the image plane into regularly spaced `` pixels '' of equal solid angle in the observer s frame , each corresponding to a single ray . following the sample rays backward in time , we calculate the original position and direction that a photon emitted from the disk would require in order to arrive at the appropriate position in the detector . the gravitational lensing and magnification of emission from the plane of the accretion disk is performed automatically by the geodesic integration of these evenly spaced photon trajectories , so that high magnification occurs in regions where nearby points in the disk are projected to points with large separation in the image plane . to model the time - varying emission from the disk , each photon path is marked with the time delay along the path from the observer to the emission point in the disk . to integrate the geodesic trajectories of photons or massive particles , we use a hamiltonian formalism that takes advantage of certain conserved quantities in the dynamics . the resulting equations of motion do not contain any sign ambiguities from turning points in the orbits , as are introduced by many classical treatments of the geodesic equations in the kerr metric . we define a hamiltonian function of eight phase space variables @xmath3 and an integration variable ( affine parameter ) @xmath4 along the path length . for a general space - time metric @xmath5 with inverse @xmath6 , we can define a hamiltonian @xmath7 quadratic in the momenta as @xmath8 where the rest mass @xmath9 is a constant ( @xmath10 for photons , @xmath11 for massive particles ) . applying hamilton s equations from classical mechanics , we reproduce the geodesic equations : @xmath12 @xmath13 for any metric , the hamiltonian @xmath7 is independent of the affine parameter @xmath4 , allowing us to use one of the coordinates as the integration parameter and reduce the dimensionality of the phase space by two . we use the coordinate @xmath14 as the time coordinate for the six dimensional phase space @xmath15 . the corresponding hamiltonian is @xmath16^{1/2}\ ] ] with equations of motion @xmath17 @xmath18 we have thus reduced the phase space to the six - dimensional tangent bundle @xmath15 . moreover , because the metric is independent of @xmath14 , @xmath19 is also an integral of motion . in boyer - lindquist coordinates @xmath20 , the kerr metric may be written @xmath21 which allows a relatively simple form of the inverse metric @xmath22 for a black hole of mass @xmath1 and specific angular momentum @xmath23 , we define ( in geometrized units with @xmath24 ) @xmath25 \sin^2\theta.\end{aligned}\ ] ] as a check , we see that the metric reduces to the well - known schwarzschild metric in the limit @xmath26 . in the limit @xmath27 , it reduces to flat space - time with hyperbolic - elliptical coordinates . with this form of the metric , the hamiltonian @xmath28 can be written @xmath29 this new hamiltonian is also independent of @xmath30 ( azimuthally symmetric space - time ) , giving the conjugate momentum @xmath31 as the second integral of motion for @xmath28 . we are now left with five coupled equations for @xmath32 . the third integral of motion , carter s constant @xcite @xmath33,\ ] ] is used as an independent check of the accuracy of the numerical integration . as mentioned above , many traditional schemes to calculate trajectories in the kerr metric use this additional integral of motion to further reduce the dimensionality of the problem @xcite and even reduce the problem to one of quadrature integration [ e. g. @xcite ] . while potentially increasing the speed of the computation , this approach also introduces significant complications in the form of arbitrary signs in the equations of motion corresponding to turning points in @xmath34 and @xmath35 . another common approach is to define effective potentials in @xmath36 and @xmath37 , which give a set of four parametric equations for @xmath38 and @xmath39 . these parametric solutions introduce the difficulty of reconstructing the coordinate momenta , which are not evolved along with the coordinate positions . in our ray - tracing algorithm , we have traded a modest improvement in computational speed for simplicity in implementation . the initial conditions for the photon or particle geodesics are determined in the local orthonormal frame of a `` zero - angular momentum observer '' ( zamo ) . the zamo basis is defined such that the spatial axes are aligned with the coordinate axes and then the time axis is determined by orthogonality @xcite . the zamo tetrad @xmath40 is @xmath41 where @xmath42 is the standard coordinate basis in boyer - lindquist coordinates . one advantage of the zamo basis is that the basis vector @xmath43 is time - like ( @xmath44 ) everywhere outside of the horizon . for a coordinate stationary observer , on the other hand , the time basis vector @xmath45 becomes space - like ( @xmath46 ) inside the ergosphere @xmath47 . for sufficiently large values of the spin parameter @xmath48 , the inner - most stable circular orbit ( often taken for the inner edge of the accretion disk ) extends within the ergosphere , emphasizing the advantage of using the zamo basis . at a point far away from the black hole , the space - time is nearly flat so euclidean spherical geometry gives the spatial direction of the photon @xmath49 , from which the initial momentum in the coordinate basis is calculated : @xmath50 where the photon energy measured by the distant zamo is @xmath51 . the photon trajectories are integrated backward in time from the image plane oriented at some inclination angle @xmath52 with respect to the axis of rotation for the black hole , where @xmath53 corresponds to a face - on view of the disk and @xmath54 is an edge - on view . the accretion disk is confined to a finite region of latitude with angular thickness @xmath55 normal to the rotation axis . the photons terminate either at the event horizon or pass through the surfaces of colatitude @xmath56 , as shown in figure [ plotone ] . as trajectories pass through the disk , the photon s position and momentum @xmath57 are recorded for each plane intersection in order to later reconstruct an image of the disk . for an infinitely thin disk @xmath58 , it is easy to show how the image plane maps onto the source plane . taking an evenly spaced grid of initial photon directions , figure [ plottwo ] shows the positions of intersection with the source plane , in pseudo - cartesian coordinates defined by @xmath59 photons that cross the black hole event horizon before intersecting the plane are not shown . as the rays are deflected by the black hole , they tend to be focused on the far side , giving a strong magnification by mapping a large area in the image plane onto a small area of the source plane , as seen here by a higher density of lattice grid points . for the flat disk geometry , rays are not allowed to pass through the plane defined by @xmath60 , so we do not see multiple images of sources `` behind '' the black hole , as is often observed in the strong gravitational lensing of distant quasars by intervening galaxies . however , for sufficiently high inclinations and spin values , single points in the equatorial plane can be mapped to different regions of the image plane , creating multiple images of certain regions of the disk . this effect is seen in the folding of the image map onto itself near the horizon in the bottom right of figure [ plottwo ] . the disk itself is modeled as a collection of mass elements moving along circular geodesics around the black hole , emitting isotropic , monochromatic light with energy @xmath61 in the emitter s rest frame . for each photon with 4-momentum @xmath62 intersecting a particle trajectory with coordinate velocity @xmath63 , the measured redshift is given by @xmath64 where for a distant observer at @xmath65 , we take @xmath66 $ ] . for disk models with finite thickness , the radiation transport equation can be solved as the ray passes through the disk . in a locally flat frame , the radiation transport equation is @xmath67 where @xmath68 is the differential path length and @xmath69 , @xmath70 , and @xmath71 are respectively the radiation intensity and the emissivity and opacity of the plasma at a frequency @xmath72 ( scattering terms can be included in both @xmath70 and @xmath71 ) . additionally , as the photon bundle propagates through the global curvature around the black hole , the spectral intensity at a given frequency evolves as the photons are gravitationally red - shifted , maintaining the lorentz invariance of @xmath73 . for most of the calculations presented in this paper , we are primarily concerned with radiation coming from a limited region of the disk , treated as a monochromatic source with @xmath74 and @xmath75 , with @xmath76 the emissivity at space - time coordinate @xmath77 . when calculating the emission from a flat , steady - state disk , the plane defined by @xmath78 is taken to be totally opaque @xmath79 $ ] so that rays can not curve around and see the `` underside '' of the accretion disk . for each pixel @xmath80 in the image plane , an observed photon bundle spectrum @xmath81 is given for each time step by integrating the contribution of the hot spot and the disk through the computational grid . this collection of incident photons can then be summed to give time - dependent light curves , spectra , or spatially resolved images . the results of this paper are based primarily on flat , steady state disks and single hot spots , yet with the computational methods described above , arbitrary disk geometries and emissivity / opacity models can be simulated as well . the ray - tracing calculation is carried out by numerically integrating equations ( [ hameq_1 ] ) and ( [ hameq_2 ] ) with a fifth - order runge - kutta algorithm with adaptive time stepping . this provides high accuracy over a large range of scales as the photon follows a long path through the relatively flat space - time between the observer and the black hole , and then experiences strong curvature over a small region close to the horizon . we typically maintain an accuracy of one part in @xmath82 , which can be independently confirmed by monitoring @xmath83 , carter s constant . the images and spectra are formed by ray - tracing a set of photon paths , usually of the order @xmath84 grid points in @xmath80 with @xmath85 latitudinal zones in @xmath35 and spectral resolution of @xmath86 . a steady - state disk can be made of a collection of massive particles moving in concentric planar circular orbits . for orbits at a radius @xmath34 in a plane orthogonal to the spin axis , a particle s specific energy and angular momentum are given analytically by @xcite : @xmath87 and @xmath88 here the top sign is taken for prograde orbits ( particle angular momentum parallel to black hole angular momentum ) and the bottom sign for retrograde orbits . inside the isco , the particles follow plunge trajectories with constant energy and angular momentum determined at the isco . in practice , when calculating emission from a steady - state disk , we take the inner edge of the disk to be the isco radius @xmath89 . for a disk made up of massive particles on circular orbits emitting isotropically from a region between @xmath90 and @xmath91 , the doppler broadening of an emission line ( typically iron k@xmath0 with @xmath92 kev ) may be used to determine the inclination of the disk with respect to the observer . disks at higher inclination will have an intense blue - shifted segment of the spectrum corresponding to the doppler - boosted photons emitted from gas moving toward the observer . the higher intensity for the blue - shifted photons is caused by relativistic beaming , determined by the lorentz invariance of @xmath73 along a photon bundle : @xmath93 figure [ plotthree ] shows the integrated spectra from a set of accretion disks with outer radius @xmath94 and inner radius at @xmath95 for a spin parameter @xmath96 , normalized such that @xmath97 . all spectra are assumed to come from a flat , opaque disk with uniform emission @xmath98 $ ] . repeating this calculation for a range of spin parameters ( and thus a range of @xmath99 ) , we find that the dependence on disk inclination is quite strong , while the dependence on black hole spin is almost insignificant . this is reasonable because , except for very close to the horizon , the spin has little effect on the orbital velocity for circular orbits : for spin parameter @xmath48 and radius @xmath34 , the observed angular frequency is given by @xmath100 where again we take the upper sign for prograde orbits and the lower sign for retrograde orbits . for the large part of the disk , the orbits have nearly keplerian orbital frequencies , as measured in coordinate time @xmath101 . for larger values of @xmath48 , the isco extends in close to the event horizon , increasing the radiative area of the disk . however , due to the strong gravitational redshift in this inner region , the observed intensity is reduced by a significant factor of @xmath102 , resulting in a weak dependence on spin for disks with uniform emission . many accretion disk models include an emissivity that scales as a power of the radius . following @xcite , we apply an emissivity factor proportional to @xmath103 , giving an added weight to the inner , presumably hotter , regions . for the inner section of an accretion disk with @xmath94 , this inverse - square dependence closely approximates the classic shakura - sunyaev steady - state disk model @xcite . as can be seen in figure 4 , this extra emission from close to the black hole serves to break the otherwise weak dependence on spin . for an inclination of @xmath104 , five different spin values are shown : @xmath105 , corresponding to inner disk boundaries at @xmath106 . since the sign of @xmath48 is defined with respect to the angular momentum of the accretion disk , negative values of @xmath48 imply retrograde orbits that do not survive as close to the black hole , plunging at larger values of @xmath99 . the disks that extend in closer produce more low - energy red - shifted photons , giving longer tails to the spectra at @xmath107 and smaller relative peaks at @xmath108 . thus , if we can determine the inclination of the disk independently ( e.g. through spectroscopic observations of the binary companion ) , the spin may be inferred from the broadening of an iron emission line . however , since the plane of the disk tends to align normal to the black hole spin axis close to the isco , the binary inclination may not coincide with the inclination of the inner disk . the problem of inclination , along with other complications , such as additional emission lines and other causes of scattering and line broadening , motivates us to look more closely at the qpo spectra as a method for determining black hole spin . given the map from the accretion disk to the image plane , with each photon bundle labeled with a distinct 4-momentum and time delay , we can reconstruct time - dependent images of the disk based on time - varying emission models . the simplest model we consider is a single region of isotropic , monochromatic emission following a geodesic trajectory : the `` hot spot '' or `` blob '' model @xcite . the hot spot is a small region with finite radius and emissivity chosen to have a gaussian distribution in local cartesian space : @xmath109.\ ] ] the spatial position 3-vector @xmath110 is given in pseudo - cartesian coordinates by the transformation defined by equation ( [ cart_boyer ] ) and @xmath111 . outside a distance of @xmath112 from the guiding geodesic trajectory , there is no emission . we typically take @xmath113 , but find the normalized light curves and qpo power spectra to be independent of spot size . because we assume all points in the hot spot have the same 4-velocity as the geodesic guiding trajectory , one must be careful not to use too large a spot or the point of emission @xmath77 can be spatially far enough away from the center @xmath114 to render the inner product @xmath115 unphysical . we have also explored a few different hot spot shapes , ranging from spherical to a disk flattened in the @xmath116 direction and similarly find no significant dependence of the spectra on spot shape . after calculating and tabulating the hot spot trajectory as a function of coordinate time @xmath101 , the ray - tracing map between the disk and the observer is used to construct a time - dependent light curve from the emission region . for each photon bundle intersection point there is a time delay @xmath117 ( where @xmath118 are the coordinate indices in the image plane and @xmath119 is the latitude index in the disk ) so for the observer time @xmath120 , we first determine where the hot spot was at coordinate time @xmath121 . if the spot centroid is close ( within @xmath112 ) to the disk intersection point @xmath122 , then the redshifted emission is added to the pixel spectrum @xmath123 , weighted by equation ( [ i_exp ] ) . in this way , a movie can be produced that shows the blob orbiting the black hole , including all relativistic effects . such a movie shows a few immediately apparent special relativistic effects such as the doppler shift and beaming as the spot moves toward and then away from the observer . for a hot spot orbiting in the clock - wise direction as seen from above ( @xmath124 with @xmath125 toward the observer ) , the point of maximum blue shift actually occurs at a point where @xmath126 because of the gravitational lensing of the light , beamed in the forward direction of the emitter and then bent toward the observer by the black hole . gravitational lensing also causes significant magnification of the emission region when it is on the far side of the black hole , spreading the image into an arc , much like distant galaxies are distorted by intervening matter in galaxy clusters . a simulated time - dependent spectrum or _ spectrogram _ for this hot spot model is shown in figure [ plotfive ] . the horizontal axis measures time in the observer s frame , with @xmath127 corresponding to the time at which the spot center is moving most directly away from the observer @xmath128 . as mentioned above , this is _ not _ the same as the point of maximum redshift , which occurs closer to @xmath129 due to gravitational deflection of the emitted light . the spectrum shown in figure [ plotfive ] can be integrated in time to give a spectrum similar to those shown in figures [ plotthree ] and [ plotfour ] , corresponding to something like a very narrow circular emitting region with @xmath130 . by integrating over frequency , we get the total x - ray flux as a function of time , i.e. the light curve @xmath131 . this time - varying signal can be added to a background intensity coming from the inner regions of a steady state disk described in section 3 . by definition the hot spot will have a higher temperature or density and thus greater emissivity than the background disk , adding a small modulation to the total flux . rxte observations find the hfqpo x - ray modulations to have typical amplitudes of 1 - 5% of the mean flux during the outburst @xcite . @xcite present a first - order argument that a 1% amplitude modulation requires a hot spot with 100% overbrightness extending over an area of 1% of the steady - state region of the disk . for @xmath94 , this requires a hot spot with radius @xmath132 , which they argue is too large to survive the viscous shearing of the disk . hydrodynamic stability aside , this reasoning ignores the important effects of disk inclination , relativistic beaming and gravitational lensing of the hot spot emission . also , assuming a shakura - sunyaev type disk with steady - state emissivity @xmath133 and a similar scaling for the hot spot emission , we find that hot spots with smaller size and overbrightness are capable of creating x - ray modulations on the order of 1% rms , defined by @xmath134 ^ 2dt } { \int i(t)dt}}.\ ] ] figure [ plotsix ] shows the required overbrightness of a flattened gaussian hot spot orbiting near the isco to produce a modulation with rms amplitude of 1% for a range of inclinations and black hole spin parameters . in the limit of a face - on accretion disk @xmath135 , even an infinitely bright spot on a circular orbit will not produce a time - varying light curve . as the inclination increases , the required overbrightness decreases , as the special relativistic beaming focuses radiation toward the observer from a smaller region of the disk , _ increasing _ the relative contribution from the hot spot . as the spin of the black hole increases , the isco moves in toward the horizon and the velocity of a trajectory near that radius increases , as does the gravitational lensing , magnifying the contribution from the hot spot . understandably , the required overbrightness is inversely proportional to the area of the hot spot so @xmath136*r_{\rm spot}^2 = \rm const$ ] . for example , from figure [ plotsix ] we see that a black hole binary with inclination @xmath137 and spin @xmath138 would require a spot size of @xmath139 with 67% overbrightness ( e.g. 14% temperature excess ) to produce a 1% rms modulation in the light curve . this is well within the range of the typical size and magnitude of fluctuations predicted by 3d mhd calculations of 3-dimensional accretion disks @xcite . the hot spot model is well - suited for simplified calculations of the x - ray emission from these random fluctuations in the accretion disk . by adding the emission from small , coherent hot spots to a steady - state shakura - sunyaev disk , we can interpret the amplitudes and positions of features in the qpo spectrum in terms of a model for the black hole mass , spin , and inclination . considering the x - ray flux from the hot spot alone , the frequency - integrated light curves for a variety of inclinations are shown in figure [ plotseven ] . all light curves are shown for one period of a hot spot orbiting a schwarzschild black hole at the isco . as the inclination increases , the light curve goes from nearly sinusoidal to being sharply peaked by special relativistic beaming . thus the shape of a qpo light curve may be used to determine the disk inclination . with current observational capabilities , it is not possible even for the brightest sources to get a strong enough x - ray signal over individual periods as short as 3 - 5 ms to be able to differentiate between the light curves in figure [ plotseven ] . instead , the fourier power spectrum can be used to identify the harmonic features of a periodic or quasi - periodic light curve over many orbits . disks with higher inclinations will give more power in the higher harmonic frequencies , due to the `` lighthouse '' effect , as the hot spot shoots a high - power beam of photons toward the observer once per orbit , approximating a periodic delta - function in time . figure [ ploteight]a shows a sample section of such a light curve , including only the x - ray flux from the hot spot , subtracting out the steady - state flux from the disk . the sharp peaks in the light curve , while unresolvable in the time domain , will give a characteristic amount of power in the higher harmonics , shown in figure [ ploteight]b . here we have normalized the rms amplitudes to the background flux from the disk with a hot spot overbrightness of 100 % . for a signal @xmath131 with fourier components @xmath140 : @xmath141 we define the rms amplitude @xmath142 in each mode @xmath143 as @xmath144 with this normalization , the rms defined in equation ( [ rms ] ) can be conveniently written @xmath145 in figure [ ploteight]b , the main peak at @xmath146 hz corresponds to the fundamental azimuthal frequency for an orbit at the isco of a @xmath147 schwarzschild black hole . in the limit where the light curve is a periodic delta - function in time , there should be an equal amount of power in all harmonic modes , because the fourier transform of a periodic delta - function is a periodic delta - function . however , even in the limit of edge - on inclination @xmath148 , unless the hot spot is infinitessimally small and ultra - relativistic , the light curve will always be a continuous function with some finite width and non - zero minimum , thus contributing less and less power to the higher harmonics . the harmonic dependence on inclination for a hot spot orbiting a schwarzschild black hole is shown in figure [ plotnine ] . predictably , as the inclination increases , we see that both the absolute and relative amplitudes of the higher harmonics increase , almost to the limit of a periodic delta - function when @xmath149 . interestingly , we find very little dependence of the harmonic structure on hot spot size or shape , as long as the total emission of the spot relative to the disk is constant ( this emphasizes the robustness of the simple hot spot model in interpreting an x - ray power spectrum , without needing to include the detailed physics of the disk perturbations ) . as the spin parameter increases for kerr black holes , the isco moves closer to the horizon , increasing the circular velocities of particles on the isco and thus the doppler shifts , giving broader time - integrated spectra , as seen in figure [ plotfour ] . the phase lag in time of the peak blueshift with respect to angular phase of the hot spot is also amplified for these smaller values of @xmath99 , giving light curves that are asymmetric in time . the hot spot orbits around kerr black holes also experience more gravitational lensing as they move through the strongly curved space - time near the horizon . this lensing becomes more significant for larger inclinations as the black hole and the emitting region approach collinearity with the observer , producing an einstein ring as @xmath149 . this somewhat counters the special relativistic beaming effect as the light curve peaks over a broader phase of each period due to lensing . thus the harmonic power dependence on inclination for a kerr light curve is in fact smaller than that for the schwarzschild geometry . however , in the event that the accretion disk is actually rotating in a retrograde fashion with respect to the black hole spin , the isco will move _ away _ from the horizon @xmath150 $ ] , reversing the above effects and giving a _ stronger _ dependence on inclination . one of the major unsolved puzzles motivating theoretical models of black hole qpos is the observation of multiple peaks in the high frequency power spectrum @xcite . as discussed above , any non - sinusoidal light curve will contribute to fourier power in harmonics at integer multiples of the fundamental kepler frequency . however , for at least two x - ray binary systems ( xte j1550 - 564 and gro j1655 - 40 ; possibly also grs 1915 + 105 ) , peaks are found with rational ( but non - integer ) frequency ratios @xcite . in these particular examples , significant power is measured around the frequencies ( 184 , 276 hz ) for xte j1550 - 564 and ( 300 , 450 hz ) for gro j1655 - 40 , almost exactly a 2:3 commensurability in frequencies , while grs 1915 + 105 has peaks at 40 and 67 hz . following the work of @xcite , we investigate the possibility of these commensurabilities coming from integral combinations of the radial and azimuthal coordinate frequencies of nearly circular geodesics around a kerr black hole . in a newtonian point mass potential , the radial , azimuthal , and vertical ( latitudinal ) frequencies @xmath151 and @xmath152 are identical , giving closed planar elliptical orbits . for the schwarzschild metric the vertical and azimuthal frequencies are identical , giving planar rosette orbits that are closed only for a discrete set of initial conditions . the kerr metric allows three unique coordinate frequencies , so geodesic orbits in general can fill out a torus - shaped region around the black hole spin axis . when these coordinate frequencies are rational multiples of each other , the trajectories will close after a finite number of orbits . while there is currently no clear physical explanation for why hot spots may tend toward such trajectories , some recent evidence suggests the possible existence of nonlinear resonances near geodesic orbits with integer commensurabilities @xcite . another important clue may come from the fact that at these special orbits are closed , perhaps enhancing the stability of non - circular trajectories . the quasi - periodic nature of the oscillations suggest the continual formation and subsequent destruction of hot spots _ near _ , but not exactly at , the resonant orbits . for the purposes of this paper , we will take the apparent preference for such orbits as given and concern ourselves primarily with calculating the resulting light curves and power spectra . in geometrized units with @xmath153 , coordinate time is measured in units of @xmath154 sec . for example , an orbit with angular frequency @xmath155 around a @xmath147 black hole would have an observed period of 3.1 ms , whereas the analogous orbit around a supermassive black hole with mass @xmath156 would have a period of 8.6 hours . in these units , the three fundamental coordinate frequencies for nearly circular orbits are given by equation ( [ omega_phi ] ) and @xcite @xmath157^{1/2}\ ] ] and @xmath158^{1/2},\ ] ] where the upper sign is taken for prograde orbits and the lower sign is taken for retrograde orbits . the radial frequency approaches zero at @xmath159 , where geodesics can orbit the black hole many times with steadily decreasing @xmath34 . in the limit of zero spin and large @xmath34 , the coordinate frequencies reduce to the degenerate keplerian case with @xmath160 . to model the 2:3 frequency commensurability , we begin by looking for perturbed circular planar orbits where the radial frequency @xmath161 is one - third the azimuthal frequency @xmath162 . since the orbits are nearly circular , the fundamental mode of the light curve should peak at the azimuthal frequency with additional power in beat modes at @xmath163 . for @xmath161:@xmath162 = 1:3 , the power spectrum should have a triplet of peaks with frequency ratios 2:3:4 . these commensurate orbits can be found easily from equations ( [ omega_phi ] ) and ( [ omega_r ] ) and solving for @xmath34 : @xmath164 figure [ plotten ] shows the radius ( solid lines ) of these special orbits as a function of spin parameter . also shown ( dashed line ) is the inner - most stable circular orbit for prograde trajectories . the position of the 1:3 commensurate orbits follows closely outside the isco , suggesting a connection between the high frequency qpos and the black hole isco . however , other integer commensurabilities such as 1:2 , 2:5 , or 1:4 also closely follow the isco curves for varying @xmath48 , so the proximity to the isco alone is probably not enough to explain the hot spot preference for these specific coordinate frequencies . it is important to note that any given black hole source will have a constant value of @xmath48 , certainly over the lifetime of our observations . thus , we may need to observe many more sources like xte j1550 - 564 and gro j1655 - 40 in order to better sample the parameter space of figure [ plotten ] and thus the connection between certain preferred orbits and the black hole isco . a 1:3 commensurate trajectory moves through three revolutions in azimuth for each radial period , forming a closed rosette of three `` layers . '' for such rosettes , the eccentricity can be defined as @xmath165 the time - dependent light curve for an prograde orbit with eccentricity @xmath166 , spin @xmath138 , and inclination @xmath167 is shown in figure [ ploteleven]a . the time axis begins at the point when the hot spot is at apocenter , moving away from the observer . thus the first and third peaks come from the hot spot moving toward the observer at a relatively larger radius , while the second , higher peak is caused by the emitter moving toward the observer through pericenter at a higher velocity , giving a larger blue - shift and thus beaming factor . the combined doppler beaming and gravitational lensing causes the peak following the pericenter peak to be slightly larger , as the emitter is moving away from the black hole yet the light is focused more toward the observer . the power spectrum for this light curve is shown in figure [ ploteleven]b , with the strongest peaks at the azimuthal frequency of @xmath168 hz and its first harmonic at @xmath169 hz for @xmath170 . even for this modest deviation from circularity , there is significant power in the frequencies @xmath163 . the beating of the fundamental @xmath162 with the radial frequency @xmath171 hz gives the set of secondary peaks at @xmath172 and @xmath173 . additional peaks occur at beats of the harmonic frequencies @xmath174 . it is interesting to note that there is not significant power in the radial mode at @xmath175 hz , but only in the beats with the fundamental azimuthal frequency and its harmonics . however , in the limit of a face - on orientation @xmath176 , the radial frequency should dominate the light curve variation as the gravitational and transverse doppler red - shift modulate the intensity as a function of the hot spot s radial coordinate . the radial mode should also be present in the limit of an edge - on orientation @xmath177 , as gravitational lensing becomes more important , and the hot spot will experience more magnification when closer to the black hole . figure [ plottwelve ] shows the dependence on inclination of the lower order harmonic and beat modes . at low inclinations , the radial frequency @xmath178 contributes significant power , while at higher inclinations , the first harmonic of the azimuthal mode begin to dominate with similar behavior to the circular orbits shown in figure [ plotnine ] . along with increasing power at @xmath179 , there is also increasing power in the radial beats of the first harmonic at @xmath180 . for a @xmath147 black hole , all these frequencies should be observable within the timing sensitivity of rxte . to further explore the constraints of our model , we investigated the effect of orbital eccentricity on the qpo power . maintaining a 1:3 commensurability between radial and azimuthal frequencies , we calculated the light curves for a range of eccentricities @xmath181 . as expected , the beat modes at @xmath182 have more power for more eccentric orbits , as the radial variation of the emitter becomes larger . at the same time , the first harmonic at @xmath183 provides relatively less power with increasing eccentricity . this is best understood as the `` picket fence '' character of the light curve becomes modulated in amplitude and frequency from peak to peak , i.e. for each 3-peak cycle , the time between peaks 1 - 2 , 2 - 3 , and 3 - 1 are not all the same , damping the harmonic overtone . interestingly , the fourier power in the beat modes appears to saturate at a moderate eccentricity of @xmath184 , suggesting highly eccentric geodesics are not necessary or even helpful for producing power in the beat frequencies . while there is significant evidence for higher frequency harmonic and beat modes in the qpo power spectrum of xte j1550 - 564 , the fourier power is clearly dominated by the two frequencies @xmath185 and @xmath186 hz @xcite , corresponding to @xmath187 and @xmath162 in our model . what are the physical mechanisms that could damp out the higher frequency modes ? one possible explanation is in the geometry of the hot spot . as explained in section 4 , in order to produce the power observed in qpos , the relative flux coming from the hot spot must be some finite fraction of that of the disk ( typically @xmath188 for a qpo amplitude of @xmath189 ) , so the hot spot must have some minimum size or it would not produce enough emission to be detected above the background disk . yet if the hot spot is too large , it would be sheared by differential rotation in the accretion disk and not survive long enough to create the coherent x - ray oscillations that are observed . as mentioned above , we find that the relative qpo power in different modes is not sensitive to the size of the hot spot @xmath190 . three - dimensional mhd simulations @xcite show a range of density and temperature fluctuations consistent with the hot spot size and overbrightness factor predicted by our model in conjunction with the observations . it also appears from simulations that as the hot spot is formed in the disk , differential rotation will tend to shear a finite region of gas as it follows a geodesic orbit around the black hole , modifying the shape of the hot spot into an arc in azimuth . in the limit that the emission region could be sheared into a ring of arc length @xmath191 , the fundamental mode and its harmonics would be essentially removed , leaving power only in the radial modulation . indeed , as shown in figure [ plotthirteen]a , for an arc length of @xmath192 , the higher frequency modes at @xmath193 and @xmath194 are strongly suppressed , while still maintaining a significant amount of power in the fundamental beat modes @xmath195 . the total qpo power also increases as the area of the emission region increases relative to the circular hot spot geometry . however , when we allow the arc to be sheared into a ring with @xmath196 , the total qpo power is actually decreased as the differential beaming is essentially eliminated by the extended emission region : there is always some portion of the arc moving toward the observer . the resulting modulation is then more weighted to the first radial beat mode at @xmath187 , as seen in figure [ plotthirteen]b . it is not intuitively obvious why the @xmath187 mode is dominant while the @xmath197 mode ( @xmath72 = 368 hz ) is much weaker in the arc geometry . if anything , this is a strong argument for the necessity of a full ray - tracing calculation of the hot spot light curves when predicting qpo power spectra , as it clearly gives information unavailable to simple analysis of the geodesic coordinate frequencies . this behavior offers a plausible explanation for the two major types of qpos described in @xcite . type a outbursts have more total power in the hfqpos , with a major peak at 276 hz and a minor peak at 184 hz . type b outbursts have most of the qpo power around 184 hz and a smaller peak around 276 hz and less overall power in the high frequency region of the spectrum . thus we propose that type a outbursts are coming from more localized hot spot regions , while type b outbursts come from a more extended ring geometry . in addition to the commensurate high - frequency qpos observed in sources like xte j1550 - 564 , there are also strong low - frequency qpos observed at the same time with frequencies in the range @xmath198hz . there have been suggestions that these low - frequency qpos may be caused by the lens - thirring precession of the disk near the isco , also known as `` frame dragging '' @xcite . for geodesic orbits out of the plane perpendicular to the black hole spin , the latitudinal frequency @xmath199 of massive particles is not equal to the azimuthal frequency @xmath200 [ see eqs.([omega_theta ] ) and ( [ omega_phi ] ) ] , leading to a precession of the orbital plane with frequency @xmath201 for black holes with the mass and spin used above @xmath202 , the frame - dragging frequency , as calculated at the radius corresponding to the commensurability @xmath161:@xmath162=1:3 , is somewhat higher than that observed in the low - frequency qpos from xte j1550 - 564 . the type a qpo peaks at 12 and 276 hz appear to be consistent with a black hole mass of @xmath203 and spin parameter of @xmath204 , quite similar to the values used throughout much of this paper . for the bh binary gro j1655 - 40 , we can fit the qpos at 18 and 450hz with a mass of @xmath205 and spin @xmath206 , slightly less than the published mass range of @xmath207 @xcite . if we relax the requirement of matching the lfqpos and only fit the hfqpos with a 1:3 coordinate frequency commensurability , there remains a 1-dimensional degeneracy in the mass - spin parameter space . based solely on the hfqpos , for xte j1550 - 564 with @xmath208 , the range of spin parameters would be @xmath209 and for gro j1655 - 40 , the spin would be in the range @xmath210 . these results are shown in table [ tableone ] . to get a more quantitative feeling for the effect of lens - thirring precession on the power spectrum , we investigated hot spot orbits with initial trajectories inclined to the plane of the disk : @xmath211 . this is much like changing the observer s inclination with a period of @xmath212 . thus we see additional modulation in the hot spot light curve at the `` double - beat '' modes @xmath213 . we find that , for modestly inclined hot spot orbits @xmath214 , the contribution to the power spectrum at lens - thirring frequencies is quite small ( @xmath215 of total power ) for the basic circular hot spot geometry . this relative contribution increases with arc length as the spot becomes a ring precessing about the spin axis , consistent with the relative power in lfqpos and hfqpos in the type a ( more high frequency power than low frequency ) and type b ( more low frequency power ) sources described above . under the premise that the hfqpo commensurate frequencies are caused by the geodesic motion of a sheared , overbright region in the disk , in table [ tabletwo ] we show the best fit parameters for the type a and type b qpo spectra from xte j1550 - 564 [ cf . table 1 in @xcite ] . guided also by the ( somewhat speculative ) assumption that the lfqpos come from the lens - thirring precession of the hot spot orbital plane , we predict a black hole mass and spin . using a fixed inclination of @xmath216 , we can match the frequencies and amplitudes of the observed hfqpo peaks ( and , equally important , the lack of power at certain frequencies ) for both type a and type b outbursts . setting constant the eccentricity @xmath217 , geodesic inclination to the disk @xmath218 , and the overbrightness to be a factor of unity , we fit the hot spot radius and arc length to match the observations . being able to match the qpo rms amplitudes of the peaks ( or lack thereof ) at 92 , 184 , 276 , and 368 hz , for at least two different types of x - ray outburst , shows the robustness of our simple model in explaining these phenomena . however , even with the sheared arc emission , many observations still show significantly more power in the lfqpos than can be explained solely from the lens - thirring precession of a geodesic near the isco . coupled with the difficulty in simultaneously fitting the mass and spin to three coordinate frequencies in a manner consistent with spectroscopic mass predictions , it seems likely that the lfqpos may be caused by some other mechanism in the disk that is related only indirectly to the high - frequency hot spot emission . another possibilty is that the thin , warped disk model breaks down near the isco , allowing more complicated emission geometries and thus amplifying the effects of lens - thirring precession . we have developed a hot spot model to help understand the high - frequency qpos observed in accreting black holes . while the model does not consider radiation or hydrodynamic pressures , it is fully relativistic and calculates exact geodesics for both massive particles and photons . treating the hot spot as a monochromatic , isotropic emitter , we have calculated the time - dependent spectra of x - ray lines produced as the hot spot orbits the black hole . the model makes a number of general predictions of the fourier power of the x - ray light curve as a function of inclination and black hole spin , and is also able to explain qpo observations from the black hole binaries xte j1550 - 564 and gro j1655 - 40 . simply by matching the locations of the low - frequency and high - frequency qpos with the coordinate frequencies ( under the condition @xmath219 ) , we can determine the black hole mass and spin . relaxing the lfqpo constraint , the spin can still be determined uniquely for a given mass , which in turn can often be measured independently with the inclination and radial velocity of the companion star . by matching the amplitudes of various peaks observed in xte j1550 - 564 , we have explored the model parameters such as the hot spot size , shape , and the overbrightness relative to a steady - state background disk . the predicted magnitude of these fluctuations are within the range predicted by 3d mhd calculations of the accretion disk . future work will investigate the effect of multiple hot spots of various size , emissivity , and lifetime , as guided by mhd calculations . observations of additional sources with commensurate frequency qpos may help us further constrain the hot spot model and better understand the connection between the lfqpos and hfqpos . some of the physical problems with the hot spot model have been raised by @xcite , as discussed above . many of these points are addressed in our model . first , unlike @xcite , we only attempt to explain a set of qpo data from _ black hole _ binaries , which differ qualitatively from neutron star binaries in many ways , e.g. lacking strong global magnetic fields and thermonuclear activity . and perhaps most significant , black holes have no rotating surface to interact with the accreting matter and provide additional confusion to the qpo power spectrum . our model produces light curves with power spectra consistent with black hole observations even with low eccentricity hot spot orbits . because they do not include ray tracing in their calculations , @xcite are unable to model many relativistic effects , including the gravitational lensing of the hot spot source , which can be quite significant for systems with moderate to high inclination angles @xmath220 . since we calculate the actual x - ray modulation from the orbiting hot spots , we predict both the location and amplitude of every peak in the light curve power spectrum , which can not be done by analyzing the bh coordinate frequencies alone . by introducing a perturbation on circular orbits near the isco , additional peaks begin to appear in the power spectrum , caused by beats of the azimuthal and radial frequencies @xmath162 and @xmath161 . the dependence of the relative power in the different peaks on inclination and spin helps to constrain the details of the hot spot model in explaining the hfqpos , particularly the 2:3 commensurability observed in the power spectra from xte j1550 - 564 and gro j1655 - 40 . as an additional parameter , we introduce a finite arc length for the emission region , motivated by the shearing of the hot spot by differential rotation in the disk . the spreading of the hot spot in azimuth leads to suppression of the higher qpo modes , in agreement with observations . we have also examined the possibility of lens - thirring precession for non - planar orbits as an explanation for the low - frequency qpos that have been observed coincident with the hfqpos yet often with even stronger fourier power . the predicted power spectra from non - planar precessing arcs are consistent with observations of xte j1550 - 564 if we associate type a outbursts with hot spot arcs of @xmath221 and type b outbursts with hot spot rings of @xmath222 . however , the difficulty in matching the lfqpos amplitude and frequency with a single hot spot geodesic suggests the low - frequency modulations may be caused by a different mechanism or perhaps our disk geometry is too simplistic . one major remaining issue with the hot spot model is the preferred location of the geodesic that gives rise to 1:3 coordinate frequencies . why should the orbital frequencies favor integer ratios , and why should the preferred ratio be 1:3 and not 1:2 or 1:4 ? it is possible that detailed radiation - hydrodynamic calculations with full general relativity will be required to answer this question . perhaps the non - circular orbits can only survive along closed orbits such as these to somehow avoid destructive intersections . or there may be magnetic interactions with the black hole itself , analogous to the blandford - znajek process , that lock the accreting gas into certain preferred trajectories @xcite . for now , we are forced to leave this as an open question unanswered by the geodesic hot spot model . a less difficult , yet still unanswered problem is the explanation of the widths of the qpo peaks . as it stands , our hot spot model predicts purely periodic light curves and thus power spectra made up of delta functions . if there _ is _ some physical mechanism that preferentially focuses accreting material onto eccentric orbits at specific radii , then it is likely that these hot spots are forming and then being destroyed as a continual processes . the superposition of many hot spots around the same orbit , all with slightly different initial trajectories , could explain the quasi - periodic nature of the power spectrum : the phase decoherence of the hot spots would cause a natural broadening of the strictly periodic signal from a single spot . with the computational framework in place , this question can be answered by modeling a whole collection of hot spots and arcs continually forming and evolving in shape and emissivity . the particular physical parameters for these hot spots can be derived from published mhd calculations such as @xcite . with the basic ray - tracing and radiation transport methods in place , it should be possible to use our code as a `` post - processor '' to analyze the results of these 3d simulations to simulate x - ray light curves and spectra from a realistic accretion disk . a final piece of the black hole qpo puzzle is the spectral behavior of the source during outburst . @xcite have shown there is a large range of x - ray fluxes with different relative contributions from a power - law component and a disk bolometric component . these relative and absolute fluxes seem to be correlated with the amount of power in both the lfqpos and the hfqpos . future work on 3-dimensional disks and more detailed radiative transfer models should give us important insights into understanding this spectral behavior and its relation to the qpo power . abramowicz , m. a. , & kulzniak , w. 2001 , a&a , 375 , l19 abramowicz , m. a. , et al . 2002 , preprint ( gr - qc/0202020 ) abramowicz , m. a. , et al . 2003 , pasj , in press ( astro - ph/0302183 ) armitage , p. j. , reynolds , c. s. , & chiang , j. 2001 , , 548 , 868 bao , g. 1992 , astron . & astro . 257 , 594 bardeen , j. m. , press , w. h. , teukolsky , s. a. 1972 , 178 , 347 blandford , r. d. , & znajek , r. l. 1977 , mnras , 179 , 433 bromley , b. c. , chen , k. , & miller , w. a. 1997 , , 475 , 57 carter , b. 1968 , phys . 26 , 331 ford , e. c. 2000 , 535 , l119 ford , e. c. et al . 2000 , 539 , 368 ghosh , p. , & lamb , f. k. 1978 , , 223 , l83 gierlinsky , m. , done , c. , barret , d. 2002 , mnras 331 , 141 hawley , j. f. , & krolik , j. h. 2001 , , 548 , 348 karas , v. , vokrouhlicky , d. , & polnarev , a. g. 1992 , mnras , 259 , 569 kojima , y. 1991 , mnras 250 , 629 kluzniak , w. , & abramowicz , m. a. 2001 , preprint ( astro - ph/0105057 ) kluzniak , w. , & abramowicz , m. a. 2002 , preprint ( astro - ph/0206063 ) lamb , f. k. 2002 , in _ x - ray binaries and gamma - ray bursts _ , ed . e. p. j. van den heuvel ( san francisco : astronomy society of the pacific ) lanzafame , g. , molteni , d. , & chakrabarti , s. k. 1998 , mnras , 299 , 799 laor , a. 1991 , , 376 , 90 lee , w. h. , & ramirez - ruiz , e. 2002 , , 577 , 893 markovic , d. , & lamb , f. k. 1998 , , 507 , 316 markovic , d. , & lamb , f. k. 2000 , preprint ( astro - ph/0009169 ) merloni , a. , vietri , m. , stella , l. , & bini , d. 1999 , mnras , 304 , 155 miller , j. m. et al . 2002 , , 570 , l69 misner , c. w. , thorne , k. s. , & wheeler , j. a. 1973 , _ gravitation _ ( new york : w. h. freeman ) pariev , v. i. , bromley , b. c. , miller , w. a. 2001 , , 547 , 649 rauch , k. p. , & blandford , r. d. 1994 , , 421 , 46 remillard , r. a. et al . 2002 , , 580 , 1030 reynolds , c. s. , & begelman , m. c. 1997 , , 488 , 109 shahbaz , t. , et al . 1999 , mnras , 306 , 89 shakura , n. i. , & sunyaev , r. a. 1973 , astron . , 24 , 337 shapiro , s. l. , & teukolsky , s. a. 1983 , _ black holes , white dwarfs , and neutron stars _ ( new york : john wiley & sons , inc . ) stella , l. , & vietri , m. 1998 , , 492 , l59 stella , l. , & vietri , m. 1999 , phys . 82 , 17 strohmayer , t. e. 2001 , , 552 , l49 strohmayer , t. e. 2001 , , 554 , l169 wagoner , r. v. , silbergleit , a. s. , & ortega - rodriguez , m. 2001 , , 559 , l25 wang , d. x. 2003 , mnras , in press ( astro - ph/0305478 ) schematic picture of ray - tracing method from distant observer through a disk of angular thickness @xmath223 . the rays either terminate at the black hole horizon or pass through the disk , with each point of intersection labeled with the photon position and momentum @xmath224 . ] projection of a uniform cartesian grid in the image plane onto the source plane of the accretion disk @xmath225 . inclination angles are @xmath53 @xmath226 and @xmath167 @xmath227 and spin parameters are @xmath228 @xmath229 and @xmath230 @xmath231 . the region inside the horizon is cut out from each picture . ] normalized spectra of a monochromatic emission line from steady - state accretion disks of varying inclination . an inclination of @xmath53 corresponds to a face - on view of the disk while @xmath54 would be edge - on . the emissivity is taken to be uniform between @xmath232 and @xmath233 . the spin is taken to be @xmath138 but the dependence on @xmath48 is negligible for uniformly emitting disks . ] normalized spectra of steady - state accretion disks with inclination @xmath234 and varying spin parameter @xmath48 ( negative values of @xmath48 correspond to retrograde disk rotation ) . the emissivity is taken to be proportional to @xmath103 between @xmath235 and @xmath233 . black holes with higher values of @xmath48 allow the inner disk to extend in closer to the horizon , giving a greater weight to the high - redshift radiation emitted there . ] spectrogram of a circular hot spot with radius @xmath236 orbiting a schwarzschild black hole at the isco @xmath237 ) , viewed at an inclination of @xmath238 . the spot is moving in the @xmath239 direction with the observer at @xmath125 . the maximum redshift occurs when @xmath240 and the maximum blueshift occurs when @xmath241 . ] overbrightness required of a hot spot on a circular orbit to produce a 1% rms modulation in x - ray flux when added to a steady - state disk with @xmath95 , @xmath94 and emissivity @xmath242 . an overbrightness of unity means the peak hot spot emissivity is twice that of the steady - state disk with no hot spot . the spot size @xmath190 is measured in gravitational radii @xmath1 , so for a black hole with @xmath138 and @xmath167 , the required overbrightness for a hot spot with @xmath243 would be 67% . ] frequency - integrated light curves of an orbiting hot spot at the isco of a schwarzschild black hole for different disk inclination angles . the spot is moving in the @xmath239 direction as in figure [ plotfive ] . for high inclination angles , the special relativistic beaming causes the light curve to become sharply peaked as the hot spot moves toward the observer . ] fourier amplitude @xmath142 in higher harmonic frequencies @xmath244 as a function of orbital inclination to the observer , normalized as in equation ( [ arms ] ) . the hot spot has size @xmath243 , an overbrightness factor of 100% , and is in a circular orbit at @xmath99 around a schwarzschild black hole . ] radius of prograde orbits with commensurate frequencies @xmath245(1:3 , 1:2 , 2:3 ) ( solid lines ) as a function of dimensionless spin parameter @xmath246 . the isco ( dashed line ) corresponds to @xmath247 . also shown are the respective orbital frequencies @xmath162 at these radii for a black hole with mass @xmath147 ( dot - dashed lines ) . ] power in low - order harmonics and beat modes with frequencies @xmath248 , as a function of disk inclination angle . the hot spot trajectory is the same as in figure [ ploteleven ] , with @xmath142 normalized as in figure [ plotnine ] . the curves are labeled by the ratio @xmath249 . ]

a model that rapidly computes the secular evolution of a gravitating disk planet system is developed . the disk is treated as a nested set of gravitating rings , with the rings/planets time evolution being governed by the classical laplace lagrange solution for secular evolution but modified to account for the disk s finite thickness . the lagrange planetary equations for this system yield a particular class of spiral wave solutions , usually denoted as apsidal density waves and nodal bending waves . there are two varieties of apsidal waves long waves and short waves . planets typically launch long density waves at the disk s nearer edge or else at a secular resonance in the disk , and these waves ultimately reflect downstream at a more distant disk edge or else at abarrier in the disk , whereupon they return as short density waves . planets also launch nodal bending waves , and these have the interesting property that they can stall in the disk , that is , their group velocity plummets to zero upon approaching a region in the disk that is too thick to support the continued propagation of bending waves . the rings model is used to compute the secular evolution of a kuiper belt having a variety of masses , and it is shown that the early massive belt was very susceptible to the propagation of low amplitude apsidal and nodal waves launched by the giant planets . for instance , these waves typically excited orbits to in a primordial kuiper belt of mass earth masses . although these orbital disturbances are quite small , the resulting fractional variations in the disk s surface density due to the short density waves is usually large , typically of order unity . this epoch of apsidal and nodal wave propagation probably lasted throughout the belt s first to years , with the waves being shut off between the time when the large km kbos first formed and when the belt was subsequently eroded and stirred up to its present configuration .

astro-ph0305602

the kuiper belt is a vast swarm of comets orbiting at the solar system s outer edge . this belt is comprised of debris that was left over from the epoch of planet formation , and this swarm s distribution of orbit elements preserves a record of events that had occurred when the solar system was still quite young . a common goal of nearly all dynamical studies of the kuiper belt is to decipher this kuiper belt record . however this record is still open to some interpretation . the dots in fig . [ compare_ei ] give the kuiper belt objects ( kbos ) eccentricities @xmath7 and inclinations @xmath8 versus their semimajor axes @xmath9 . this figure reveals the kbos three major dynamical classes : the plutinos which inhabit neptune s 3:2 resonance at @xmath10 au , the main belt kbos which are the non resonant kbos orbiting between @xmath11 au , and the more distant scattered kbos that live in eccentric , nearly neptune crossing orbits . the figure also shows that the plutinos and the scattered kbos have inclinations that span @xmath12 , while the main belt kbos appear to have a bimodal distribution of inclinations centered on @xmath13 and @xmath14 @xcite . note that accretion models show that these large @xmath15 km kbos must have first formed from much smaller planetesimal seeds that were initially in nearly circular and coplanar orbits having @xmath7 and @xmath16 @xcite . however gravitational self stirring can not account for the kuiper belt s current excited state , so one or more mechanisms must also have stirred up the kuiper belt since the time of formation . the orbits of the scattered kbos are perhaps the most easily understood . these objects likely have had one or more close encounters with neptune which lofted these bodies into eccentric , inclined orbits @xcite . repeated encounters with neptune cause these objects semimajor axes and eccentricities to evolve stochastically along the neptune crossing curve shown in fig . [ compare_ei ] , and most of these bodies are ultimately ejected or accreted by the giant planets . however neptune has numerous weak , high order mean motion resonances that thread the kuiper belt , and these resonances permit some of these scattered objects to diffuse to lower eccentricities . this allows a small percentage of the scattered objects to persist over the age of the solar system at eccentricities just below the neptune crossing curve seen fig . [ compare_ei ] @xcite . this diffusion to lower eccentricities may also have been more pronounced had neptune s orbit also migrated outwards . in particular , @xcite shows that during the epoch of planet migration , mean motion and secular resonances act as pathways that allow some scattered kbos to descend _ irreversibly _ into lower eccentricity orbits that are far from the neptune crossing curve and hence stable . since scattered kbos have large inclinations of @xmath17 , this process might also account for the main belt s bimodal inclination distribution , with the @xmath18 component representing the main belt s native population of low inclination kbos , and the @xmath19 being due to scattered invaders who were deposited in the main belt by neptune . however this process is quite inefficient since only @xmath20 of these scattered kbos manage to find orbits that are stable over a solar age @xcite . the possibility that neptune s orbit had expanded outwards is also supported by the cluster of kbos that inhabit neptune s 3:2 resonance ( see figure [ compare_ei ] ) . this may have occurred when neptune had first formed and began to vigorously scatter the local planetesimal debris . this process can drive an exchange of angular momentum between the planets and the planetesimal disk @xcite , so this episode of disk clearing can result in a rearrangement of all of the giant planets orbits over a timescale of @xmath4 years @xcite . outward planet migration will also cause neptune s mean motion resonances to sweep out across the primordial kuiper belt , and these migrating resonances are quite effective at capturing kbos and pumping up their eccentricities @xcite . models of this process show that if neptune s orbit had in fact smoothly expanded some @xmath21 au over a timescale longer than a few million years , then resonance capture would have deposited numerous kbos in the 3:2 and the 2:1 resonances with eccentricities distributed over @xmath22 @xcite . although numerous kbos do indeed inhabit neptune s 3:2 resonance , only a handful are known to live near the 2:1 resonance at @xmath23 au , and many of these bodies have eccentricities of @xmath24 which puts them quite near the neptune crossing curve ( see fig . [ compare_ei ] ) . thus it is possible that some or all of the bodies orbiting near @xmath23 au might instead be members of the scattered belt . thus if planet migration did indeed occur , then the apparent low abundance of kbos at the 2:1 resonance is a mystery that can only be partly due to the observational bias that selects against the discovery of lower eccentricity objects at the 2:1 @xcite . but of particular interest here is the main belt which preserves additional evidence for another mechanism having stirred up the kuiper belt . again , if neptune did indeed migrate outwards @xmath21 au , then the entire main belt was swept by the advancing 2:1 resonance . models of planet migration show that the efficiency of resonance capture is no more than @xmath25 @xcite , so the current members of the main belt evidently avoided permanent capture by slipping through the advancing 2:1 resonance . however this planet migration scenario is utterly unable to account for the high inclinations of @xmath12 observed in the main belt ( see fig . [ compare_ei ] ) , since the n body simulations show that the advancing 2:1 resonance typically excites main belt inclinations by only a few degrees @xcite . evidently , an additional mechanism is also responsible for exciting the main kbo belt . it has been shown that sizable kbo excitation can occur if a recently formed neptune had been scattered outwards by jupiter and/or saturn into a belt crossing orbit @xcite . another possible source of kbo excitation is the invasion of the main belt by the scattered belt @xcite . however this latter process is a very inefficient mechanism having @xmath26 ; an alternate mechanism that is possibly more efficient will be explored below . it has also been suggested that secular resonance sweeping may have been responsible for exciting the kuiper belt @xcite . note that the locations of the secular resonances are very sensitive to the solar system s mass distribution . consequently , the depletion of the solar nebula ( which includes perhaps @xmath27 of the solar system s initial mass content ) could have driven these secular resonances across vast tracts of the solar system . indeed , the model by @xcite suggests that if the nebula was depleted on a timescale of @xmath28 years , main belt inclinations of @xmath29 can be excited due to the passage of the @xmath30 secular resonance that sweeps outwards to infinity as the nebula is depleted from the ecliptic , then almost no excess excitation results @xcite . we also note that the nebula models of @xcite as well as @xcite both treat the gas disk as a rigid slab of gas . however a more realistic treatment would allow the nebula disk to flex and warp in response to the planets secular perturbations . it is suspected that this additional degree of freedom will substantially alter the secular resonance sweeping ; indeed , it can be argued that the @xmath30 never did sweep across the kuiper belt on account of this flexure ( e. chiang and w. ward , 2002 , private communication ) , so perhaps secular resonance sweeping of the kuiper belt is actually a moot issue . ] . however this model is rather idealized in that it treats the kuiper belt as massless , which is a concern since a primordial kuiper belt having some mass is also susceptible to the propagation of very long wavelength spiral waves that can be launched at a secular resonance in the disk @xcite . this issue is worth further examination since wave action can alter the magnitude of resonant excitation considerably @xcite . but even if there is no resonance in the disk , planets orbiting interior to a particle disk can still launch these spiral waves at the disk s inner edge @xcite . preliminary results from a model of secular resonance sweeping also reveals that a kuiper belt having only a modest amount of mass is utterly awash in these waves once the nebula is depleted @xcite . however the purpose of the present study is to first characterize the properties of these waves in the simpler , post nebula environment , and to explore their cosmogonic implications . thus the following will consider a suite of models of the secular evolution of the outer solar system for primordial kuiper belts having a variety of masses . accretion models tell us that the primordial kuiper belt must have had a mass of @xmath31 m@xmath32 in the @xmath33 au interval in order for pluto and its cohort of kbos to have formed and survive over the age of the solar system @xcite . a similar kuiper belt mass is also needed to drive neptune s orbital migration of @xmath21 au @xcite . however the current mass is @xmath34 m@xmath32 @xcite , so the kuiper belt appears to have been eroded by a factor of @xmath35 . this may be due to a dynamical erosion of the belt by neptune or possibly by other perturbers that may once have been roaming about the outer solar system , as well as due to the collisional erosion that has since ground all of the smaller kbos down to dust grains that are then removed from the solar system by radiation forces @xcite . the following will give results obtained from models of the secular evolution of the outer solar system for kuiper belts having masses in the interval @xmath36 m@xmath32 . section [ model ] derives the so called rings model that will be used to study the secular evolution of disk planet systems ; the reader uninterested in these details might skip ahead to section [ waves ] or [ results ] . since spiral density and bending waves appear prominently in the model results , their properties are examined in section [ waves ] . section [ results ] will describe the model s application to the primordial kuiper belt , and a summary of results is then given in section [ summary ] . the following shall treat the disk as a collection of nested gravitating rings in orbit about the sun . their mutual perturbations will cause these rings to slowly flex and tilt over time , and this evolution is governed by the lagrange planetary equations . begin with the gravitational potential that a single perturbing ring of mass @xmath37 exerts at the point @xmath38 on another ring of mass @xmath39 : @xmath40 where @xmath41 is the gravitation constant , @xmath42 is the mass density of the differential volume element @xmath43 , @xmath44 is the separation between the perturbing mass element @xmath45 at @xmath46 and the field point @xmath38 , and the integration proceeds over the three dimensional extent of ring @xmath37 . in the following , primed quantities will refer to the perturbing ring @xmath37 and unprimed quantities will refer to the perturbed ring @xmath39 . each ring can be thought of as a swarm of numerous particles all having a common semimajor axis @xmath9 and an identical mean orbital eccentricity @xmath7 , inclination @xmath8 , longitude of periapse @xmath47 , and longitude of ascending node @xmath48 . it is also assumed that these particles have an isotropic dispersion velocity @xmath49 that gives rise to the ring s finite radial as well as vertical half thickness @xmath50 , where @xmath51 is the ring s mean motion . it shall also be assumed that the density @xmath42 varies only in the azimuthal direction due to the keplerian motion of the ring s particles ; in this case the density is @xmath52 where @xmath53 is the ring s azimuthal mass per unit length @xcite . in cylindrical coordinates @xmath46@xmath54 and @xmath55 . if the slight radial variations in equation ( [ phi1 ] ) s integrand are ignored ( _ i.e. _ , @xmath56 ) , the potential becomes @xmath57 where @xmath58 is the longitude dependent height of the perturbing ring s midplane from the @xmath59 plane . of course the perturbed ring @xmath39 also has a radial and vertical half thickness @xmath0 , and it is useful to form an effective potential by averaging @xmath60 over the radial and vertical extent of ring @xmath39 : @xmath61 where @xmath62 where again the slight variations in the integrand with @xmath63 are ignored , and that only @xmath44 is assumed to be sensitive to the variations in @xmath64 and @xmath65 . note that this averaging of @xmath66 is essential in order for the algorithm developed below to conserve angular momentum . the next task is to evaluate the double integrals in @xmath1 . first note that the separation @xmath44 between the perturbing mass element @xmath45 at @xmath46 and the field point @xmath38 obeys @xmath67 where @xmath68 . setting @xmath69 , @xmath70 , and @xmath71 , then @xmath72 to second order in the @xmath73 s , which are of order the rings inclinations and are assumed small . inserting this into the expression for @xmath1 yields @xmath74^{-1/2}\\ = \int_{\beta_o'-\mathfrak{h'}}^{\beta_o'+\mathfrak{h ' } } \frac{\ln(\upsilon)d\beta ' } { 4\mathfrak{h}\mathfrak{h}'\sqrt{\alpha\cos\delta\phi}}\end{gathered}\ ] ] where @xmath75 and @xmath76 are the fractional half thicknesses of rings @xmath39 and @xmath37 , @xmath77 and @xmath78 are the rings midplane latitudes , @xmath79 , and the right hand side of equation ( [ q1 ] ) is the result of doing the integration in @xmath73 , where @xmath80 with @xmath81 and @xmath82 . the @xmath73 s and the @xmath83 s are assumed small , so @xmath84 is second order in the small quantities and is negligible when compared to other terms . thus @xmath85/ \sqrt{\alpha\cos\delta\phi\gamma } } { 1-[\beta'\alpha-\beta_o\alpha\cos\delta\phi+ \mathfrak{h}\alpha\cos\delta\phi]/ \sqrt{\alpha\cos\delta\phi\gamma } } \simeq1 + 2\mathfrak{h}\sqrt{\frac{\alpha\cos\delta\phi}{\gamma}}\ ] ] so @xmath86 . this is inserted back into equation ( [ q1 ] ) and the remaining integral over @xmath87 is evaluated similarly , yielding @xmath88 where @xmath89/ \sqrt{\alpha\cos\delta\phi(\psi+{\cal z}+\xi ) } } { 1-[\beta_o\alpha-\beta_o'\alpha\cos\delta\phi+ \mathfrak{h'}\alpha\cos\delta\phi]/ \sqrt{\alpha\cos\delta\phi(\psi+{\cal z}-\xi ) } } \simeq1 + 2\mathfrak{h'}\sqrt{\frac{\alpha\cos\delta\phi } { \psi+{\cal z}}}\ ] ] with @xmath90 , @xmath91 , @xmath92 and @xmath93 is another negligible term . inserting @xmath94 back into @xmath1 yields @xmath95 a fourier expansion of @xmath96 will be useful , _ i.e. _ , @xmath97 where @xmath98 is the softened laplace coefficient . the usual unsoftened form is @xmath99 . these two coefficients are nearly equal when @xmath100 is far from unity , but the softened form is finite at @xmath101 whereas the unsoftened form diverges . writing @xmath1 in terms of softened laplace coefficients thus gives @xmath102\tilde{b}^{(m)}_{3/2}\right\},\ ] ] which is then inserted in equation ( [ phi2 ] ) to get the perturbing ring s potential @xmath103 \tilde{b}^{(m)}_{3/2}\right\}.\ ] ] the final task of this section is to write the rings coordinates in terms of its orbit elements [ orbits ] @xmath104 where @xmath105 is the true anomaly of a ring element at @xmath38 and @xmath106 is the corresponding mean anomaly . inserting equations ( [ orbits ] ) into the potential @xmath107 , expanding to second order in the @xmath7 s and @xmath8 s , doing the @xmath108 integration in equation ( [ phi3 ] ) , and time averaging the resulting expression over the orbital period of ring @xmath39 then yields the time averaged potential @xmath109 experienced by ring @xmath39 due to ring @xmath37 assuming small eccentricities and inclinations : @xmath110 \end{split}\ ] ] where the @xmath111 and @xmath112 functions are [ fg ] @xmath113 where @xmath114 and @xmath100 has been redefined as @xmath115 . the right hand side of equations ( [ fg ] ) are derived in appendix [ appendixb ] , and it is shown in appendix [ appendixc ] that the softened laplace coefficients can be written in terms of complete elliptic integrals . consequently , functions @xmath111 , @xmath112 , and @xmath116 all can be rapidly evaluated without relying upon a numerical integration of equation ( [ b_soft_int ] ) . it is also noted that the @xmath116 , @xmath111 , and @xmath112 functions obey the following reciprocal relations [ reciprocal ] @xmath117 these relations will be used in section [ lconservation ] to show that the equations of motion developed below conserve angular momentum . the laborious procedure of expanding , integrating , and then time averaging @xmath107 is not included here since a similar analysis can be found in @xcite exerts on @xmath39 , which they insert into the gauss equations to obtain a set of dynamical equations equivalent to that obtained here when @xmath118 . ] . since the terms proportional to @xmath119 and @xmath120 as well as the first term in equation ( [ phi4 ] ) do not contribute to the resulting dynamical equations , they may be neglected . the disturbing function @xmath121 for ring @xmath39 due to another ring @xmath37 is @xmath122 the surviving terms in @xmath109 , _ i.e. _ , @xmath123.\ ] ] note that when @xmath124 the disturbing function for a point mass @xmath39 perturbed by point mass @xmath37 is recovered @xcite , which is to be expected since both a point mass as well as a thin ring have the same disturbing function to this degree of accuracy @xcite . it should also be noted that many celestial mechanics texts develop two distinct expressions for the disturbing function @xmath121 , one due to a perturber in an interior orbit having @xmath125 , and another expression due to an exterior perturber having @xmath126 [ c.f . , @xcite ] . however this pairwise development is unnecessary in this application since equation ( [ r_soft ] ) is valid for @xmath127 as well as for @xmath128 . indeed , it is straightforward to show that these pairs of disturbing functions , such as equations ( 7.6 ) and ( 7.7 ) in @xcite , are in fact equivalent to equation ( [ r_soft ] ) with @xmath124 ; they only appear distinct since one is a function of @xmath100 and the other is actually a function of @xmath129 . in terms of the variables @xmath130 the disturbing function can then be written @xmath131 \end{split}\ ] ] where the mean motion @xmath132 with @xmath133 being the solar mass . for the more general problem of @xmath134 perturbing rings , replace the perturbing ring mass @xmath37 with @xmath135 and give all the other primed quantities the subscript @xmath136 . the disturbing function @xmath137 for the perturbed ring having a mass @xmath138 is sum of equation ( [ r_hk ] ) over all the other rings @xmath139 . setting @xmath140 , @xmath141 , and @xmath142 that sum becomes @xmath143\ ] ] in the notation of @xcite . the @xmath144 and the @xmath145 can be regarded as two @xmath146 matrices @xmath147 and @xmath148 whose entries describe the magnitude of the mutual gravitational interactions that are exerted among the @xmath134 rings . in the following , quantities having a @xmath149 subscript will always refer to the perturbed ring in question while the @xmath136 subscript will always refer to another perturbing ring . variation of the rings orbit elements is given by the lagrange planetary equations ; to lowest order in the @xmath7 s and @xmath8 s these are @xcite [ hkpq_dot ] @xmath150 and their well known laplace lagrange solution is [ solution ] @xmath151 where @xmath152 is the @xmath153 eigenvalue of the @xmath147 matrix , @xmath154 is the @xmath153 eigenvalue of @xmath148 , @xmath155 is the @xmath146 matrix formed from the @xmath134 eigenvectors to @xmath147 and @xmath156 is the matrix of eigenvectors to @xmath148 , and @xmath157 and @xmath158 are integration constants . to apply this rings model , first assign to the @xmath134 rings their masses @xmath159 , their semimajor axes @xmath160 , and their fractional half thickness @xmath161 . planets are represented as thin rings having @xmath162 . then construct the system s @xmath147 and @xmath148 matrices and compute their eigenvalues @xmath152 , @xmath154 and the eigenvector arrays @xmath155 and @xmath156 . the rings initial orbits @xmath163 and @xmath164 are then used to determine the integration constants @xmath157 and @xmath158 as well as to rescale the eigenvectors @xmath155 and @xmath156 such that equations ( [ solution ] ) agree with the initial conditions . a handy recipe for this particular task is given in @xcite . equations ( [ solution ] ) are then used to compute the system s time history , and orbit elements are recovered via @xmath165 , @xmath166 , @xmath167 , and @xmath168 . the rather laborious derivation given above thus confirms the assertion by @xcite that one only needs to soften the laplace coefficients in order to use the laplace lagrange solution for calculating the secular evolution of a continuous disk . however section [ lconservation ] will show that this softening must be done judiciously such that equation ( [ reciprocal b ] ) is obeyed in order for the solution to preserve the system s angular momentum . several tests have been devised in order to demonstrate that the rings model described above behaves as expected . the equations developed above conserve the system s total @xmath64component of angular momentum @xmath169 . to show this , expand @xmath170 to second order in the @xmath171 and @xmath172 , which is the same degree of precision to which the disk potential is developed above . this gives @xmath173 where [ lei ] @xmath174 and @xmath175 is a constant . it is shown in appendix [ appendixd ] that the dynamical equations ( [ hkpq_dot ] ) conserve angular momenta to second order in the @xmath7 s and @xmath8 s , that is , @xmath176 and @xmath177 . when the rings model is used to calculate the secular evolution of a ` sparse ' system , such as the jupiter saturn system described below or one composed of all four giant planets , the angular momenta @xmath178 and @xmath179 are preserved to near machine limits to a fractional precision of @xmath180 in this single floating point implementation . this is true regardless of whether the planets are represented as thin rings having @xmath162 or as thick rings with @xmath181 . however @xmath170 conservation is a little worse , @xmath182 , in the ` crowded ' systems described in section [ results ] that consist of a few planets plus a disk comprised of many closely packed rings . however these errors are always smaller , by a factor of @xmath183 to @xmath184 , than the angular momenta associated with the disturbances and waves seen in these disks . thus the wave like disk behavior reported below is real and is not due to a diffusion of some numerical error . @xcite use the laplace lagrange planetary solution , equations ( [ solution ] ) , to study a system composed of jupiter and saturn . the rings model developed here reproduces that system s @xmath147 and @xmath148 matrices , eigenvector arrays @xmath155 and @xmath156 , eigenvalues @xmath152 and @xmath154 , and integration constants @xmath157 and @xmath158 , to the same precision quoted by @xcite , provided one sets @xmath185 and @xmath162 . the rings model also reproduces the figures in @xcite that show this system s orbital history , as well as the figures showing the forced orbit elements of numerous other massless ` test rings ' orbiting throughout this system . however is should be noted that rigorous angular momentum conservation actually requires setting @xmath141 instead . @xcite points out that a massless test particle orbiting in a smooth , axisymmetric disk experiences a _ regression _ of its longitude of periapse , _ i.e. _ , @xmath186 , which is the opposite of the usual prograde apse precession that occurs throughout the solar system . as @xcite shows , it is the nearby disk parcels whose orbits actually cross the test particle s orbit that drives periapse regression at a rate that exceeds the prograde contribution from the more distant parts of the disk . the rings code also reproduces this phenomenon . an annulus having a surface density @xmath187 , mass @xmath188 , a semimajor axis @xmath189 , and a fractional half thickness @xmath190 contributes @xmath191 \delta m'/8m_\odot$ ] to the particle s disturbing function [ see equation ( [ r_soft ] ) with @xmath192 . adopting a power law in the disk surface density , @xmath193 where @xmath115 , the total disturbing function integrated across a semi infinite disk is @xmath194\ ] ] where @xmath195 is the so called normalized disk mass and [ ii ] @xmath196 note that @xmath197 and @xmath198 are double integrals according to the definitions of @xmath116 and @xmath111 , equations ( [ b_soft_int ] ) and ( [ f ] ) . however these integrals are analytic for selected power laws @xmath199 . for instance , setting @xmath200 or @xmath201 and then instructing symbolic math software such as maple to do the radial integration first and the angular integration second yields @xmath202 and @xmath203/\sqrt{(1+\mathfrak{h'}^2/4 ) ( 1+\mathfrak{h'}^2/2)}\simeq1/\mathfrak{h'}$ ] for small @xmath190 . the lagrange planetary equations then give the test ring s precession rates : [ precession ] @xmath204 very similar precession rates were previously derived in @xcite and @xcite . note that the model rings start to overlap when their fractional radial thickness @xmath205 exceeds the rings fractional separation @xmath206 , so a massless test ring should precess at the above rates when the disk rings are sufficiently overlapping . these expectations are tested by constructing a 50 m@xmath32 disk having an @xmath201 power law surface density using 200 circular , coplanar rings arranged over @xmath207 au , with a number of thin , massless ` test rings ' also orbiting within this disk . several simulations are performed with the massive rings having a variety of thicknesses @xmath190 . as expected , those test rings that reside far from the disk s edges precess at rates given by equation ( [ precession ] ) whenever the disk rings are sufficiently overlapping , namely , when @xmath208 . although a particle s periapse will experience retrograde precession when embedded in an axisymmetric disk , prograde precession is possible in a non axisymmetric disk . for instance , prograde precession is evident in the n body simulations of an eccentric stellar disk orbiting the putative black hole at the center of the andromeda galaxy m31 @xcite . these simulated disks have masses @xmath209 times the central mass , with the interior parts of the disk being progressively more eccentric . these simulations reveal a long lived overdense region in the direction of apoapse that persists due to a coherent alignment of the particles periapse , with the density pattern rotating in a prograde sense at rates that increase with the disk mass . it should be noted that the disk particles eccentricities are not always small , so these n body simulations can not be used as a quantitative benchmark for the rings code . nonetheless , it is comforting to find that the rings code does indeed reproduce the density patterns seen in the n body disks that precess at rates very similar to that reported in @xcite . section [ results ] will use the preceding rings model to demonstrate that apsidal density waves and nodal bending waves may have once propagated throughout the kuiper belt . an apsidal wave is a one armed spiral density wave that slowly rotates over a periapse precession timescale . similarly , a nodal wave is a one armed spiral bending wave that rotates over a nodal precession timescale . a brief review of spiral wave theory is in order . many of the waves properties , such as their wavelength and propagation speed , are readily extracted from the waves dispersion relations . these dispersion relations are usually obtained from solutions to the poisson and euler equations for the disk . however the following will show that these dispersion relations may also be derived from the lagrange planetary equations . the disturbing function @xmath210 for the disk material orbiting at a semimajor axis @xmath9 is obtained from equation ( [ r_soft ] ) with the perturbing mass @xmath37 replaced by the differential mass @xmath211 whose contributions are integrated across a semi infinite disk : @xmath212d\alpha.\ ] ] where @xmath213 and @xmath214 are the orbit elements of the perturbing parts of the disk , the unprimed quantities refer to the perturbed annulus at @xmath9 , @xmath199 is the power law for the disk s surface density variation , and a constant fractional thickness @xmath83 is assumed throughout the disk . the lagrange planetary equations then give the disk s periapse precession rate at @xmath9 : @xmath215 where @xmath216 is from the left term in equation ( [ rdisk ] ) and is the contribution by an undisturbed disk to its own precession [ see equations ( [ ii][precession ] ) ] , and the right term is @xmath217 which is the relative rate at which the density wave drives its own precession . it is expected that the eccentricities associated with a spiral density wave will vary only slowly with distance @xmath9 such that @xmath218 . the spiral wave will also organize the disk s longitude of periapse @xmath47 such that it varies as @xmath219 , where @xmath220 is the wavenumber and @xmath221 is the radial wavelength . if the spiral wave pattern is tightly wound such that @xmath222 and @xmath223 , then the dominant contributions to @xmath224 are largely due to the nearby parts of the disk where @xmath225 and @xmath226 , while the contributions from the more distant parts of the disk tend to cancel owing to the rapid oscillation of the cosine factor . thus we can set @xmath101 in equation ( [ i_wave1 ] ) except where it appears as the combination @xmath227 where @xmath228 . in this case the softened laplace coefficients that are present in the @xmath112 function can be replaced with the approximate forms that are valid for @xmath228 [ see equation ( [ b_local ] ) ] , so @xmath112 becomes @xmath229 it is also permitted to extend the lower integration limit in equation ( [ i_wave1 ] ) to @xmath230 in the tight winding limit , so @xmath231 and finally , if the wave is to remain coherent across this disk , this self precession must occur at the same constant rate @xmath232 throughout the disk , so @xmath233 . note that this disturbing frequency @xmath232 , which is also called the pattern speed , can also be identified as any one of the eigenfrequencies @xmath152 that appear in equations ( [ solution ] ) . usually it is another perturber that is responsible for launching the wave and causing the disk to precess in concert at the rate @xmath232 , and this is usually at a rate that dominates over the non wave contribution to the disk s precession , _ i.e. _ , @xmath234 . this then yields the dispersion relation for tightly wound apsidal density waves : @xmath235 which has the dimensionless form @xmath236 where @xmath237 is the dimensionless frequency and @xmath238 the dimensionless wavelength ; figure [ disp_rel ] plots @xmath239 versus @xmath240 . a more general dispersion relation for an @xmath39armed spiral wave in a stellar disk is given in @xcite , and in the limit that @xmath241 the resulting formula with @xmath242 behaves qualitatively quite similar where the reduction factor @xmath243 is a more complicated function of @xmath240 . however a numerical evaluation of this function shows that @xmath244 has the same form as the @xmath245 curve shown in figure [ disp_rel ] . ] to equation ( [ w_e ] ) . note that equation ( [ edispersion ] ) also recovers the usual dispersion relation @xmath246 for apsidal waves in an infinitesimally thin disk when @xmath118 . note that @xmath247 and that it also has a maximum at @xmath248 where @xmath249 . since @xmath239 is a function of semimajor axis @xmath9 , the restriction @xmath250 indicates that apsidal waves can only propagate in a restricted interval in @xmath9 . this restriction can also be viewed as a constraint on the disk thickness , namely , @xmath251 alternatively , equation ( [ hq ] ) can also be viewed as an upper limit on the frequency @xmath232 or a lower limit on the disk mass @xmath252 wherein wave action is permitted . waves having a wavenumber @xmath253 are called long waves since they have a wavenumber @xmath254 and the longer wavelength @xmath255 , while short waves have wavenumbers @xmath256 or @xmath257 and the shorter wavelength @xmath258 . these long and short density waves also correspond to the @xmath112 and @xmath259 modes , respectively , of @xcite . writing @xmath260 in term of @xmath261 and then requiring @xmath262 also means that apsidal wave can propagate wherever @xmath263 . the rate at which apsidal waves propagate across the disk is given by their group velocity @xmath264 where @xmath265 . waves with @xmath266 are called trailing waves , and their longitude of perihelia @xmath47 decreases with increasing semimajor axis @xmath9 , while @xmath267 are leading waves whose longitude increases with @xmath9 . since the waves group velocity @xmath268 is proportional to the slope of the @xmath245 curve , the site where @xmath248 is a turning point where wave reflection occurs ; in galactic dynamics this reflection site is known as a @xmath1 barrier . a long wave that approaches the @xmath1 barrier from the @xmath253 side of figure [ disp_rel ] thus reflects as a short wave as it continues along the @xmath256 side of the curve . the simulations shown in section [ results ] also show that long trailing waves that instead strike a disk edge also reflect as short trailing waves . the compression or rarefaction among the disk s rings , or streamlines , is @xmath269 , which is also be the relative change in the disk s surface density @xmath270 associated with a density wave [ c.f . , @xcite ] . for a small amount of compression , @xmath271 where the fractional surface density variation is @xmath272 to lowest order in @xmath7 . the second term dominates over the first in the tight winding limit so @xmath273 . density waves are nonlinear when @xmath274 , and these large density variations are a consequence of overlapping streamlines . for long density waves having a wavenumber @xmath275 , the disk s streamlines will cross when the waves eccentricities exceed @xmath276 , while streamline crossing occurs among short waves when eccentricities exceed @xmath277 . as the simulations of section [ results ] will show , a dynamically cool kuiper belt is very susceptible to the propagation of short nonlinear density waves that facilitate streamline crossing . depending upon the relative velocities of these crossed streamlines , apsidal wave action might encourage accretion or else enhance collisional erosion among kbos . the derivation of the dispersion relation for nodal bending waves , and its analysis , proceeds similarly . the disk s integrated disturbing function is @xmath278d\alpha.\ ] ] so the lagrange planetary equation gives @xmath279 where @xmath280 and @xmath281d\alpha \simeq\frac{e^{-\sqrt{2}\mathfrak{h}|ka|}}{\sqrt{2}\mathfrak{h}}\ ] ] is the bending wave s contribution to its own precession . the contribution from the undisturbed disk is @xmath282 , with the additional @xmath283 factor the result of changing the middle argument in equation ( [ iomega ] ) from 0 to @xmath83 . since @xmath284 is a constant for a coherent wave , the dispersion relation for tightly wound nodal bending waves is @xmath285 note that the usual dispersion relation for nodal waves in an infinitesimally thin disk , @xmath286 , is obtained when @xmath287 . the dimensionless form of the dispersion relation is @xmath288 where @xmath289 , which is plotted in figure [ disp_rel ] . as the figure shows , nodal bending waves can propagate only in regions where @xmath290 which similarly limits the disk thickness to @xmath291 . the nodal waves group velocity is @xmath292 which indicates that nodal waves tend to stall , _ i.e. _ , @xmath293 as they approach the @xmath294 boundary . note that this dispersion relation only admits a long wavelength solution having a wavenumber @xmath295 and a wavelength @xmath296 for waves far from the stall zone . since @xmath291 , the disk can sustain nodal waves wherever @xmath297 . but if an outward traveling long leading wave with @xmath267 encounters a disk edge , it will reflect as an @xmath266 long trailing wave . examples of nodal wave reflection and stalling are also given below . using the recipe given in section [ n rings ] , the rings model is used to compute the secular evolution of the primordial kuiper belt as it is perturbed by the four giant planets . in these simulations the giant planets are represented by thin @xmath118 rings whose initial orbits are their current orbits , while the kuiper belt is represented by 500 rings whose semimajor axes extend from 36 au out to 5070 au . the location of the belt s inner edge is chosen such that only the outermost radial and vertical secular resonances , the @xmath298 and the @xmath299 , reside in this disk near 40 au when of low mass . the semimajor axes of each belt ring increases as @xmath300 where the rings fractional separation @xmath301 is typically @xmath302 . the rings fractional half thickness @xmath83 is always in excess of @xmath303 , as is required to get the correct apse precession in an axisymmetric disk ( see section [ axisymmetric disk ] ) . the belt rings initial eccentricities and inclinations are zero , with all inclinations being measured with respect to the system s invariable plane . the mass of each ring is chosen such that the belt s surface density @xmath304 varies as @xmath305 . for this configuration , if the total kuiper belt mass over the @xmath306 au zone is @xmath307 , then the total mass in the ` observable ' @xmath308 au zone that is presently accessible to astronomers would be @xmath309 had the above surface density law extended inwards to 30 au . for these systems the normalized disk mass is @xmath310 . figure [ m_kb=10 ] shows a snapshot of apsidal density waves as they propagate across an @xmath311 m@xmath32 kuiper belt having a half thickness @xmath312 . since @xmath313 , the necessary disk conditions for the propagation of apsidal and nodal waves are well satisfied . initially , a long trailing density wave is launched at the belt s inner edge . this wave is really more like a pulse @xmath314 au wide , and figure [ m_kb=10 ] shows that by time @xmath315 years the wave has just started to reflect at the disk s outer edge at 70 au . the greyscale map of the disk s surface density variation , @xmath316 , is obtained using equation ( [ delta - sigma ] ) , and this map also provides a historical record of this system s wave action . it should be noted that equation ( [ delta - sigma ] ) is quantitatively correct only when @xmath317 , a condition that is rarely satisfied by the results obtained here . nonetheless , equation ( [ delta - sigma ] ) is still useful in a qualitative sense since it will indicate the disk regions where large surface density variations as well as orbit crossing can be expected . as the outer edge of the @xmath316 map shows , the main apsidal density wave pulse at @xmath318 au has just reflected at a short trailing wave , and this nonlinear wave having @xmath319 will completely dominate the disk s appearance at later times as it propagates inwards . but until this happens , the bulk of the disk s appearance over @xmath320 au is still dominated by lower amplitude long waves that are following behind the main density pulse . note also that short leading waves were also emitted at the disk s inner edge , but as the density greyscale shows , they have only propagated out to @xmath321 au thus far owing to their slower group velocity [ see equation ( [ density c_g ] ) ] . these short waves are well resolved in the sense that their radial wavelength of @xmath322 au spans about 15 disk rings . although these short waves are seen in the @xmath323 and @xmath324 plots as only tiny wiggles over @xmath325 au that are superimposed on top of the disk s longer wavelength behavior , the density greyscale shows that the short waves dominate the inner disk s appearance . the online edition of fig [ m_kb=10 ] is also linked to an animated sequence of these figures that give the system s complete time history . that animation shows that by time @xmath326 years , nonlinear short waves will have swept across the entire disk , and they result in large surface density variations @xmath319 over radial wavelengths of @xmath322 au . figure [ m_kb=10 ] also shows that by time @xmath315 years , a long leading nodal bending wave pulse has already propagated across the disk where it has reflected at the disk s outer edge and just started to return inwards as a long trailing bending wave . but when that pulse reaches the disk s inner edge , a portion of the wave s angular momentum content will continue to propagate further inwards where it will give a small kick to the giant planets orbit elements , while the remaining wave pulse reflects again and propagates outwards . the same phenomenon also occurs among the apsidal density waves . thus after a few reflections , a single wave pulse will lose its initial spatial coherence by spawning multiple wave trains that , in this friction free model , forever roam about the belt . this ultimately results in a rather wobbly looking standing density wave pattern that varies over the short wavelength scale @xmath322 au , as well as a standing bending wave pattern that varies over a much longer scale @xmath327 au . a dynamical ` spectrum ' of this kuiper belt is shown in figure [ eigenvectors ] , which plots all of the eccentricity eigenvector elements @xmath328 for each of the disk rings versus their eigenfrequency @xmath152 , as well as inclination eigenvector elements @xmath329 versus eigenfrequency @xmath330 . the upper figure is quite reminiscent of the findings of @xcite , who showed that a gravitating disk tends to exhibit its strongest response to slow radial perturbations via modes having discrete patterns speeds @xmath232 that can be identified with any of the peak frequencies @xmath152 seen in figure [ eigenvectors ] . figures such as these are quite useful for identifying the patterns speeds @xmath232 that are associated with the density and bending waves that propagate in the disk . simulations have been performed with kuiper belts having masses @xmath332 , 0.08 , 0.2 , 2 , 10 , and 30 m@xmath32 in the observable @xmath333 au zone ( again this would be the belt s mass assuming its surface density @xmath334 were to extend solely over that region ) . the results are summarized in figure [ compare_ei ] which shows the maximum eccentricities @xmath335 and maximum inclinations @xmath336 achieved by the rings in each simulation . the belt s radial width is indicated by the breadth of the curves in figure [ compare_ei ] , which ranges from 34 au in the higher mass belts to 14 au for the @xmath337 m@xmath32 system . each simulation uses 500 disk rings having a fractional half thickness @xmath338 , so the three higher mass systems have @xmath339 while the lower mass disk @xmath340 m@xmath32 is somewhat thinner with @xmath341 and the @xmath337 m@xmath32 system has @xmath342 . the rings in these lower mass disks are more closely packed so that their shorter wavelength density waves are well resolved , and they are also made thinner so as to push their @xmath1barriers a bit further downstream . the disk s initial orbits are @xmath343 , except for the @xmath332 system which instead adopts the forced orbit elements appropriate for a massless disk [ e.g. , @xcite ] . these systems are evolved until their initial density and bending wave pulses have had the opportunity to reflect multiple times and have largely dissolved into standing wave patterns . the lower mass disks necessarily have longer run times due to their waves slower propagation speeds [ see equation ( [ density c_g ] ) ] ; these run times are @xmath344 years , @xmath345 years , @xmath346 years , @xmath347 years , and @xmath348 years , respectively , for the @xmath337 , 0.2 , 2 , 10 , and 30 m@xmath32 systems . once the standing wave pattern has emerged , the rings instantaneous eccentricities and inclinations range over @xmath349 and @xmath350 . as figure [ compare_ei ] shows , there are no peaks in the @xmath335 and @xmath336 curves for the higher mass disks having @xmath351 m@xmath32 , indicating that there are no secular resonances in the disk itself ; any such resonances likely lie between the orbits of neptune and the disk s inner edge at 36 au . the simulations of these higher mass disks show long trailing density waves and long leading bending waves being launched at the disk s inner edge . these waves sweep across the disk , reflect at the disk s outer edge , and return as short trailing density waves and long trailing bending waves , similar to the history described in section [ example ] and seen in figure [ m_kb=10 ] . however secular resonances at @xmath352 au are quite prominent in the lower mass disks having @xmath337 and 0.2 m@xmath32 . these are sites that launch long density and bending waves , and as figure [ m_kb=10 ] shows , the density waves are able to propagate out as far as @xmath321 and 49 au , respectively , where they are reflected by a @xmath1barrier and return as short density waves . these reflection sites occur where @xmath353 and @xmath354 , which shows that equation ( [ hq ] ) is an approximate yet useful indicator of where apsidal wave are allowed to propagate . note also the large amplitudes of the density waves in these low mass disks , @xmath355 , which clearly violates the model s assumption of low eccentricities . thus these particular curves should not be taken at face value . nonetheless , they do indicate that apsidal waves in a low mass kuiper belt may result in large eccentricities , and figure [ m_kb=10 ] shows that sizable inclinations can also result due to nodal bending waves . indeed , in the @xmath337 m@xmath32 disk , maximum inclinations are typically @xmath29 , which is comparable to the main belt s high inclination component . the larger @xmath7 s and @xmath8 s seen in the lower mass disk s are a consequence of the giant planets transmitting a small fraction of their initial angular momentum deficit and @xmath179 of equations ( [ lei ] ) . ] ( amd ) into the disk in each simulation . in each simulation , the giant planets deposit @xmath356 and @xmath357 into the disk s inner edge , which waves then transport and smear out across a vast swath of the kuiper belt . since the amd deposited in the disk is roughly constant in each simulation , the lower mass disks exhibit larger @xmath7 and @xmath8 excitations . it should also be noted that computational limitations in dynamical studies of the kuiper belt , particularly n body simulations , are often require treating the belt as a swarm of massless test particles . however the comparison of the @xmath332 curve to the @xmath358 curves in figure [ m_kb=10 ] shows that the endstate of a kuiper belt having even just a modest amount of mass can be radically different from one naively treated as massless . figure [ m=10 ] shows how the response of an @xmath311 m@xmath32 belt varies with increasing disk thickness @xmath359(0.0027 , 0.027 , 0.04 , 0.08 , 0.13 ) . according to equation ( [ hq ] ) , @xmath360 in these @xmath361 disks , so the @xmath1barrier will move inwards as the disk s @xmath83 is increased , as is evident in figure [ m=10 ] . all of these disks have a normalized disk mass @xmath362 , a mean motion @xmath363 radians / year , and a pattern speed that is typically @xmath364 radians / year , so wave action is shut off when the disk thickness @xmath83 exceeds @xmath365 . in the thinnest disk with @xmath339 , the @xmath1barrier lies beyond the disk s outer edge at 70 au , so long and then short apsidal density waves are able to slosh about the disk s full extent . however the @xmath1barrier lies in the disk at @xmath366 au when @xmath367 ( red curve ) , at @xmath368 au when @xmath369 ( green curve ) , and apsidal waves are prohibited in the disks with @xmath370 ( blue and purple curves ) . figure [ m=10 ] also shows that the nodal bending waves are shut off when the disk thickness @xmath83 exceeds the somewhat more relaxed criterion @xmath371 . note also the peak at @xmath368 au in the @xmath367 disk ( red curve ) and at @xmath372 au in the @xmath369 disk ( green curve ) . these particular disks admit two spiral patterns , a higher amplitude spiral that corotates with neptune s dominant eigenmode at the pattern speed of @xmath373 radians / year , and a lower amplitude mode that corotates with uranus at the faster rate @xmath374 radians / year . since @xmath375 , the faster spiral pattern has @xmath376 , which is why this wave stalls at @xmath377 au in the @xmath378 disks while the slower spiral mode is able to propagate the full breadth of these disks . the behavior of a lower mass disk with @xmath340 m@xmath32 is shown in figure [ m=0.2 ] for disks having @xmath379(0.0015 , 0.0037 , 0.0074 , 0.011 ) . a pair of secular resonances lie near @xmath380 au , and they launch apsidal and nodal waves having pattern speeds @xmath381 radians / year . these disks have @xmath382 and @xmath383 , and as the upper plot shows , further increases in @xmath83 brings the @xmath1barrier inwards until the wave propagation zone has shrunk down to zero . the lower plot also shows that nodal waves forever slosh about the model kuiper belt in the @xmath341 disk ( orange curve ) , whereas the peaks in the other curves show that these waves stall at sites ever closer to resonance in progressively thicker disks . these figures also show that when @xmath384 , the motions of a non gravitating disk are recovered , namely , the disk s @xmath7 s and @xmath8 s are twice the forced motions seen in the @xmath332 disk ( black curves ) , with the factor of two being a consequence of these disks initial conditions @xmath343 . the primordial kuiper belt likely had an initial mass of @xmath3 m@xmath32 ( see section [ introduction ] ) , and accretion models show that the initial kbo swarm must have had dispersion velocities less than @xmath385 keplerian @xcite , so @xmath386 , @xmath387 , and thus @xmath388 assuming the spiral waves have pattern speeds of @xmath389 radians / year . since @xmath262 , the primordial kuiper belt readily sustained apsidal and nodal waves . figure [ compare_ei ] shows that in this high mass environment , these will be rather low amplitude waves having @xmath2 . these waves will quickly propagate across a kuiper belt of width @xmath390 in time @xmath391 , so the wave propagation time is @xmath392 in the friction free model employed here , the outward bound long density waves eventually reflect , either at a @xmath1barrier in the disk ( which might lie downstream where @xmath393 ) or else at the disk s outer edge . the reflected waves then propagate inwards as short density waves , and such waves are nonlinear in the sense that their surface density variations @xmath316 typically exceed unity . figure [ m_kb=10 ] shows a snapshot of long and short density waves in a @xmath311 m@xmath32 disk . note that the long waves completely dominate the disk s orbit elements @xmath323 and @xmath324 that vary over a wavelength of @xmath394 au , while the short waves are the tiny variations in @xmath323 and @xmath324 that occur over a @xmath322 au scale at the disk s inner and outer edge . even though the short waves are almost imperceptible in the orbit elements plots , the greyscale map shows that these nonlinear short waves completely dominate the disk s surface density structure . figure [ compare_ei ] also shows that the waves in lower mass systems have higher amplitudes . this suggests that wave action may tend to drive disk planet systems towards an equipartition of angular momentum deficits , since the angular momentum content of the waves seen in all simulations are @xmath395 and @xmath396 of the planets @xmath178 and @xmath179 . the large amplitude waves seen in the @xmath340 m@xmath32 disk also suggests that apsidal and nodal wave action might account for much of the kuiper belt s excited state ( see figure [ compare_ei ] ) . however the viability of this scenario depends critically upon the timescales that govern the belt s mass loss as well as the kbos velocity evolution . first note that accretion models show that as soon as the large @xmath6 km kbos form , further kbo growth is halted as they initiate an episode of more vigorous collisions that steadily grind the belt s smaller members down to dust which radiation forces then transport away @xcite . accretion models show that the @xmath397 km kbos form at @xmath398 au over a @xmath399 year timescale , and that the belt s subsequent collisional erosion occurs over a @xmath400 year timescale @xcite . and if neptune s orbit had migrated substantially , the advancing 2:1 resonance would also have swept across the main belt , which further enhances the stirring as well the collisional / dynamical erosion . the resulting grinding and erosion thus makes it ever more difficult for the belt to sustain waves as the initially massive kuiper belt mass is eroded by a factor of @xmath35 , which also lowers the disk s @xmath401 . note that the removal of the smaller kbos also turns off the dynamical friction that was once keeping the particle disk quite thin . the gravitational stirring by the surviving larger kbos is then free to raise their dispersion velocities @xmath49 up to their surface escape velocity @xmath402 @xcite , which results in a thicker disk with @xmath403 where @xmath121 is the kbos characteristic size . since this is substantially larger than the current belt s @xmath261 , it seems quite likely that this stirring shut off the waves while they were still of low amplitude ( see fig . [ compare_ei ] ) long before the kuiper belt was ground down to its present mass . consequently , apsidal and nodal waves were likely able to propagate throughout the kuiper belt during a timescale @xmath404 that is bounded by the moment when the large kbos first formed and when the belt eroded away , _ i.e. _ , @xmath405 . it should also be noted that this implementation of the rings model does not account for any viscous damping of spiral waves which , as @xcite point out , may be more important as bending waves approach a disk s outer edge . unlike the sharp edged disks employed here , a more realistic disk likely has a ` fuzzy ' edge where the disk s surface density gently tapers to zero . @xcite show that bending waves entering such a lower density zone tend to excite substantially larger inclinations there , and such sites will be considerably more susceptible to the viscous dissipation of these possibly nonlinear waves . @xcite suggests that disks having a tapered edge might also cause bending waves to stall there since the group velocity @xmath406 as the surface density @xmath407 smoothly goes to zero . however this assertion was not confirmed by the rings model , which tapered a disk s outer surface density by multiplying it by the factor @xmath408 where @xmath390 is the distance from the outer edge and @xmath63 is the tapering scale length . these experiments adopt values of @xmath63 that are smaller , comparable , and larger than the bending wavelength @xmath260 , and in all cases the bending wave reflects at or near the disk edge , with considerably larger inclinations being excited in this tapered zone . a model that rapidly computes the secular evolution of a gravitating disk planet system is developed . the disk is treated as a nested set of rings , with the rings/planets time evolution being governed by the lagrange planetary equations . it is shown that the solution to the dynamical equations is a modified version of the classical laplace lagrange solution for the secular evolution of the planets @xcite , with the modification being due to a ring s finite thickness @xmath409 that is a consequence of the dispersion velocity @xmath49 of that ring s constituent particles . since the ring s finite thickness @xmath0 softens its gravitational potential , this also softens the laplace coefficients appearing in the laplace lagrange solution over a scale @xmath410 . it is shown that the lagrange planetary equations admit spiral wave solutions when the tight winding approximation is applied . there are two types of spiral density ( or apsidal ) waves , long waves of wavelength @xmath411 and short waves of wavelength @xmath412 where @xmath413 is the disk s fractional thickness and @xmath232 is the angular rate at which the spiral pattern rotates . the simulations presented here show that the giant planets launch long waves at either a resonance in the disk or else at the disk s nearest edge , and that these waves propagate away until they reflect at the disk s far edge or else at a @xmath1barrier in the disk which resides where @xmath414 where @xmath415 is the maximum disk thickness that can sustain apsidal waves , with @xmath416 being the normalized disk mass . of course all of these findings may be derived from the stellar dispersion relation given in @xcite in the limit that the pattern speed @xmath232 is much smaller than the disk s mean motion @xmath51 . nonetheless , it is satisfying to see that the theory of unforced apsidal waves is readily obtained from the lagrange planetary equations ; with a little more effort the theory for forced apsidal waves [ e.g. , @xcite ] should also be recoverable . however new results are obtained for the nodal wave problem , which admits only a long wavelength solution @xmath260 to the planetary equations in the tight winding limit . in particular , it is shown that these waves can stall , that is , the waves group velocity plummets to zero as they approach a site in the disk where @xmath417 . if , however , these waves instead encounter a disk edge , they will reflect and return as long waves . in the limit that @xmath287 , the results for nodal waves propagating in a infinitesimally thin disk is recovered @xcite , but note that the wave stalling phenomenon does not appear in a @xmath118 treatment of the disk . the rings model is also used to examine the propagation of apsidal and nodal waves that are launched by the giant planets into a variety of kuiper belts having a mass @xmath418 m@xmath32 ( the estimated primordial mass ) down to @xmath337 m@xmath32 ( which is @xmath419 of the belts current mass estimate ) . in each simulation the giant planets deposit roughly the same fraction of their initial angular momentum deficits , @xmath395 and @xmath396 of the planets @xmath178 and @xmath179 , into the disk in the form of spiral waves . and since the waves angular momentum content is roughly the same in each simulation , the lower mass kuiper belts thus experience higher amplitude waves . indeed , the waves seen in the @xmath420 m@xmath32 simulations are of sufficient amplitude that they could in principle account for much of the dynamical excitation that is observed in the kuiper belt . however wave action in a @xmath421 m@xmath32 belt also requires its fractional scale height to quite thin , namely , @xmath422 . most likely , apsidal and nodal waves were shut off , due to self stirring by large kbos as well as by other external perturbations , long before the belt eroded down to its current mass , in which case the excitation by wave action would have been quite modest . the rings model developed here has many other applications . one issue of great interest is to determine whether apsidal and nodal waves may be propagating in saturn s rings . of particular interest are the short apsidal waves since their detection could yield the ring s dispersion velocity @xmath49 via a measurement of the short wavelength @xmath423 . although the ring particles dispersion velocity is of fundamental importance to ring dynamics , it is less than well constrained at saturn . of course , the differential precession due to planetary oblateness also needs to be included in the model [ c.f . , @xcite ] since this effect may actually defeat this form of wave action . the rings model can also be used to examine the forced motions of a relatively massless but much thicker circumstellar dust disk like @xmath73 pictoris . the warps and brightness asymmetries seen in this system are usually attributed to secular perturbations exerted by an unseen planetary system , and the code developed here can be used to very rapidly explore the wide range of planetary parameters . this rings model will be used to study these and other problems in greater detail in the near future . * acknowledgments * this paper is contribution xxxx from the lunar and planetary institute , which is operated by the universities space research association by cooperative agreement ncc5679 with the national aeronautics and space administration . this research was also supported by nasa via the origins of solar systems grant no . nag510946 issued through the office of space science . differentiating the @xmath424 appearing in the @xmath111 and @xmath112 functions , equations ( [ fg ] ) , yields can be used to simplify equations ( [ fg1 ] ) further . equations ( [ recursion1 ] ) are derived in @xcite for the case where @xmath427 . however the more general relations given above are readily obtained by replacing the combination @xmath428 appearing in the @xcite recursion relations with @xmath429 . so for @xmath242 and @xmath430 equation ( [ recursion1a ] ) is @xmath431 and @xmath432 for @xmath433 and @xmath430 while equation ( [ recursion1b ] ) yields @xmath434 for @xmath242 and @xmath430 . inserting equations ( [ recursion2][recursion4 ] ) into ( [ fg1 ] ) then yields the symbolic mathematics software maple has been used to write the needed softened laplace coefficients @xmath116 , equation ( [ b_soft_int ] ) , in term of complete elliptic integrals @xmath240 and @xmath436 . setting @xmath114 , @xmath437 , and @xmath438 , then [ b ] @xmath439 } { \pi\sqrt{2\alpha(\gamma+1)}}\\ \tilde{b}^{(0)}_{3/2}(\alpha,\mathfrak{h},\mathfrak{h'})= & \frac{2e(\chi)}{\pi\alpha(\gamma-1)\sqrt{2\alpha(\gamma+1)}}\\ \tilde{b}^{(1)}_{3/2}(\alpha,\mathfrak{h},\mathfrak{h'})= & \frac{2\left[-(\gamma-1)k(\chi)+\gamma e(\chi)\right ] } { \pi\alpha(\gamma-1)\sqrt{2\alpha(\gamma+1)}}\\ \tilde{b}^{(2)}_{3/2}(\alpha,\mathfrak{h},\mathfrak{h'})= & \frac{2\left[-4\gamma(\gamma-1)k(\chi)+ ( 4\gamma^2 - 3)e(\chi)\right ] } { \pi\alpha(\gamma-1)\sqrt{2\alpha(\gamma+1)}}\\ \tilde{b}^{(0)}_{5/2}(\alpha,\mathfrak{h},\mathfrak{h'})= & \frac{4\left[-(\gamma-1)k(\chi)+4\gamma e(\chi)\right ] } { 3\pi(2\alpha)^{5/2}(\gamma+1)^{3/2}(\gamma-1)^{2}}\\ \tilde{b}^{(1)}_{5/2}(\alpha,\mathfrak{h},\mathfrak{h'})= & \frac{4\left[-\gamma(\gamma-1)k(\chi)+ ( \gamma^2 + 3)e(\chi)\right ] } { 3\pi(2\alpha)^{5/2}(\gamma+1)^{3/2}(\gamma-1)^{2}}\end{aligned}\ ] ] where @xmath440 is the complete elliptic integral of the first kind and @xmath441 is the complete elliptic integral of the second kind . the series expansions for @xmath442 and @xmath443 given in @xcite then permit the rapid calculation of the softened laplace coefficients @xmath116 without requiring a numerical integration of equation ( [ b_soft_int ] ) . however equations ( [ b ] ) can give unreliable results for extreme values of @xmath100 due to numerical roundoff errors . in this case it is preferable to factor the @xmath444 term out of the integrand in ( [ b_soft_int ] ) and expand the denominator for the case of large @xmath445 : another useful form for @xmath116 is obtained for regions where @xmath452 where @xmath228 . in this ` local ' approximation , the dominant contribution to the integral , equation ( [ b_soft_int ] ) , occurs where @xmath453 . thus we can set @xmath454 and @xmath455 except where it appears as @xmath456 , extend the upper integration limit to infinity , and set @xmath457 in the numerator [ cf . @xcite ] : @xmath458 where @xmath460 . let @xmath461 where @xmath462 is the sum over the @xmath463 terms and @xmath464 the sum over the @xmath465 . swap the @xmath149 and @xmath136 indices in @xmath464 so that it becomes @xmath466 equation ( [ g_reciprocal ] ) shows that @xmath467 , and with @xmath468 , @xmath464 becomes @xmath469 which is @xmath462 since the sums obey @xmath470 . consequently @xmath471 , and a similar analysis will also show that @xmath177 .

we present a pedagogical review of some of the methods employed in eulerian computational fluid dynamics ( cfd ) . fluid mechanics is governed by the euler equations , which are conservation laws for mass , momentum , and energy . the standard approach to eulerian cfd is to divide space into finite volumes or cells and store the cell - averaged values of conserved hydro quantities . the integral euler equations are then solved by computing the flux of the mass , momentum , and energy across cell boundaries . we review both first - order and second - order flux assignment schemes . all linear schemes are either dispersive or diffusive . the nonlinear , second - order accurate total variation diminishing ( tvd ) approach provides high resolution capturing of shocks and prevents unphysical oscillations . we review the relaxing tvd scheme , a simple and robust method to solve systems of conservation laws like the euler equations . a 3-d relaxing tvd code is applied to the sedov - taylor blast wave test . the propagation of the blast wave is accurately captured and the shock front is sharply resolved . we apply a 3-d self - gravitating hydro code to simulating the formation of blue straggler stars through stellar mergers and present some numerical results . a sample 3-d relaxing tvd code is provided in the appendix .

astro-ph0210611

astrophysical structure formation and the dynamics of astrophysical systems involve nonlinear gas dynamical processes which can not be modeled analytically but require numerical methods . one would like to address the challenging problem of star formation and how this process produces planetary systems . observations of the x - ray emission from hot gas in galaxy clusters , the sunyaev - zeldovich effect in the cmb spectrum , and the lyman alpha forest in the spectra of quasars are only meaningful if we understand the gas dynamical processes involved . the evolution of complex systems is best modeled using numerical simulations . a large class of astrophysical problems involve collisional systems where the mean free path is much smaller than all length scales of interest . hence , one can appropriately adopt an ideal fluid description of matter where the thermodynamical properties of the fluid obey well known equations of state . conservation of mass , momentum , and energy allows one to write down the euler equations which govern fluid mechanics ( see * ? ? ? this formalism is an ideal basis for simulating astrophysical fluids . hydrodynamical simulations are faced with challenging problems , but advancements in the field have made it an important tool for theoretical astrophysics . one of the main challenges in simulating complex fluid flows is the capturing of strong shocks , which frequently occur and play an important role in gas dynamics . the differential euler equations are ill - defined at shock discontinuities where derivatives are infinite . much effort has been devoted to solving this problem and a field of work has resulted from it . computational fluid dynamics ( cfd ) is a powerful approach to simulating fluid flow with emphasis on high resolution capturing of shocks and prevention of numerical instabilities . both eulerian and lagrangian methods have been developed . lagrangian methods based on smoothed particle hydrodynamics ( sph ; * ? ? ? * ; * ? ? ? * ) consider a monte - carlo approximation to solving the fluid equations , somewhat analogous to @xmath0-body methods for the vlasov equation . sph schemes follow the trajectories of particles of fixed mass which represent fluid elements . the lagrangian forms of the euler equations are solved to determine smoothed fluid variables like density , velocity , and temperature . the particle formulation does not naturally capture shocks and artificial viscosity is added to prevent unphysical oscillations . however , the addition of artificial viscosity broadens shocks over several smoothing lengths and degrades the resolution . the lagrangian approach has a large dynamic range in length but not in mass . it achieves good spatial resolution in high density regions but does poorly in low density regions . sph schemes must smooth over a large number of neighbouring particles , making it computationally expensive and challenging to implement in parallel . the standard approach to eulerian methods is to discretize the problem and solve the integral euler equations on a cartesian grid by computing the flux of mass , momentum , and energy across grid cell boundaries . in conservative schemes , the flux taken out of one cell is added to the neighbouring cell and this ensures the correct shock propagation . flux assignment schemes based on the _ total variation diminishing condition _ @xcite have been designed for high order accuracy and high resolution capturing of shocks , while preventing unphysical oscillations . the eulerian approach has a large dynamic range in mass but not in length , opposite to that of lagrangian schemes . in general , eulerian algorithms are computationally faster by several orders of magnitude . they are also easy to implement and to parallelize . the purpose of this paper is to present a pedagogical review of some of the methods employed in eulerian computational fluid dynamics . in @xmath1 we briefly review the euler equations and discuss the standard approach to discretizing conservation laws . we describe traditional central differencing methods such as the lax - wendroff scheme in @xmath2 and more modern flux assignment methods like the tvd scheme in @xmath3 . in @xmath4 we review the relaxing tvd method for systems of conservation laws like the euler equations , which has been successfully implemented for simulating cosmological astrophysical fluids by @xcite . in @xmath5 we apply a self - gravitating hydro code to simulating the formation of blue straggler stars through stellar mergers . a sample 3-d relaxing tvd code is provided in the appendix . the euler equations which govern hydrodynamics are a system of conservation laws for mass , momentum , and energy . in differential conservation form , the continuity equation , momentum equation , and energy equation are given as : @xmath6 \frac{{\partial}(\rho v_i)}{{\partial t}}+\frac{{\partial}}{{\partial x}_j}(\rho v_iv_j+p\delta_{ij})=0\ , \\[8pt ] \frac{{\partial}e}{{\partial t}}+\frac{{\partial}}{{\partial x}_j}[(e+p)v_j]=0\ .\end{gathered}\ ] ] we have omitted gravitational and other source terms like heating and cooling . the physical state of the fluid is specified by its density @xmath7 , velocity field @xmath8 , and total energy density , @xmath9 in practice , the thermal energy @xmath10 is evaluated by subtracting the kinetic energy from the total energy . for an ideal gas , the pressure @xmath11 is related to the thermal energy by the equation of state , @xmath12 where @xmath13 is the ratio of specific heats . another thermodynamic variable which is of importance is the sound speed @xmath14 which is given by @xmath15 the thermodynamical properties of an ideal gas obey well known equations of state , which we do not fully list here . the differential euler equations require differentiable solutions and therefore , are ill - defined at jump discontinuities where derivatives are infinite . in the literature , nondifferentiable solutions are called _ weak solutions_. the differential form gives a complete description of the flow in smooth regions , but the integral form is needed to properly describe shock discontinuities . in integral conservation form , the rate of change in mass , momentum , and energy is equal to the net flux of those conserved quantities through the surface enclosing a control volume . for simplicity of notation , we will continue to express conservation laws in differential form , as a shorthand for the integral form . the standard approach to eulerian computational fluid dynamics is to discretize time into discrete steps and space into finite volumes or cells , where the conserved quantities are stored . in the simplest case , the integral euler equations are solved on a cartesian cubical lattice by computing the flux of mass , momentum , and energy across cell boundaries in discrete time steps . consider the euler equations in vector differential conservation form , @xmath16 where @xmath17 contains the conserved physical quantities and @xmath18 represents the flux terms . in practice , the conserved cell - averaged quantities @xmath19 and fluxes @xmath20 are defined at integer grid cell centres @xmath21 . the challenge is to use the cell - averaged values to determine the fluxes @xmath22 at cell boundaries . in the following sections , we describe flux assignment methods designed to solve conservation laws like the euler equations . for ease of illustration , we begin by considering a 1-d scalar conservation law , @xmath23 where @xmath24 and @xmath25 is a constant advection velocity . equation ( [ eqn : advect ] ) is referred to as a linear advection equation and has the analytical solution , @xmath26 the linear advection equation describes the transport of the quantity @xmath27 at a constant velocity @xmath25 . in integral flux conservation form , the 1-d scalar conservation law can be written as @xmath28 where @xmath29 and @xmath30 for our control cells . let @xmath31 denote the flux of @xmath27 through cell boundary @xmath32 at time @xmath33 . note then that the second integral is simply equal to @xmath34 . the rate of change in the cell - integrated quantity @xmath35 is equal to the net flux of @xmath27 through the control cell . for a discrete time step , the discretized solution for the cell - averaged quantity @xmath36 is given by @xmath37 the physical quantity @xmath27 is conserved since the flux taken out of one cell is added to the neighbouring cell which shares the same boundary . note that equation ( [ eqn : conservation ] ) has the appearance of being a finite difference scheme for solving the differential form of the 1-d scalar conservation law . this is why the differential form can be used as a shorthand for the integral form . central - space finite - difference methods have ease of implementation but at the cost of lower accuracy and stability . for illustrative purposes , we start with a simple first - order centered scheme to solve the linear advection equation . the discretized solution is given by equation ( [ eqn : conservation ] ) where the fluxes at cell boundaries , @xmath38 are obtained by taking an average of cell - centered fluxes @xmath39 . the discretized first - order centered scheme can be equivalently written as @xmath40 in this form , the discretization has the appearance of using a central difference scheme to approximate spatial derivatives . hence , centered schemes are often referred to as central difference schemes . in practice when using centered schemes , the discretization is done on the differential conservation equation rather than the integral equation . this simple scheme is numerically unstable and we can show this using the _ von neumann _ linear stability analysis . consider writing @xmath41 as a discrete fourier series : @xmath42 where @xmath0 is the number of cells in our periodic box . in plane - wave solution form , we can write this as @xmath43\ , \ ] ] where @xmath44 are the fourier series coefficients for the initial conditions @xmath45 . equivalently , the time evolution of the fourier series coefficients in equation ( [ eqn : dfte ] ) can be cast into a plane - wave solution of the form , @xmath46 where the numerical dispersion relation @xmath47 is complex in general . the imaginary part of @xmath48 represents the growth or decay of the fourier modes while the real part describes the oscillations . a numerical scheme is linearly stable if @xmath49 . otherwise , the fourier modes will grow exponentially in time and the solution will blow up . the exact solution to the linear advection equation can be expressed in the form of equation ( [ eqn : advectsol ] ) or by a plane - wave solution where the dispersion relation is given by @xmath50 . the waves all travel at the same phase velocity @xmath51 in the exact case . the centrally discretized linear advection equation ( equation [ eqn : cfde ] ) is exactly solvable . after @xmath52 times steps , the time evolution of the independent fourier modes is given by @xmath53 where @xmath54 and @xmath55 . the dispersion relation is given by @xmath56\ , \ ] ] for any time step @xmath57 , the imaginary part of @xmath48 will be @xmath58 . the fourier modes will grow exponentially in time and the solution will blow up . hence , the first - order centered scheme is numerically unstable . the lax - wendroff scheme @xcite is second - order accurate in time and space and the idea behind it is to stabilize the unstable first - order scheme from the previous section . consider a taylor series expansion for @xmath59 : @xmath60 + \frac{{\partial u}}{{\partial t}}{\delta t}+\frac{{\partial}^2u}{{\partial t}^2}\frac{{\delta t}^2}{2}+{\cal o}({\delta t}^3)\ , \end{gathered}\ ] ] and replace the time derivatives with spatial derivatives using the conservation law to obtain @xmath60 -\frac{{\partial f}}{{\partial x}}{\delta t}+\frac{{\partial}}{{\partial x}}\left(\frac{{\partial f}}{{\partial u}}\frac{{\partial f}}{{\partial u}}\frac{{\partial u}}{{\partial x}}\right)\frac{{\delta t}^2}{2}+{\cal o}({\delta t}^3)\ . \label{eqn : taylor}\end{gathered}\ ] ] for the linear advection equation , the eigenvalue of the flux jacobian is @xmath61 . discretization using central differences gives @xmath62 + \left(\frac{f_{n+1}^t - f_n^t}{{\delta x}}-\frac{f_n^t - f_{n-1}^t}{{\delta x}}\right)\frac{v{\delta t}^2}{2{\delta x}}\ . \label{eqn : lwe}\end{gathered}\ ] ] in conservation form , the solution is given by equation ( [ eqn : conservation ] ) , where the fluxes at cell boundaries are defined as @xmath63 compare this with the boundary fluxes for the first - order scheme ( equation [ eqn : cfdf ] ) . the lax - wendroff scheme obtains second - order fluxes , @xmath64 by modifying the first - order fluxes @xmath65 with a second - order correction . the stability of the lax - wendroff scheme to solve the linear advection equation can also be determined using the von neumann analysis . the discretized lax - wendroff equation ( equation [ eqn : lwe ] ) is exactly solvable and after @xmath52 time steps , the fourier modes evolve according to @xmath66^mc_k^\circ\ , \label{eqn : lwce}\ ] ] where @xmath54 is called the _ courant _ number and @xmath55 . the dispersion relation is given by @xmath67 \\[8pt ] + \frac{in}{4\pi{\delta t}}\ln\,\left[1 - 4\lambda^2(1-\lambda^2)\sin^4\left(\frac{\phi}{2}\right)\right]\ . \label{eqn : lwdr}\end{gathered}\ ] ] it is important to note three things . first , the lax - wendroff scheme is conditionally stable provided that @xmath49 , which is satisfied if @xmath68 this constraint is a particular example of a general stability constraint known as the _ courant - friedrichs - lewyor _ ( cfl ) condition . the courant number @xmath69 is also referred to as the cfl number . second , for @xmath70 the dispersion relation is exactly identical to that of the exact solution and the numerical advection is exact . this is a special case , however , and it does not test the ability of the lax - wendroff scheme to solve general scalar conservation laws . normally , one chooses @xmath71 to satisfy the cfl condition . lastly , for @xmath71 the dispersion relation @xmath47 for the lax - wendroff solution is different from the exact solution where @xmath50 . the dispersion relation relative to the exact solution can be parametrized by @xmath72 the second - order truncation of the taylor series ( equation [ eqn : taylor ] ) results in a phase error @xmath73 which is a function of frequency . in the lax - wendroff solution , the waves are damped and travel at different speeds . hence the scheme is both diffusive and dispersive . in figure ( [ fig : lwdispersion ] ) we plot the phase error @xmath73 and the amplification term @xmath74 for the lax - wendroff scheme with parameters @xmath75 , @xmath76 , and @xmath77 . a negative value of @xmath73 represents a lagging phase error while a positive value indicates a leading phase error . for the chosen cfl number , the high frequency modes have the largest phase errors but they are highly damped . some of the modes having lagging phase errors are not highly damped . we will subsequently see how this becomes important . a rigourous test of the 1-d lax - wendroff scheme and other flux assignment schemes we will discuss is the linear advection of a square wave . the challenge is to accurately advect this discontinuous function where the edges mimic riemann shock fronts . in figure ( [ fig : lax ] ) we show how the lax - wendroff scheme does at advecting the square wave once ( dashed line ) and ten times ( dotted line ) through a periodic box of 100 grid cells at speed @xmath76 and @xmath77 . note that this scheme produces numerical oscillations . recall that a square wave can be represented by a sum of fourier or sine waves . these waves will be damped and disperse when advected using the lax - wendroff scheme . figure ( [ fig : lwdispersion ] ) shows that the modes having lagging phase errors are not damped away . hence , the lax - wendroff scheme is highly dispersive and the oscillations in figure ( [ fig : lax ] ) are due to dispersion . we leave it as an exercise for the reader to advect a sine wave using the lax - wendroff scheme . since there is only one frequency mode in this case , there will be no spurious oscillations due to dispersion , but a phase error will be present . for a comprehensive discussion on the family of lax - wendroff schemes and other centered schemes , see @xcite and @xcite . upwind methods take into account the physical nature of the flow when assigning fluxes for the discrete solution . this class of flux assignment schemes , whose origin dates back to the work of @xcite , has been shown to be excellent at capturing shocks and also being highly stable . we start with a simple first - order upwind scheme to solve the linear advection equation . consider the case where the advection velocity is positive and flow is to the right . the flux of the physical quantity @xmath27 through the cell boundary @xmath32 will originate from cell @xmath78 . the upwind scheme proposes that , to first - order , the fluxes @xmath31 at cell boundaries be taken from the cell - centered fluxes @xmath39 , which is in the upwind direction . if the advection velocity is negative and flow is to the left , the boundary fluxes @xmath31 are taken from the cell - centered fluxes @xmath79 . the first - order upwind flux assignment scheme can be summarized as follows : @xmath80 f_{n+1}^t & \text{if $ v<0 $ . } \end{cases}\ ] ] unlike central difference schemes , upwind schemes are explicitly asymmetric . the cfl condition for the first - order upwind scheme can be determined from the von neumann analysis . we consider the case of a positive advection velocity . after @xmath52 time steps , the fourier modes evolve according to @xmath81^mc_k^\circ\ , \label{eqn : uwce}\ ] ] where @xmath54 and @xmath55 . the dispersion relation is given by @xmath82 \\[8pt ] + \frac{in}{4\pi{\delta t}}\ln\,\left[1 - 4\lambda(1-\lambda)\sin^2\left(\frac{\phi}{2}\right)\right]\ . \label{eqn : updr}\end{gathered}\ ] ] the cfl condition for solving the linear advection equation with this scheme is to have @xmath83 , identical to that for the lax - wendroff scheme . for @xmath71 the dispersion relation @xmath47 for the first - order upwind scheme is different from the exact solution where @xmath50 . this scheme is both diffusive and dispersive . since it is only first - order accurate , the amount of diffusion is large . in figure ( [ fig : uwdispersion ] ) we compare the dispersion relation of the upwind scheme to that of the lax - wendroff scheme . the fourier modes in the upwind scheme also have phase errors but they will be damped away . the low frequency modes which contribute to the oscillations in the lax - wendroff solution are more damped in the upwind solution . hence , one does not expect to see oscillations resulting from phase errors . in figure ( [ fig : uw ] ) we show how the first - order upwind scheme does at advecting the riemann shock wave . this scheme is well - behaved and produces no spurious oscillations , but since it is only first - order , it is highly diffusive . the first - order upwind scheme has the property of having monotonicity preservation . when applied to the linear advection equation , it does not allow the creation of new extrema in the form of spurious oscillations . the lax - wendroff scheme does not have the property of having monotonicity preservation . the flux assignment schemes that we have discussed so far are all linear schemes . @xcite showed that all linear schemes are either diffusive or dispersive or a combination of both . the lax - wendroff scheme is highly dispersive while the first - order upwind scheme is highly diffusive . godunov s theorem also states that linear monotonicity preserving schemes are only first - order accurate . in order to obtain higher order accuracy and prevent spurious oscillations , nonlinear schemes are needed to solve conservation laws . @xcite proposed the _ total variation diminishing _ ( tvd ) condition which guarantees that a scheme have monotonicity preservation . applying godunov s theorem , we know that all linear tvd schemes are only first - order accurate . in fact , the only linear tvd schemes are the class of first - order upwind schemes . therefore , higher order accurate tvd schemes must be nonlinear . the tvd condition is a nonlinear stability condition . the total variation of a discrete solution , defined as @xmath84 is a measure of the overall amount oscillations in @xmath27 . the direct connection between the total variation and the overall amount of oscillations can be seen in the equivalent definition @xmath85 where each maxima is counted positively twice and each minima counted negatively twice ( see * ? ? ? . the formation of spurious oscillations will contribute new maxima and minima and the total variation will increase . a flux assignment scheme is said to be tvd if @xmath86 which signifies that the overall amount of oscillations is bounded . in linear flux - assignment schemes , the von neumann linear stability condition requires that the fourier modes remain bounded . in nonlinear schemes , the tvd stability condition requires that the total variation diminishes . we now describe a nonlinear second - order accurate tvd scheme which builds upon the first - order monotone upwind scheme described in the previous section . the second - order accurate fluxes @xmath31 at cell boundaries are obtained by taking first - order fluxes @xmath87 from the upwind scheme and modifying it with a second order correction . first consider the case where the advection velocity is positive . the first - order upwind flux @xmath87 comes from the averaged flux @xmath88 in cell @xmath78 . we can define two second - order flux corrections , @xmath89 { \delta f}_{n+1/2}^{r , t}&=\frac{f_{n+1}^t - f_n^t}{2}\ , \end{aligned}\ ] ] using three local cell - centered fluxes . we use cell @xmath78 and the cells immediately left and right of it . if the advection velocity is negative , the first - order upwind flux comes from the averaged flux @xmath90 in cell @xmath91 . in this case , the second - order flux corrections , @xmath92 { \delta f}_{n+1/2}^{r , t}&=-\frac{f_{n+2}^t - f_{n+1}^t}{2}\ .\end{aligned}\ ] ] are based on cell @xmath91 and the cells directly adjacent to it . near extrema where the corrections have opposite signs , we impose no second - order correction and the flux assignment scheme reduces to first - order . a flux limiter @xmath93 is then used to determine the appropriate second - order correction , @xmath94 which still maintains the tvd condition . the second - order correction is added to the first - order fluxes to get second - order fluxes . the first - order upwind scheme and second - order tvd scheme will be referred to as _ monotone upwind schemes for conservation laws _ ( muscl ) . time integration is performed using a second - order runge - kutta scheme . we first do a half time step , @xmath95 using the first - order upwind scheme to obtain the half - step values @xmath96 . a full time step is then computed , @xmath97 using the tvd scheme on the half - step fluxes @xmath98 . the reader is encouraged to show that is second - order accurate . we briefly discuss three tvd limiters . the minmod flux limiter chooses the smallest absolute value between the left and right corrections : @xmath99\min(|a|,|b|)\ .\ ] ] the superbee limiter @xcite chooses between the larger correction and two times the smaller correction , whichever is smaller in magnitude : @xmath100 minmod(2a , b ) & \text{otherwise . } \end{cases}\ ] ] the van leer limiter @xcite takes the harmonic mean of the left and right corrections : @xmath101 the minmod limiter is the most moderate of all second - order tvd limiters . in figure ( [ fig : mm ] ) we see that it does not do much better than first - order upwind for the square wave advection test . superbee chooses the maximum correction allowed under the tvd constraint . it is especially suited for piece - wise linear conditions and is the least diffusive for this particular test , as can be seen in figure ( [ fig : sb ] ) . note that no additional diffusion can be seen by advecting the square wave more than once through the box . it can be shown that the minmod and superbee limiters are extreme cases which bound all other second - order tvd limiters . the van leer limiter differs from the previous two in that it is analytic . this symmetrical approach falls somewhere inbetween the other two limiters in terms of moderation and diffusion , as can be seen in figure ( [ fig : vl ] ) . it can be shown that the cfl condition for the second - order tvd scheme is to have @xmath71 . for a comprehensive discussion on tvd limiters , see @xcite and @xcite . we now describe a simple and robust method to solve the euler equations using the monotone upwind scheme for conservation laws ( muscl ) from the previous section . the relaxing tvd method @xcite provides high resolution capturing of shocks using computationally inexpensive algorithms which are straightforward to implement and to parallelize . it has been successfully implemented for simulating cosmological astrophysical fluids by @xcite . the muscl scheme is straightforward to apply to conservation laws like the advection equation since the velocity alone can be used as a marker of the direction of flow . however , applying the muscl scheme to solve the euler equations is made difficult by the fact that the momentum and energy fluxes depend on the pressure . in order to determine the direction upwind of the flow , it becomes necessary to calculate the flux jacobian eigenvectors using riemann solvers . this step requires computationally expensive algorithms . the relaxing tvd method offers an attractive alternative . we first present a motivation for the relaxing scheme by again considering the 1-d scalar conservation law . the muscl scheme for solving the linear advection equation is explicitly asymmetric in that it depends on the sign of the advection velocity . we now describe a symmetrical approach which applies to a general advection velocity . the flow can be considered as a sum of a right - moving wave @xmath102 and a left - moving wave @xmath103 . for a positive advection velocity , the amplitude of the left - moving wave is zero and for a negative advection velocity , the amplitude of the right - moving wave is zero . in compact notation , the waves can be defined as : @xmath104 u^l&=\left(\frac{1-v / c}{2}\right)u\ , \end{aligned}\ ] ] where @xmath105 . the two waves are traveling in opposite directions with advection speed @xmath106 and can be described by the advection equations : @xmath107 @xmath108 the muscl scheme is straightforward to apply to solve equations ( [ eqn : rmw ] ) and ( [ eqn : lmw ] ) since the upwind direction is left for the right - moving wave and right for the left - moving wave . the 1-d relaxing advection equation then becomes @xmath109 where @xmath110 and @xmath111 . for the discretized solution given by equation ( [ eqn : conservation ] ) , the boundary fluxes @xmath31 are now a sum of the fluxes @xmath112 and @xmath113 from the right - moving and left - moving waves , respectively . note that the relaxing advection equation will correctly reduce to the linear advection equation for any general advection velocity . using this symmetrical approach , a general algorithm can be written to solve the linear advection equation for an arbitrary advection velocity . this scheme is indeed inefficient for solving the linear advection equation since one wave will have zero amplitude . however , the euler equations can have both right - moving and left - moving waves with non - zero amplitudes . we now discuss the 1-d relaxing tvd scheme and later generalize it to higher spatial dimensions . consider a 1-d system of conservation laws , @xmath114 where for the euler equations , we have @xmath115 and @xmath18 the corresponding flux terms . we now replace the vector conservation law with the relaxation system @xmath116 @xmath117 where @xmath118 is a free positive function called the freezing speed . the relaxation system contains two coupled vector linear advection equations . in practice , we set @xmath119 and use it as an auxiliary vector to calculate fluxes . equation ( [ eqn : relax1 ] ) reduces to our 1-d vector conservation law and equation ( [ eqn : relax2 ] ) is a vector conservation law for @xmath120 . in order to solve the relaxed system , we decouple the equations through a change of variables : @xmath121 { \boldsymbol{u}}^l&=\frac{{\boldsymbol{u}}-{\boldsymbol{w}}}{2}\ , \end{aligned}\ ] ] which then gives us @xmath122 @xmath123 equations ( [ eqn : w1 ] ) and ( [ eqn : w2 ] ) are vector linear advection equations , which can be interpreted as right - moving and left - moving flows with advection speed @xmath106 . note the similarity with their scalar counterparts , equations ( [ eqn : rmw ] ) and ( [ eqn : lmw ] ) . the 1-d vector relaxing conservation law for @xmath124 becomes @xmath125 where @xmath126 and @xmath127 . the vector relaxing equation can now be solved by applying the muscl scheme to equations ( [ eqn : w1 ] ) and ( [ eqn : w2 ] ) . again , note the similarity between the vector relaxing equation and its scalar counterpart , equation ( [ eqn : rlae ] ) . the relaxed scheme is tvd under the constraint that the freezing speed @xmath106 be greater than the characteristic speed given by the largest eigenvalue of the flux jacobian @xmath128 . for the euler equations , this is satisfied for @xmath129 @xcite considered the freezing speed to be a positive constant in their relaxing scheme while we generalize it to be a positive function . time integration is again performed using a second - order runge - kutta scheme and the time step is determined by satisfying the cfl condition , @xmath130 note that a new freezing speed is computed for each partial step in the runge - kutta scheme . the cfl number @xmath131 should be chosen such that @xmath132 will be larger than @xmath133 and @xmath134 . we now summarize the steps needed to numerically solve the 1-d euler equations . at the beginning of each partial step in the runge - kutta time integration scheme , we need to calculate the cell - averaged variables defined at grid cell centres . first for the half time step , we calculate the fluxes @xmath135 and the freezing speed @xmath136 . we then set the auxiliary vector @xmath137 and construct the right - moving waves @xmath138 and left - moving waves @xmath139 . the half time step is given by @xmath140 where @xmath141 the first - order upwind scheme is used to compute fluxes at cell boundaries for the right - moving and left - moving waves . for the full time step , we construct the right - moving waves @xmath142 and left - moving waves @xmath143 , using the half - step values of the appropriate variables . the full time step , @xmath144 is computed using the second - order tvd scheme . this completes the updating of @xmath145 to @xmath146 . we have found that a minor modification to the implementation described above gives more accurate results . consider writing the flux of the right - moving and left - moving waves as : @xmath147 { \boldsymbol{f}}^l = c{\boldsymbol{g}}^l\ , \end{aligned}\ ] ] where @xmath148 is the flux of @xmath149 and @xmath150 is the flux of @xmath151 . the linear advection equations for @xmath152 and @xmath153 are similar to equations ( [ eqn : w1 ] ) and ( [ eqn : w2 ] ) , but where we replace @xmath154 with @xmath152 and @xmath155 with @xmath153 . for each partial step in the runge - kutta scheme , the net fluxes at cell boundaries are then taken to be @xmath156 where we use @xmath157 . in practice , this modified implementation has been found to resolve shocks with better accuracy in certain cases . note that the two different implementations of the relaxing tvd scheme are identical when a constant freezing speed is used . the 1-d relaxing tvd scheme can be generalized to higher dimensions using the dimensional splitting technique by @xcite . in three dimensions , the euler equations can be dimensionally split into three separate 1-d equations which are solved sequentially . let the operator @xmath158 represent the updating of @xmath159 to @xmath160 by including the flux in the @xmath161 direction . we first complete a forward sweep , @xmath162 and then perform a reverse sweep @xmath163 using the same time step @xmath164 to obtain second - order accuracy . we will refer to the combination of the forward and reverse sweeps as a double sweep . a more symmetrical sweeping pattern can be used by permutating the sweeping order when completing the next double time step . the dimensional splitting or operator splitting technique can be summarized as follows : @xmath165 u^{t_3}&=u^{t_2 + 2{\delta t}_2}=l_zl_xl_yl_yl_xl_zu^{t_2}\ , \\[8pt ] u^{t_4}&=u^{t_3 + 2{\delta t}_3}=l_yl_zl_xl_xl_zl_yu^{t_3}\ , \end{aligned}\ ] ] where @xmath166 , @xmath167 , and @xmath168 are newly determined time steps after completing each double sweep . the cfl condition for the 3-d relaxing tvd scheme is similarly given by equation ( [ eqn : relaxcfl ] ) , but with @xmath169\ .\ ] ] where @xmath170 . note that since @xmath171 is on average a factor of @xmath172 smaller than @xmath173 , a dimensionally split scheme can use a longer time step compared to an un - split scheme . the dimensional splitting technique also has other advantages . the decomposition into a 1-d problem allows one to write short 1-d algorithms , which are easy to optimize to be cache efficient . a 3-d hydro code is straightforward to implement in parallel . when sweeping in the x direction , for example , one can break up the data into 1-d columns and operate on the independent columns in parallel . a sample 3-d relaxing tvd code , implemented in parallel using openmp directives , is provided in the appendix . a rigourous and challenging test for any 3-d eulerian or lagrangian hydrodynamic code is the sedov - taylor blast wave test . we set up the simulation box with a homogeneous medium of density @xmath174 and negligible pressure and introduce a point - like supply of thermal energy @xmath175 in the centre of the box at time @xmath176 . the challenge is to accurately capture the strong spherical shock wave which sweeps along material as it propagates out into the ambient medium . the sedov - taylor test is used to model nuclear - type explosions . in astrophysics , it is often used as a basic setup to model supernova explosions and the evolution of supernova remnants ( see * ? ? ? the analytical sedov solution uses the self - similar nature of the blast wave expansion ( see * ? ? ? consider a frame fixed relative to the centre of the explosion . the spherical shock front propagates outward and the distance from the origin is given by @xmath177 where @xmath178 for an ideal gas with @xmath179 . the velocity of the shock @xmath180 is given by @xmath181 since the ambient medium has negligible pressure , the shocks will be very strong . the density @xmath182 , velocity @xmath183 , and pressure @xmath184 directly behind the shock front are : @xmath185 v_2&=\left(\frac{2}{\gamma+1}\right)v_{sh}\ , \\[8pt ] \ p_2&=\left(\frac{2}{\gamma+1}\right)\rho_1v_{sh}^2\ .\end{aligned}\ ] ] the immediate post - shock gas density is constant in time , while the shocked gas velocity @xmath183 and pressure @xmath184 decrease as @xmath186 and @xmath187 , respectively . the full analytical sedov - taylor solutions can be found in @xcite . the 3-d relaxing tvd code based on the van leer flux limiter is applied to capturing the sedov - taylor blast wave . we set up a box with @xmath188 cells and constant initial density @xmath189 . at time @xmath176 , we inject a supply of thermal energy @xmath190 into one cell at the centre of the box . the simulation is stopped at time @xmath191 , in which the shock front has propagated out to a distance of @xmath192 cells from the centre . in figure ( [ fig : st1 ] ) and ( [ fig : st2 ] ) we plot the radial distributions of density , momentum , and pressure , normalized to @xmath182 , @xmath193 , and @xmath184 respectively . the data points are taken from a random subset of cells and the solid lines are the analytical sedov - taylor solutions . despite solving a spherically symmetric problem on an explicitly non - rotationally invariant cartesian grid , the anisotropic scatter in the results is small . the distance of the shock front from the centre of the explosion as a function of time is indeed given by equation ( [ eqn : stre ] ) , demonstrating that the 3-d relaxing tvd code ensures the correct shock propagation . the resolution of the shock front is roughly two grid cells . the numerical shock jump values of @xmath182 , @xmath183 , and @xmath184 are resolution dependent and come close to the theoretical values for our test with @xmath188 cells . we leave it as an exercise for the reader to test the code using the minmod and superbee flux limiters . for astrophysical applications , both hydrodynamical and gravitational forces are included . the gravitational forces arise from the self - gravity of the fluid and can also come from an external field . the euler equations with the gravitational source terms included are given as : @xmath6 \frac{{\partial}(\rho v_i)}{{\partial t}}+\frac{{\partial}}{{\partial x}_j}(\rho v_iv_j+p\delta_{ij})=-\rho\frac{{\partial}\phi}{{\partial x}_i}\ , \\[8pt ] \frac{{\partial}e}{{\partial t}}+\frac{{\partial}}{{\partial x}_j}[(e+p)v_j]=-\rho v_i\frac{{\partial}\phi}{{\partial x}_i}\ .\end{gathered}\ ] ] where @xmath93 is the gravitational potential . poisson s equation , @xmath194 relates the gravitational potential to the density field . the general solution can be written as @xmath195 where the kernel is given by @xmath196 in the discrete case , the integral in equation ( [ eqn : poisson ] ) becomes a sum and poisson s equation can be solved using fast fourier transforms ( fft ) to do the convolution . the forces are then calculated by finite differencing the potential ( see * ? ? ? the addition of gravatitational source terms in the euler equations is easily handled using the operator splitting technique described in @xmath197[sec : mdscl ] . consider the double sweep : @xmath198 where the operator @xmath158 represents the updating of @xmath27 by including the flux in the @xmath161 direction and the operator @xmath199 represents the gravitational acceleration of the fluid . during the gravitational step , the flux terms in the euler equations are ignored . the density distribution does not change and only the fluid momenta and total energy density are updated . the stellar density in the cores of globular and open clusters is high enough for stellar collisions to take place with significant frequency @xcite . current observations and simulations suggest that the merger of two main sequence stars produces a blue straggler @xcite . the blue stragglers are out - lying main sequence stars which lie beyond the main sequence turnoff in the colour - magnitude diagram ( cmd ) of a star cluster . the blue stragglers are more massive , brighter , and bluer than the turnoff stars . since more massive stars evolve faster than lower mass stars and are not expected to lie beyond the turnoff , this suggests that blue stragglers must have formed more recently . in principle the merger of two main sequence stars can produce a young remnant star provided that significant mixing occurs in the process . the mixing produces a higher hydrogen fraction in the core of the remnant than that of the parent stars which have already burnt most of the hydrogen to helium in their cores . @xcite used low resolution sph simulations with @xmath200 particles to simulate the merging of @xmath201 polytropes and found that they fully mixed . however , medium resolution sph simulations with @xmath202 particles of @xmath201 or @xmath203 polytropes showed only weak mixing @xcite . it is worth noting that @xmath201 polytropes are more representative of low mass main sequence stars with large convective envelopes while @xmath203 polytropes resemble main sequence stars near the turnoff which have little mass in their convective envelopes . high resolution sph simulations involving @xmath204 particles have now been applied to simulating stellar collisions @xcite . the merging stars process is mostly subsonic and strong shocks are not expected . in the absence of shocks , sph particles will follow flow lines of constant entropy due to the lagrangian nature of the method . as a result , the particles may experience sedimentation . in addition , the mixing can also depend on the adopted smoothing length and the form of artificial viscosity . for a sph fluid , the reynolds number is of order @xmath205 , where @xmath206 is the total number of particles and @xmath207 is the number of particles over which the smoothing is done . for @xmath208 and @xmath209 , the reynolds number is @xmath210 . however , a fluid with a low reynolds number will tend to experience laminar flow . hence , sph may under mix . it is a worthwhile exercise to model the merging process using eulerian hydrodynamical simulations . the differences between eulerian and lagrangian approaches may lead to very different results on mixing . as of present , no such work has been reported in the literature . we consider the off - axis collision of two main sequence stars with @xmath211 and @xmath212 , which are modeled using @xmath203 polytropes . a polytrope with polytropic index @xmath78 has equilibrium density and pressure profiles which are related by @xmath213 the density profile is determined by solving the lane - emden equation ( see * ? ? ? we adopt an ideal gas equation of state with adiabatic index @xmath179 . note that for an @xmath203 polytrope , 90% of the total mass is contained within @xmath214 . we define the dynamical time to be , @xmath215 where @xmath216 is the average density . for the chosen parent stars , the dynamical time is approximately one physical hour . the collision is simulated in a box with @xmath217 cells and the orbital plane coincides with the @xmath218 plane . initially , each parent star has a radius of 96 grid cells . the stars are set up on zero - energy parabolic orbits with a pericentre separation equal to @xmath219 . the initial trajectories are calculated assuming point masses . in an eulerian simulation , the vacuum can not have zero density . we set the minimum density of the cells to be @xmath220 of the central density of the parent stars . the hydrodynamics is done in a non - periodic box with vacuum boundary conditions . a non - trivial test of a self - gravitating eulerian hydro code is the advection of an object in hydrostatic equilibrium . the challenge is to maintain the equilibrium profile over a large number of time steps . one of the parent stars is placed in a periodic box with @xmath188 cells and given some initial momentum . we make the test rigorous by having the polytrope move in all three directions . in figure [ fig : poly ] we compare the mass and entropy profiles of the initial and advected polytrope . the entropic variable @xmath221 is used in place of the specific entropy . the parameter @xmath222 is defined to be the minimum entropy of the parent polytrope . after 1000 timesteps in which the polytrope has moved 256 cells in each direction , the advected polytrope has still retained its equilibrium profile . shock heating can occur in the outer envelope as the polytrope moves through the false vacuum . however , by setting the density of the false vacuum to be @xmath220 of the central density of the polytrope , we can minimize the spurious shock heating . in figure [ fig : merge ] we show four snapshots of the merging process taken at time @xmath176 , 2 , 4 , and @xmath223 . the 2-d density maps are created by averaging over 4 planes taken about the orbital mid - plane . the density contours are spaced logarithmically with 2 per decade and covering 3 decades down from the maximum . the parent stars are initially separated by @xmath224 and placed on zero - energy orbits with a pericentre separation of @xmath225 . during the collision process , the outer envelopes of the parent stars are shock heated and material gets ejected . in less than @xmath226 , the merger remnant establishes hydrostatic equilibrium . the merger remnant is a rotating oblate with mass approximately 90% of the combined mass of the parent stars . a large fraction of the mass loss is due to the vacuum boundary conditions . ejected material do not have the opportunity to fall back onto the merger remnant . however , the additional mass loss in the envelope does not present a problem since we are interested in the question of mixing in the interior of the star . in figure [ fig : profiles ] we plot the thermodynamic profiles of the merger remnant and the parent stars . the central density and pressure in the core of the merger remnant is lower than the corresponding values in the parent stars by approximately half . the entropy floor has risen by a factor of 1.6 . shock heating is expected to be minimal in the core so a change in entropy suggests that some mixing has taken place . however , it is difficult to quantify the amount of mixing from examining the thermodynamic profiles alone . to help address the question of mixing , we are implementing a particle - mesh ( pm ) scheme where test particles can be used to track passively advected quantities such as chemical composition . initially , each parent star is assigned a large number of particles with known chemical composition . the test particles are passively advected along velocity field lines . for each time step , the velocity of each particle is interpolated from the grid using a `` cloud - in - cell '' ( cic ) scheme @xcite and the equations of motions are solved using second - order runge - kutta integration . the cic interpolation scheme is also used to determine the local density , pressure , and entropy associated with each particle . with this setup , we have the benefit of being able to track thermodynamic quantities like in an sph scheme but avoid the under mixing problem since the fluid equations are solved using the eulerian scheme . future work ( trac , sills , & pen 2003 ) will have higher resolution simulations . collisions will be simulated in a box with @xmath227 cells . each parent star will have a radius of 192 grid cells and be assigned @xmath188 test particles . we will also be doing a detailed comparison between eulerian and sph simulations of stellar mergers . the self - gravitating hydro code used for the simulations is very memory friendly . for the @xmath228 grid , 10 gb is required to store the hydro variables , 2 gb for the potential , and less than 1 gb for the test particles . for every double timestep , approximately 1000 floating point operations per grid cell is needed to carry out the tvd hydro calculations . the potential is computed once for every double step and this requires two ffts . since eulerian codes are very memory friendly , have low floating point counts , are easily parallelized , and scale very well on shared - memory , multiple - processor machines , they can be used to run very high resolution simulations . we have presented several numerical schemes for solving the linear advection equation and given the cfl stability conditions for each scheme . we have implemented the relaxing tvd scheme to solve the euler system of conservation laws . the second - order accurate tvd scheme provides high resolution capturing of shocks , as can be seen in the riemann shock test and the sedov - taylor blast wave test . the 1-d relaxing tvd scheme can be easily generalized to higher dimensions using the dimensional splitting technique . a dimensionally split scheme can use longer time steps and is straightforward to implement in parallel . we have presented a sample astrophysical application . a 3-d self - gravitating eulerian hydro code is used to simulate the formation of blue straggler stars through stellar mergers . we hope to have convinced the reader that eulerian computational fluid dynamics is a powerful approach to simulating complex fluid flows because it is simple , fast , and accurate . we thank joachim stadel and norm murray for comments and suggestions on the writing and editing of this paper . we also thank alison sills , phil arras , and chris matzner for discussions on stellar mergers . benz , w. & hills , j. g. , 1987 , , 323 , 614 chandrasekhar , s. , 1957 , an introduction to the study of stellar structure ( new york : dover publications ) courant , r. , isaacson , e. , & reeves , m. , 1952 , comm . pure and applied math . , 5 , 243 gingold , r. a. & monaghan , j. j. , 1977 , , 181 , 375 godunov , s. k. , 1959 , math . sbornik , 47 , 271 harten , a. , 1983 , j. comp . phys . , 49 , 357 hill , j. g. & day , c. a. , 1976 , , 17 , 87 hirsch , c. , 1990 , numerical computation of internal and external flows , vol . 2 : computational methods for inviscid and viscous flows ( new york : john wiley ) hockney , r. w. & eastwood , j. w. , 1988 , computer simulation using particles ( philadelphia : iop publishing ) jin , s. & xin , z. , 1995 , comm . pure and applied math . , 48 , 235 landau , l. d. & lifshitz , e. m. , 1987 , fluid mechanics ( 2nd ed . ; oxford : pergamon press ) laney , c. b. , 1998 , computational gas dyanamics ( cambridge : cambridge university press ) lax , p. d. & wendroff , b. , 1960 , comm . pure and applied math . , 10 , 537 lombardi , j. c. , jr . , rasio , f. a. , & shapiro , s. l , 1996 , , 468 , 797 lucy , l. b. , 1977 , , 82 , 1013 pen , u. , 1998 , , 115 , 19 roe , p. l.,1985 , in large - scale computations in fluid mechanics , lectures in applied mathematics , eds . b. e. engquist , s. osher , & r. c. j. somerville ( providence , ri : american mathematical society ) sandquist , e. l. , bolte , m. , & hernquist , l. , 1997 , , 477 , 335 shu , f. h. , 1992 , gas dynamics ( mill valley : university science books ) sills , a. , adams , t. , davies , m. b. , & bate , m. r. , 2002 , , 332 , 49 sills , a. , lombardi , j. c. , jr . , bailyn , c. d. , demarque , p. d. , rasio , f. a. , & shapiro , s. l. , 1997 , , 487 , 290 strang , g. , 1968 , siam j. num . , 5 , 506 van leer , b. , 1974 , j. comp . we provide a sample 3-d relaxing tvd code written in fortran 90 . the code is implemented using openmp directives to run in parallel on shared memory machines . the code is fast and memory friendly . the array u(a , i , j , k ) stores the five conserved hydro quantities @xmath229 for each cell ( i , j , k ) in the cartesian cubical lattice with side length nc . for each sweep , we first call the subroutine timestep to determine the appropriate time step dt which satisfies the cfl condition . the updating of u by including the flux in the x direction is performed by the sweepx subroutine . the data array u is divided into 1-d array sections u1d(a , i ) which are operated on by the relaxing subroutine . the independent columns are distributed amongst multiple processors on a shared memory machine by the openmp directives . the relaxing tvd subroutine in this sample code is written for ease of readability and therefore , is not fully optimized . at the beginning of each partial step in the runge - kutta time integration scheme , the cell - averaged variables defined at grid cell centres are calculated by the averageflux subroutine . the fluxes at cell boundaries for the right - moving and left - moving waves are stored in fr and fl , respectively . we have implemented the minmod , superbee , and van leer flux limiters and the user of the code can easily switch between them . we have provided some initial conditions for the sedov - taylor blast wave test . the reader is encouraged to test the code and compare how the various flux limiters do at resolving strong shocks . this sample code does not implement the modified relaxing tvd scheme described at the end of @xmath230 , which has been found work very well with the van leer flux limiter but unstable with superbee for the 3-d sedov taylor test . we have found that the superbee limiter is often unstable for 3-d fluid simulations . please contact the authors regarding any questions on the implementation of the relaxing tvd algorithm . + ` program main ` + ` implicit none ` + ` integer , parameter : : nc=64,hc = nc/2 ` + ` real , parameter : : gamma=5./3,cfl=0.9 ` + ` ` + ` real , dimension(5,nc , nc , nc ) : : u ` + + ` integer nsw , stopsim ` + ` real t , tf , dt , e0,rmax ` + + ` t=0 ` + ` dt=0 ` + ` nsw=0 ` + ` stopsim=0 ` + + ` e0=1e5 ` + ` rmax=3*hc/4 ` + ` tf = sqrt((rmax/1.15)**5/e0 ) ` + + ` call sedovtaylor ` + ` do ` + ` call timestep ` + ` call sweepx ` + ` call sweepy ` + ` call sweepz ` + ` call sweepz ` + ` call sweepy ` + ` call sweepx ` + ` if ( stopsim .eq . 1 ) exit ` + ` call timestep ` + ` call sweepz ` + ` call sweepx ` + ` call sweepy ` + ` call sweepy ` + ` call sweepx ` + ` call sweepz ` + ` if ( stopsim .eq . 1 ) exit ` + ` call timestep ` + ` call sweepy ` + ` call sweepz ` + ` call sweepx ` + ` call sweepx ` + ` call sweepz ` + ` call sweepy ` + ` if ( stopsim .eq . 1 ) exit ` + ` enddo ` + ` call outputresults ` + + + ` contains ` + + + ` subroutine sedovtaylor ` + ` implicit none ` + ` integer i , j , k ` + + ` do k=1,nc ` + ` do j=1,nc ` + ` do i=1,nc ` + ` u(1,i , j , k)=1 ` + ` u(2:4,i , j , k)=0 ` + ` u(5,i , j , k)=1e-3 ` + ` enddo ` + ` enddo ` + ` enddo ` + ` u(5,hc , hc , hc)=e0 ` + ` return ` + ` end subroutine sedovtaylor ` + + + ` subroutine outputresults ` + ` implicit none ` + ` integer i , j , k ` + ` real r , x , y , z ` + + ` open(1,file=sedovtaylor.dat,recl=200 ) ` + ` do k=1,nc ` + ` z = k - hc ` + ` do j=1,nc ` + ` y = j - hc ` + ` do i=1,nc ` + ` x = i - hc ` + ` r = sqrt(x**2+y**2+z**2 ) ` + ` write(1 , * ) r , u(:,i , j , k ) ` + ` enddo ` + ` enddo ` + ` enddo ` + ` close(1 ) ` + ` return ` + ` end subroutine outputresults ` + + + ` subroutine timestep ` + ` implicit none ` + ` integer i , j , k ` + ` real p , cs , cmax ` + ` real v(3 ) ` + + ` cmax=1e-5 ` + ` ! ` @xmath231`omp end parallel do ` + ` ` + ` dt = cfl / cmax ` + ` if ( t+2*dt .gt . tf ) then ` + ` dt=(tf - t)/2 ` + ` stopsim=1 ` + ` endif ` + ` t = t+2*dt ` + ` nsw = nsw+1 ` + ` write(*,````(a7,i3,a8,f7.5,a6,f8.5 ) ` '' ` ) nsw = ,nsw, dt = ,dt, t = ,t ` + ` return ` + ` end subroutine timestep ` + + + ` subroutine sweepx ` + ` implicit none ` + ` integer j , k ` + ` real u1d(5,nc ) ` + + ` ! ` @xmath232`omp end parallel do ` + ` return ` + ` end subroutine sweepx ` + + + ` subroutine sweepy ` + ` implicit none ` + ` integer i , k ` + ` real u1d(5,nc ) ` + ` ` + ` ! ` @xmath233`omp end parallel do ` + ` return ` + ` end subroutine sweepy ` + + + ` subroutine sweepz ` + ` implicit none ` + ` integer i , j ` + ` real u1d(5,nc ) ` + ` ` + ` ! ` end parallel do ` + ` return ` + ` end subroutine sweepz ` + + + ` subroutine relaxing(u ) ` + ` implicit none ` + ` real , dimension(nc ) : : c ` + ` real , dimension(5,nc ) : : u , u1,w , fu , fr , fl , dfl , dfr ` + + ` ! ! do half step using first - order upwind scheme ` + ` call averageflux(u , w , c ) ` + ` fr=(u*spread(c,1,5)+w)/2 ` + ` fl = cshift(u*spread(c,1,5)-w,1,2)/2 ` + ` fu=(fr - fl ) ` + ` u1=u-(fu - cshift(fu,-1,2))*dt/2 ` + + ` ! ! do full step using second - order tvd scheme ` + ` call averageflux(u1,w , c ) ` + + ` ! ! right - moving waves ` + ` fr=(u1*spread(c,1,5)+w)/2 ` + ` dfl=(fr - cshift(fr,-1,2))/2 ` + ` dfr = cshift(dfl,1,2 ) ` + ` call vanleer(fr , dfl , dfr ) ` + ` ! call minmod(fr , dfl , dfr ) ` + ` ! call superbee(fr , dfl , dfr ) ` + + ` ! ! left - moving waves ` + ` fl = cshift(u1*spread(c,1,5)-w,1,2)/2 ` + ` dfl=(cshift(fl,-1,2)-fl)/2 ` + ` dfr = cshift(dfl,1,2 ) ` + ` call vanleer(fl , dfl , dfr ) ` + ` ! call minmod(fl , dfl , dfr ) ` + ` ! call superbee(fl , dfl , dfr ) ` + + ` fu=(fr - fl ) ` + ` u = u-(fu - cshift(fu,-1,2))*dt ` + ` return ` + ` end subroutine relaxing ` + + + ` subroutine averageflux(u , w , c ) ` + ` implicit none ` + ` integer i ` + ` real p , v ` + ` real u(5,nc),w(5,nc),c(nc ) ` + + ` ! ! calculate cell - centered fluxes and freezing speed ` + ` do i=1,nc ` + ` v = u(2,i)/u(1,i ) ` + ` p = max((gamma-1)*(u(5,i)-sum(u(2:4,i)**2)/u(1,i)/2),0 . ) ` + ` c(i)=abs(v)+max(sqrt(gamma*p / u(1,i)),1e-5 ) ` + ` w(1,i)=u(1,i)*v ` + ` w(2,i)=(u(2,i)*v+p ) ` + ` w(3,i)=u(3,i)*v ` + ` w(4,i)=u(4,i)*v ` + ` w(5,i)=(u(5,i)+p)*v ` + ` enddo ` + ` return ` + ` end subroutine averageflux ` + + + ` subroutine vanleer(f , a , b ) ` + ` implicit none ` + ` real , dimension(5,nc ) : : f , a , b , c ` + + ` c = a*b ` + ` where ( c .gt . 0 ) ` + ` f = f+2*c/(a+b ) ` + ` endwhere ` + ` return ` + ` end subroutine vanleer ` + + + ` subroutine minmod(f , a , b ) ` + ` implicit none ` + ` real , dimension(nc ) : : f , a , b ` + + ` f = f+(sign(1.,a)+sign(1.,b))*min(abs(a),abs(b))/2 . ` + ` return ` + ` end subroutine minmod ` + + + ` subroutine superbee(f , a , b ) ` + ` implicit none ` + ` real , dimension(5,nc ) : : f , a , b ` + + ` where ( abs(a ) .gt . abs(b ) ) ` + ` f = f+(sign(1.,a)+sign(1.,b))*min(abs(a),abs(2*b))/2 . ` + ` elsewhere ` + ` f = f+(sign(1.,a)+sign(1.,b))*min(abs(2*a),abs(b))/2 . ` + ` endwhere ` + ` return ` + ` end subroutine superbee ` + + + ` end program main ` +

we have assembled a sample of 64 late - type spiral galaxies ( types 6.09.0 , corresponding to hubble types scd sm ) with archival _ chandra _ data . at a signal - to - noise ( s / n ) threshold of 3 , we find 12 objects with x - ray point - source detections in close proximity with the optical or near - infrared position of the nucleus ( median offset = 16 ) , suggestive of possible low - luminosity active galactic nuclei ( agns ) . including measurements with 3 s / n 1.5 , our detections increase to 18 . these x - ray sources range in luminosity from(210 kev ) = to ergs s . considering possible contamination from low - mass x - ray binaries ( lmxbs ) , we estimate that detections are possible lmxbs instead of true agns , based on the probability of observing a lmxb in a nuclear star cluster typically found in these late - type spiral galaxies . given the typical ages of nuclear star clusters , contamination by high - mass x - ray binaries is unlikely . this agn fraction is higher than that observed in optical surveys , indicating that active nuclei , and hence central black holes , are more common than previously suggested . the incidence of agn activity in such late - type spiral galaxies also suggests that nuclear massive black holes can form and grow in galaxies with little or no evidence for bulges . follow - up multiwavelength observations will be necessary to confirm the true nature of these sources .

0809.1443

there is now overwhelming evidence that the vast majority of massive early - type galaxies harbor a supermassive black hole ( bh ) at their centers , and that bhs play an important role during the assembly and evolution of their host galaxies ( see reviews in @xcite ) . the existence of the @xmath8 correlation between bh mass and central stellar velocity dispersion links the growth of the bh with the formation of the host bulge @xcite , perhaps through feedback mechanisms associated with an active galactic nucleus ( agn ) . the detection of bhs can be achieved by several methods , including direct kinematics of surrounding gas and stars ( e.g. , @xcite ) , and optical spectroscopic identification of nuclear emission lines originating from an agn ( e.g. , @xcite ) . the first method is limited to nearby , inactive or mildly active galaxies , however , while optical spectroscopic identification is generally limited to moderately bright nuclei that are not strongly confused by circumnuclear star formation or heavily obscured by dust . an extensive spectroscopic survey of nearby galaxies has revealed that approximately @xmath9% of galaxies of hubble type e sb contain an agn @xcite , broadly consistent with the idea that bhs are ubiquitous in these systems @xcite . for types later than sc , where classical bulges disappear @xcite , only @xmath10% of galaxies show optical agn activity . this leads to the notion that classical bulges or ellipticals are necessary for bh formation . there are , however , a few striking counter examples . ngc4395 , an essentially bulgeless sm galaxy , exhibits all the hallmarks of a seyfert 1 nucleus @xcite , and it has a bh mass of @xmath11 @xcite . pox52 shares an almost identical optical spectrum to ngc4395 , but is hosted in a spheroidal galaxy @xcite , a visually similar but distinct morphological type compared to elliptical galaxies or classical bulges . more recently , an agn was discovered in ngc3621 , a late - type sd galaxy , using infrared spectroscopic observations @xcite . on the other hand , the nearby scd galaxy m33 has a firm upper limit on the mass of any central dark object of @xmath12 @xcite , far below the typical @xmath13 bhs found in galaxies such as ngc4395 . thus bhs can indeed exist in bulgeless systems , but exactly how common such systems are remains unknown . from a spectral analysis of sdss data , @xcite have found a population of low - mass bhs ( @xmath14@xmath15 ) accreting at appreciable eddington ratios . the host galaxies from the initial sample of 19 sources @xcite , using _ hst _ imaging , are found to be mid- to late - type spirals or spheroidals similar to pox52 @xcite . none are as late - type as ngc4395 , however , and so the incidence of bhs in very late - type systems remains uncertain . for such systems , there is also the possibility that bhs are replaced by nuclear star clusters ( ncs ) . both @xcite and @xcite suggest that bhs in the @xmath8 relation are replaced by compact massive objects at low mass , with a generalized correlation @xmath16 . such compact objects could include ncs instead of bhs , commonly found in late - type spirals @xcite . it is therefore not immediately obvious that such late - type spirals _ should _ contain bhs . establishing a bh census for late - type spirals is important to determine if bhs are indeed formed in earlier - type galaxies only ( ellipticals or early - type spirals ) , or if they also exist in very late - type galaxies . one complication in late - type spirals is the abundance of gas and dust , with strong levels of star formation . if these systems contain bhs , they are likely to be of relatively low mass , and hence the agn , even when fully active , will be of quite low luminosity . while the mere presence of an agn indicates a bh , energetically weak or heavily embedded agns are extraordinarily difficult to observe , particularly at optical wavelengths where most of these observations have historically occurred . dust obscuration severely attenuates optical and ultraviolet flux , and central star formation can dominate the optical / infrared emission . x - ray observations allow us to peer through the obscuring gas and dust column , since high column densities are required to suppress x - ray flux ( particularly energetic x - rays ) . x - ray detections are therefore a useful indicator of agn activity , and since the launch of the _ chandra x - ray observatory _ , the sensitivity with which to carry out such studies has been greatly improved . in particular , the advanced ccd imaging spectrometer ( acis ; @xcite ) provides better than 05 resolution with very low background contamination ( @xmath17 counts pix@xmath6 sec@xmath6 between 0.5 and 8 kev ) . this allows low flux limits to be achieved with very short ( few ks ) exposures , enabling surveys of non - x - ray selected samples @xcite . @xcite have used x - ray detections successfully to identify bhs in low - mass bulges , and have begun to establish statistics on agn x - ray activity as a function of host bulge mass . x - ray observations have also been useful in searching for nuclear activity in quiescent ( i.e. optically faint ) bhs residing in massive elliptical galaxies @xcite , and in star - forming early - type spiral galaxies @xcite . in this paper we investigate the nuclear activity in a sample of late - type spiral galaxies using archival _ chandra _ data , which provide a clearer method of detecting agn activity in such star formation - dominated , potentially heavily obscured environments . systems with nuclear x - ray detections coincident with the optical nucleus are identified as agn candidates , and after considering the possible contamination of our results from x - ray binaries , we estimate the prevalence of agns in late - type spiral galaxies . we discuss our sample selection in [ sec : sample ] and our results in [ sec : results ] . we conclude with a discussion of possible non - agn sources and directions for future work in [ sec : disc ] . all data were obtained from the _ chandra _ data archive , with the full list summarized in table [ tab : log ] . we initially selected objects from the palomar spectroscopic survey of nearby galaxies @xcite with morphological @xmath0 types between 6.0 and 9.0 ( representing hubble types scd to sm ; i.e. bulgeless or nearly bulgeless spiral galaxies ) , which were available in the archive . the palomar survey contains the most sensitive and comprehensive statistics on local agns to date , but the reliability of optically derived statistics in late - type spirals is uncertain due to dust and star formation contamination . we can therefore complement the survey with x - ray data . out of 63 objects in the palomar survey with @xmath0 types between 6.0 and 9.0 , 33 are present in the _ chandra _ archives . the 33 palomar objects in our sample have a median absolute magnitude of @xmath18 ( corrected for galactic and internal extinction ) , and a median distance of 7.8 mpc . if we do not restrict ourselves to palomar objects , we can expand our sample significantly . from the master list of all public _ chandra _ observations ( as of may 2008 ) , we searched for any galaxies in the hyperleda database within 4 of the aimpoint ( i.e. half a chip size for acis - s ) . after restricting our search to objects with @xmath0 type between 6.0 and 9.0 @xcite and within 30 mpc @xcite , our sample grows to 64 objects . these palomar and non - palomar subsamples are denoted by `` p '' and `` c , '' respectively , in table [ tab : log ] . when multiple observations exist for a given object , we generally chose the longest acis - s exposure if possible , otherwise opting for the newest observation . in two cases acis - s images were not available . for m33 , we chose the hrc - s observation , since the detector does not suffer from pileup as a result of the high count rate ( see below for more discussion on pileup ) . because our sample is selected only on the basis of availability within the _ chandra _ data archive , there is a potential for selection biases ( based on , for instance , the selection criteria of the original proposals)although these objects were not necessarily targeted for their nuclear properties . in figure [ fig : ttypes ] we compare the distributions of @xmath0 type between our sample and the original sample of 63 palomar objects from @xcite . the distributions agree reasonably well , although our sample is perhaps weighted slightly toward the most late - type spirals ( i.e. @xmath0 = 9.0 ) . with two exceptions , all data were obtained using the _ chandra _ acis detector , in either the i- or s - array . the i - array consists of 4 ccds arranged for imaging , while the s - array consists of 6 ccds used for either imaging or grating spectroscopy . the mirror assembly has a point - spread function ( psf ) with a full width at half maximum ( fwhm ) of 05 , equal to the acis pixel size . at a distance of 10 mpc , this corresponds to a physical scale of @xmath725 pc . two objects were observed with the high resolution camera ( hrc ) , which has 04 pixels . we used type 2 event files output by standard pipeline processing . we used 0.28 kev counts in our analysis , limiting background contamination at high energies . all data were corrected for known astrometry offsets , following standard ciao threads[multiblock footnote omitted ] ( using ciao version 3.3 ) . we then used celldetect , with default parameters , to automatically find all point sources within each image and determine the centroids . these detections have absolute astrometry accurate to 1 , although it can be marginally worse for off - axis objects with larger psfs and few counts . from these centroids , we measured background - subtracted counts in a 2 radius aperture using the ciao task dmextract , with a background annulus ranging from 5 to 10 . in some cases , the background region was slightly altered to avoid nearby point sources . for ngc3079 and 4945 , the aperture is 1 to avoid nearby sources and strong background , and for ngc3184 the aperture is 3 since this object is far off - axis ( and thus has a large psf ) . a 2 aperture ensures that we measure virtually all of the point source counts , since the on - axis psf of _ chandra _ contains 95% of the encircled energy with 1 . finally , we used two micron all - sky survey ( 2mass ; @xcite ) positions ( obtained from the nasa / ipac extragalactic database , ned ) , with typical @xmath19 uncertainties of @xmath705 for the photometric center of _ extended _ sources @xcite , to determine whether x - ray point - source detections were coincident with the nucleus . while the nominal accuracy for 2mass is often quoted as 01 , this only applies to point sources . the distribution of offsets between the nearest x - ray detection and photometric center from 2mass is highly bimodal , and we chose to consider objects belonging to the smaller - offset group as agn candidates . our formal criterion requires that the 2mass positions and x - ray detections lie within 5 of each other . in practice almost all candidates have better than 25 agreement , while nearly all non - detections have no x - ray point sources within 10 of the 2mass position . ( ngc4625 has an x - ray source within 6 of the optical nucleus , while ngc3395 , ngc3556 , ngc4038 , and ngc4631 have an x - ray source within 10 of the optical nucleus . ) the minimum x - ray signal - to - noise ( s / n ) threshold accepted was 1.5 , although in practice only three sources were detected with such low s / n . we detected 18 x - ray point sources in our sample of 64 objects ; 12 sources with s / n @xmath2 3 , three sources with 3 @xmath2 s / n @xmath2 2 , and three sources with 2 @xmath2 s / n @xmath2 1.5 . for all agn candidates except ic342 , m33 , ngc55 , and ngc4395 , we estimate the total x - ray luminosity similarly to @xcite , assuming a photon index of @xmath20 , typical of most low - luminosity agns ( see @xcite ) , and an average galactic interstellar medium column density of @xmath21 @xmath22 . this gives @xmath3(210 kev ) @xmath23 ergs s@xmath6 ( acis counts ks@xmath6 ) ( @xmath24/10 mpc)@xmath25 . this is only valid , however , for acis - s data . for ic342 and ngc55 ( i.e. hrc - i and acis - i data ) , we use the pimms tool ( version 3.9d ) with the same assumptions to obtain the 210 kev luminosity . the luminosities would increase by a factor of 1.8 and 6.7 for intrinsic column densities of obscuring gas and dust of @xmath26 and @xmath27 @xmath22 , respectively . for m33 ( hrc - s data ) , we again use pimms but assume @xmath28 @xmath22 @xcite . ngc4395 is a known agn with strong absorption ( @xmath29 @xmath22 ; @xcite ) , and so we use this value instead to estimate @xmath3 . none of our acis objects suffer from significant pileup ( multiple detections of lower energy photons within a single frame time that is registered as a single event at higher energy , possibly even rejected as a cosmic ray hit ) . ngc4395 has the highest count rate of @xmath30 counts s@xmath6 , which corresponds to a pileup of only 2% ( calculated using pimms ) , roughly equal to the measured uncertainty in total counts . we obtained an upper limit for those sources without an x - ray point - source detection . we used a circular aperture with a 2 radius , centered at the optical position of the nucleus . we measured the background counts with its uncertainty , and determined the counts necessary for a theoretical source to have a s / n equal to 1.5 and thus be detected by celldetect . these counts were then converted to a luminosity as described above . in some cases , because of the very low exposure times , no counts were detected in this aperture size . in these cases , we used larger apertures and scaled down to a 2 aperture , up to a 12 aperture in the most extreme cases . fields were sparse enough that no contamination from nearby x - ray point sources was present . for objects with a detected x - ray point source , the limiting luminosity was similarly found using the background measurements , scaled to a 2 aperture . as a proxy for spectral shape , we compute x - ray colors for those objects with detected x - ray sources , with counts measured in three bands : soft ( @xmath31 , 0.31 kev ) , medium ( @xmath32 , 12 kev ) , and hard ( @xmath33 , 28 kev ) . the soft and hard colors are then defined to be @xmath34 and @xmath35 , respectively , where @xmath0 is the total counts in all three bands . this was only achievable with acis data , which have accurate registration of incoming photon energy , whereas the hrc instrument has a spectral resolution of only @xmath36 . four objects in our sample have sufficient counts ( i.e. @xmath37 ) to do a preliminary spectral analysis : ngc4039 , ngc4395 , ngc4490 , and ngc6946 . ngc4945 has most of its counts in a strong fe k@xmath38 feature at 6.4 kev , resulting in few continuum photons . to accomplish this analysis , we used the ciao task psextract to extract the x - ray spectra along with the associated response files . the aperture and background used are the same as those for the x - ray photometry . bins are selected to have a minimum of 20 counts . using xspec @xcite , we then fit simple models to the data , ignoring bins with energies below 0.3 kev and above 5 kev ( where there are few counts ) . in the case of ngc4395 , a strong source , we ignored bins with energies below 1 kev ( there is a soft excess at low energies ; see also @xcite ) and above 10 kev . we tested various models , including an absorbed power law , a blackbody , and an absorbed power law + blackbody . out of our sample of 64 objects , we detect 18 sources coincident with the optical position of the nucleus . we consider these as agn candidates ( except for m33 , which we discuss in section [ sec : disc ] ) . our full sample is summarized in table [ tab : log ] , while the agn candidates are summarized in table [ tab : agn ] , along with their measured x - ray colors . their images are presented in figure [ fig : images ] . of the 17 candidates , three objects are known agns from optical diagnostics ( ngc3079 , ngc4395 , and ngc4945 , although they are weak sources ) , and one more is a known low - ionization nuclear emission - line region ( liner ; ngc6503 ) . ngc4654 , ngc4701 and ngc5457 were detected above a s / n of 1.5 . ngc925 , ngc4713 , ngc6503 were detected above a s / n of 2 . the rest of the agn candidates were detected with a s / n above 3 . ngc45 has a possible nuclear x - ray source coincident with the optical position , but it was not detected above a s / n of 1.5 , and thus is not included in our candidate list . the 17 agn candidates have 210 kev luminosities ranging from @xmath39 to @xmath5 ergs s@xmath6 . the distribution of @xmath3 is shown in figure [ fig : lx ] . the three known agns ( ngc3079 , ngc4395 , and ngc4945 ) have @xmath40 , @xmath5 , and @xmath41 ergs s@xmath6 , respectively . the median offset between the x - ray detection and the optical / near - infrared position is @xmath1 = 16 . for those objects with enough counts , our x - ray spectral fits are summarized in table [ tab : spectra ] and figure [ fig : spectra ] , and are discussed below . for the agn candidates with available measurements of @xmath42 in the literature , we estimate a possible bh mass using the @xmath8 relation of @xcite . we also estimate the corresponding x - ray luminosity to eddington luminosity ratio . these are presented in table [ tab : agn ] . the evidence for the majority of these candidates being true agns is circumstantial . for most of these systems , the physical scale of an x - ray point source in _ chandra _ data is of order tens of parsecs , making it extremely unlikely that the vast majority of detections are serendipitous field x - ray binaries along the line of sight to or near the nucleus . furthermore , the probability of detecting a background / foreground x - ray source within the _ chandra _ psf is extremely low , of order @xmath43 @xcite . with a few exceptions , the positional agreement between the x - ray source and the 2mass position of the nucleus is excellent ( less than 2 , compared to the _ chandra _ and 2mass positional uncertainties of @xmath71 ) . ngc4945 has a larger offset ( @xmath44 32 ) , but the presence of bright diffuse x - ray emission makes point - source detection more challenging , and once again this object is a known seyfert 2 . many of these galaxies have nuclei as identified in the palomar survey , but most of the optical observations have slit apertures of @xmath45 @xcite , much larger than the _ the palomar data are therefore probing a much larger region than the x - ray point sources , and we must be careful when making comparisons . as these are gas- and dust - rich systems , there is clearly some level of star formation in the circumnuclear region . if we consider , for example , typical luminosities emitted by such low - luminosity agns , they are significantly lower than the measured luminosities in the palomar data . assuming the bolometric , x - ray , and luminosities of agns are related by @xmath46 @xcite , a range @xmath47 @xmath48 ergs s@xmath6 corresponds to @xmath49 @xmath50 ergs s@xmath6 . palomar objects in our sample with nuclei have @xmath51 @xmath52 ergs s@xmath6 , which can explain why such low - luminosity agns , if present , were missed in the palomar data . the galaxies in our sample are low - mass , late - type spirals , and thus likely harbor a nc at the center of the gravitational potential well . @xcite found that @xmath775% of late - type spirals ( @xmath0 = 69 ) host a compact , photometrically distinct source located near the photometric center of the galaxy , with the surface brightness profile deviating from a pure disk profile . these ncs have mean effective radii of @xmath53 pc and @xmath54-band luminosities of @xmath55 . @xcite found that for ncs in galaxies within a similar range in @xmath0 type , the average cluster mass is @xmath56 . the increased stellar density at the center therefore implies an increased chance of detecting , in particular , a low - mass x - ray binary ( lmxb ) as a nuclear x - ray point source , confusing our results . m33 , for example , likely has no central massive bh @xcite , but does possess an x - ray source coincident with the optical nucleus , with properties suggestive of an x - ray binary @xcite . this is why we do not included m33 as part of our agn candidates . ngc2403 also hosts an x - ray binary in its known nuclear star cluster @xcite , although the x - ray luminosity is not high enough to be confused with an agn ( in this work , no detection was found above the limiting luminosity of @xmath57 ergs s@xmath6 ) . x - ray binaries must therefore be carefully considered when attempting to detect low - luminosity agns . although ncs also occur frequently in more massive early - type spiral galaxies @xcite and intermediate - luminosity elliptical galaxies @xcite , lmxb contamination is not important in these cases because at these larger bh masses @xmath58 ergs s@xmath6 due to nuclear activity , much greater than typical lmxb x - ray luminosities . an x - ray detection in a nc is much more likely to be a lmxb rather than a high - mass x - ray binary ( hmxb ) because the mean age of ncs in late - type hosts is @xmath59 yrs @xcite , older than typical high - mass stars in hmxbs . although many late - type galaxies have nuclei @xcite , the slit aperture used , in most cases , is significantly larger than the _ chandra _ psf . given that nearly all galaxies in the survey exhibit some emission lines in their nuclear spectra , and that late - type galaxies in particular are gas and dust rich with strong levels of star formation , the prevalence of nuclei is not surprising . this does not necessarily imply , however , that the ncs have ongoing star formation ( and therefore hmxbs are more likely ) , as evidenced by the inferred nc ages from @xcite . for our purposes , we will approximate the ncs as globular clusters ( gcs ) in order to estimate the incidence of lmxbs . although the environments of both types of clusters are widely different , ncs and gcs share remarkably similar structural and dynamical properties . viewed in the fundamental plane ( relating effective radius , surface density , and velocity dispersion ) , ncs occupy a region very distinct from classical bulges and elliptical galaxies , and fall on the high - mass extension of the gc sequence @xcite . with some important exceptions ( see below ) , approximating ncs as gcs is therefore well justified . the quantity @xmath63 specifies the corrected half - mass radius , where @xmath64}$ ] . since this depends rather weakly on color , we can approximate @xmath65 pc as a mean value for ncs in late - type spirals . @xcite find a mean @xmath66 color for such ncs of order @xmath67@xmath68 mag . in order to get @xmath61 colors , we use approximate color transformations from @xcite by approximating the stellar population in a nc as that between an scd and an sbc galaxy , which have @xmath66 colors similar to those of ncs . this gives @xmath61 colors ranging from 1.05 to 1.17 mag ; we adopt @xmath69 mag . therefore the mean number of lmxbs we expect is @xmath70 . this estimate is specifically for sources brighter than @xmath71 ergs s@xmath6 , which corresponds to the completeness limit of @xcite . the limiting luminosity for our x - ray detections varies from @xmath72 to @xmath73 ergs s@xmath6 . for objects in the @xcite sample with a lower limiting luminosity ( @xmath74 ergs s@xmath6 or lower ) , however , they detected only a factor of a few more sources . an important caveat to this estimate is the question of stellar age . gcs generally have old stellar populations , formed via a single burst of star formation , whereas ncs have repeated star formation episodes since the reservoir of gas can be replenished . this leads to , for a given cluster mass , a younger mean stellar age for ncs compared to gcs , and thus bluer colors @xcite . as a result , the expected number of lmxbs estimated above is an upper limit , since it was derived for gcs , and the true number of lmxbs in ncs is likely lower . this is due to the @xmath75 gyr timescale necessary for the stellar companion to come into roche lobe contact with its bh companion ( and thus turn on the lmxb ) , either by stellar evolution or via orbital decay @xcite . using @xmath70 as our estimate , this would imply only @xmath76 x - ray detections from lmxbs in a sample of 64 late - type spirals ( assuming 75% have ncs with mean properties ) , of which m33 is a typical example . this is a factor of 34 lower than our observed detection rate . the observed x - ray luminosities in our sample are similar to the known agns , making our other detections reasonable agn candidates . this implies that ncs and bhs need not be necessarily mutually exclusive , if the majority of our agn candidates are true agns . indeed , @xcite find that over all hubble types , roughly 10% of all galaxies hosting a nc also host an agn , based on optical spectroscopy , with a fraction that increases strongly with galaxy and nc mass . it is important to note that these detections are agn _ candidates _ , and are worthy of further detailed observations to properly identify these sources . @xcite provide some evidence from a multiwavelength analysis that ngc3184 , ngc4713 , and ngc5457 are true agns . since most of our candidates have very few counts , long - integration x - ray spectroscopy will be useful in differentiating true agns from lmxbs , particularly at very soft and hard energies . as described above , agns have x - ray spectra composed of an absorbed power law ( with @xmath77 ) , whereas x - ray binaries have spectra that can be separated into three states : a high / soft state with a @xmath71 kev thermal component and a weak nonthermal tail at high ( @xmath2 10 kev ) energies ; a very high state with a steep power law ( @xmath78 ) combined with a @xmath71 kev thermal blackbody , which can constitute a significant amount of the total flux ; and a low / hard state with a single power law ( @xmath79 ) @xcite . of these three states , only the low / hard state could be confused with an agn , but x - ray binaries in these states typically have extraordinarily low luminosities ( @xmath80 @xmath81 ergs s@xmath6 ) , which would distinguish them from true agns . our spectral fits ( table [ tab : spectra ] and figure [ fig : spectra ] ) are insufficient to definitively rule out one model over another , though they remain very suggestive . for ngc4039 , a pure thermal blackbody is likely ruled out , whereas an absorbed power law is plausible , and an absorbed power law + blackbody model ( indicative of a very high state lmxb ) is consistent with the data . given that the reduced @xmath82 of the latter model is @xmath831 , however , we may be overfitting the data , and so we must be cautious . although it may appear as though this is more likely an x - ray binary ( of extremely high luminosity @xmath84 ergs s@xmath6 ) , longer - integration data are needed to distinguish between the two models . ngc4490 and ngc6946 are both well fit by simple absorbed power law models ( @xmath85 ) , consistent with the agn interpretation . in the case of ngc4490 , the unabsorbed blackbody model results in a poorer fit , particularly at low energies , while for ngc6946 , both models are consistent with the data . unfortunately , the data are not of high enough quality to differentiate between models with more parameters , such as an absorbed power law + blackbody , since @xmath86 in all cases . once again , these results underscore the need for higher s / n data if we are to use a @xmath82 statistic to differentiate between various models .. as a check , we also analyzed ngc4395 , a known seyfert 1 , which is unsurprisingly well fit by an absorbed power law , whose parameters are consistent with those given by @xcite . in figure [ fig : colors ] we show the x - ray colors of those agn candidates with sufficient counts . we also include various absorbed power - law models and a 1 kev blackbody ( representative of the high / soft state from @xcite ) , computed for observations with acis - s using xspec . all agn candidates are consistent with absorbed power - law models , and inconsistent with the 1 kev blackbody . although the uncertainties are large , it appears as though most of these candidates exhibit softer x - ray spectra than is typically assumed for low - luminosity agns . low - mass agns can exhibit a range of x - ray spectra , however , with @xmath87 1.0 to 3.0 @xcite . clearly , detailed x - ray spectroscopy is needed to further characterize these systems . once agns are positively identified , determining a bh mass estimate will be challenging . dynamical measurements are currently not feasible in these objects . reverberation mapping @xcite or virial estimates @xcite based on the strength of optical emission lines ( if detected ) will likely require excellent seeing conditions with a narrow slit or adaptive optics . as these are all nearly bulgeless systems , using the bulge velocity dispersion @xmath42 or the bulge luminosity @xmath88 @xcite to determine @xmath89 is challenging and highly uncertain . ncs with bhs , however , fall on the @xmath8 relation for example ngc4395 @xcite and g1 @xcite consistent with the extrapolation of @xmath8 to lower masses @xcite . the velocity dispersions of ncs @xcite could therefore be used to estimate the bh mass . bh masses can also be estimated from x - ray variability timescales derived from complete power density spectra ( @xcite ) , obtained over long monitoring campaigns . alternatively , the bh mass can be estimated from the excess variance measured in shorter x - ray data sets @xcite . our detection rate suggests an agn fraction in late - type spiral galaxies of @xmath720%25% ( depending on the lmxb contamination ) , higher than that observed by @xcite . ( we should caution , however , that several detections are at low s / n . using only s / n@xmath23 detections , our agn fraction becomes @xmath717% . ) this would imply that we are indeed missing many low - mass agns in optical surveys , and that bhs in such systems are more prevalent than previously suggested . optical surveys of low - mass agns can only find systems with high accretion rates , such that their bolometric luminosity @xmath90 approaches a substantial fraction of the eddington luminosity @xmath91 ergs s@xmath6 @xcite . @xcite found a median @xmath92 and a minimum @xmath93 in a sample of 174 optically selected low - mass agns . in a pilot study drawn from this sample , @xcite measured @xmath3(0.52 kev ) @xmath94@xmath95 ergs s@xmath6 , much higher than any of our measurements . most of our candidates with available @xmath89 estimates exhibit @xmath96 ratios orders of magnitude lower than those of @xcite , assuming a correction factor of 16 to convert from @xmath3 to @xmath90 @xcite . the one exception is ngc4395 , a known seyfert nucleus , with @xmath97 , consistent with the value of @xmath98 to @xmath99 found by @xcite . ( in comparison , pox52 has @xmath100 ; @xcite . ) our estimated eddington ratios are similar , however , to those seen in massive bhs in earlier - type hosts with low - luminosity agns @xcite , suggesting that we are detecting the low - mass analogs of weakly accreting bhs . x - ray surveys therefore allow us to probe much fainter down the luminosity function of low - mass agns , detecting weakly accreting systems that would be completely swamped by galactic emission at optical wavelengths , and obtain a much higher detection rate . the @xcite sample , for instance , identified only 174 low - mass agns out of 8435 optically identified agns ( from a parent sample of @xmath7600,000 galaxies ) . given the low estimated accretion rates of our agn candidates , radio observations can help to clarify the nature of these x - ray sources . if these are indeed low - mass and low - accretion - rate systems , we should expect them to have a radio signature , much like the more massive low - luminosity bhs @xcite . extrapolating from higher masses , we would expect 2 cm radio fluxes of order 10 to 100 @xmath101jy @xcite . if these x - ray sources are x - ray binaries instead , however , they would be accreting at eddington or super - eddington rates to achieve such high luminosities . x - ray binaries in such states do no have a radio counterpart @xcite . @xcite have conducted a similar survey of agns in the infrared , searching for emission - line signatures of agn activity ( in particular , emission ; see @xcite ) in late - type galaxies that are optically quiescent . this should identify weakly accreting systems similar to our candidates . in a sample of 32 objects , nine of which overlap with our _ chandra _ sample , they find seven agn candidates , a similar detection fraction . interestingly , of the nine overlap objects , six ( ic342 , ngc925 , ngc3184 , ngc4490 , ngc4559 , and ngc6946 ) are identified in this paper as agn candidates , but not by @xcite it is possible that the weakly active nuclei picked out by our x - ray search method are too faint to emit much emission . we can estimate the total luminosity @xmath102 for our sample assuming the bolometric luminosity @xmath103 and @xmath104 @xcite , where @xmath105 is the continuum luminosity at 5100 , coupled with the correlation between @xmath105 and @xmath102 found by @xcite . for our range in x - ray luminosities ( @xmath47 @xmath48 ergs s@xmath6 ) , this corresponds to luminosities of @xmath106 @xmath107 ergs s@xmath6 . these luminosities are orders of magnitude fainter than those measured by @xcite . even ngc3184 , which appears to be a _ bona fide _ agn @xcite , was not detected . the @xcite detections of are in excess of the @xcite relation , however , thus it is possible these relations do not hold for very late - type spiral galaxies . another factor that might be relevant is the ionization parameter ; since is a high - ionization line , its strength decreases dramatically when the ionization parameter decreases @xcite . by analogy with the local population of liners and seyferts @xcite , the very low eddington ratios of our low - mass systems should correspond to low ionization parameters , and thus plausibly weak emission . finally , we note that ngc3556 is identified by @xcite as an agn candidate but has no x - ray detection in this work . @xcite identify an x - ray source as a candidate for the galactic nucleus in ngc3556 ( which they number source 35 ) ; we chose not to include this source in our candidate list as the offset from the optical nucleus is 10 . more detailed study is clearly warranted , and it will likely take the combined efforts of large multiwavelength surveys to find such low - mass and low - luminosity agns , and to properly assess the global agn fraction in these late - type systems . l .- b.d . would like to thank eliot quataert for the many helpful and insightful discussions , and jenny greene for advice and guidance . we also thank an anonymous referee for careful , helpful comments that greatly improved this manuscript . support for this work was provided by nasa through chandra award number 06700184 issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 , and through grant hst - go-11130.01 from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas5 - 26555 . 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 . we also acknowledge the usage of the hyperleda database ( http://leda.univ-lyon1.fr ) . , g. p. , bautz , m. w. , ford , p. g. , nousek , j. a. , & ricker , jr . , g. r. 2003 , in proceedings of the spie , vol . 4851 , x - ray and gamma - ray telescopes and instruments for astronomy , ed . j. e. truemper & h. d. tananbaum , 28 lcccccccccccrrc ic239 & 14.2 & 6.0 & sab(rs)cd & l2 : : & & 02 36 27.83 & + 38 58 08.5 & 7131 & acis - s & 4.5 & @xmath835 & @xmath8337.9 & & p + ic342 & 3.9 & 6.0 & sab(rs)cd & h & y ( b97 ) & 03 46 48.51 & + 68 05 45.9 & 6902 & hrc - i & 12.1 & 58 @xmath108 9 & 37.9 & 37.0 & p + ic1727 & 6.4 & 9.0 & sb(s)m & t2/l2 & & 01 47 29.88 & + 27 20 00.0 & 1634 & acis - s & 1.7 & @xmath835 & @xmath8337.7 & & p + ic3635 & 13.1 & 6.0 & scd ? & & & 12 40 13.33 & + 12 52 29.9 & 8104 & acis - s & 5.1 & @xmath835 & @xmath8337.8 & & c + ic5332 & 8.4 & 7.0 & s(s)d & & & 23 34 27.48 & @xmath10936 06 03.8 & 2067 & acis - s & 55.2 & @xmath836 & @xmath8336.4 & & c + m33 & 0.7 & 6.0 & s(s)cd & h & y ( km93 ) & 01 33 50.90 & + 30 39 35.7 & 3948 & hrc - s & 5.2 & 3638 @xmath108 61 & 38.9 & 36.3 & p + ngc45 & 5.9 & 8.0 & s(s)dm & & & 00 14 03.98 & @xmath10923 10 55.5 & 4690 & acis - s & 34.4 & @xmath8311 & @xmath83 36.6 & & c + ngc55 & 1.3 & 9.0 & sb(s)m : spin & & n ( s06 ) & 00 14 53.60 & @xmath10939 11 47.8 & 2255 & acis - i & 59.4 & @xmath837 & @xmath8335.1 & & c + ngc625 & 3.9 & 9.0 & sb(s)m ? spin & & & 01 35 04.63 & @xmath10941 26 10.3 & 4746 & acis - s & 60.3 & @xmath837 & @xmath8335.8 & & c + ngc672 & 7.5 & 6.0 & sb(s)cd & h & & 01 47 54.52 & + 27 25 58.0 & 7090 & acis - s & 2.2 & @xmath835 & @xmath8337.7 & & p + ngc925 & 9.4 & 7.0 & sab(s)d & h & & 02 27 16.91 & + 33 34 43.9 & 7104 & acis - s & 2.2 & 13 @xmath108 5 & 38.3 & 37.9 & p + ngc959 & 10.1 & 8.0 & sdm : & h & & 02 32 23.94 & + 35 29 40.8 & 7111 & acis - s & 2.2 & @xmath835 & @xmath8337.9 & & p + ngc988 & 18.3 & 6.0 & sb(s)cd : & & & 02 35 27.72 & @xmath10909 21 21.6 & 3552 & acis - s & 5.3 & @xmath835 & @xmath8338.1 & & c + ngc1003 & 10.7 & 6.0 & s(s)cd & & & 02 39 16.89 & + 40 52 20.2 & 7116 & acis - s & 2.7 & @xmath835 & @xmath8337.9 & & c + ngc1313 & 3.7 & 7.0 & sb(s)d & & n ( b02 ) & 03 18 16.04 & @xmath10966 29 53.7 & 2950 & acis - s & 19.9 & @xmath837 & @xmath8336.2 & & c + ngc1493 & 11.3 & 6.0 & sb(r)cd & & y ( b02 ) & 03 57 27.38 & @xmath10946 12 38.6 & 7145 & acis - s & 10.0 & 50 @xmath108 8 & 38.4 & 37.4 & c + ngc1507 & 10.6 & 9.0 & sb(s)m pec ? & & n ( p96 ) & 04 04 27.21 & @xmath10902 11 18.9 & 7115 & acis - s & 2.8 & @xmath836 & @xmath83 37.9 & & c + ngc2403 & 4.2 & 6.0 & sab(s)cd & h & y ( dc02 ) & 07 36 51.39 & + 65 36 09.1 & 2014 & acis - s & 35.6 & @xmath838 & @xmath8336.1 & & p + ngc2500 & 10.1 & 7.0 & sb(rs)d & h & y ( p96 ) & 08 01 53.22 & + 50 44 13.5 & 7112 & acis - s & 2.6 & @xmath838 & @xmath8338.1 & & p + ngc2541 & 10.6 & 6.0 & s(s)cd & t2/h : & & 08 14 40.07 & + 49 03 41.1 & 1635 & acis - s & 1.9 & @xmath835 & @xmath8338.0 & & p + ngc2552 & 10.0 & 9.0 & s(s)m ? & & y ( b02 ) & 08 19 20.55 & + 50 00 35.1 & 7146 & acis - s & 7.9 & @xmath835 & @xmath8337.4 & & c + ngc3079 & 20.4 & 7.0 & sb(s)c spin & s2 & & 10 01 57.92 & + 55 40 48.0 & 2038 & acis - s & 26.6 & 73 @xmath108 10 & 38.6 & 37.8 & p + ngc3184 & 8.7 & 6.0 & sab(rs)cd & h & & 10 18 16.98 & + 41 25 27.7 & 804 & acis - s & 42.1 & 133 @xmath108 13 & 37.9 & 36.6 & p + ngc3274 & 5.9 & 7.0 & sabd ? & & & 10 32 17.28 & + 27 40 07.5 & 7085 & acis - s & 1.7 & @xmath836 & @xmath8337.7 & & c + ngc3299 & 5.4 & 8.0 & sab(s)dm & & & 10 36 23.80 & + 12 42 26.6 & 7831 & acis - s & 5.1 & @xmath835 & @xmath8337.0 & & c + ngc3395 & 27.4 & 6.0 & sab(rs)cd pec : & h & & 10 49 50.11 & + 32 58 58.2 & 2042 & acis - s & 19.5 & @xmath8310 & @xmath8338.2 & & p + ngc3432 & 7.8 & 9.0 & sb(s)m spin & h & & 10 52 31.13 & + 36 37 07.6 & 7091 & acis - s & 1.9 & @xmath836 & @xmath8337.8 & & p + ngc3495 & 12.8 & 7.0 & sd : & h : & & 11 01 16.16 & + 03 37 40.9 & 7126 & acis - s & 4.1 & @xmath836 & @xmath8337.9 & & p + ngc3556 & 14.1 & 6.0 & sb(s)cd spin & h & & 11 11 30.96 & + 55 40 26.8 & 2025 & acis - s & 59.4 & @xmath8312 & @xmath8337.2 & & p + ngc3985 & 8.3 & 9.0 & sb(s)m : & & & 11 56 42.10 & + 48 20 02.2 & 7099 & acis - s & 1.7 & @xmath836 & @xmath8337.9 & & c + ngc4020 & 8.0 & 7.0 & sbd ? spin & & & 11 58 56.93 & + 30 24 49.0 & 7093 & acis - s & 1.7 & @xmath835 & @xmath8337.8 & & c + ngc4038 & 25.5 & 9.0 & sb(s)m pec & & & 12 01 52.98 & @xmath10918 52 09.9 & 3043 & acis - s & 67.1 & @xmath8324 & @xmath8337.9 & & c + ngc4039 & 25.3 & 9.0 & s(s)m pec & & & 12 01 53.64 & @xmath10918 53 10.8 & 3043 & acis - s & 67.1 & 571 @xmath108 27 & 39.3 & 37.8 & c + ngc4204 & 7.9 & 8.0 & sb(s)dm & & y ( b02 ) & 12 15 14.35 & + 20 39 32.6 & 7092 & acis - s & 2.0 & @xmath836 & @xmath8337.8 & & c + ngc4244 & 3.1 & 6.0 & s(s)cd : spin & h & y ( s06 ) & 12 17 29.65 & + 37 48 25.5 & 942 & acis - s & 49.2 & @xmath837 & @xmath8335.7 & & p + ngc4395 & 4.2 & 9.0 & s(s)m : & s1.8 & y ( fh03 ) & 12 25 48.92 & + 33 32 48.2 & 5302 & acis - s & 28.2 & 2755 @xmath108 54 & 39.6 & 37.0 & p + ngc4411b & 16.8 & 6.0 & sab(s)cd & & y ( b02 ) & 12 26 47.21 & + 08 53 04.4 & 7840 & acis - s & 4.9 & @xmath835 & @xmath8338.0 & & c + ngc4490 & 7.8 & 7.0 & sb(s)d pec & h & & 12 30 36.36 & + 41 38 37.0 & 4726 & acis - s & 39.6 & 1575 @xmath108 41 & 38.9 & 36.6 & p + ngc4559 & 9.7 & 6.0 & sab(rs)cd & h & & 12 35 57.68 & + 27 57 35.1 & 2686 & acis - s & 3.0 & 27 @xmath108 6 & 38.5 & 37.8 & p + ngc4561 & 12.3 & 8.0 & sb(rs)dm & & & 12 36 08.13 & + 19 19 21.3 & 7125 & acis - s & 3.5 & 100 @xmath108 11 & 39.2 & 37.9 & c + ngc4592 & 9.6 & 8.0 & s(s)dm : & & & 12 39 18.77 & @xmath10900 31 54.6 & 7105 & acis - s & 2.1 & @xmath835 & @xmath8337.9 & & c + ngc4618 & 7.3 & 9.0 & sb(rs)m & h & y ( b02 ) & 12 41 32.84 & + 41 09 02.7 & 7147 & acis - s & 9.3 & @xmath835 & @xmath8337.0 & & p + ngc4625 & 8.2 & 9.0 & sab(rs)m pec & & y ( b02 ) & 12 41 52.70 & + 41 16 25.3 & 7098 & acis - s & 1.7 & @xmath835 & @xmath8337.9 & & c + ngc4631 & 6.9 & 7.0 & sb(s)d spin & h & n ( s06 ) & 12 42 08.00 & + 32 32 29.4 & 797 & acis - s & 59.2 & @xmath838 & @xmath8336.4 & & p + ngc4654 & 16.8 & 6.0 & sab(rs)cd & h & & 12 43 56.63 & + 13 07 34.8 & 3325 & acis - s & 4.9 & 12 @xmath108 5 & 38.4 & 38.1 & p + ngc4688 & 17.1 & 6.0 & sb(s)cd & h & & 12 47 46.46 & + 04 20 09.8 & 7859 & acis - s & 4.7 & @xmath836 & @xmath8338.1 & & p + ngc4701 & 20.5 & 6.0 & s(s)cd & & y ( b02 ) & 12 49 11.56 & + 03 23 19.4 & 7148 & acis - s & 9.0 & 12 @xmath108 5 & 38.3 & 38.0 & c + ngc4713 & 17.9 & 7.0 & sab(rs)d & t2 & & 12 49 57.89 & + 05 18 41.1 & 4019 & acis - s & 4.9 & 10 @xmath108 4 & 38.4 & 38.1 & p + ngc4945 & 5.2 & 6.0 & sb(s)cd : spin & s & & 13 05 27.27 & @xmath10949 28 04.4 & 864 & acis - s & 49.1 & 379 @xmath108 22 & 37.9 & 36.7 & c + ngc5068 & 6.7 & 6.0 & sab(rs)cd & & y ( b02 ) & 13 18 54.80 & @xmath10921 02 20.7 & 7149 & acis - s & 6.7 & @xmath835 & @xmath8337.1 & & c + ngc5204 & 4.8 & 9.0 & s(s)m & h & & 13 29 36.50 & + 58 25 07.4 & 3933 & acis - s & 4.8 & @xmath835 & @xmath8337.0 & & p + ngc5457 & 5.4 & 6.0 & sab(rs)cd & h & & 14 03 12.59 & + 54 20 56.7 & 2065 & acis - s & 9.6 & 20 @xmath108 6 & 37.3 & 36.8 & p + ngc5474 & 6.0 & 6.0 & s(s)cd pec & h & & 14 05 01.60 & + 53 39 43.9 & 7086 & acis - s & 1.7 & @xmath835 & @xmath8337.6 & & p + ngc5585 & 7.0 & 7.0 & sab(s)d & h & y ( b02 ) & 14 19 48.20 & + 56 43 44.5 & 7150 & acis - s & 5.3 & @xmath835 & @xmath8337.2 & & p + ngc5774 & 26.8 & 7.0 & sab(rs)d & & y ( b02 ) & 14 53 42.75 & + 03 34 56.0 & 2940 & acis - s & 58.2 & @xmath836 & @xmath8337.5 & & c + ngc6503 & 6.1 & 6.0 & s(s)cd & t2/s2 : & & 17 49 26.51 & + 70 08 39.6 & 872 & acis - s & 13.2 & 13 @xmath108 5 & 37.1 & 36.8 & p + ngc6689 & 12.2 & 7.0 & sd ? spin & h & n ( p96 ) & 18 34 50.24 & + 70 31 25.9 & 7123 & acis - s & 3.5 & @xmath835 & @xmath8337.9 & & p + ngc6946 & 5.5 & 6.0 & sab(rs)cd & h & ? ( b02 ) & 20 34 52.33 & + 60 09 13.2 & 1043 & acis - s & 58.3 & 766 @xmath108 29 & 38.2 & 36.2 & p + ngc7424 & 11.5 & 6.0 & sab(rs)cd & & y ( b02 ) & 22 57 18.36 & @xmath10941 04 14.0 & 3496 & acis - s & 23.9 & @xmath836 & @xmath8337.1 & & c + ngc7741 & 12.3 & 6.0 & sb(s)cd & h & n ( b02 ) & 23 43 54.37 & + 26 04 32.0 & 7124 & acis - s & 3.5 & @xmath835 & @xmath8337.9 & & p + ngc7793 & 2.8 & 7.0 & s(s)d & & y ( b02 ) & 23 57 49.82 & @xmath10932 35 27.7 & 3954 & acis - s & 48.9 & @xmath837 & @xmath8335.6 & & c + ugc05460 & 19.9 & 7.0 & sb(rs)d & & & 10 08 09.30 & + 51 50 38.0 & 7835 & acis - s & 5.1 & @xmath835 & @xmath8338.2 & & c + ugc08041 & 14.2 & 7.0 & sb(s)d & & & 12 55 12.68 & + 00 07 00.0 & 7132 & acis - s & 4.7 & @xmath835 & @xmath8337.9 & & c + ugc09912 & 19.7 & 8.0 & sbdm & & & 15 35 10.50 & + 16 32 58.0 & 7832 & acis - s & 5.1 & @xmath835 & @xmath8338.2 & & c lllcccccrl ic342 & 1.96 & 37.0 & 33 @xmath108 3 ( 1 ) & 6.5 & 37.9 & @xmath1096.7 & & & colors n / a + ngc925 & 1.65 & 75.3 & & & 38.3 & & & & too few counts for colors + ngc1493 & 0.68 & 37.4 & 25.0 @xmath108 3.8 ( 2 ) & 4.5 & 38.4 & @xmath1094.3 & 0.08 @xmath108 0.16 & @xmath1090.31 @xmath108 0.15 & + ngc3079 & 1.29 & 127.6 & 150 @xmath108 10 ( 3 ) & 7.6 & 38.6 & @xmath1097.1 & & 0.35 @xmath108 0.15 & bright diffuse emission ; + & & & & & & & & & no soft source ; known seyfert 2 + ngc3184 & 2.19 & 92.1 & & & 37.9 & & 0.09 @xmath108 0.10 & @xmath1090.34 @xmath108 0.09 & double source ? + ngc4039 & 2.08 & 255.0 & & & 39.3 & & 0.01 @xmath108 0.04 & @xmath1090.30 @xmath108 0.04 & + ngc4395 & 1.08 & 22.0 & @xmath8330 ( 4 ) & @xmath834.8 & 39.6 & @xmath1103.3 & 0.06 @xmath108 0.01 & 0.74 @xmath108 0.02 & known seyfert 1 + ngc4490 & 1.91 & 72.2 & & & 38.9 & & 0.47 @xmath108 0.02 & @xmath1090.16 @xmath108 0.03 + ngc4559 & 1.03 & 48.7 & & & 38.5 & & & @xmath1090.42 @xmath108 0.29 & no soft source + ngc4561 & 1.60 & 95.5 & & & 39.2 & & 0.07 @xmath108 0.10 & @xmath1090.16 @xmath108 0.10 + ngc4654 & 1.59 & 129.8 & & & 38.4 & & & & too few counts for colors + ngc4701 & 2.92 & 290.5 & & & 38.3 & & & & too few counts for colors + ngc4713 & 0.30 & 26.3 & & & 38.4 & & & & too few counts for colors + ngc4945 & 3.17 & 79.8 & 134 @xmath108 20 ( 5 ) & 7.4 & 37.9 & @xmath1097.7 & & & bright diffuse emission ; no soft + & & & & & & & & & or medium source ; known seyfert + ngc5457 & 0.98 & 25.6 & 80 @xmath108 20 ( 6 ) & 6.5 & 37.3 & @xmath1097.3 & & & too few counts for colors + ngc6503 & 0.95 & 28.1 & 46 @xmath108 3 ( 8) & 5.6 & 37.1 & @xmath1096.5 & & & too few counts for colors ; + & & & & & & & & & known nuclear activity + ngc6946 & 1.28 & 34.2 & & & 38.2 & & 0.47 @xmath108 0.04 & @xmath1090.22 @xmath108 0.04 & double source ? + [ tab : agn ] lccccccc ngc4039 + abs . power law & 0.20 @xmath108 0.05 & 2.7 @xmath108 0.2 & @xmath111 & & & 1.38 & 24 + bbody & & & & 0.35 @xmath108 0.02 & @xmath112 & 2.54 & 25 + abs . pl + bbody & 0.7 @xmath108 0.2 & 2.8 @xmath108 0.4 & @xmath113 & 0.09 @xmath108 0.01 & @xmath114 & 0.83 & 22 + + ngc4395 + abs . power law & 2.7 @xmath108 0.2 & 0.8 @xmath108 0.1 & @xmath115 & & & 1.15 & 110 + bbody & & & & 4.5 @xmath108 0.3 & @xmath116 & 3.36 & 111 + + ngc4490 + abs . power law & 0.80 @xmath108 0.06 & 2.3 @xmath108 0.1 & @xmath117 & & & 1.10 & 60 + bbody & & & & 0.78 @xmath108 0.02 & @xmath118 & 1.56 & 61 + + ngc6946 + abs . power law & 0.50 @xmath108 0.09 & 1.9 @xmath108 0.2 & @xmath119 & & & 0.59 & 30 + bbody & & & & 0.73 @xmath108 0.03 & @xmath120 & 0.73 & 31 [ tab : spectra ]

homogeneous cosmological models with non - vanishing intrinsic curvature require a special treatment when they are quantized with loop quantum cosmological methods . guidance from the full theory which is lost in this context can be replaced by two criteria for an acceptable quantization , admissibility of a continuum approximation and local stability . a quantization of the corresponding hamiltonian constraints is presented and shown to lead to a locally stable , non - singular evolution compatible with almost classical behavior at large volume . as an application , the bianchi ix model and its modified behavior close to its classical singularity is explored .

gr-qc0311004

since observations show that space is homogeneous at large scales to a very good approximation , homogeneous models have been studied for many decades in classical and quantum cosmology @xcite . they provide insights into the behavior of our universe while avoiding all the complicated field theoretic details of full gravity by their restriction to finitely many degrees of freedom . despite the huge reduction by infinitely many degrees of freedom , a large variety of different models is left which allow a detailed investigation of various non - trivial issues such as : dirac observables , the embedding of symmetric models ( e.g. , isotropic ) into less symmetric ones , dynamics , the approach to classical singularities and cosmological phenomenology . loop quantum cosmological methods for those models have recently been developed @xcite and shown to simplify considerably after a diagonalization of the connection and triad degrees of freedom @xcite . as an example the dynamics of the bianchi i model has been studied and shown to be singularity - free . in this case the hamiltonian constraint resembles that of the full theory @xcite ( albeit it is much simpler ) , and thus the quantization can be regarded as a reliable test . it turned out that the extension from the isotropic case @xcite to this anisotropic case , requires certain features which are in fact present in quantum geometry . in particular , the mechanism of a singularity - free evolution in anisotropic models relies on the fact that the evolution extends through the classical singularity which can be identified with a submanifold in minisuperspace . this requires the classical singularity to lie in the interior of minisuperspace , rather than at the boundary ( including infinity ) . here it is essential that quantum geometry is based on densitized triad variables where in fact the kasner singularity is in the interior ( all densitized triad components being zero ) , while it would be on the infinite boundary in co - triad variables . the isotropic case is not sensitive to this issue since both densitized and co - triad variables have the classical singularity in the interior of minisuperspace ( here , it is essential to use any ( co-)triad rather than metric variables ) ; thus the existence of the extension of the methods to a homogeneous model is non - trivial . an extension to the full theory of the general class of inhomogeneous models seems much more complicated at this point . however , it has been conjectured @xcite that at the classical level , the most general homogeneous behavior is that of the bianchi - ix model and furthermore that it describes the approach to the singularities even of _ inhomogeneous _ models . it is therefore important to study the more general bianchi class a models . homogeneous models other than bianchi i , however , present an additional complication since they have non - zero intrinsic curvature and , as a consequence , their spin connection can not be zero . this is in contrast even to the full theory where the spin connection can be made arbitrarily small locally by choosing appropriate coordinates . in a homogeneous model the freedom of choosing coordinates is restricted to those which preserve ( manifest ) homogeneity , and this implies that the spin connection is a covariant object and that it has to have a certain size in a given model . this has to be taken into account properly when one quantizes the hamiltonian constraint . in particular , in a classical regime only the extrinsic curvature can be assumed to be small , but not necessarily the intrinsic curvature which determines the spin connection . this is essential to understand the semiclassical behavior . since a special treatment is required which is not necessary in the full theory , one has to be more careful when interpreting the results . it introduces more quantization ambiguities which have to be shown not to influence main results . on the other hand , more possibilities for phenomenology emerge which , at least qualitatively , can often be seen to be insensitive to ambiguities . having a more distant relation to the full theory , it is helpful to have a set of admissibility criteria . one such criterion is of course to accommodate a semiclassical approximation to the quantum dynamics . since a semiclassical description is based on continuum geometry while the quantum dynamics is in terms of discrete quantum geometry , this criterion is formulated here in terms of a _ continuum approximation_. a further criterion is provided by the requirement of a _ locally stable evolution _ @xcite . such a requirement arises because the evolution equation derived along the lines of @xcite is usually a high order difference equation which at large volume is well approximated by the second order wheeler dewitt differential equation . hence , one can always construct solutions by choosing initial data , at large volume , to be that provided by solutions of the wheeler dewitt equation . perturbation of such an initial data will generate solutions which will also include surplus solutions of the high order difference equation . generically these will have planck scale oscillations . there is then the possibility that such extra solutions can become dominant under the evolution . in fact these are expected to become dominant as one gets to smaller volume since there are huge differences between the continuum and the discrete formulations in the planck regime . however as one evolves to larger volumes , the perturbations must not grow too rapidly since this would imply domination by solutions with planck scale oscillation even in the classical regime . the requirement we are looking for , local stability , prohibits this behavior by demanding that the local behavior of solutions to the difference equation around a large value of the evolution parameter is not exponentially increasing . in this paper we present a quantization of the hamiltonian constraint for bianchi models with non - zero intrinsic curvature which fulfills this condition and has the correct semiclassical limit . the first quantization given in @xcite was not admissible in this sense ; thus , the quantization given there is valid only for the bianchi i model ( for which we have the same quantization here ) . it turns out that the local stability condition is selective : it requires all roots of a high order polynomial to have unit norm which is not easy to achieve randomly . the selectivity is increased by the fact that the same strategy of quantizing a hamiltonian constraint should work in all homogeneous models , using different procedures in different models would imply that not all of them can be related to the full theory . the stability condition can be side - stepped by quantizing connection components in the constraint by hand such that only a second order difference equation results . while this would eliminate local stability as a selection criterion , it would also imply that the quantization is even more distant to that in the full theory where such a quantization can not be possible . further clues as to the necessary order of the evolution equation can come from studying dirac observables , which will be pursued elsewhere . in section [ s : models ] , we briefly describe the diagonal , homogeneous models addressed in this paper . we recall the classical framework and the corresponding loop quantization , including a quantization of inverse triad components and general aspects of the hamiltonian constraint equation . here we also discuss the two criteria of admissibility of the continuum approximation and local stability . in section [ hami ] , we present a quantization of the hamiltonian constraint . since the hamiltonian constraint now also has a potential term depending on the spin connection , a quantization of the spin connection is required . this is done using the quantization of inverse triad components . we derive the ( partial ) difference equation satisfied by the physical states and show that in the continuum approximation , to the leading order , the approximating differential equation is precisely the wheeler - dewitt equation with a specific factor ordering dictated by the underlying loop quantization . we also show that the requirement of local stability is satisfied . we demonstrate that the singularity avoidance mechanism found in the case of bianchi i @xcite continues to hold for these more general models . we conclude that the proposed quantization of the hamiltonian constraint is satisfactory with respect to all our requirements . in section [ bianchiix ] , we focus attention on the modification of the potential implied by the quantization of the spin connection . the non - trivial and non - perturbative behavior of the inverse triad components may be expected to lead to substantial modifications of the potential at small triad components . in view of the conjectured central role of the bianchi ix potential , we present a brief description of this potential and its implications for the modified approach to classical singularities . in the following , we restrict attention to the so called diagonal , bianchi class a models . among the class of homogeneous models i.e. models whose symmetry group acts transitively on the spatial manifold , is the sub - class of the bianchi models for which the symmetry group contains a subgroup with simply transitive action on the spatial manifold . the simply transitive subgroups are classified in terms of three integers , @xmath0 , parameterizing the structure constants as @xmath1 only class a models for which @xmath2 admit a canonical formulation . in the connection formulation , bianchi class a models are described in terms of the invariant connections and densitized triads given by , @xmath3 with constant and canonically conjugate @xmath4 , @xmath5 , where @xmath6 are left - invariant ( with respect to the subgroup acting simply transitively ) 1-forms on the spatial manifold @xmath7 and @xmath8 are corresponding dual density - weighted vector fields . the 1-forms satisfy the maurer - cartan equations @xmath9 in the metric formulation , _ diagonal _ homogeneous models are those for which the spatial metric can be written in a diagonal form such that only three degrees of freedom remain . in the connection formulation the situation is analogous . to begin with one has nine degrees of freedom , @xmath4 but with a three parameter gauge freedom of su(2 ) rotations of the index @xmath10 . a further freedom , depending upon the parameters @xmath0 , would have been available if the left - invariant 1-forms were not thought of as a background structure in parameterizing the invariant connections . to get the same three degrees of freedom as in the diagonal metric formulation , one then restricts the @xmath4 to a diagonal form " , @xmath11 and correspondingly @xmath12 where the so(3)-matrix @xmath13 includes gauge degrees of freedom ( see @xcite for details ) . such a restriction specifies the diagonal models in the connection formulation . this is _ not _ a symmetry reduction in the same sense as restriction to isotropic models is . nevertheless , these restricted models can be analyzed by following procedures similar to those in the context of symmetry reductions . as in reductions to isotropic models , this leads to significant simplifications in such diagonalized models because the gauge parameters in @xmath13 and ( almost ; see below ) gauge invariant parameters , @xmath14 , are neatly separated @xcite . the basic variables for diagonal bianchi class a models are specified via @xmath15 in terms of the ` gauge invariant ' coefficients @xmath14 with the pure gauge degrees of freedom contained in the @xmath16-matrix @xmath13 . the components @xmath14 are not completely gauge invariant but subject to residual discrete gauge transformations which change the sign of two of the three components simultaneously . a diagonal densitized triad has the form @xmath17 where @xmath8 are left - invariant densitized vector fields dual to @xmath6 . being an so(3)-matrix , @xmath13 satisfies @xmath18 the triad components @xmath19 are subject to the same residual gauge transformations as the connection components @xmath14 to which they are conjugate with basic poisson bracket @xmath20 where @xmath21 is the barbero immirzi parameter and @xmath22 the gravitational constant . we will use the value @xmath23 fixed by the black hole entropy calculations @xcite . a diagonal co - triad has the form @xmath24 with @xmath25 note that the components @xmath19 as well as @xmath26 can take negative values , but only the overall sign @xmath27 ( i.e. , the orientation ) and the absolute values @xmath28 are gauge invariant . with this form of the co - triad , we obtain , in fact , a diagonal metric @xmath29 the extrinsic curvature also has diagonal form with components @xmath30 which appear in the relation @xmath31 defining the ashtekar connection components @xmath14 in terms of the spin connection components @xmath32 and the extrinsic curvature . the form of the spin connection introduces a dependence of the framework on the particular model via the structure constants @xcite : @xmath33 note that in contrast to the full theory , the spin connection is a covariant object within a homogeneous model since coordinate transformations have to respect the symmetry . consequently , if non - zero , it can not be made small by choosing appropriate local coordinates . the only model which has identically vanishing spin connection is the bianchi i model with @xmath34 . otherwise , the spin connection is non - zero and can even depend on the triad ( it is a constant @xmath35 in the closed isotropic model , a submodel of bianchi ix , with @xmath36 and @xmath37 ) . the structure constants also appear explicitly in the hamiltonian constraint @xcite @xmath38a_1\right.\nonumber\\ & & + \left[(c_1\gamma_3+c_3\gamma_1-\gamma_1\gamma_3 ) ( 1+\gamma^{-2})- n^2c_2-\gamma^{-2}c_1c_3\right]a_2\nonumber\\ & & \left.+\left[(c_1\gamma_2+c_2\gamma_1-\gamma_1\gamma_2 ) ( 1+\gamma^{-2})- n^3c_3-\gamma^{-2}c_1c_2\right]a_3\right\}\\ & = & 2\kappa^{-1}\left[(\gamma_2\gamma_3-n^1\gamma_1)a_1 + ( \gamma_1\gamma_3-n^2\gamma_2)a_2 + ( \gamma_1\gamma_2-n^3\gamma_3)a_3\right.\nonumber\\ & & \left . -{{\textstyle \frac{1}{4}}}(a_1\dot{a}_2\dot{a}_3 + a_2\dot{a}_1\dot{a}_3 + a_3\dot{a}_1\dot{a}_2)\right]\ , . \label{hh}\end{aligned}\ ] ] in order to derive the classical field equations ( for which we consider only positive @xmath26 and @xmath19 ) it is advantageous to transform to new canonical variables @xmath39 where the prime denotes a derivative with respect to a new time coordinate @xmath40 related to @xmath41 by @xmath42 ( corresponding to the lapse function @xmath43 ) . with these new variables we have @xmath44 and @xmath45\\ & = & { { \textstyle \frac{1}{2}}}\left[(n^1)^2\left(\frac{p^2p^3}{p^1}\right)^2 + ( n^2)^2\left(\frac{p^1p^3}{p^2}\right)^2 + ( n^3)^2\left(\frac{p^1p^2}{p^3}\right)^2\right]\nonumber\\ & & - n^1n^2(p^3)^2- n^1n^3(p^2)^2- n^2n^3(p^1)^2 -{{\textstyle \frac{1}{2}}}(\pi_1\pi_2+\pi_1\pi_3+\pi_2\pi_3)\\ & = & { { \textstyle \frac{1}{2}}}\left[(n^1)^2a_1 ^ 4 + ( n^2)^2a_2 ^ 4 + ( n^3)^2a_3 ^ 4\right]- n^1n^2a_1 ^ 2a_2 ^ 2- n^1n^3a_1 ^ 2a_3 ^ 2- n^2n^3a_2 ^ 2a_3 ^ 2\nonumber\\ & & -{{\textstyle \frac{1}{2}}}(\pi_1\pi_2+\pi_1\pi_3+\pi_2\pi_3)\,.\end{aligned}\ ] ] now one can easily derive the equations of motion , @xmath46 which yield @xmath47 or , using @xmath48 @xmath49 for the bianchi i model we have @xmath34 so that @xmath50 . this implies the kasner behavior @xmath51 where @xmath52 and the constraint requires @xmath53 . the coefficients @xmath54 can be rescaled by choosing a different @xmath55 which can be fixed by requiring the conventional parameterization @xmath56 . as usual , these equations can be solved only if one coefficient , say @xmath57 , is negative and the other two are positive . correspondingly , one direction , the first , contracts whereas the other two expand toward larger time . when we approach the classical singularity at @xmath58 , space shrinks only in two directions while the third one expands unboundedly . the total volume however continues to approach zero according to @xmath59 . in @xcite it has been shown that the kasner singularity disappears when the model is quantized along the lines of loop quantum cosmology . for this result it was important that quantum geometry is based on _ densitized triad _ variables rather than co - triad variables . in those variables , all @xmath60 decrease to zero when one approaches the classical singularity . by contrast , for the minisuperspace described in terms of the scale factors or co - triad variables ( @xmath26 ) , the singularity is reached when at least one of them goes to infinity . we paraphrase this by saying that the classical singularity lies in the _ interior _ of the triad minisuperspace ( containing triads of both orientations and also degenerate ones ) . bianchi models other than bianchi i have non - vanishing structure constants and thus the evolution of the three triad components can be described as motion in a non - trivial potential given by @xmath61 which has infinite walls at small @xmath19 owing to the divergence of the spin connection components ( see fig . [ potclass ] ) . the evolution can then be described approximately as a succession of kasner epochs with intermediate reflections at the potential @xcite . for the bianchi ix model the reflections never stop and the classical evolution is believed to be chaotic @xcite . diagonal homogeneous loop quantum cosmology @xcite is first formulated in the connection representation where an orthonormal basis is given by the @xmath62-eigenstates @xmath63 with @xmath64 ( the full hilbert space of loop quantum cosmology is non - separable , but for our purposes it is sufficient to use the separable subspace spanned by the states ( [ n ] ) ; see @xcite for a detailed discussion . ) the eigenvalues of the triad operators can be read off from @xmath65 using the basic operators @xmath62 one can define the volume operator @xmath66 which will be used later . its eigenstates are also @xmath67 with eigenvalues @xmath68 a kinematical state @xmath69 is described in the triad representation by coefficients @xmath70 defined via , @xmath71 for a state to be gauge invariant under the residual gauge transformations , the coefficients @xmath72 have to satisfy @xmath73 these states are left invariant by the gauge invariant triad operators @xmath74 and the orientation operator @xmath75 . in calculations it is often easier to work with non - gauge invariant states in intermediate steps and project to gauge invariant ones in the end . together with the basic derivative operators @xmath62 we need multiplication operators which usually arise from ( point ) holonomies @xmath76 with action @xmath77 and correspondingly for @xmath78 and @xmath79 . from the basic operators we can build more complicated ones . we will later need a quantization of the spin connection which is a composite operator containing several triad operators . in particular , it also contains inverse powers of triad components which classically diverge at the singularity . since the triad operators have a discrete spectrum containing zero , they do not have an inverse . however , general methods of quantum geometry and loop quantum cosmology @xcite imply that there exist well - defined operators quantizing inverse triad components . to obtain these operators one makes use of a classical reformulation , e.g. @xmath80 which can then be quantized to @xmath81^{-2 } \left({\mathop{\mathrm{tr}}}\nolimits_j\lambda_1^i\tau_i h_1[h_1^{-1},\sqrt{|\hat{p}^1|}]\right)^2\,.\ ] ] here we have indicated that there are quantization ambiguities @xcite when one quantizes composite operators . the one , most relevant for this paper , is indicated by the subscript @xmath82 of the trace and corresponds to the choice of representation while writing holonomies as multiplicative operators . this operator acts as @xmath83 the discrete values @xmath84 _ decrease toward lower values _ for @xmath85 @xcite . thus , one can see that the classical divergence of the inverse of @xmath28 at vanishing @xmath19 is explicitly absent in the quantized operator . this will be seen to have further consequences for the approach to the classical singularity . furthermore , the state @xmath67 with @xmath86 ( on which the classical inverse triad would diverge ) is actually annihilated by @xmath87 due to @xmath88 . this allows us to define the inverse triad operator ( not just its absolute value ) by @xmath89 without ambiguity in defining the sign of zero . these inverse triad operators will be used later to find well - defined operators quantizing the spin connection . their eigenvalues are @xmath90 since the classical hamiltonian constraint contains the structure constants , it must have a model dependent quantization . the procedure in the full theory @xcite motivates to model the holonomy of a closed loop by using the combination @xmath91 in order to quantize connection components in the constraint @xcite . this works in fact for the bianchi i model , but it must be modified for models with non - vanishing intrinsic curvature . in these models a curve which is formed from four pieces of integral curves of vector fields @xmath8 generating the symmetry action , which would have holonomy @xmath91 , does not close to form a loop . one needs a correction which in @xcite has been done by adding a holonomy for the remaining curve multiplicatively . however this does not result in an admissible evolution for the closed isotropic model derived from bianchi ix @xcite ( an admissible quantization for this model has recently been presented in @xcite ) . we thus have to find a quantization which satisfies certain criteria for a well - defined evolution . to appreciate the criteria we recall that the general form of the hamiltonian constraint yields a _ difference _ equation for the wave function in the triad representation because of the action of holonomy operators @xcite . this fact is directly related to the discreteness of quantum geometry . however in the classical regime of large volume and small extrinsic curvature , the geometry is the familiar continuum ( metric ) geometry . thus the difference equation must admit solutions which can be very well approximated by solutions of _ differential _ equations , namely the wheeler dewitt equation which comes from the ( schrodinger ) quantization of continuum geometry . such solutions are slowly varying ( in the classical regime ) and can be interpolated by slowly varying solutions of the wheeler dewitt equation and are termed as _ pre - classical_. since quantum geometry depends on the barbero - immirzi parameter while the continuum geometry does not , absence of @xmath21 dependence in arriving at the differential approximation provides non - trivial constraints on the coefficients of the difference equation . admissibility of _ a continuum approximation _ is our first criterion . however , admissibility of a continuum approximation alone is not enough . typically the order of the difference equation is larger than that of the wheeler dewitt equation . consequently there are solutions of the difference equation which are not interpolable by solutions of differential equation . therefore , if in some local region in the classical regime we have a small " admixture of these extra solutions ( small so that the continuum approximation is valid ) , there is no guarantee that contribution of the extra solutions remains small under subsequent evolution , i.e. the continuum approximation may get invalidated . that this does _ not _ happen is our second criterion . it is thus necessary for the _ stability _ of the continuum approximation . a detailed formulation of these criteria is given in @xcite . as a first step , one divides the classical regime ( large eigenvalues of the triad operators ) into smaller cells ( local regions ) which are still large compared to the order of the difference equation . in each of these cells , the coefficients of the difference equations are approximately constant . this leads to approximating the difference equation by an equation with constant coefficients . in the case of ordinary difference equations that result in the isotropic models , the solutions are controlled by the roots of a polynomial . the admissibility of continuum approximation puts a set of conditions on these constant coefficients which translate into conditions on the roots . for the isotropic case in particular , the criterion requires existence of a root with value 1 and multiplicity 2 . the second criterion of local stability further requires that all roots must have absolute value less than or equal to 1 in order to prevent exponential growth of the non - pre - classical solutions . a further complex conjugation property of the coefficients of the difference equations then implies that all roots in fact must have absolute value 1 . the quantization of the hamiltonian constraint presented in the next section will be shown to satisfy both these criteria . as noted in the introduction , models with non - vanishing intrinsic curvature must have non - zero spin connection which , thanks to homogeneity , can not be made to vanish . therefore , the ashtekar connection , a linear combination of the spin connection and the extrinsic curvature , is not necessarily small even in a semiclassical regime corresponding to small extrinsic curvature . while the hamiltonian constraint in quantum geometry is formulated naturally in terms of the connection , for a semiclassical approximation it is desirable to perform an expansion in the small extrinsic curvature . this can only lead to the correct result if the spin connection also appears explicitly in the constraint . we will first propose the quantized constraint operator , show how it reproduces the classical expression and heuristically argue how it can admit a continuum approximation with local stability . the properties of the proposed operator will be elaborated in detail in the following subsections . the proposed hamiltonian constraint operator is : @xmath92 \right\}~ \right ] \label{hop}\end{aligned}\ ] ] where @xmath93 \label{gammaopr}\end{aligned}\ ] ] and the @xmath94 are defined in equation ( [ invtriad ] ) . since the inverse triad operators commute with the triad operators , there are no ordering ambiguities in the spin connection ( [ gammaopr ] ) . note that the holonomy operators @xmath95 appearing in the second set of braces do not need to contain the spin connection since it commutes with @xmath96 and would drop out . in ( [ hop ] ) the quantized spin connection appears explicitly both in the diagonal ` potential ' term and in holonomies @xmath97 . for vanishing @xmath32 the constraint operator reproduces the one of the bianchi - i model @xcite . it is easy to see that the term @xmath98 comes from the curvature 2-form of the spin connection @xmath99 for a bianchi model . in fact , an expression via the curvature components would be more general since it could also be used for kantowski sachs models . the holonomies @xmath97 are products of holonomies @xmath100 and @xmath101 . since we have to choose a quantization of the spin connection components @xmath32 , which contain inverse triads , and also an ordering of the non - commuting @xmath102- and @xmath99-holonomies , there are many more quantization ambiguities compared to the bianchi - i model . however , qualitative effects can often be seen to be independent of such ambiguities . to see that the proposed hamiltonian operator goes over to the classical expression in the classical limit , we can ignore the fact that connection components and the spin connection do not commute since this will only lead to effects of the order @xmath103 which disappear in the classical limit . the holonomy product @xmath104 can then be written formally as @xmath105 which for small @xmath106 yields @xmath107 on states ( [ n ] ) with large labels @xmath108 , the commutator containing the volume operator turns into @xmath109\sim -{{\textstyle \frac{1}{2}}}i\gamma{\ell_{\mathrm p}}^2 \lambda_ka_k\ ] ] so that the classical constraint in the form ( [ hh ] ) is reproduced . with regard to the continuum approximation and the local stability conditions , we have to take into account the fact that the connection does not commute with the spin connection because even small terms can have significant effects should they lead to instabilities . the products of holonomies then do not depend on the differences @xmath110 only , as could be expected if they were to commute . while ignoring the non - commutativity we have seen that expressions like @xmath111 appear which , however , are not well - defined in this form since there do not exist @xmath14-operators in quantum geometry ( only holonomies of the ashtekar connection are well - defined operators ) . motivated by the form of the holonomies in ( [ holonomyopr ] ) , @xmath112 we define @xmath113 and @xmath114 operators by @xmath115 with action @xmath116 and analogously for @xmath114-operators . this means that , by definition , the spin connection in trigonometric expressions of @xmath110 , always acts first . in the final expression we used eigenvalues @xmath117 of the spin connection operators , @xmath118 operators of the form ( [ sindiff ] ) however , do not produce the basic difference operator in the triad representation since they contain the spin connection eigenvalues as coefficients . nevertheless , for large @xmath108 we have in the triad representation for a state @xmath69 @xmath119 where @xmath120 i.e. , @xmath121 for large @xmath108 , therefore , the operators ( [ sindiff ] ) do act as basic difference operators on the wave function @xmath122 up to phase factors , and analogously operators @xmath123 act as basic mean operators . using the wave function @xmath122 instead of @xmath124 , the constraint operator ( [ hop ] ) contains , besides diagonal ( potential ) terms , a sum of products of a basic difference operator and a basic mean operator for two of the three discrete parameters @xmath108 . the resulting difference equation for @xmath125 will thus be of the same form as the difference equation of the bianchi - i model . consequently , the quantization ( [ hop ] ) can be expected to yield a locally stable evolution equation as in the case of bianchi - i . we will discuss this in detail in the following subsections . we now return to the constraint operator ( [ hop ] ) and obtain the difference equation in detail . from the definitions it follows that , @xmath126 & = & \left\ { \hat{v } - \cos({{\textstyle \frac{1}{2}}}c_k ) \hat{v } \cos({{\textstyle \frac{1}{2}}}c_k ) - \sin({{\textstyle \frac{1}{2}}}c_k ) \hat{v } \sin({{\textstyle \frac{1}{2}}}c_k ) \right\ } \nonumber \\ & & - 2 \lambda_k \left\ { \sin({{\textstyle \frac{1}{2}}}c_k ) \hat{v } \cos({{\textstyle \frac{1}{2}}}c_k ) - \cos({{\textstyle \frac{1}{2}}}c_k ) \hat{v } \sin({{\textstyle \frac{1}{2}}}c_k ) \right\}\end{aligned}\ ] ] note that the connection components @xmath14 do not have a ` hat ' on them as these do not exist as operators . their exponentials however are well defined , multiplicative operators . for notational convenience these ` hats ' have been suppressed . the ` sin ' , @xmath127 , has been defined _ without _ the factor of @xmath10 also for future notational convenience . then the trace in ( [ hop ] ) gives ( suppressing the hats on the operators ) : @xmath128 doing the sum over @xmath129 finally yields , @xmath130 the terms containing @xmath96 and @xmath131 are both diagonal when acting on the basis vectors @xmath132 . they give : @xmath133 using the notation of ( [ invopract ] ) . next , @xmath134 @xmath135 @xmath136 on the basis states can be expressed as : @xmath137 type of terms and two of @xmath138 types of terms . we will just have different factors of the @xmath139 . thus , for example , @xmath140 @xmath141 @xmath142 @xmath143 @xmath144 @xmath142 @xmath145 @xmath141 @xmath146 and @xmath147 @xmath144 @xmath148 it is convenient to define different basis vectors at this stage , namely , @xmath149 corresponding to the transformation ( [ stilde ] ) . this introduces additional phases which will turn out to be useful in our discussion of the continuum approximation . the transformation to the triad representation via , @xmath150 : @xmath151 where the coefficients are rather complicated expressions given by : @xmath152 @xmath153 and @xmath154 with obvious abbreviations , e.g. , @xmath155 , @xmath156 and + @xmath157 etc . , we have @xmath158 and @xmath159\nonumber \\ & & - 2 \kappa \gamma \ell_p^2 \left ( e^{\frac{i}{2 } \sum_i m_i \gamma_i(\vec{m } ) } \right ) \hat{h}_{m_1 , m_2 , m_3}^{\text{matter}}\end{aligned}\ ] ] the matter hamiltonian acting on the matter field dependence of the wave function has also been incorporated in the coefficient @xmath160 . noting that the coefficients @xmath161 involve @xmath162 and that the volume eigenvalues have a separable product structure , we can absorb the volume factors in the wave functions as follows . define , @xmath163 then , @xmath164 now defining @xmath165 and @xmath166 for @xmath167 , one sees that all volume eigenvalue factors are absorbed in the wave function . this preserves the gauge invariance condition on the @xmath168 but now in addition , due to the explicit volume eigenvalues , we also have that @xmath169 if any of the @xmath170 s equal zero . for bianchi i , the spin connection is zero . the phases @xmath171 vanish . the coefficient @xmath172 reduces to the matter term only , while @xmath173 reduces to : @xmath174 equation ( [ diffeqn ] ) then becomes , @xmath175 which matches with the equation derived in @xcite . we will first derive a simplified difference equation which approximates the exact one for large @xmath108 . then using an interpolating function , we will show that the approximate difference equation gives a differential equation for the interpolating function , which matches with the wheeler dewitt equation at the level of leading terms . this will verify that the continuum approximation is admissible . consider first @xmath176 . then , for @xmath177 , @xmath178 which is _ independent _ of @xmath179 to the leading order . this implies that @xmath180 the phases @xmath181 etc . all become equal to the common phase appearing in the coefficient @xmath172 and cancel out . the coefficients @xmath182 become exactly the same as in bianchi - i with @xmath183 and the equation becomes , @xmath184 to arrive at a differential equation , we look for a function @xmath185 such that @xmath186 satisfies the approximate difference equation . here , @xmath187 and thus @xmath188 is the change induced when the @xmath108 change by 1 . in particular , this implies : @xmath189 it follows that terms of order @xmath190 cancel out leaving a second order differential expression to the leading order . we also have the pre - factors @xmath191 which can be eliminated using @xmath192 . the spin connection then becomes just the classical expression without any factors of @xmath193 . _ all _ factors of @xmath21 cancel out . multiplying by @xmath194 the final equation becomes : @xmath195 ~ + ~ \mbox{cyclic}\ ] ] @xmath196 we have thus verified that the constraint operator ( [ hop ] ) , admits a continuum approximation . note that using an interpolating function , @xmath185 , one can always arrive at some differential expression . that _ all _ dependence on @xmath21 disappears from the differential equation is a non - trivial property of the locally approximated difference equation and is essential for matching with the wheeler dewitt equation which knows nothing about the barbero - immirzi parameter . had the phases not canceled , we would have acquired terms from the taylor expansion of the @xmath32 , which could have produced lower order differentials with a @xmath21-dependence left over , destroying the hope of matching with the wheeler dewitt equation . that the spin connection controls the phases in a cancelable manner is thus also a non - trivial property of the constraint operator . finally , the fact that we get a purely second order differential expressions is determined by the structure of the coefficients in the locally approximated difference equation @xcite . this is directly responsible for getting the wheeler - dewitt equation in a _ particular _ factor ordering appearing in ( [ wdw ] ) . the wave function @xmath197 has been obtained after some redefinitions . since @xmath198 in the pre - classical limit , @xmath197 is related to @xmath168 as @xmath199 . thus , the wheeler dewitt equation for our original pre - classical wave function @xmath168 is @xmath200 ~ + ~ \mbox{cyclic}\ ] ] @xmath201 note that the ordering of an isotropic sub - model is not determined uniquely by using @xmath202 because @xmath203 and @xmath204 , say , commute in the anisotropic model but not after using isotropy . however , there is also a unique ordering for isotropic models @xcite which can be derived _ after _ doing the reduction in loop quantum cosmology . our next requirement is about local stability of the evolution equation around pre - classical solutions . this is most directly formulated and illustrated in the context of an _ ordinary _ difference equation @xcite . the equation we have however is a _ partial _ difference equation . although , we do get a partial difference equation with constant coefficients in a local approximation , their solutions are not as easily analyzable in terms of roots of some polynomial . nevertheless , the property of local stability can still be analyzed . since we have three @xmath108 s , even if all are taken to be large , it is possible to have a subset of them to be much larger compared to the remaining ones . to begin with , we consider a neighborhood of a point where all @xmath108 are large and of the same order ; other regions can be explored similarly . for @xmath21 we assume the value derived from black hole entropy calculations which is of order @xmath205 . since all @xmath108 s are large , it follows exactly as before , that @xmath32 are given by the equation ( [ gamasy ] ) and all phases cancel out giving the locally approximated difference equation ( [ asy ] ) . furthermore , since we assume the @xmath108 to be about the same size , the @xmath32 are approximately equal and taking the bianchi ix model for definiteness , @xmath206 . we obtain @xmath207 in this equation , @xmath208 and @xmath209 all play the same role , but without loss of generality we can select @xmath209 to be the evolution parameter in terms of which we analyze local stability . to express it clearly , let us introduce the notation @xmath210 which allows us to write the equation as @xmath211 where @xmath212 is a small constant ( in the local approximation ) characterizing the size of the matter contribution . for @xmath108 s of the same order @xmath213 , the equation can be further simplified as @xmath214 we are interested in checking if perturbation of slowly varying ( along all three directions ) solutions , continue to remain as perturbations under the @xmath209-evolution . to this end , we assume that we have some fixed solution which is slowly varying . then the first bracket on the right hand side is small while the second one is suppressed by @xmath215 ( @xmath216 is small even for large @xmath213 because the energy density decreases at least as @xmath217 ) . this is a linear equation among four combinations analogous to the equation of a plane in @xmath218 passing close to the origin . it can be solved approximately by introducing three , fixed orthonormal vectors , @xmath219 , each of which is orthogonal to the vector @xmath220 in @xmath218 . explicitly , @xmath221 in the above , corrections of the order @xmath222 as well as of the order @xmath223 are ignored and @xmath224 are three functions of @xmath209 . the @xmath224 can be assumed to be small for @xmath225 to represent a solution close to a pre - classical one . thus we obtain ordinary , _ non - homogeneous _ difference equations with constant coefficients for the four @xmath226 . all the equations have the same structure , and their solutions consist of a combination of the particular solution and linear combinations of solutions of the corresponding homogeneous equations . the _ difference _ between a perturbed solution and the fixed solution of course satisfies the _ homogeneous _ equations . the local stability now requires that these differences do not grow exponentially . equivalently , the characteristic roots for the homogeneous equations must have absolute value less than or equal to one . in our case , the characteristic polynomials are the same for all four equations namely , @xmath227 , with roots equal to @xmath228 . these obviously satisfy the condition of local stability . because of local stability , perturbations of slowly varying solutions continue to remain close to them under evolution along the @xmath209 direction . this also ensures _ self - consistency _ when we consider evolutions along the @xmath208 directions in a similar manner by keeping the corresponding right hand sides close to zero . had we not neglected the terms on the right hand side of eq . ( [ approxeqn ] ) , we would have got an extra , common non - homogeneous term which is 1/2 of the right hand side of eq.([approxeqn ] ) . clearly , this does not affect the behavior of the differences . to summarize , although we got a partial difference equation , we were able to implement the idea behind the local stability criterion by obtaining a system of ordinary difference equations . that these equations are non - homogeneous , does not affect the stability properties @xcite . we have now verified that the proposed quantization ( [ hop ] ) has a valid continuum approximation and is locally stable around pre - classical solutions . thus , it is possible to start with initial data at large volume which are close to a continuum solution and to follow the evolution to smaller volume , toward the classical singularity . in this regime , quantum effects of the discreteness and also ordering issues become more and more important and the full equation has to be considered without approximations . fortunately , it is possible to see essential features of the absence of singularities quite generally in a way similar to the isotropic case @xcite . since the constraint equation is used as a recurrence relation for the wave function , the term with lowest order in the labels @xmath108 should have non - vanishing coefficient , except when one of the labels vanishes such that the corresponding component of the wave function , i.e. its value at singular surfaces in minisuperspace , will drop out ( see @xcite for details ) . it can readily be verified that this general behavior is true for the evolution equation of an arbitrary bianchi class a model . the difference equation resulting from ( [ hop ] ) is more complicated than in the isotropic @xcite or bianchi i @xcite case ; but since the exponentiated spin connection never vanishes and the additional potential term containing the spin connection is diagonal , only the commutator of holonomies with the volume operator determines whether or not a lowest order coefficient vanishes . the structure resulting from this term is the same as in the bianchi i case , and thus the evolution is singularity - free in the same way . this shows that the wave function can be extended uniquely from the region of minisuperspace with positive orientation to the region with negative orientation and that the orientation again provides us with a branch ` before ' the classical singularity . for bianchi models other than bianchi i , however , the _ classical approach to the singularity _ is more complicated than the simpler kasner behavior . this is most dramatic in the bianchi ix model where different kasner epochs succeed each other infinitely many times . it turns out , as we will discuss in the next section , that the classical approach to singularities is modified at small scales by quantum geometry effects which is expected to lead to a simpler picture . even after imposing the condition of local stability the constraint expression is far from being unique . there may be other choices for the kinetic term which lead to a locally stable behavior , but they would have to be very special and difficult to find . the advantage of the quantization presented here is that it is rather simple and provides a general formula for all diagonal bianchi class a models at once without having to find an expression in each case separately . as for factor ordering ambiguities , the main issue is that of ordering the commutator to the right or to the left . this ordering is the same here as in the bianchi i case and also in the full theory . thus , the special issues one has to deal with in models with non - vanishing intrinsic curvature do not change this result . since the non - singular evolution depends significantly on this kind of ordering , it is robust . the main purpose of this paper is to prove the existence of one quantization for all models in our class which has the correct semiclassical limit , results in a locally stable evolution and is also non - singular in the quantum regime . that such a general quantization fulfilling all the different requirements exists is a non - trivial fact and provides a further check of the methods employed in the full theory . the second issue , besides the singularity - free evolution , which is robust under the choices we have is the semiclassical behavior and corrections to it . the main effects in this context are insensitive to factor ordering and other choices . as we will discuss in the next section , there are non - perturbative effects coming from the potential term which lead to drastic differences to the classical behavior . since they come from the potential term and are based on general features of quantizations of inverse triad components , they have to be present in any quantization . the quantization ( [ hop ] ) of the hamiltonian constraint yields consistent quantizations of the dynamics of bianchi class a models which in the limit of large triad components reproduce the correct classical behavior . at finite values of @xmath19 , however , there are always correction terms to the classical equations of motion ( [ motion ] ) which can change the classical evolution . it is possible to derive different types of corrections and to include them in effective classical equations of motion . there are the following sources of correction terms : first , we can keep more terms beyond the leading order in a taylor expansion of difference operators which results in a higher order wheeler dewitt equation . these correction terms would correspond to higher curvature terms in an effective action . the second source is the non - commutativity of the @xmath102-holonomy @xmath100 and the @xmath99-holonomy @xmath229 in ( [ hop ] ) . since we have , for example , @xmath230 which in a quantization would appear as the diagonal operator @xmath231 \ , \ ] ] the coefficients of the wheeler dewitt equation receive corrections of the order @xmath232 . thirdly , we have to use well - defined inverse metric operators when we quantize the spin connection components in ( [ hop ] ) which leads to a modified behavior at small scales ( as in the isotropic case where it implies inflation @xcite ) . since the last effect is non - perturbative while the first two are perturbative ( in @xmath21 ) , it can be expected to have the most dramatic consequences . therefore , we will focus on this type of correction in our discussion of a modified approach to classical singularities . as in @xcite , we exploit the fact that there are quantization ambiguities @xcite which can extend the range of a modified classical behavior @xcite . if we choose the ambiguity parameter large enough it is reasonable to use effective classical equations of motion to describe the trajectory of a quantum wave packet while still experiencing the modified behavior at small scales . in particular , the spin connection components will always decrease when we approach the classical singularity close enough whereas they always have some diverging components in the purely classical scenario . the diverging components lead to infinitely high walls in the potential ( [ potential ] ) on minisuperspace and to an infinite number of reflections in the bianchi ix model leading to chaotic behavior @xcite . with the modified spin connection , the walls do not grow arbitrarily high but instead shrink once a small triad component is reached . consequently , one expects that the reflections will cease after a finite number of times and the wave packet will simply move through the classical singularity in a final kasner epoch . here , we will make this explicit by deriving the potential and its changes due to the modified spin connection . a detailed analysis of the modified approach to the singularity will be presented elsewhere @xcite . ( solid line ) compared to its large-@xmath235 limit @xmath236 ( dashed ) . [ f],width=453,height=302 ] limiting behaviors for the eigenvalues of @xmath237 are @xmath238 here @xmath239 are the eigenvalues of @xmath62 . the eigenvalues of the spin connection operators are obtained as , @xmath240 \\ & & \nonumber \\ & \sim & \frac{1}{2 } \left [ n^j \frac{p^k(m_k)}{p^j(m_j ) } + n^k \frac{p^j(m_j)}{p^k(m_k)}- n^i \frac{p^j(m_j ) p^k ( m_k)}{(p^i)^2(m_i ) } \right ] \mbox{\hspace{1.3 cm } $ \left(\frac{|p^i|}{\gamma { \ell_{\mathrm p}}^2 j } \gg 1 \right)$ } \\ & & \nonumber \\ & \sim & \frac{1}{4 j^3 } \left [ \ , n^j m_k m_j^2 { \mathop{\mathrm{sgn}}}(m_j ) + n^k m_j m_k^2 { \mathop{\mathrm{sgn}}}(m_k ) - n^i \frac{m_j m_k m_i^4}{2 j^3 } \ , \right ] \nonumber \\ & \approx & 2 \left(\gamma \ell_p^2 j\right)^{-3 } p^j(m_j ) p^k(m_k ) \left [ ~ n^j |p^j(m_j)| + n^k |p^k(m_k)| ~ \right ] \mbox{\hspace{0.4 cm } $ \left(\frac{|p^i|}{\gamma { \ell_{\mathrm p}}^2 j } \ll 1 \right)$ } \end{aligned}\ ] ] in the last equation we have dropped the sub - leading third term . evidently , the eigenvalues of the spin connection operators approach their classical expressions for large values of the triad components while for small values the quantum modification shows up . to find the potential in this regime , we must use the first line of eqn . ( [ potential ] ) since the second follows by using the classical expression for @xmath32 . for @xmath241 this leads to , @xmath242 ~ + ~ \mbox{cyclic } \ .\end{aligned}\ ] ] dropping the @xmath243 term compared to the @xmath244 , for @xmath245 , we get @xmath246\end{aligned}\ ] ] by contrast , for @xmath247 , the potential takes the form , @xmath248\end{aligned}\ ] ] specializing to bianchi - ix this is : @xmath249 & \left(\frac{|p^i|}{\gamma \ell_p^2 j } \ll 1 \right ) \nonumber \\ & \\ \frac{1}{2 } \left [ \left(\frac{p^1 p^2}{p^3}\right)^2 + \left(\frac{p^2 p^3}{p^1}\right)^2 + \left(\frac{p^3 p^1}{p^2}\right)^2 - 2 \left\ { ( p^1)^2 + ( p^2)^2 + ( p^3)^2 \right\ } \right ] & \left(\frac{|p^i|}{\gamma \ell_p^2 j } \gg 1 \right ) \nonumber \\ \end{array } \right.\ ] ] a sample snapshot of the modified potential is shown in figs . [ pot ] and [ pot4 ] together with the classical potential ( fig . [ potclass ] ) for comparison . and @xmath250 with @xmath251 . [ potclass],width=453,height=302 ] and @xmath252 ( with @xmath253 ) . [ pot],width=453,height=302 ] and @xmath252 ( diagonal canyons ) with @xmath253 . also regions for negative @xmath19 are shown to which a test particle in the potential can move ; with the original potential only the lower quadrant is allowed . [ pot4],width=453,height=302 ] with the modified potential , the classical hamiltonian ( [ classham ] ) becomes @xmath254 we have thus derived an _ effective potential _ modified by the _ non - perturbative _ quantum effects . other correction terms , such as higher order contributions to the kinetic part coming from the discreteness , are perturbative and less important as remarked earlier . note that classical regime is identified by @xmath255 . the quantum modifications however are dominant in the _ small @xmath256 _ regimes . choosing larger values of the ambiguity parameter @xmath82 , thus allows us to access the quantum effects even while staying in the classical regime . we can thus study the qualitative effects of quantum modifications within a classical framework by using the effective potential instead of using a fully quantum description in terms of a wave function on minisuperspace . effective classical equations of motion , in terms of an external time parameter , can be derived from the hamiltonian ( [ heff ] ) as we did in section [ modclass ] . the corresponding classical trajectories can be thought of as ( approximately ) describing the motion of a quantum wave packet , giving suggestions regarding the full quantum behavior . alternatively , one can also analyze the modified behavior in terms of an internal time namely the volume . the hamiltonian ( [ heff ] ) is in fact hyperbolic . the potential now has a non - factorizable time dependence and the analysis of motion is more complicated . figs . [ potclassv ] and [ potva ] show the classical and the modified potential , respectively , at fixed volume @xmath257 rather than at fixed @xmath258 as in figs . [ potclass ] and [ pot ] . one can already expect from the figures that the behavior will change dramatically once the universe reaches a small triad component . for a detailed analysis see @xcite . , @xmath259 , and @xmath260 ( from left to right ) . while moving in the triangular valley the universe is pushed toward the classical singularity ( @xmath261 ) by reflections on the moving left wall . [ potclassv],title="fig:",width=188,height=151 ] , @xmath259 , and @xmath260 ( from left to right ) . while moving in the triangular valley the universe is pushed toward the classical singularity ( @xmath261 ) by reflections on the moving left wall . [ potclassv],title="fig:",width=188,height=151 ] , @xmath259 , and @xmath260 ( from left to right ) . while moving in the triangular valley the universe is pushed toward the classical singularity ( @xmath261 ) by reflections on the moving left wall . [ potclassv],title="fig:",width=188,height=151 ] ) at constant volume @xmath262 , @xmath259 , and @xmath260 ( from left to right ) . [ potva],title="fig:",width=188,height=151 ] ) at constant volume @xmath262 , @xmath259 , and @xmath260 ( from left to right ) . [ potva],title="fig:",width=188,height=151 ] ) at constant volume @xmath262 , @xmath259 , and @xmath260 ( from left to right ) . [ potva],title="fig:",width=188,height=151 ] in this paper we have extended the methods of homogeneous loop quantum cosmology to models with non - zero spin connection . these models have non - trivial intrinsic curvature from the symmetric background which requires a special treatment compared to the bianchi i model as well as compared to the full theory . initially , therefore , part of the guidance from the full theory is lost and there are more ambiguities when one quantizes the hamiltonian constraint . in @xcite two conditions which have to be imposed for a reasonable quantization , have been introduced and discussed in detail . these stem primarily from the requirement that the loop quantization admit physical states which , in a semiclassical regime , have much less sensitivity to variations on the planck scale so that the idealization of the continuum geometric formulation is well justified . this has been made more explicit by the formulation of the continuum approximation . in section [ hami ] we introduced a quantization of the hamiltonian constraint for all diagonalized bianchi class a models , and showed that the two conditions are satisfied . it was important to take the freedom of phases in the triad representation into account which we did by transforming to a new wave function @xmath122 in ( [ stilde ] ) . the transformation was dictated by the requirement of reproducing the wheeler dewitt equation as the leading approximation . it is satisfying to note that the same transformation also appeared in @xcite where it was used in the context of relating the wave functions obtained in the schrdinger quantization based on connection variables and the wave functions of the wheeler dewitt quantization based on the metric variables . since the two sets of basic variables differ significantly only when the spin connection is non - zero which happens when the intrinsic curvature is non - zero , non - trivial phases show up in this situation . the same transformation then appears whether one uses a loop quantization or a schrdinger quantization in the connection formulation and compares with the wheeler dewitt quantization . we also showed that the mechanism for the absence of singularities continues to hold , despite the phases , exactly as in the case of the bianchi - i model @xcite . thus , as the main result of this paper we proved that for all diagonal bianchi class a models there is a loop quantization which satisfies all the conditions for a good continuum limit as well as that of a singularity - free evolution . that such a quantization exists at all , and even one which provides a general form for all the models , is a non - trivial fact and gives further credence to the viability of loop methods and the physical applications obtained so far . in section [ bianchiix ] we carried out a preliminary exploration of the consequences of the quantized spin connection and the correspondingly modified potential . we focused on the bianchi ix case and pointed out that unlike the classical potential which has infinite walls as the singularity is approached , the quantum modified potential has finite walls . since the infinite walls are responsible for the chaotic approach to the singularity , we expect that the finite walls of the modified potential will significantly alter the approach to the classical singularity @xcite . finally a remark about the relevance of the results is in order . since the connection of the quantization discussed here to the full theory is weaker , it is legitimate to question what this quantization may have to do with the full theory . firstly , we observe that the methods adopted as well as the two admissibility criteria work uniformly for all homogeneous models ( including isotropic sub - models ) which do not have a non - zero spin connection . the direct loss of connection to the full theory is only in the models with spin connection . the treatment of the spin connection enters in two distinct places , firstly in using the inverses of triad components and secondly in the modification of holonomies . the first one is quite natural and inherits the ambiguities in the definition of inverses @xcite . the second one has additional ambiguities due to factor ordering . it is already clear that the first type of ambiguity is _ not _ fixed by the two admissibility criteria . our emphasis in this work has been to demonstrate the existence of _ a uniform method _ of quantization which satisfies the two admissibility criteria _ and _ is singularity free . the uniqueness type of issues , while important , requires a systematic classification of possible ambiguities and is beyond the scope of the present work . likewise the general question of how many of the results of a minisuperspace quantization can survive the full theory is also beyond the scope of this paper . while the spin connection is seen to require a careful treatment of the continuum approximation and also to lead to a modified approach to the bkl singularity in the context of homogeneous models , its role in the full theory is more complicated and is an open issue . for instance , the phase factor in @xmath122 which has the form , @xmath263 is not even well - defined in the full theory due to the transformation properties of the spin connection ( which is no longer a covariant object ) . since the spin connection can be made arbitrarily small locally , by choosing appropriate coordinates , one can expect that the diffeomorphism constraint and its algebra with the hamiltonian constraint will play a role in this issue . requirements such as the local stability condition can also be expected to play a role in analyzing this issue , at least with a local version in the absence of a global internal time . m. b. and g. d. are grateful to madhavan varadarajan and naresh dadhich for invitations to and hospitality at the rri , bangalore , and iucaa , pune , where part of the work has been done . m. b. also thanks the imsc , chennai , for hospitality . the work of m. b. and k. v. was supported in part by nsf grant phy00 - 90091 and the eberly research funds of penn state .

we perform an extensive analysis of neutrino diffusion in quark matter within the mit bag model at arbitrary temperature and degeneracy . we examine in detail the contribution of each of the relevant weak interaction processes to the total neutrino opacity and evaluate the effect of the strange quark mass , the bag constant , and the qcd perturbative corrections to the mit bag model . we also investigate the anisotropic contribution to the neutrino distribution function in scatterings , the mean energy transfer and the mean scattering angle . the density and temperature dependence of the diffusion coefficients that govern the cooling and deleptonization of a compact star is shown in detail . finally , our numerical results for the neutrino mean free paths are compared against known analytic approximations . we conclude that neutrino scattering constitutes a significant portion of the total neutrino opacity in leptonized quark matter and neutrino - quark scattering is , in general , very similar to neutrino - electron scattering with respect to both mean energy transfer per scattering and mean scattering angle .

1405.3294

compact stellar objects containing deconfined quark matter have been envisaged since long ago @xcite . according to theoretical studies such objects could be formed if the density inside a purely hadronic star goes beyond a critical density @xcite . this may happen due to accretion onto a cold hadronic star in a binary system , as a consequence of cooling , deleptonization and fallback accretion during the protoneutron star phase of a just born compact star , due to spin down of a fast rotating star or by more exotic mechanisms such as strangelet contamination @xcite . the conversion of the star presumably begins through the formation of a small quark matter seed that grows at the expenses of the gravitational energy extracted from the contraction of the object and/or through strongly exothermic combustion processes @xcite . all these scenarios lead to the formation of hot and neutrino rich quark matter occupying the whole compact object ( strange star ) or its core ( hybrid star ) . a full comprehension of the above mechanisms requires a detailed knowledge of the neutrino transport in quark matter . in fact , neutrino interactions in neutron star matter have been studied in detail for a long time now . most of the early contributions were given in the form of approximate treatments focused on a particular state of neutrino degeneracy : neutrinos were considered to be either nondegenerate or highly degenerate . from these works , those which are particularly important for the present analysis have been summarized by iwamoto @xcite , with emphasis in neutrino interactions in quark matter . extensive numeric analyses for arbitrary temperatures , densities and neutrino level of degeneracy , however , are not so abundant . in this case , we may cite the works of reddy & prakash @xcite and reddy , prakash & lattimer @xcite , where neutrino interactions in hadronic neutron star matter with the possible presence of hyperons have been studied . works focused on neutrino interactions in quark matter at arbitrary conditions are even rarer and the reference for this topic is usually the paper of steiner , prakash & lattimer @xcite . in that work , the authors analyse the neutrino diffusion coefficients in the interior of a hybrid star containing a mixed quark - hadron phase and quark matter was described using the mit bag model with a fixed bag constant @xmath1mev/@xmath2 . in the present work , we intend to fill the gap in the literature represented by the lack of studies exploring general aspects of neutrino interactions and diffusive neutrino transport in quark matter in general conditions . we describe quark matter using the mit bag model and analyze the influence of its three possible free parameters @xmath3 the bag constant @xmath4 , the mass of the strange quark @xmath5 and the strong coupling constant @xmath6 @xmath3 on the neutrino mean free paths and energy - averaged diffusion coefficients . we pay particular attention to the relative contributions of absorption and scattering to the total neutrino opacity and how neutrino - quark scattering compares to neutrino - electron scattering . having time - consuming simulations in mind , we explore different expressions for the neutrino scattering opacity and the possible penalties in precision associated to more simplified treatments . we also explore here the actual range of validity of the known analytic approximate forms for the neutrino mean free paths . neutrino transport is described in terms of the boltzmann transport equation for massless particles , which dictates the kinetic evolution of its invariant distribution function @xmath7 : @xmath8 where , in the general relativistic case , @xmath9 denotes the operator @xmath10 , with @xmath11 being the christoffel symbols . the right - hand side of eq . ( [ eq : bte_general ] ) designates the changes in @xmath12 due to particle interaction processes ( collisions " ) and it can be separated into different contributions @xmath13 + b_{\mbox{\scriptsize{s}}}[f]+\cdots,\end{aligned}\ ] ] where ae stands for absorption / emission of neutrinos and s designates all forms of scattering . here , we will focus only on these two kind of processes , since those are the dominant ones on neutron star matter at the typical values of temperature and lepton degeneracy we are interested in . for processes of the form @xmath14 , each contribution to eq . ( [ eq : colint_b ] ) can be explicitly written in the generic collision integral " @xmath15=-\int\frac{d^3p_2}{(2\pi)^3}\int\frac{d^3p_3}{(2\pi)^3 } \int\frac{d^3p_4}{(2\pi)^3}\big\{\nonumber\\ & \times g_\nu g_2f(e_\nu)f_2(e_2)[1-f_3(e_3)][1-f_4(e_4)]w_{fi}\nonumber\\ & -g_3g_4[1-f(e_\nu)][1-f_2(e_2)]f_3(e_3)f_4(e_4)w_{if}\big\}\end{aligned}\ ] ] where @xmath16 denotes the distribution function for particle species @xmath17 , @xmath12 is the complete neutrino distribution function and the @xmath18 denote the phase - space spin degeneracy for each particle . here , @xmath19 represents the transition rate between the initial and the final states for the given process . we can define the neutrino emissivity @xmath20 and the absorptivity @xmath21 through @xmath22&=\left[1-f(e_\nu)\right]j_a-\frac{f(e_\nu)}{\lambda_a},\end{aligned}\ ] ] and the scattering contribution to the collision integral can be written as @xmath23=&\left[1-f(e_\nu)\right]\int\frac{d^3p_4}{(2\pi)^3}f(e_\nu^\prime ) r^{in}\nonumber\\ & -f(e_\nu)\int\frac{d^3p_4}{(2\pi)^3}\left[1-f(e_\nu^\prime)\right]r^{out},\end{aligned}\ ] ] where @xmath24 are the scattering kernels , @xmath25 being the scattering angle . when the pairs of reactions of emission / absorption and scattering in / out are balanced , the transition rates satisfy the reciprocity relations @xmath26 and we have the relations of detailed balance @xmath27 _ i.e. _ , eqs . ( [ eq : bae ] ) and ( [ eq : bscatt ] ) have only one independent contribution each . the _ diffusive approximation _ consists of assuming that matter is at thermodynamic equilibrium and neutrinos are only slightly out of equilibrium . since the equilibrium states of our mit bag model are those of a gas non - interacting fermions , we automatically set each @xmath16 in eq . ( [ eq : collisionint ] ) to the corresponding fermi - dirac distribution function , @xmath28 . on this regime , neutrinos are expected to _ relax _ in a very short time scale and their distribution function differs from its equilibrium value only by a small anisotropic factor . in terms of a legendre expansion of @xmath12 , the diffusion approximation can be written as @xmath29 where @xmath30 is the cosine of the angle between the neutrino propagation direction and the radial vector , and @xmath31 . all the quantities we are interested in in the study of neutrino transport will be defined in terms of @xmath12 and its angular moments of the form @xmath32 @xcite . with the use of the approximation given by eq . ( [ eq : diffusiveapprox ] ) , it is possible to write the angular moments of eq . ( [ eq : colint_b ] ) in a very convenient form . the @xmath33-th moment will be associated with neutrino source terms @xcite and , since we are assuming the detailed balance of eqs . ( [ eq : detbalance ] ) , it is identically zero . the @xmath34st moment , on the other hand , contains the traditional _ opacities _ and it is given by @xmath35d\mu&=-\frac{f_1(e_\nu)}{3}\left(j_a+\frac{1}{\lambda_a } + \kappa_1\right),\end{aligned}\ ] ] where @xmath36}{\left[1-f_0(e_\nu)\right]}r^{out}_0\right.\nonumber\\ & \left.-\frac{f_0(e_\nu)}{f_1(e_\nu)}\frac{f_1(e_\nu^\prime)}{f_0(e_\nu^\prime ) } r_1^{out}\right\}.\end{aligned}\ ] ] we have defined here the legendre moments of the scattering kernel in terms of the scattering angle @xmath25 @xmath37 with @xmath38 , where @xmath39 is the azimutal angle of one neutrino with respect to each other . the first term between the curly braces in eq . ( [ eq : kappa_0 ] ) gives the inverse neutrino scattering mean free path when neutrinos are in thermodynamic equilibrium with the rest of the matter , and we denote it here by @xmath40 : @xmath41^{-1}}{(2\pi)^2}\int_0^\infty de_\nu^\prime e_\nu^{\prime 2}\left[1-f_0(e_\nu^\prime)\right]\nonumber\\ & \times\int_{-1}^1 d \cos\theta r^{out}(e_\nu , e_\nu^\prime,\cos\theta).\end{aligned}\ ] ] eq . ( [ eq : lambda_scatt ] ) is just eq . ( [ eq : collisionint ] ) with @xmath28 for the neutrino distribution function and it can be written without one explicitly defining the scattering kernel @xmath42 . when the scattering process is isoenergetic ( meaning that neutrinos and matter do not exchange energy ) , regardless the actual neutrino distribution function , eq . ( [ eq : kappa_0 ] ) specializes to @xmath43 which is a common form of scattering opacity used in different applications . in a general situation , however , @xmath44 depends explicitly on @xmath45 , which is , _ a priori _ unknown . from the energy - dependent diffusion coefficient @xmath46 , we define the energy - averaged coefficients @xcite @xmath47d(e_\nu),\end{aligned}\ ] ] where @xmath48 . in particular , the rosseland energy - averaged neutrino mean free path @xcite can be defined as @xmath49}.\end{aligned}\ ] ] all neutrino processes under consideration are listed on table [ table : cconsts ] . electron neutrino absorption on quarks @xmath50 and @xmath51 involve the exchange of a @xmath52 boson , while neutrino scattering on quarks or on leptons may involve either the exchange of @xmath52 or @xmath53 bosons . at the energy density regime we are interested in , the neutrino energy is always much smaller than the @xmath52 and @xmath53 masses and the interactions may be described in terms of current - current couplings @xcite . weak charged currents are associated to both neutrino absorption and neutrino scattering on leptons of the same generation , while weak neutral currents take part in neutrino scattering on either leptons or quarks . given that it is possible to write the charged current contribution of a scattering process in terms of a neutral current ( and vice - versa ) , we may write the current - current interaction lagrangian in a combined form as @xmath54 \nonumber\\ & \times\left[\bar{u}(2)\gamma^\mu(c_a - c_v\gamma_5)u(3)\right ] + \mbox{h.c.},\end{aligned}\ ] ] with @xmath55 and @xmath56 being the appropriate axial and vector coupling constants to be read from table [ table : cconsts ] and @xmath57 is the fermi weak coupling constant ( @xmath58gev@xmath59 ) . the transition rates @xmath60 appearing in eq . ( [ eq : collisionint ] ) are obtained from the matrix element by means of femi s golden rule : @xmath61 where @xmath62 denotes the squared matrix element summed over final spins and averaged over the initial spins . from eq . ( [ eq : lint ] ) we may then derive an expression for @xmath60 also valid for both absorption and scattering processes : @xmath63\nonumber\\ & \times(2\pi)^4\delta^4\left(p_\nu+p_2-p_3-p_4\right),\end{aligned}\ ] ] where @xmath64 is the mass of the corresponding particle . the explicit form of the expressions used for the inverse mean free paths and for the scattering kernels can be found on appendices [ sec : ap_mfp ] and [ sec : scatt_kernel ] , respectively . l @ cc + @xmath14 & @xmath65 & @xmath66 + + @xmath67 & @xmath68 & @xmath68 + @xmath69 & @xmath70 & @xmath70 + + @xmath71 & @xmath68 & @xmath72 + @xmath73 & @xmath70 & @xmath74 + + @xmath75 & @xmath76 & @xmath77 + @xmath78 & @xmath79 & @xmath80 + @xmath81 & @xmath79 & @xmath80 + @xmath82&@xmath83 & @xmath77 + @xmath84 & @xmath85 & @xmath80 + + we describe quark matter constituted of _ u _ , _ d _ and _ s _ quarks , plus electrons and electron - neutrinos in terms of the mit bag model @xcite . the thermodynamics follows from the grand potential ( per unit volume ) @xmath86 where @xmath87 is the thermodynamic potential for a gas of ideal fermions @xmath88.\end{aligned}\ ] ] antiparticles are included through @xmath89 and all derived thermodynamic quantities must be understood as net quantities , containing the contributions of both particles and antiparticles . the inclusion of the vacuum energy density @xmath4 ( the bag constant ) , requires that both the pressure and energy density are modified according to @xmath90 where the entropy density , @xmath91 , and the particle number densities , @xmath92 , are derived from @xmath93 by the usual thermodynamic relations . the five chemical potentials ( @xmath94 ) at given temperature , baryon number density , @xmath95 , and fixed net electron - type lepton fraction , @xmath96 , are determined by means of the conditions of @xmath97-equilibrium @xmath98 and electric charge neutrality @xmath99 when quark matter does nt have trapped neutrinos , or when the trapped neutrinos are nondegenerate , the electron chemical potential is always much smaller than those of quarks . quark matter in these circumstances requires only a small fraction of electrons to be electrically neutral . when the rate of occurrence of the weak interaction processes is high enough for neutrinos to become trapped and degenerate in quark matter , an increase in the abundance of electrons follows , as it can be seen from the behavior of the electron chemical potential in fig . [ fig:1 ] . in fact , as shown in fig . [ fig:1 ] , the degeneracy of electrons in quark matter follows closely that of neutrinos . this moderate uniformity between the electron and electron - neutrino chemical potentials , allows us to use the total electron - type lepton fraction @xmath100 as an indicative of both the electron and electron - neutrino state of degeneracy , as we show in fig . [ fig:2 ] . equilibrium chemical potentials of quark matter constituted only of the three lightest quarks at @xmath101mev as a function of the baryon number density . the panel at the left corresponds to neutrino - free quark matter . the panel at the right represents quark matter with trapped neutrinos , corresponding to a total electron - type lepton fraction of @xmath102 . for this figure , @xmath103mev/@xmath2 and @xmath104mev . ] electron and electron - neutrino equilibrium chemical potentials in quark matter with trapped neutrinos at @xmath105 for two different temperatures . the smallest value of @xmath100 shown in this figure always represent neutrino - free quark matter , for which @xmath106 . the largest value of @xmath100 has been chosen to be @xmath102 in both panels . note how an increasing neutrino degeneracy is followed by an increasing electron degeneracy for fixed temperature and baryon density . for this figure , @xmath103mev/@xmath2 and @xmath104mev . ] qcd corrections to the thermodynamic potential of eq . ( [ eq : bag_omega ] ) for arbitrary temperatures , quark masses and chemical potentials have been derived for the orders of @xmath6 and @xmath107 in perturbation theory @xcite . closed - form expressions are known only on the approximate regimes . for degenerate massless quarks , one has , to first order in @xmath108 @xcite @xmath109,\end{aligned}\ ] ] and the thermodynamic potential to second order is obtained through @xmath110 . in the context of the mit bag model , the strong coupling constant , @xmath6 , is regarded as a numeric constant . together , @xmath4 , @xmath5 and @xmath6 constitute the three free parameters of the model . throughout this paper we use mainly the set of parameters @xmath103mev/@xmath2 , @xmath104 mev and @xmath111 , which falls inside the stability window of ref . @xcite and , consequently , describes absolutely stable ( strange ) quark matter , with an energy per baryon at zero pressure and temperature smaller than the neutron s mass . however , we employ also other parameter sets , some of which represent standard ( non - absolutely stable ) quark matter . in particular , at the end of sec . [ subsec : mfpana ] , we show how each of @xmath4 , @xmath5 and @xmath6 affect the neutrino mean free paths through their influence on the equilibrium chemical potentials . scattering contribution to the total rosseland mean free path as a function of baryon density at several temperatures for both nondegenerate and degenerate neutrinos . all electron - neutrino processes of table [ table : cconsts ] have been included in the total opacity . ] influence of the bag constant , @xmath4 , on the total rosseland neutrino mean free path of degenerate neutrinos ( @xmath102 ) in quark matter . here , @xmath112 and @xmath113 represent @xmath103mev/@xmath2 and @xmath1mev/@xmath2 , respectively . ] influence of the strong coupling constant , @xmath6 , on the total rosseland neutrino mean free path of nondegenerate ( left panel ) and of degenerate neutrinos ( right panel , @xmath102 ) in quark matter . in the panel at the right , only the cases @xmath111 and @xmath114 are considered . the curves corresponding to @xmath115 and @xmath116 lie between the two curves shown and have been omitted for clarity purposes . ] influence of the strange quark mass , @xmath5 , on the total rosseland neutrino mean free path of nondegenerate neutrinos in quark matter with @xmath103mev/@xmath2 . ] the same as fig . [ fig:12 ] but for degenerate neutrinos ( @xmath102 ) . ] the same as fig . [ fig:12 ] but for degenerate neutrinos ( @xmath102 ) . ] mean scattering angle @xmath117 of nondegenerate neutrinos in quark matter . here , @xmath118 , @xmath50 , @xmath51 and @xmath119 indicate the target particle . ] the same as fig . [ fig:16 ] but for @xmath102 . ] electron - neutrino inverse absorption mean free paths in quark matter at two times nuclear saturation density . dashed lines indicate the approximations given in eqs . ( [ eq : iw1982_abs_dnu ] ) ( right panel ) and ( [ eq : iw1982_abs_ndnu ] ) ( left panel ) . ] our results are shown separately in terms of either nondegenerate or highly degenerate neutrinos . to nondegenerate neutrinos we associate a chemical potential @xmath120 . degenerate neutrinos , on the other hand , must have @xmath121 . we represent the latter category in terms of a high net electron - neutrino lepton fraction @xmath122 , where @xmath123 and @xmath124 represent , respectively , electron and electron - neutrino net number densities ( number density of particles minus number density of the corresponding antiparticles ) and @xmath125 is the baryon number density . in fig . [ fig:3 ] we show the inverse neutrino mean free paths associated to the six electron - neutrino processes listed in table [ table : cconsts ] . it must be noted that the factor @xmath126^{-1}$ ] present in eq . ( [ eq : lambda_scatt ] ) has been kept so the curves can be better distinguished in the region @xmath127 . when neutrinos are nondegenerate , neutrino - quark scattering dominates the scattering opacity with respect to neutrino - electron scattering by orders of magnitude . this discrepancy can be explained by the tiny density of electrons present in quark matter . the presence of trapped degenerate neutrinos also increases the electron density and hence the neutrino - electron contribution to the scattering opacity also increases , as can be seen in the lower panel at the left . in both the upper and the lower panels at the left in fig . [ fig:3 ] , it can be seen that neutrino - strange quark scattering differs from the other scatterings mainly at lower neutrino energies . in the panels at the right in fig . [ fig:3 ] we show the neutrino inverse mean free paths associated with absorption . in general , the process of neutrino absorption on @xmath50 quarks dominates the opacity . the contribution of neutrino absorption on @xmath51 quarks , on the other hand , tends to be smaller than that of scatterings , with exception of the very low neutrino energy regime , in which this process dominates the opacity . these results are in agreement with the results found in ref . @xcite . the behavior of the neutrino inverse mean free paths with respect to baryon number density for fixed neutrino energies is shown in figs . [ fig:4 ] and [ fig:5 ] . it can be seen that the contribution of neutrino - electron scattering is always small in comparison with that of neutrino - quark scattering when neutrinos are nondegenerate . it is only when neutrinos are trapped and highly degenerate , implying in an also higher abundance of electrons , that neutrino - electron scattering becomes an important contribution to the total opacity , as it is shown in the lower panels of fig . [ fig:5 ] . it is noticed that , for fixed @xmath128 , the inverse mean free paths vary within one order of magnitude in the baryon number density range of interest . in contrast , the temperature dependence is stronger , e.g. there is a variation of more than two orders of magnitude when @xmath128 goes from @xmath129 mev to a value above @xmath130 mev . the combined scattering processes amount to at least 10% of the total rosseland mean free path , as shown in fig . [ fig:6 ] . in the case of leptonized quark matter , neutrino scattering represents @xmath131 of the total neutrino opacity . it is only when neutrinos are less degenerate in colder quark matter that neutrino absorption becomes dominant . even in this case , as we already mentioned , neutrino scattering will contribute to at least @xmath132 of the total opacity in lower density regions of the star where neutrinos are completely nondegenerate . in figs . [ fig:7 ] and [ fig:8 ] we show the diffusion coefficients @xmath133 , @xmath134 and @xmath135 for both electron - neutrinos and electron - antineutrinos . when neutrinos are nondegenerate , all diffusion coefficients decrease with increasing density . as it can be seen in the panels at the middle , the combined coefficient @xmath136 will be negative . in fig . [ fig:8 ] it can be seen that the behavior of @xmath133 , @xmath134 and @xmath135 for degenerate neutrinos is not the same with respect to increasing density . notoriously , @xmath133 decreases , while @xmath135 increases . this behavior can be anticipated by means of analytic approximations for degenerate neutrinos , as we show in appendix [ ap : dn ] . in figs . [ fig:9 ] through [ fig:11 ] we analyze separately how @xmath5 , @xmath4 and @xmath6 indirectly influence the neutrino rosseland mean opacity through their effects on the equilibrium composition of quark matter as a function of pressure for a given temperature . in eq . ( [ eq : pbag ] ) , it is seen that , for a given @xmath137 , a larger @xmath4 will imply in a larger absolute value of @xmath93 . for fixed @xmath5 and @xmath6 , it follows that the equilibrium chemical potentials will increase accordingly . as verified in fig . [ fig:9 ] , this implies in smaller mean free paths . from eq . ( [ eq : omega_alphac ] ) , it follows that the same reasoning applies for increases in @xmath6 and this is shown in fig . [ fig:10 ] . for last , we analyze the influence of the strange quark mass on the neutrino mean free paths . since @xmath5 enters not only in the equation of state but also directly in the computation of the mean free paths in a nontrivial way , its influence is inferred directly from the numeric results shown in fig . [ fig:11 ] , where it can be seen that an increase in @xmath5 leads to larger mean free paths . it follows then that increases in both @xmath4 and @xmath6 lead to decreases in the neutrino mean free paths . the effect of increasing @xmath6 , however , is distinct from that of increasing @xmath4 in the sense that its effect is more prominent at higher pressures . it can also be noted that the mean free paths of highly degenerate neutrinos in highly degenerate quark matter is less sensitive to changes in @xmath5 or @xmath6 , since in these situations the chemical potentials are already very high . we have seen that scattering may represent a considerable fraction of the total neutrino opacity in quark matter . it is evident then that neutrino scattering must be treated with proper care . now we wish derive an estimate of the error committed when the term containing @xmath45 in eq . ( [ eq : kappa_0 ] ) is neglected and eq . ( [ eq : lambda_scatt ] ) is used as the neutrino inverse scattering mean free path . even though we do not know an explicit form for @xmath45 , it is expected to be much smaller than @xmath28 in magnitude on the regions of higher density , where we expect neutrinos to be almost in thermodynamic equilibrium with the remaining of the matter . in this case , the ratio @xmath138 tends to unity , given that the term containing @xmath45 in eq . ( [ eq : kappa_0 ] ) becomes vanishingly small compared to the first term and eq . ( [ eq : kappa_0 ] ) tends to eq . ( [ eq : lambda_scatt ] ) . on the other hand , on the regions of extremely small neutrino opacity , neutrinos will stream almost freely and , in the limit of free stream , @xmath139 and the ratio @xmath138 deviates maximally from the unity . a fairly general expression relating @xmath45 and @xmath28 on the surface of a stellar core may be written in the form @xcite @xmath140 where @xmath141 is responsible for smoothly shifting the neutrino flux from isotropic to radially outward as the optical depth increases . in fact , as the optical depth becomes larger , eq . ( [ eq : f1_surface ] ) will hold not only on the stellar surface but in a whole thick layer where neutrinos stream freely . in principle , @xmath141 should depend on the neutrino energy , however , to avoid further complications we consider it as a constant geometric factor . in this case , ( [ eq : kappa_0 ] ) reduces to @xmath142 and the ratio @xmath143 represents a bound to the maximum deviation from unity possible . in other words , the ratio @xmath144 will always be closer to @xmath34 than it is @xmath143 . this information can be used as an estimate of the largest possible error committed on the computation of the scattering contribution to the opacity when @xmath145 is used in the place of @xmath44 . when neutrinos are nondegenerate , it can be seen in fig . [ fig:12 ] that which between @xmath145 and @xmath146 is greater depends on the incident neutrino energy . for @xmath147 , @xmath145 tends to be @xmath148 smaller than @xmath146 . for @xmath149 , we see that the contribution of the term containing @xmath150 in eq . ( [ eq : kmaxerr ] ) becomes larger and @xmath146 becomes many times smaller than @xmath145 as the energy of the incident neutrino increases . in fig . [ fig:13 ] we see that when neutrinos are degenerate , there s almost no distinction between the two calculated scattering opacities when @xmath151 . however , for neutrinos with @xmath152 , there is a noticeable difference , that increases with increasing neutrino degeneracy . for instance , when @xmath153mev , @xmath145 is @xmath154 smaller than @xmath146 for @xmath102 . for higher neutrino energies with respect to @xmath155 , @xmath145 becomes several times greater than @xmath146 , similarly to what happens when neutrinos are nondegenerate . in diffusive transport schemes , one is generally interested in the energy - averaged diffusion coefficients as in eqs . ( [ eq : dndef ] ) and ( [ eq : lambda_ross ] ) , whose integrands are weighted with the factor @xmath156 , which quickly cuts off the higher energies , and features a peak centered at @xmath157 when neutrinos are degenerate . it means that in practice , @xmath145 will give scattering opacities at most @xmath148 smaller than the the ones that would be obtained if @xmath158 was used . we have seen that the scattering contribution to the total rosseland mean opacity is in the range @xmath159 for temperatures in the range @xmath160mev when neutrinos are nondegenerate ( fig . [ fig:6 ] ) . it follows then that the error committed in the total neutrino opacity , may be estimated to have an upper bound in the range @xmath161 . in regions where neutrinos are very close to thermodynamic equilibrium , and the anisotropy in the neutrino distribution function is vanishingly small , this error is expected to be much smaller . the situation is very similar when neutrinos are degenerate . in this case , the largest errors will be associated to those shown in fig . [ fig:13 ] for @xmath162 , which we see to be of about @xmath163 when @xmath153mev and smaller for larger temperatures . in this case , neutrino scattering amounts to about @xmath164 of the total neutrino opacity ( fig . [ fig:6 ] ) and the error committed may be estimated to have an upper bound in the range @xmath165 for temperatures increasing in the range @xmath166mev . if neutrinos are not expected to be too strongly degenerate in quark matter with temperatures as low as @xmath129 mev , then in this case the estimated upper bound for the error will be closer to @xmath167 . neutrino - electron scattering has long been known to be an important thermalizing agent in neutron star matter matter @xcite , while neutrino - baryon scattering is usually regarded as an isoenergetic process @xcite . in ref . @xcite , prakash & lattimer work in detail neutrino scattering in neutron star matter with and without hyperons , nevertheless , no attention has been paid to the aspects of neutrino / matter energy exchange or neutrino mean scattering angles . a detailed analysis of neutrino scattering in quark matter , however , is missing in the literature . given that quarks are very degenerate in quark matter and considering , for instance , the large mass of the strange quark with respect to that of electrons , it remains the question of whether neutrino - quark scattering is similar to neutrino - baryon or to neutrino - electron scattering with respect to neutrino - matter energy transfer and mean scattering angles . in this section we show that neutrino - quark scattering is always similar to neutrino - electron scattering with respect to energy exchange and hence non - isoenergetic . both the mean energy transfer and the mean scattering angle are defined in appendix [ ap : etransfer ] . the first thing to be noticed in figs . [ fig:14 ] and [ fig:15 ] , is that neutrino - matter energy exchange vanishes on average for a particular incident neutrino energy , @xmath168 , depending on both the neutrino and matter state of degeneracy . for @xmath169 , neutrinos tend to gain energy on average , while for @xmath170 , they tend to lose energy . this is not to say that neutrinos with @xmath171 do nt exchange energy with matter , but rather , that the net energy exchange vanishes on average . we see that the energy - exchange in neutrino - strange quark scattering is very similar to that in neutrino - electron scattering . the curves representing neutrino - up quark and neutrino - down quark scattering have been omitted since the difference between those and that of neutrino - electron scattering would be barely visible . when neutrinos are nondegenerate , @xmath168 lies between @xmath172 and @xmath173 . when neutrinos are degenerate , @xmath168 lies between @xmath155 and @xmath174 . the exact value depends on the state of degeneracy of the matter , which determines both the mean energy of the particles participating on the reaction and the phase space available for the scattered particles @xcite . regardless of the neutrino / matter particular state of degeneracy , it is a general result that highly energetic neutrinos ( @xmath175 ) will lose on average half their initial energy in each scattering process , a result previously known for the case of neutrino - electron scattering when neutrinos are nondegenerate @xcite , and which we extend here for degenerate neutrinos as well . on the other hand , very low - energetic neutrinos can gain on average several times their initial energy in a given scattering process and it is in this energy regime that neutrino - strange quark and neutrino - electron scatterings differ . it can be seen that low - energetic neutrinos tend to gain more energy from electrons ( and from up and down quarks ) than from strange quarks . figs . [ fig:16 ] and [ fig:17 ] show the neutrino mean scattering angle . less energetic neutrinos will scatter on average by an angle of about @xmath176 degrees . in association with figs . [ fig:14 ] and [ fig:15 ] , this indicates not only high energy transfer but also high momentum transfer . the mean scattering angle decreases as the energy of the incident neutrino increases , as shown in fig . [ fig:16 ] , meaning a more isotropic scenario . when @xmath177 approaches the target s chemical potential , we see a local peak in @xmath117 . more highly energetic neutrinos will on average scatter almost equally on any direction , resulting in small mean scattering angles . when the matter is known to be completely degenerate , it is possible to find approximate expressions for the neutrino mean free paths with respect to the processes of absorption and of scattering , considering that the momenta of all matter constituents are fixed at their fermi surface values . neutrinos , on their turn , are considered to be either completely degenerate or nondegenerate . for the absorption of degenerate neutrinos , it is found @xcite @xmath178&\nonumber\\ & \times\left[1+\frac{1}{2}\frac{p_{f_e}}{p_{f_u}}+\frac{1}{10 } \left(\frac{p_{f_e}}{p_{f_u}}\right)^2\right],\end{aligned}\ ] ] where @xmath179 is the fermi momentum of particle species @xmath17 . ( [ eq : iw1982_abs_dnu ] ) is suitable for the case @xmath180 . in the opposite situation , i.e. , when @xmath181 , the appropriate mean free path is obtained through the replacements @xmath182 and @xmath183 . for the absorption of nondegenerate neutrinos , it is found @xmath184 the strong coupling constant , @xmath6 , makes it appearance in eq . ( [ eq : iw1982_abs_ndnu ] ) through the lowest - order corrections to the quarks chemical potentials due to quark - gluon strong interaction @xmath185 necessary to make the reaction kinematically allowed , since the masses of the @xmath118 and @xmath50 quarks are small and the neutrino momentum is not expected to play any significant role in the momentum conservation when they are nondegenerate , and hence , excluded from the momentum - conservation delta function @xcite . the influence of the neutrino state of degeneracy on neutrino - electron scattering has been analyzed to great extent . the inverse neutrino mean free path for degenerate neutrinos for this process has been found to be approximately @xcite @xmath186\sqrt{\frac{\mu_\nu}{p_{f_e}}x_e } & \nonumber\\ & \times\left[\left(c_{v , e}^2+c_{a , e}^2\right)\left(10+x_e^2\right)+10c_{v , e}c_{a , e}x_e\right],\end{aligned}\ ] ] for @xmath149 . here , @xmath187 @xmath188 and @xmath189 . approximate expressions for neutrino - electron scattering for nondegenerate neutrinos were found for different neutrino energy regimes @xcite @xmath190 \frac{n_e\sigma_0}{6m_e^2}e_\nu p_{f_e}\end{aligned}\ ] ] for @xmath191 , and @xmath192 \frac{n_i\sigma_0}{40 m_e^2}\frac{e_\nu^3}{p_{f_e}},\end{aligned}\ ] ] for @xmath193 . neglecting the quark masses , the same results of eqs . ( [ eq : iw1982_scatt_dnu])-([eq : iw1982_scatt_ndnu_low ] ) are expected to be valid for neutrino - quark scattering , with appropriate coupling constants . it is important to note that in all of eqs . ( [ eq : iw1982_abs_dnu])-([eq : iw1982_scatt_ndnu_low ] ) it is assumed that also electrons are completely degenerate . tubbs & schramm show approximations for neutrino - electron scattering when both neutrinos and electrons are nondegenerate @xcite . in fig . [ fig:18 ] we show a comparison between full numeric results and analytic approximations of eqs . ( [ eq : iw1982_abs_dnu ] ) and ( [ eq : iw1982_abs_ndnu ] ) for the process @xmath194 . in the case of degenerate neutrinos , we have divided the numeric results by the factor @xmath195 , already included in the analytic results . in the nondegenerate case , we have used @xmath196 in eq . ( [ eq : iw1982_abs_ndnu ] ) ( with @xmath111 in the thermodynamic potential of quark matter ) . larger values for @xmath6 may result in better agreement with the numeric results for @xmath197 for this particular combination of temperature and density , but the same value for the strong coupling constant will not always give the better results for higher temperatures or densities for a given bag constant . using eq . ( [ eq : iw1982_abs_ndnu ] ) requires then fixing @xmath6 to an optimal value , whenever this is physically acceptable , so the approximation may give the best results in its domain of validity . ( [ eq : iw1982_abs_dnu ] ) , for absorption of degenerate neutrinos , on the other hand is shown to give good results only on the vicinity of @xmath157 and the results are better the higher the neutrino degeneracy is . in the upper panels of fig . [ fig:19 ] we compare the results given by eqs . ( [ eq : iw1982_scatt_ndnu_high ] ) and ( [ eq : iw1982_scatt_ndnu_low ] ) with the numeric results for the indicated scattering processes of nondegenerate neutrinos . it is clear that the range of energies @xmath198 is in general not covered by the approximations . for the case of neutrino - electron scattering , eq . ( [ eq : iw1982_scatt_ndnu_high ] ) gives the better results for the whole range of energies , even though its domain of validity is @xmath199 . nonetheless , it must be noticed that , in this case , results share only the same order of magnitude and , roughly , the same qualitative behavior . in the lower panels of of fig . [ fig:19 ] , we show how the results of eq . ( [ eq : iw1982_scatt_dnu ] ) compare to the exact results . as in the case of absorption , the results are better for @xmath177 near @xmath155 , nonetheless , the approximate results are in good qualitative agreement and same order of magnitude as the exact results also for @xmath200 . in the panel of the middle , we show how the strange quark mass affects the scattering cross section . using the approximate mean free paths given by eqs . ( [ eq : iw1982_abs_dnu])-([eq : iw1982_scatt_ndnu_low ] ) , we calculate the neutrino rosseland mean free path and compare the results against exact calculations . since we considered only the cases of either highly degenerate or completely nondegenerate neutrinos , we did not use any interpolating procedure to join the different regimes , such as the one suggested by @xcite . also , it must be noted that , for scatterings of nondegenerate neutrinos , only eq . ( [ eq : iw1982_scatt_ndnu_low ] ) was used . even though an interpolating scheme to join smoothly eqs . ( [ eq : iw1982_scatt_ndnu_high ] ) and ( [ eq : iw1982_scatt_ndnu_low ] ) can be proposed , this is not expected to improve the results too much , since this energy range is strongly cut - off by the factor @xmath201 . the results shown in fig . [ fig:20 ] reflect the fact that the approximations are , in general , good only on a limited range of neutrino energies . as one would expect , the analytic approximations of eqs . ( [ eq : iw1982_abs_dnu])-([eq : iw1982_scatt_ndnu_low ] ) imply in generally poor agreement with exact results of energy - averaged opacities . notoriously , the results are better for degenerate neutrinos . this can be understood from the fact that the approximations to the mean free paths tend to be better for neutrino energies near @xmath155 and the integrands of eq . ( [ eq : lambda_ross ] ) are highly weighted in the same region . nevertheless , even in this case , the results agree only by order of magnitude and they share roughly the same qualitative behavior . we have analyzed in detail the neutrino mean free paths in quark matter . all the electron - neutrino weak interaction processes listed in table [ table : cconsts ] were considered . we have shown that , when neutrinos are nondegenerate , neutrino - quark scattering dominates the scattering contribution to the total neutrino opacity with respect to neutrino - electron scattering . on the other hand , when neutrinos are degenerate , the different scattering processes contribute almost the same , given that in this case , quark matter has a larger abundance of electrons . in general , neutrino absorption constitutes the largest contribution to the neutrino opacity . however , neutrino scattering can represent a considerable fraction , amounting to about @xmath164 in the highly leptonized scenario . when neutrinos are nondegenerate , scattering represents no less than @xmath202 of the total opacity , for temperatures in the range @xmath160mev and baryon number densitiy in the range @xmath203 . we estimate that using @xmath40 , given in eq . ( [ eq : kappa_0 ] ) instead of @xmath44 given by eq . ( [ eq : lambda_scatt ] ) in the computation of the neutrino scattering mean free paths will represent an error of less than @xmath202 in the total neutrino opacity . this upper bound is associated with the threshold of neutrino free - streaming , where @xmath45 is known to be proportional to @xmath28 and both @xmath40 and @xmath44 can be calculated and compared . in regions where neutrinos are closer to thermodynamic equilibrium , and @xmath45 is vanishingly small , this error is expected to be much smaller . for baryon number densities in the range @xmath203 , the energy - averaged diffusion coefficients @xmath0 , show a greater dependency on temperature than on density when @xmath128 varies in the range @xmath160mev . when neutrinos are nondegenerate , all of @xmath133 , @xmath134 and @xmath135 decrease with baryon density , while when neutrinos are degenerate they show different behaviors . in this case , @xmath133 decreases , while @xmath134 show no significant variation and @xmath135 increases with increasing density . we have also shown that the diffusion coefficients associated with antineutrinos are larger than those of neutrinos when the latter are nondegenerate . when neutrinos are degenerate , on the other hand , the diffusion coefficients of neutrinos are greater than those of antineutrinos . since we have not considered the presence of muons and taus in quark matter , the corresponding curves associated to the diffusion of muon and tau neutrinos have been left outside of the present work . nevertheless , the following general remarks are worth addressing . considering muon and tau neutrinos present in quark matter only in the form of thermally produced pairs , this type of neutrinos can take part only in the neutral - current scattering processes listed in table [ table : cconsts ] , given that the corresponding charged - current processes are kinematically suppressed . it follows then that the total muon and tau neutrino mean free paths are generally increased in comparison with the total electron neutrino mean free paths . in particular , considering that the scattering opacity is the same for both muon and tau neutrinos , their contribution to the combined diffusion coefficient @xmath204 makes explicit their role in the energy transport in neutron stars , mainly in the regime of nondegenerate electron neutrinos , as we show in fig . [ fig:21 ] . diffusion coefficient @xmath135 for electron neutrinos ( full lines ) and muon neutrinos ( dashed lines ) . the topmost pairs of curves correspond to @xmath153mev and those at the bottom correspond to @xmath205mev . note that @xmath120 and @xmath206 correspond to electron neutrinos . muon neutrinos always have @xmath207 here . ] the influence of free parameters of the mit bag model on the neutrino mean free paths have been analyzed . it is verified that increases in @xmath4 , and @xmath6 lead to in increases in the total neutrino opacity , while a higher @xmath5 decreses it . in general , when neutrinos are nondegenerate , increasing @xmath4 from @xmath208mev/@xmath2 to @xmath209mev/@xmath2 or increasing @xmath5 from @xmath33 to @xmath210mev will affect more the neutrino opacity than does increasing @xmath6 from @xmath33 to @xmath211 in the approximation of massless quarks . the same occurs when neutrinos are degenerate , with the difference that the effects are less prominent . it should be noted that we have not considered here how many - body interactions may alter the neutrino cross - sections through correlations or collective behavior in general . in specific , we have neglected the possibility of quark pairing and color superconductivity and how the neutrino mean free paths are altered by it . in fact , at asymptotically high densities it has been shown from first principles that color superconductivity occurs in qcd . however , in the density regime relevant for neutron star physics such weak - coupling calculations are unreliable and most conclusions are obtained from phenomenological models that are claimed to capture the essential physics of qcd . as a consequence , the actual occurrence of color superconductivity inside neutron stars is still an open question . nevertheless , it is important to examine the impact of color superconductivity on the neutrino mean free paths and in fact this has been addressed in several works . for more details on the subject , we refer the reader to refs . @xcite . for last , neutrino - quark scattering share remarkable similarities with neutrino - electron scattering in what comes to neutrino / matter energy transfer and mean scattering angles . the general scenario is : low - energetic neutrinos tend to gain large amounts of energy and to scatter by larger angles . on the other hand , high - energetic neutrinos will lose on average half their initial energy and scatter almost isotropically . we would like to thank fapesp and universidade federal do abc for the financial support . with @xmath19 given in eq . ( [ eq : wfi ] ) , eq . ( [ eq : collisionint ] ) can be cast into several different forms for integration . to calculate the absorption mean free paths and the scattering mean free paths as in eq . ( [ eq : lambda_scatt ] ) it is convenient to use the total momentum as an integration variable . with that , all angular integrations can be performed and it only remains a double integration do be performed numerically @xcite : @xmath212,\end{aligned}\ ] ] where @xmath213[1-f_0(e_1+e_2-e_3)]$ ] and the integrals @xmath214 , @xmath215 and @xmath216 are given by @xmath217,\\ \nonumber\\ i_b=&i_a(e_2 \leftrightarrow -e_3),\\ \nonumber\\ i_c=&\frac{2\pi^2}{3}m_2m_3\left[\left(q_{\mbox{\scriptsize{max}}}^3 -q_{\mbox{\scriptsize{min}}}^3\right)\right.\nonumber\\ & \quad\quad\left . + 6k^{\nu 3}_-\left(q_{\mbox{\scriptsize{max } } } -q_{\mbox{\scriptsize{min}}}\right)\right],\end{aligned}\ ] ] with @xmath218 here , @xmath219 and the particle labels follow the convention of table [ table : cconsts ] . we define the neutrino scattering kernels @xmath42 as @xmath220w_{fi},\end{aligned}\ ] ] where @xmath221 denotes the total phase - space degeneracy factor and the integrations involve only the target particles . with @xmath19 given in eq . ( [ eq : wfi ] ) , all integrations except one can be performed exactly . the results can be found in references @xcite ( with the remark that our definition of @xmath42 does not carry the overall @xmath222 ) : @xmath223,\end{aligned}\ ] ] where @xmath224 $ ] and the integrals @xmath225 , @xmath226 and @xmath226 are given by @xmath227,\\ h_b&=h_a(e_\nu\leftrightarrow -e_\nu^\prime),\\ h_c&=\frac{gg_f^2}{(2\pi)e_\nu e_\nu^\prime } \frac{m_2 ^ 2e_\nu e_\nu^\prime(1-\cos\theta)}{\alpha^{1/2}},\end{aligned}\ ] ] where @xmath228 , @xmath25 is the scattering angle . the quantities @xmath229 , @xmath4 and @xmath230 are given by @xmath231,\\ \nonumber\\ b=&\frac{\gamma^2}{\alpha^{5/2}}e_\nu \left[2e_\nu ^2+(3-\cos\theta)e_\nu e_\nu^\prime\right.\nonumber\\ & \qquad\left.-(1 + 3\cos\theta)e_\nu^{\prime 2}\right],\\ \nonumber\\ c=&\frac{\gamma^2}{\alpha^{5/2}}e_\nu ^2\left[(e_\nu -e_\nu^\prime\cos\theta)^2 -\frac{e_\nu^{\prime 2}\sin^2\theta}{2}\right.\nonumber\\ & \qquad\left.-\frac{m_2 ^ 2(1+\cos\theta)}{2e_\nu ^2(1-\cos\theta)}\alpha\right],\end{aligned}\ ] ] where we have also defined @xmath232 . the lower integration limit in eq . ( [ eq : rout_ap ] ) , is given by @xmath233 from fermi s golden rule , the neutrino interaction cross section can be written in terms of the squared matrix element , summed over final spins and averaged over the initial spins @xmath234 for interactions happening on the vacuum . in thermodynamic systems , the particles involved in such scattering reactions have specific energy distributions and we must average the cross sections over their distribution functions and the availability of the initial and final states . in this case , we write @xmath235\nonumber\\ & \times[1-f_0(e_4)]\frac{\langle\mathcal{m}\rangle^2}{4\left(p_\nu\cdot p_2\right ) } \delta(p_\nu+p_2-p_3-p_4).\end{aligned}\ ] ] we define the mean energy transfer per collision between neutrinos and matter and the mean scattering angle as the averages [ eq : etransf ] e_=&de_^ ( e_-e_^ ) ( ) , & and [ eq : sangle ] 1-= & de_^ & + & d(1- ) ( ) , & where we have defined @xmath236[1-f_0(e_\nu^\prime)].\end{aligned}\ ] ] note that we have added the blocking factor @xmath237 , to account for neutrino degeneracy , in contrast to the results of @xcite . also , our definition of @xmath238 is only a shorthand for the quantity defined in eq . ( [ eq : sigbar ] ) and not a properly averaged statistical cross section as the one found in the same reference . for convenience , we have added a minus overall sign in eq . ( [ eq : etransf ] ) for the mere fact that we want negative values to represent energy loss . substituting @xmath239 on eqs . ( [ eq : etransf ] ) and ( [ eq : sangle ] ) , we obtain eqs . of the form @xmath240\end{aligned}\ ] ] and @xmath241\end{aligned}\ ] ] with the @xmath242 and @xmath243 given by eqs . ( 21a)-(22c ) of @xcite with the inclusion of @xmath244 $ ] in all @xmath245 integrations of the cited reference . we would like to point slightly different definitions for @xmath246 and @xmath247 that may be derived directly from eq . ( [ eq : lambda_scatt ] ) : @xmath248,\end{aligned}\ ] ] and @xmath249.\end{aligned}\ ] ] eq . ( [ eq : scattang_alt1 ] ) can then be written in terms of the legendre moments of the scattering kernel @xmath42 : [ eq : scattang_alt ] & 1-=1-de_^e_^2 r^out_1 . & even though eqs . ( [ eq : etransf_alt ] ) and ( [ eq : scattang_alt ] ) differ from eqs . ( [ eq : etransf ] ) and ( [ eq : sangle ] ) by the lack of the factor @xmath250 on the integrands ( coming from the definition of cross section ) , they give fairly similar results and they can be more easily evaluated when @xmath42 is available . in the case of highly degenerate neutrinos , the analytic approximations to the neutrino mean free paths with respect to both the neutral and charged currents , eqs . ( [ eq : iw1982_abs_dnu])-([eq : iw1982_scatt_dnu ] ) , can be combined in the form @xmath251\frac{1}{g(\eta_\nu)},\end{aligned}\ ] ] where @xmath252 with the coefficients given by @xmath253 12\cos^2\theta_c \eta_u^2\eta_e^3\left[1+\frac{1}{2}\frac{\eta_e}{\eta_u}\right . \nonumber\\ & \qquad\left.+\frac{1}{10}\left(\frac{\eta_e}{\eta_u}\right)^2\right],\\ { } b&=\left[\frac { g_f^2 t^5}{3\pi^3 \left(c\hbar\right)^7}\right]\sum_{i = u , d , s , e}g_i \left(c_{v , i}^2+c_{a , i}^2\right)\eta_i^2,\\ { } c&=\left[\frac { g_f^2 t^5}{3\pi^3 \left(c\hbar\right)^7}\right]\sum_{i = u , d , s , e}g_ic_{v , i}c_{a , i } \eta_i,\\ { } d&=\left[\frac { g_f^2 t^5}{3\pi^3 \left(c\hbar\right)^7}\right]\frac{1}{10}\sum_{i = u , d , s , e}g_i \left(c_{v , i}^2+c_{a , i}^2\right).\end{aligned}\ ] ] in these equations , @xmath254 . coefficients @xmath255 to @xmath50 are related only to the scattering reactions . coefficient @xmath256 has a counterpart as mentioned after eq . ( [ eq : iw1982_abs_dnu ] ) , however , @xmath257 , tends to be the case when neutrinos are highly degenerate . note also that we used @xmath258 in eq . ( [ eq : iw1982_scatt_dnu ] ) . in this regime , the diffusion coefficients @xmath0 read @xmath259 } { \left(x-\eta_\nu\right)^2+\pi^2}\right\}.\end{aligned}\ ] ] the expression between curly braces can be approximated by @xmath260}{\left(x-\eta_\nu\right)^2+\pi^2}\simeq \frac{e^{-\left(x-\eta_\nu\right)^2/4}}{4\pi^2},\end{aligned}\ ] ] and so eq . ( [ eq : dn_analytic_first ] ) can be solved by gamma functions . one then arrives at @xmath261\\ \times\gamma\left(\frac{k+1}{2}\right)\eta_\nu^{n - k}.\end{gathered}\ ] ] explicitly : @xmath262 to perform a quick analysis , we assume , for simplicity , that @xmath263 ( which is , in fact , a reasonable assumption when neutrinos are highly degenerate ) . with this , the denominator on eqs . ( [ eq : d2_approx])-([eq : d4_approx ] ) reduces to @xmath264 19.78 \eta_\nu^5.\end{aligned}\ ] ] it is clear then that coefficients @xmath133 , @xmath134 and @xmath135 have different behavior with respect to @xmath265 and , in particular , @xmath135 will increase with increasing neutrino degeneracy . 22ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty a. bodmer , phys . d * 4 * , 1601 ( 1971 ) . h. terazawa , ins rep . 336 , univ . tokyo , ins ( 1979 ) . e. witten , phys . d * 30 * , 272 ( 1984 ) . c. alcock , e. farhi , and a. olinto , astrophys . j. * 310 * , 261 ( 1986 ) . p. haensel , j.l . zdunik , and r. schaefer , astron . astrophys . * 160 * , 121 ( 1986 ) . i. bombaci , d. logoteta , p. k. panda , c. providncia , and i. vidaa , phys . b * 680 * , 448 ( 2009 ) ; g. lugones and a. g. grunfeld , phys . d * 84 * , 85003 ( 2011 ) ; and references therein . f. weber , _ pulsars as astrophysical laboratories for nuclear and particle physics _ , iop publishing ltd . n. k. glendenning , _ compact stars : nuclear physics , particle physics , and general relativity _ , springer ( 2000 ) . p. haensel , a. y. potekhin and d. g. yakovlev , _ neutron stars 1 : equation of state and structure _ , springer ( 2007 ) . g. lugones , c. r. ghezzi , e. m. de gouveia dal pino , and j. e. horvath , astrophys . j. * 581 * , l101 ( 2002 ) . p. keranen , r. ouyed , and p. jaikumar , astrophys . j. * 618 * , 485 ( 2005 ) . b. niebergal , r. ouyed , and p. jaikumar , phys . c * 82 * , 062801 ( 2010 ) . t. fischer , i. sagert , g. pagliara , m. hempel , j. schaffner - bielich , t. rauscher , f. k. thielemann , r. kppeli , g. martnez - pinedo , and m. liebendrfer , astrophys . j. supp . * 194 * , 39 ( 2011 ) . g. pagliara , m. herzog , and f. k. rpke , phys . d * 87 * , 103007 ( 2013 ) . @noop * * , ( ) g. w. carter and s. reddy , phys . rev . d * 62 * , 103002 ( 2000 ) . s. reddy , m. sadzikowski and m. tachibana , nucl . a * 721 * , 309 ( 2003 ) . j. kundu and s. reddy , phys . 055803 ( 2004 ) . t. schfer and k. schwenzer , phys . d * 70 * , 114037 ( 2004 ) . p. jaikumar , c. d. roberts and a. sedrakian , phys . rev . c * 73 * , 055803 ( 2006 ) . k. pal and a. k. dutt - mazumder , phys . d * 84 * , 034004 ( 2011 ) .

the spin - dependent transport of the electrons tunneling through a resonant tunneling structure with ferromagnetic multi - terminal under dc and ac fields is explored by means of the nonequilibrium green function technique . a general formulation for the time - dependent current and the time - averaged current is established . as its application the systems with two and three terminals in noncollinear configurations of the magnetizations under dc and ac biases are investigated , respectively . the asymmetric factor of the relaxation times for the electrons with different spin in the central region is uncovered to bring about various behaviours of the tmr . the present three - terminal device is different from that discussed in literature , which is coined as a spin transistor with source . the current - amplification effect is found . in addition , the time - dependent spin transport for the two - terminal device is studied . it is found that the photonic sidebands provide new channels for the electrons tunneling through the barriers , and give rise to new resonances of the tmr , which is called as the photon - asisted spin - dependent tunneling . the asymmetric factor of the relaxation times is observed to lead to additional resonant peaks besides the photon - asisted resonances .

cond-mat0406221

giant magnetoresistance ( gmr ) effect@xcite discovered in fe / cr multilayers has motivated much research on the spin - dependent transport in hybrid magnetic multilayers in recent years@xcite . parkin et al.parkin has observed that the interlayer exchange coupling oscillates damply with the cr or ru spacer layer thickness in co / ru , co / cr and fe / cr superlattice systems , which suggests that the quantum - well states be formed in the spacer layer . this is because the electrons may meet multi - reflection at the interfaces between the ferromagnets and the spacer layer , leading to the oscillation of density of states ( dos ) of electrons at the fermi level , and thereby giving rise to the oscillation of the interlayer exchange coupling with the spacer thickness@xcite . on the other hand , the magnetic tunnel junction ( mtj ) such as the ferromagnet / insulator / ferromagnet ( fm / i / fm ) structure was found to display the spin - valve effect@xcite . the mtj with double tunnel barriers such as fm / i / fm / i / fm structure was shown to reveal more interesting behaviors on the variation of the tunnel magnetoresistance ( tmr ) with bias@xcite . recently , advances in nanolithography and thin - film processing make it possible to fabricate very small double tunnel junctions called the single electron transistor ( set ) . with a very narrow spacer , the coulomb interaction is important and may lead to new phenomena such as coulomb blockade . further investigation about the spin - dependent transport in a ferromagnetic set@xcite was performed intensively in the sequential tunneling regime@xcite@xcite and in the cotunneling regime@xcite . in the former case , the oscillation of the tmr with a dc bias is observed ; while the tmr in the latter case is enhanced in the strong tunneling regime , and the coulomb blockade region is squeezed by the spin accumulation caused by the cotunneling . it would be interesting to consider another limit where the size of the spacer is not so small that the coulomb interaction is weak , and the quantum well states can be formed in the spacer . if an ac field is applied to the leads and the central spacer , one would expect that the interesting features of the tmr might appear due to new effective tunneling channels from the photonic sidebandssidebands induced by the ac field . on the other hand , johnson@xcite demonstrated that the multi - terminal system is as important as the two - terminal system . in the so - called spin dipolar switching device ( i.e. the fm / nonmagnetic metal(nm)/fm structure ) , the current flows from one fm layer and out of the nm layer , while the other fm layer referenced to the nm layer serves as a voltmeter to give an output voltage which depends on the magnetization configuration of the two ferromagnets@xcite . fert and lee@xcite interpreted johnson s experiment by means of the boltzmann equation , and at almost the same time , hershfield et al.@xcite proposed a set of weak coupling equations to describe johnson s experiment . by invoking the framework of the scattering matrix theory ( smt ) developed by landauer and bttiker@xcite for the mesoscopic devices with multi - terminal , brataas et al.@xcite discussed a three - terminal device which is to some extent different from johnson s experimental layout , where the current does not flow out of the nm layer but out of another fm layer , and the nm layer serves as a central spacer region , while a third fm layer is introduced to measure the chemical potential and no net current flows through it . in other words , the current entering the third fm lead is equal to that flowing out of it . it is found that the direction of the magnetization of the third fm lead can affect remarkably the current flowing through the first and the second fm leads when the magnetizations of these two leads align antiparallel . in this case , the intrinsic spin relaxation time in the central spacer was assumed to be sufficiently shorter than the order of the time between successive tunneling events such that the spin accumulation effect@xcite could be negligible . furthermore , jedema _ _ et al.__@xcite investigated the spin injection and spin accumulation in an all - metal lateral mesoscopic spin valve with multi - terminal . all these works were performed under dc biases . as mentioned before , if an ac electric field is applied to the magnetic hybrid junction with multi - terminal , one would expect that different spin - dependent transport behaviors of the electrons might appear . in this paper , the time - dependent spin - polarized transport of a resonant tunneling structure with ferromagnetic multi - terminal will be investigated by means of the nonequilibrium green function technique ( negft ) . it is presumed that every terminal and the central spacer are applied by the ac fields . in our model , although the spin accumulation could present , as we shall pay our attention to the current and tmr in response to ac fields , the spin accumulation will not be considered , which will be given elsewhere . a general current formalism for such a system with @xmath0 terminals which are made of ferromagnetic materials or normal metals in the presence of an ac electrical field will be developed . a three - terminal device , to be coined as a spin transistor with source ( sts ) , in which every terminal is applied by a source bias , and there is net current flowing through each of them , will be proposed . the amplification effect of the electrical curent by changing the direction of the magnetization of the third fm terminal in a sts will be discussed . the photon - assisted spin - dependent tunneling in a two - terminal device under an ac bias voltage will be explored . the rest of this paper is organized as follows . in sec . ii , a model is proposed and a general formalism for the time - dependent current in a resonant tunneling system with ferromagnetic multi - terminal is derived . in sec . iii , the systems with two and three terminals in dc steady states are discussed . in sec . iv , the spin - dependent transport under an ac bias for a system with two ferromagnetic terminals in which the magnetic moments are noncollinearly aligned is investigated . in sec . v , a brief summary will be presented . consider a resonant tunneling structure with ferromagnetic multi - terminal ( rtsfmt ) with the hamiltonian given by @xmath1where @xmath2 ( @xmath3 ) is the hamiltonian of the ferromagnetic leads with @xmath4 the hamiltonian of the @xmath5th lead , @xmath6 is the hamiltonian of the central spacer with @xmath7 the single - particle energy in the central spacer modulated by a time - varying external field where @xmath8 is the single - particle level in the absence of the external field , @xmath9 is the energy from the time - varying field applied to the central region , and @xmath10 is the creation ( annihilation ) operator of an electron with spin @xmath11 in the central region . here we presume that the single - particle energy level , say @xmath12 , which can be viewed as a single - impurity level or a single energy level of quantum well states@xcite , is spin - degenerate . this case corresponds to a nonmagnetic island where the numbers of electrons with spin up @xmath13 and down @xmath14 are equal . in the equilibrium ( e.g. in the absence of an external bias ) , the net moment at the nonmagnetic island is zero , while in the nonequilibrium ( e.g. in the presense of an external field ) the numbers of electrons with spin up and down have different tunneling rates entering into the island , and also have different tunneling rates escaping from the island . thus , if the spin relaxation time is not very short ( for example , longer than the tunneling time ) , the spin accumulation will occur . in the present model , we shall pay attention to the spin - dependent transport of the current and the tmr of the structure , and we will not discuss the effect of spin accumulation here for clarity@xcite . @xmath15 is the coupling interaction between the leads and the central region , defined by @xmath16@xmath17 , \label{hac}\]]where @xmath18 is the creation operator of an electron in the @xmath5th lead with momentum @xmath19 and spin @xmath11 , @xmath20 is the element of the tunneling matrix which describes the coupling between the leads and the central region , and depends on time , spin and momentum . we assume that all leads are ferromagnetic , and one of them is chosen as the lead 1 . the magnetization of the lead 1 is presumed to align along the @xmath21 axis , and the magnetizations of the other leads are supposed to deviate the @xmath21 axis by angles @xmath22 , @xmath23 , @xmath24 etc . the hamiltonian of the lead 1 can be written as @xmath25where @xmath26 is the creation ( annihilation ) operator of an electron of the lead 1 with momentum @xmath27 and spin @xmath11 , @xmath28 is the single - particle energy modulated by a time - dependent external field with @xmath29 @xmath30 @xmath31 with @xmath32 the land factor , @xmath33 the bohr magneton , and @xmath34 the molecular field of the lead 1 . for the noncollinear configuration of the magnetic moments of the leads , the hamiltonians of the other leads read @xmath35\widetilde{b}_{k_{b}\sigma } ^{\dagger } \widetilde{b}_{k_{b}\sigma } -m_{b}\sin \theta _ { b}\widetilde{b}_{k_{b}\sigma } ^{\dagger } \widetilde{b}_{k_{b}\overline{\sigma } } \ } , \label{bha}\]]where @xmath36 , @xmath37 , @xmath38 denotes the lead @xmath39 , @xmath37 , @xmath38 , etc . , @xmath40 is the electron momentum in the lead @xmath41 , @xmath42 is the creation ( annihilation ) operator of an electron in the lead @xmath41 with spin @xmath11 @xmath43 , @xmath44with @xmath45 the energy from the time - dependent external field applied to the lead @xmath41 , and @xmath46 with @xmath47 the molecular - field of the lead @xmath41 . in terms of eq.([hac ] ) , we have the coupling hamiltonian between the lead @xmath41 ( @xmath48 , @xmath37 ... ) and the central region @xmath49 . \label{bch}\]]by performing the @xmath50 transformation @xmath51to @xmath52 , we find that @xmath52 becomes @xmath53with @xmath54 . consequently , the coupling hamiltonian becomes @xmath55 . \label{dhtb}\]]to this end , the hamiltonian of the system is well defined . in a time - dependent case , the current conservation can be expressed as @xmath56where @xmath57 is the tunneling current flowing from the @xmath58th lead to the central spacer , while @xmath59 is the displacement current , with @xmath60 being the occupation number of electrons with the energy level @xmath12 in the central spacer . recently , there are a few works towards partitioning the displacement current into each terminal@xcite . likewise , the displacement current in the present case is supposed to be partitioned into every terminal in such a way , i.e. @xmath61 , that we have @xmath62=\sum_{\alpha } j_{\alpha } ( t)=0 $ ] , where @xmath63 is the total current flowing from the @xmath5th lead into the central scattering region , and @xmath64 \right\rangle = -\frac{ie}{\hbar } \left\langle \left [ h_{t\alpha } , n_{\alpha } \right ] \right\rangle $ ] with @xmath65 the particle occupation number operator in the @xmath5th lead . by means of the nonequilibrium green function techniquedatta , jauho , after some calculations , we obtain @xmath66,\ ] ] where @xmath67 @xmath68 @xmath69 @xmath70 is the lesser function in spin space with @xmath71 , and @xmath72 is the angle between the magnetization direction of the @xmath5th lead and the @xmath21 axis . by using the equation of motion of the green function , and then noting the langreth theorem@xcite , we get @xmath73 , \label{jca}\ ] ] where @xmath74 is the lesser function of the central spacer in spin space , with @xmath75 being the retarded green function of the central spacer in spin space : @xmath76 where @xmath77 . in eq.([jca ] ) , @xmath78 @xmath79 is corresponding to the lesser , greater , retarded and advanced self - energy , respectively , which is defined as @xmath80where @xmath81 is the corresponding green functions of uncoupled leads , with the retarded ( advanced ) green function given by @xmath82 and the lesser function of the uncoupled leads given by @xmath83 where @xmath84 is the fermi function . when the central region is uncoupled to the leads , the retarded green function of the central region can be obtained exactly : @xmath85 \right\ } .\ ] ] from eq.([selfenergy ] ) , we shall calculate the self - energy . the off - diagonal elements of the coupling matrix describe the spin - flip scattering processes , which was discussed in ref.@xcite . to focus on the low - temperature properties of the transport , where the available electrons near the fermi level are dominant in the transport process , it is reasonable to ignore the effect of the spin - flip scattering during the tunneling process , and the coupling parameters are presumed to be independent of momentum . therefore , we have @xmath88 . to perform the momentum summation of the electrons , we may first sum the momenta of electrons in the uncoupled green functions of leads . then , for the retarded and the advanced green functions of the uncoupled leads , we will obtain the real part @xmath89 and the imaginary part @xmath90 . because @xmath89 makes the energy levels shifted and can be absorbed into the energy level of the central spacer , we only consider the imaginary part @xmath90 . with this consideration , one may get @xmath91 where the line - width function @xmath92 is @xmath93with @xmath94 , \ ] ] @xmath95 , \ ] ] @xmath96 , \ ] ] @xmath97 .\ ] ] then , the lesser self - energy is @xmath98 where @xmath99 is the fermi function of the @xmath5th lead . as a result , the current can be expressed as @xmath100 \right\ } . \notag\end{aligned}\ ] ] it is nothing but the current flowing out of the @xmath5th terminal . let us first specify the forms of the energy shifts caused by application of ac biases to the leads and the central spacer . they bear the usual forms : @xmath101 @xmath102 where @xmath103 is the frequency of the corresponding external ac bias to the @xmath5th lead ( the central spacer ) , and @xmath104 is assumed . by applying the nonequilibrium green function technique , we get the equation of motion of the retarded green function @xmath105where @xmath106 with @xmath107 being defined above . performing the gauge transformation@xcite @xmath108 and introducing the double - time fourier transform @xmath109,\]]@xmath110,\]]eq.([motion ] ) is transformed into @xmath111@xmath112 can be gained by applying the same gauge transformation and the double - time fourier transform to @xmath113 . then substituting it into eq.([centerdy ] ) , we can in principle get the green function @xmath114 . to simplify the treatment , we will assume without loss of generality that the elements of the tunneling matrix are independent of time and @xmath92 is taken at the fermi energy@xcite . by changing the summation into an integral , we obtain @xmath115\widetilde{\mathbf{g}}_{d}^{r}(\omega , \omega ^{\prime } ) = 2\pi \delta ( \omega -\omega ^{\prime } ) , \label{gcengr}\ ] ] where @xmath116 with @xmath117with @xmath118 , @xmath119 , @xmath120 , @xmath121 and @xmath122 where @xmath123 is the dos of spin up ( down ) subband of the @xmath5th ferromagnetic lead , while @xmath124 is the polarization of the @xmath5th lead . so the retarded green function of the central region becomes @xmath125^{-1}\delta ( \omega -\omega ^{\prime } ) , \label{tui}\ ] ] where the superscript @xmath126 means the inverse of the matrix . by tranforming @xmath114 back , one gets @xmath127}^{-1}. \label{grdt}\ ] ] @xmath128 can be obtained by the keldysh equation @xmath129where @xmath130 . with the presumption of @xmath87 being real , we obtain the current of the @xmath5th terminal under an ac external bias : @xmath131 \label{mulcu } \\ & & + \re e\text { } \mathbf{\gamma } _ { \alpha } ( 0)\sum_{\beta } f_{\beta } ( \omega ) \mathbf{a}_{\beta } ( \omega , t)\mathbf{\gamma } _ { \beta } ( 0)\mathbf{a}_{\beta } ^{\dagger } ( \omega , t)\ } , \notag\end{aligned}\]]where the function @xmath132 of the @xmath5th lead is defined as @xmath133 the displacement current , @xmath134/dt$ ] , can be obtained by @xmath135 with @xmath136 . \label{dd}\]]the current conservation now becomes @xmath137=\sum_{\alpha } j_{\alpha } ( t)=0,$ ] where @xmath138 is the total current of the @xmath5th lead . ( [ mulcu ] ) and ( [ dd ] ) are the main results of this paper . based on them , we can investigate the spin - dependent transport in a hybrid junction system with multi - terminal under either a dc bias or an ac bias . in the following , we shall discuss the case in a steady state under a dc bias , and then , we shall consider the spin - dependent transport in a system with two and three ferromagnetic terminals whose magnetizations are noncollinearly aligned under dc and ac biases . let us consider the structure whose schematic layout is shown in fig . we assume that the molecular field of the left fm lead is aligned along the @xmath21 axis which is perpendicular to the current direction , while the magnetization direction of the right fm lead deviates the @xmath21 axis by an angle @xmath139 . a dc bias is applied between the left and the right ferromagnets , which causes a difference of chemical potential of the left and the right lead by @xmath140 . in a steady state , there should be no charge accumulation in the central scattering region . so the displacement current is zero , and @xmath141 holds . for the convenience of calculation , we then define the current flowing through the system as @xmath142 according to eq.([mulcu ] ) , we need to calculate the function @xmath143 . in the steady state , we have @xmath144 where @xmath145 in addition , we have @xmath146}^{-1 } , \label{sgr}\ ] ] where @xmath147 . by noting that the pole of @xmath148 is on the lower half - plane , while the pole of the integral function of @xmath143 is on the upper half - plane ( @xmath149 ) , one can make use of the residual theorem and perform a contour integral on the upper half - plane , and obtain the function @xmath150 in a steady state : @xmath151it is obvious that the function @xmath152 is the fourier transform of the retarded green function in a steady state . from eq.([stj ] ) , we arrive at @xmath153t_{eff}(\omega ) , \label{jjj}\ ] ] where @xmath154 $ ] is the effective transmission coefficient ( etc ) , and from eq.([gamatt ] ) , @xmath155 have forms of @xmath156 @xmath157 with @xmath158 ( recall that @xmath87 is supposed to be real ) . before making the numerical calculation , it is better to specify the parameters . first , we assume that the tunnel junction under interest is symmetric , say , the two fm terminals are made of the same ferromagnets . then , we may take @xmath159 , and @xmath160 , @xmath161 . defining a parameter which is called as the spin asymmetry factor @xmath162 , which is similar but different to that in ref.@xcite , we may write @xmath163 , @xmath164 , @xmath165 with @xmath166 as an energy scale hereafter . at low temperature and a small bias , the tunnel conductance @xmath167 can be obtained from eq.([jjj ] ) : @xmath168 with @xmath169 the fermi energy . the tunnel magnetoresistance @xmath170 can be defined as @xmath171 , where @xmath172 is the tunnel conductance when the magnetizations of the two fm leads are aligned parallel . the @xmath139 dependences of the etc and @xmath170 for different @xmath173 are presented in figs . 2(a ) and ( b ) . it is seen that the etc ( @xmath170 ) increases ( decreases ) with increasing @xmath139 when @xmath174 , while it decreases ( increases ) with increasing @xmath139 when @xmath175 . @xmath170 becomes negative when @xmath173 is small . the tunnel conductance and the @xmath176 are remarkably affected by the spin asymmetry factor @xmath173 which is the ratio of @xmath177 and @xmath178 , while @xmath177 ( @xmath178 ) is the tunneling rate of electrons with spin up ( down ) from the left or right electrode to the energy level in the central island . different @xmath173 means different tunneling rate for up- and down - spin electrons . the @xmath139 dependence of tmr exhibits the conventional spin - valve effect , i.e. , the conductance is larger in the antiparallel configuration than that in the parallel configuration for larger @xmath173 . a small @xmath173 ( for example @xmath179 ) implies that @xmath180 ( i.e. the tunneling probability of electrons from the spin up subband to spin up subband is less than that from down to down subbands ) , leading to an inverse tmr ( i.e. the conductance in the antiparallel case is larger than that in the parallel case ) , as shown in fig . when @xmath173 is around @xmath181 , the @xmath176 is negligibly small and displays a nonmonotonic behavior , as shown in fig . it suggests that the coupling strength of the spin subbands leads to the nonmonotonic change of the @xmath176 . when the polarization @xmath182 , i.e. if the leads are made of half - metals , the @xmath170 will be always positive , and @xmath183 , which recovers the perfect spin - valve effect . here it should be pointed out that bratkovsky@xcite has considered a resonant tunnel diode ( rtd ) with electrons tunneling through a single - impurity level which is similar to the present structure , where , by invoking a different method , the @xmath139- and bias - dependences of the conductance were obtained . however , the asymmetry of the tunneling rates @xmath184 was not considered there while it is included in the present case , which gives rise to interesting behaviors , as manifested in fig . the effect of the spin polarization @xmath185 of the leads on the etc as well as the @xmath170 as a function of @xmath139 is presented in figs . 3(a ) and ( b ) . it can be seen that the etc and the @xmath170 are affected considerably by @xmath185 . at @xmath186 , there is a crossing point on the curve of the etc , showing that the etc , thus the tunnel conductance , does not depend on the polarization when the magnetic moments of the leads are aligned perpendicular . as @xmath187 , the etc increases with increasing the polarization at a given @xmath139 ; when @xmath188 , the etc decreases with increasing the polarization , as shown in fig . this is easy to understand , because @xmath189 . when @xmath182 and @xmath190 , the etc is zero , showing again a spin - valve phenomenon . this is because in this situation the majority ( spin - up ) subband of the left lead is fully filled , while the minority ( spin - down ) subband is empty , if the spin - flip scattering is ignored in the tunneling process , the transport of the minority electrons in the left lead tunneling into the minority subband of the right lead is prohibited@xcite . at a given @xmath139 , @xmath170 increases with increasing the polarization @xmath185 , as shown in fig . this behavior can be simply understood by the fact @xmath191\sin ^{2}\theta /2 $ ] . to be consistent with the common definition of the tmr in literature , we define a quantity @xmath192 , where @xmath193 is the tunnel conductance when the magnetizations of the ferromagnets are aligned antiparallel . in fact , such a definition is nothing but @xmath194 . in fig . 4 , we show the @xmath176 as a function of the incident energy of electrons . it can be observed that the tmr reveals different behaviors for different @xmath173 . first , for any @xmath173 there is a peak at @xmath195 , and the curves have two crossing points at @xmath196 @xmath197 and @xmath198 , suggesting that the @xmath176 at @xmath195 , @xmath197 and @xmath198 is independent of @xmath173 . this property may be caused by the resonant tunneling . second , when @xmath199 and @xmath200 , the @xmath176 changes from negative to positive with increasing @xmath173 , and larger @xmath173 , larger the @xmath176 ; when @xmath201 except @xmath195 , larger @xmath173 , smaller the @xmath176 . it is obvious that the @xmath176 is symmetrical to the axis @xmath195 . this is a characteristic of the resonant tunneling structure . so far , we have discussed the spin - dependent transport of the structure at zero temperature . let us now look at what happens at finite but low temperature . to proceed , we set @xmath202 and @xmath203 . in the rest of this subsection , we follow the treatment of rudziski and barna@xcite , and consider the bias - dependent energy level of the central region as @xmath204 , where @xmath205 , @xmath206 , @xmath207 and @xmath208 are the thickness of the left and the right barrier . this treatment is corresponding to introduce a constant hartree potential in the central scattering region as discussed in ref.@xcite . as the structure under consideration is symmetrical , we have @xmath209 in the present case . the dc bias dependence of the current and the differential conductance is depicted in figs . 5(a ) and ( b ) . ( note that throughout this paper any voltage is all scaled by @xmath210 . ) it can be observed that the current increases with increasing dc bias for different energy level of the central region . when @xmath211 , the current is a monotonic function of @xmath212 ; when @xmath8 becomes larger , the curvature of the current alters with @xmath213 , and the current rises dramatically at @xmath214 to saturation , as displayed in fig . 5(a ) . as the energy of the central spacer is presumed to be positive , its role is like a barrier . thus , one may see that with increasing @xmath8 the current is remarkably suppressed . when the external dc bias lifts the fermi level of the left lead to meet with the resonant energy level , the resonance may occur , which causes the current rapidly rising around the resonant position . this character can be clearly seen in fig . 5(b ) , in which the differential conductance exhibits resonant peaks at @xmath215 , i.e. @xmath216 , a typical character of the resonant tunneling . at low bias , the differential conductance exhibits an ohmic behavior especially for large @xmath8 , i.e. , is independent of the bias voltage ; as @xmath213 goes higher , the conductance deviates the ohmic behavior , which is more obvious when @xmath8 is small . since the appearance of the resonant tunneling , the overall behavior of the different conductance in this junction structure differs dramatically from the classical behavior , as manifested in fig . the @xmath176 as a function of the dc bias voltage for different @xmath8 is shown in fig . it is found that the tmr decreases with increasing dc bias at low bias but increases at high bias . this feature is consistent with the experimental observation@xcite although the system investigated in the experiment is a single - barrier structure . one may note that for @xmath217 or @xmath218 the @xmath176first decreases to a minimum , then slowly rises with the dc bias . before reaching the minimum , there is a shoulder structure for the @xmath176 . this behavior is also noted by wang et al.@xcite , in which they attribute it to the quantum resonance . in the present case we believe that it may be caused by the coupling between the central region and the leads . this coupling makes the sharp energy levels in the central region extended to those with a finite width , leading to that the electrons in the central region can relax to the leads . it may be the finite width of the energy levels in the central region that results in the shoulderstructure . in practice , the energy levels in the central region can be tuned by applying a gate voltage @xmath219 . we may assume that the energy levels in the central region can be modulated in a form of @xmath220 . the dependence of the current and the @xmath176 on the gate voltage @xmath219 for different bias voltages is depicted in fig . 7 , where @xmath221 . it can be observed that with increasing gate voltage the current first rises rapidly , and goes to the maximum at @xmath222 , then decreases rapidly towards zero , as shown in fig . for a large @xmath212 , the current shows a hump behavior . it can be understood as follows . when a dc bias @xmath212 is applied to the leads , the fermi energies of the left and the right leads are split by @xmath223 , i.e. @xmath224 . when the resonant energy level @xmath225 is in the region @xmath226 , the resonant energy level provides a tunneling channel and the resonance occurs , leading to that the current increases rapidly ; when @xmath227 , the energy level in the central spacer serves as a barrier and the current decreases with increasing @xmath219 . when the dc bias becomes larger , the width of the resonant peaks become wider , giving rise to a hump behavior . the gate voltage dependence of the tmr is shown in fig . when the gate voltage is small , for instance @xmath228 , the tmr goes to the minimum at @xmath222 where the resonant peaks of the current just appear . around the minimum of the tmr , there appear two `` shoulders '' on both sides of @xmath222 for different @xmath212 . when @xmath212 becomes larger , the `` shoulder '' becomes flatter . with increasing the external dc bias voltage , the minimum of the tmr becomes a round peak while the `` shoulders '' become two minima . the origin of the shoulder behavior is the same as that explained above . now we turn to discuss a three - terminal device under a dc bias . the multi - terminal device has been extensively investigated within a framework of the mesoscopic device . for instance , a three - terminal device was discussed in terms of the landauer - bttiker scattering matrix theory ( smt)@xcite , where a third terminal is introduced to measure the chemical potential @xmath229 and the current flowing out of it is set to zero , say , @xmath230 . in this way , the chemical potential of the third terminal can be expressed in terms of the chemical potentials of the first and the second terminals . when electrons enter into the third terminal and then flow out of it into the scattering region , their phases are randomized . for a spin - dependent three - terminal device , johnson@xcite observed the spin bottleneck effect in a spin transistor . in a structure such as a fm / nm / fm multilayer , the polarized current is injected from one of the fm layers into the nm layer and flows out of it ; the other fm layer and the nm layer serve as a voltmeter to give a voltage output which depends on the magnetization configurations of the two ferromagnets . johnson and silsbee@xcite proposed the nonequilibrium thermodynamic theory to investigate the charge transport and the nonequilibrium magnetization . fert and lee@xcite investigated johnson s spin transistor device and discussed the effect of junction resistance on the spin accumulation based on the boltzmann equation . brataas et al.@xcite extended the smt to a ferromagnetic three - terminal system fm / nm / fm , where the current flows from the first fm terminal into the nm spacer region , then into a second fm terminal , but the net current flowing out of the third fm terminal is set to zero . when the magnetizations in the first and the second fm terminals are antiparallel , the current flowing through the first and second terminals can be varied with the magnetization direction of the third fm terminal through which there is no net current flowing , though . when the magnetization in the third terminal @xmath231 deviates by @xmath232 to the first one @xmath233 , the current goes to its maximum . here we propose a three - terminal spintronic device which is somewhat different from the one discussed in refs.@xcite . we suppose that the third fm terminal is applied by an external dc bias too , enabling the net current to flow through it , just like a usual semiconductor transistor . we call it a spin transistor with source , whose schematic layout is shown in fig . 8 . in the following , we shall investigate the spin - dependent electrical transport in such a sts . from eq.([mulcu ] ) , the current of the @xmath5th lead is @xmath234tr_{\sigma } [ \mathbf{\gamma } _ { \alpha } \mathbf{g}_{d}^{r}(\omega ) \mathbf{\gamma } _ { \beta } \mathbf{g}_{d}^{a}(\omega ) ] , \label{thrcu}\]]where @xmath235 runs over the other leads besides @xmath5 . we re - arrange the current flowing in the leads as a vector form : @xmath236@xmath237with @xmath238a constant vector , and @xmath239where @xmath240with @xmath241 , and @xmath242where the matrix elements are given by @xmath243 $ ] with @xmath5 , @xmath235 the indices of the leads . in general , the formulas can be extended to the system with @xmath0 terminals , and the current will be a vector with @xmath0 dimensions where @xmath244 is a @xmath245 matrix . when one or some of the @xmath0 terminals are made of nm , then @xmath246 and the tunneling matrix reduces to a number . in the present assumption , @xmath244 is a symmetrical matrix , i.e. @xmath247 , @xmath248 , and @xmath249 . in the subsequent discussion , we call the parallel ( antiparallel ) configuration when the magnetizations of the first and second terminals are parallel ( antiparallel ) . we set the angle between the magnetizations of the third terminal and the first one is @xmath139 , as indicated in fig . 8 . then , the current flowing through the second terminal is @xmath250tr_{\sigma } [ \mathbf{\gamma } _ { 2}^{p}\mathbf{g}_{d}^{r , p}(\omega ) \mathbf{\gamma } _ { 1}\mathbf{g}_{d}^{a , p}(\omega ) ] \label{p2 } \\ & & + [ f_{2}(\omega ) -f_{3}(\omega ) ] tr_{\sigma } [ \mathbf{\gamma } _ { 2}^{p}\mathbf{g}_{d}^{r , p}(\omega ) \mathbf{\gamma } _ { 3}\mathbf{g}_{d}^{a , p}(\omega ) ] \ } , \notag\end{aligned}\ ] ] where the superscript @xmath185 means the parallel configuration . in the antiparallel configuration , the current flowing through the second terminal is @xmath251tr_{\sigma } [ \mathbf{\gamma } _ { 2}^{ap}\mathbf{g}_{d}^{r , ap}(\omega ) \mathbf{\gamma } _ { 1}\mathbf{g}_{d}^{a , ap}(\omega ) ] \label{ap2 } \\ & & + [ f_{2}(\omega ) -f_{3}(\omega ) ] tr_{\sigma } [ \mathbf{\gamma } _ { 2}^{ap}\mathbf{g}_{d}^{r , ap}(\omega ) \mathbf{\gamma } _ { 3}\mathbf{g}_{d}^{a , ap}(\omega ) ] \ } , \notag\end{aligned}\]]where the superscript @xmath252 means the antiparallel configuration . at low temperature and small bias voltage , the difference of the conductance in these two configurations can be obtained by @xmath253 , \notag\end{aligned}\]]define the @xmath254 as usual @xmath255}{tr_{\sigma } [ \mathbf{\gamma } _ { 2}^{p}\mathbf{g}_{d}^{r , p}(e_{f},\theta _ { 3}=0)\mathbf{\gamma } _ { 1}\mathbf{g}^{a , p}(e_{f},\theta _ { 3}=0)]}. \notag\end{aligned}\ ] ] we shall consider the sts device at finite temperature . the terminal current is determined by @xmath256 . negative @xmath257 indicates that the positive charges tunnel through the barrier and enter into the central region ( i.e. the electrons tunnel through the barrier from the central region to the leads ) , and the minus means the reduction of the positive charges ( i.e. the increase of the negative charges ) in the @xmath5th terminal . when @xmath257 is positive , the things become opposite . for simplicity , we set @xmath258 and @xmath259 , while choose @xmath260 . in this case , the electrons flow from the first and third terminals to the second terminal , suggesting that the positive charges tunnel through the barrier into the central region , and @xmath261 is negative . as the flowing direction of the positive charges is the current direction , and noting @xmath262 , i.e. @xmath263 , we define the current flowing from the first and third terminal into the second terminal as @xmath264\{tr_{\sigma } [ ( \mathbf{\gamma } _ { 1}+\mathbf{\gamma } _ { 3})\mathbf{g}_{d}^{r}(\omega ) \mathbf{\gamma } _ { 2}\mathbf{g}_{d}^{a}(\omega ) ] \\ & & + tr_{\sigma } [ \mathbf{\gamma } _ { 2}\mathbf{g}_{d}^{r}(\omega ) ( \mathbf{\gamma } _ { 1}+\mathbf{\gamma } _ { 3})\mathbf{g}_{d}^{a}(\omega ) ] \}. \notag\end{aligned}\]]based on this equation , we could obtain the current @xmath265 at the parallel ( antiparallel ) configuration numerically . the @xmath139 dependence of @xmath266 under different polarizations is shown in fig . 9(a ) , where the parameters are assumed the same as those in the proceding subsection , the temperature is taken as @xmath267 , and the three terminals are supposed to be made of the same ferromagnets . in the parallel configuration , the current @xmath268 decreases with increasing @xmath139 . this is because of the spin - valve effect , the current flowing from the third terminal into the second terminal decreases with increasing @xmath139 , leading to the total current into the second terminal decreases with increasing @xmath139 . while in the antiparallel case , the current flowing from the third terminal into the second terminal increases with increasing @xmath139 because @xmath231 rotates from the antiparallel to parallel configuration with respect to @xmath269 , resulting in that the total current of the second terminal increases with @xmath139 . we note that the current in the antiparallel configuration is equal to that in the parallel configuration when @xmath270 . because in this case , i.e. the magnetizations of the first and third terminals are antiparallel , the state of @xmath271 but @xmath231 antiparallel to @xmath269 is symmetrical to the state of @xmath272 but @xmath269 antiparallel to @xmath233 , the current in the two states are equal , and the tmr would be zero . besides , we find that regardless of the magnetization configuration , the polarization makes the absolute value of the current increases except @xmath270 . to illustrate the effect of the tunnel magnetoresistance , for this three - terminal device we define the current ratio as @xmath273 this quantity reflects the relative change of the tunnel current with respect to the change of the magnetization configuration , whose role is similar to the tmr . the angle dependence of the current ratio @xmath274 for the second terminal is shown in fig . it can be noted that @xmath275 is negative for the parallel configuration and the absolute value of @xmath275 increases with increasing @xmath139 . for the antiparallel configuration , @xmath276 is positive and increases with increasing @xmath139 . @xmath277 is larger than the absolute value of @xmath278 . this difference becomes larger for a larger polarization . for example , @xmath279 , @xmath280 @xmath281 . it suggests that the change of the current for the antiparallel configuration is larger than for the parallel configuration . so , the current flowing out of the second terminal can be enlarged or reduced by changing the relative orientation of the magnetic moments of the third terminal . the so - obtaind current amplification effect is considerably large for a larger polarization at the antiparallel configuration . for example , @xmath282 , the change of the current ratio can be as high as @xmath283 . such a current amplification effect is quite different from that in a usual semiconductor transistor . the present results show that in a sts , by merely changing the magnetization direction of the third terminal the current flowing through the first and second terminals can be amplified or reduced . besides , we can define another current ratio , @xmath284 , for the sts to investigate the change of the current flowing out of the second terminal when the magnetization configuration varies from parallel to antiparallel : @xmath285the @xmath139 dependence of the @xmath284 for different polarizations is shown in fig . 10 . one may see that the tmr decreases with increasing @xmath139 . as mentioned above , the states of the parallel and antiparallel configuration are symmetrical at @xmath270 , leading to the currents equal , and then giving rise to the @xmath286 zero . when the polarization is larger , the tmr becomes larger . here we would like to point out that the present three - terminal device is different from that discussed by brataas et al.@xcite . in their device , the third terminal is introduced as an inelastic scattering resource , and there is no net current flowing through this terminal . in the present sts device , there is net current flowing through the third terminal , and the chemical potential of the third terminal is not changed ( i.e. its chemical potential can not be expressed in terms of that of the first and second terminals ) . owing to the spin - dependent scatterings , we have found that the current flowing out of the second terminal can be enlarged by changing @xmath139 when the magnetization configuration of the first and second terminals is antiparallel . the enlargement is more remarkable with increasing polarization . this three - terminal device is similar to the usual semiconductor transistor . in the latter case , the current flowing out of the collector is enlarged by varying the voltage bias between the emitter and the base , while in the present case the enlargement of the current can be realized by changing the direction of the magnetization of the third terminal . this is of course of the quantum - mechanical origin . another difference between our device and that of brataas et al . is that the current can be modified in a sts by changing @xmath139 at the parallel configuration , but it can not be tuned in the case of refs.brataas,brataas1 . however , the result presented in refs.@xcite can be recovered from our analysis . suppose that there is no net current flowing through the third terminal in the sts , like the treatment of bttiker . this implies that the chemical potential of the third terminal @xmath229 can be determined by the chemical potentials of the other two terminals . as a result , we get @xmath287\frac{c_{12}(c_{13}+c_{23})+c_{13}c_{23}}{c_{13}+c_{23}}. \label{j30}\]]in addition , it is assumed that @xmath288 is applied to the first terminal and @xmath289 to the second terminal . the current ratio @xmath290 can be obtained in terms of eq . ( [ j30 ] ) , and the result is shown in fig . one may find that our result is fairly consistent with that shown in refs.@xcite , although quite different methods are used . let us turn to investigate the spin - dependent transport in the magnetic tunnel junction with two terminals under an ac bias , where the noncollinear configuration of magnetizations of the fm leads is presumed . under an ac bias , the total current of each terminal is a sum of the terminal current and the displacement current with the ac bias determined by eqs.([le ] ) and ( [ ce ] ) . the total current is @xmath291 . \label{ttca}\]]again , we need to calculate the function @xmath292 . from eq.([grdt ] ) , we have @xmath293}\mathbf{g}_{d}^{r}(\omega ^{\prime \prime } ) \notag \\ & = & i\sum_{mnm^{\prime } n^{\prime } } j_{\alpha } ( _ { m^{\prime } n^{\prime } } ^{mn})\int \frac{d\omega ^{\prime \prime } } { 2\pi } \frac{\mathbf{g}_{d}^{r}(\omega ^{\prime \prime } ) e^{-i[(m - n)\omega _ { \alpha } + ( m^{\prime } -n^{\prime } ) \omega _ { 0}]t}}{\omega -\omega ^{\prime \prime } + m\omega _ { \alpha } -n^{\prime } \omega _ { 0}+i0^{+ } } , \label{afb}\end{aligned}\ ] ] with @xmath294 , where use has been made of @xmath295 with @xmath296 is the @xmath297-th order bessel function . in the same way , we can get @xmath298t}}{(\omega -\omega ^{\prime \prime } + m\omega _ { \alpha } -n^{\prime } \omega _ { 0}+i0^{+})(\omega -\omega ^{\prime } + n\omega _ { \alpha } -m^{\prime } \omega _ { 0}-i0^{+})}. \notag\end{aligned}\]]the time - average of a time - dependent physical quantity @xmath299 is defined by @xmath300the time - averaged displacement current can be in principle obtained by means of eqs.([dd ] ) and ( [ bs ] ) . although the integral on time can be nonzero only when the term in the square bracket of the exponent in eq.(bs ) is zero , to get the displacement current a differential of eq.([bs ] ) with respect to @xmath301 should be made , leading to that the time - averaged displacement current is zero . this is also mentioned in ref.@xcite . if @xmath299 is a periodic function of time , its integral over time should be finite . when @xmath302 tends to infinity , the time - averaged @xmath299 will be zero . when @xmath303 holds , the first term of the time - averaged @xmath5th terminal current in eq.(mulcu ) is @xmath304}{(\omega ^{\prime \prime } -\omega -m\omega _ { \alpha } + n^{\prime } \omega _ { 0}-i0^{+ } ) } , \label{ajc1}\]]and the second term is @xmath305}{(\omega ^{\prime \prime } -\omega -m\omega _ { \alpha } + n^{\prime } \omega _ { 0}-i0^{+})(\omega -\omega ^{\prime } + n\omega _ { \alpha } -m^{\prime } \omega _ { 0}-i0^{+})}. \notag\end{aligned}\]]the time - averaged terminal current is zero for other cases . to get @xmath306 , we shall perform the integral over @xmath307 first . by noting that the pole of @xmath308 is on the lower - half plane , while the pole of the kernel function is on the upper - half plane @xmath309 , we may perform a contour integral along the upper - half plane , and get @xmath310 , \label{ijc}\]]where @xmath311^{-1}$ ] , and @xmath312 . to evaluate @xmath313 , one should first perform the integral over @xmath307 , then make a contour integral on the lower - half plane , and get @xmath314where @xmath315^{-1}$ ] . as @xmath316 , leading to @xmath317 , we write @xmath318 as @xmath319 . after some algebra , the time - averaged current of the @xmath5th terminal can be obtained @xmath320 \label{jalc } \\ & & -[\sum_{\beta } j_{\beta } ( _ { m^{\prime } n^{\prime } } ^{mn})f_{\beta } ( \omega ) \mathbf{\gamma } _ { \alpha } \mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{\beta } ) \mathbf{\gamma } _ { \beta } \mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{\beta } ) ] \ } , \notag\end{aligned}\]]for the time - averaged displacement current is zero , by noting @xmath321 , one may obtain @xmath322 \\ & & + tr_{\sigma } [ \mathbf{\gamma } _ { r}\mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{l})\mathbf{\gamma } _ { l}\mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{l } ) ] ) \\ & & -j_{r}(_{m^{\prime } n^{\prime } } ^{mn})f_{r}(\omega ) ( tr_{\sigma } [ \mathbf{\gamma } _ { r}\mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{r})\mathbf{\gamma } _ { l}\mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{r } ) ] \\ & & + tr_{\sigma } [ \mathbf{\gamma } _ { l}\mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{r})\mathbf{\gamma } _ { r}\mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{r})])\}.\end{aligned}\]]define @xmath323 , then @xmath324 . because @xmath325 , @xmath326 , @xmath327 and @xmath328 are all symmetrical matrices , we find @xmath329 . with this observation , we get @xmath330 \label{cuact } \\ & & -j_{r}(_{m^{\prime } n^{\prime } } ^{mn})f_{r}(\omega ) tr_{\sigma } [ \mathbf{\gamma } _ { r}\mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{r})\mathbf{\gamma } _ { l}\mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{r})]\}. \notag\end{aligned}\]]this is the time - averaged current of the system with two fm terminals whose magnetizations are arranged in a noncollinear configuration under an ac bias , which can be used to study the spin - dependent transport of the spin - polarized electrons in the presence of an ac field . in this subsection , we shall consider a particular situation in which only a dc bias is applied between the two terminals while an ac bias voltage is applied to the central spacer region such as @xmath331 . in this case , @xmath332 , @xmath333 and @xmath334 . the time - averaged current becomes @xmath335tr_{\sigma } [ \mathbf{\gamma } _ { l}\mathbf{g}_{d}^{r}(\omega , \varepsilon _ { 0}^{\prime } ) \mathbf{\gamma } _ { r}\mathbf{g}_{d}^{a}(\omega , \varepsilon _ { 0}^{\prime } ) ] , \label{twoacc}\]]where @xmath336 is the @xmath297-th order bessel function . the effective transmission coefficient is given by @xmath337 $ ] . at low temperature and under a small dc bias , the integral can be performed by noting @xmath338 . as a result , we can get the tunnel conductance @xmath339 as well as the @xmath340 @xmath341 . based on these equations , the numerical calculation can be carried out . the parameters are taken as those in sec . the etc and the @xmath176 as a function of @xmath8 is shown in figs . 12(a)-(d ) . it is seen that the etc and the tmr exhibit resonant peaks at certain values of @xmath342 , and the curves are symmetric to the axis @xmath343 . these resonant peaks occur at the positions of the photonic sidebands characterized by @xmath344 , @xmath345 , @xmath346 , @xmath347 which correspond to the shifts of the energy levels caused by the ac field in the central spacer . we have found that the resonant peaks of the etc become sharper and higher as the orientations of the magnetizations of the left and the right ferromagnetic leads vary from antiparallel to parallel . with increasing @xmath139 , the amplitudes of the resonant peaks are noticed to become small . in addition , we have observed that the resonant magnitude of the etc is sensitive to the ratio of the strength and the frequency of the ac field , say , @xmath348 , which appears as an argument of the bessel function . with increasing @xmath5 , the main resonant peak of the etc at @xmath343 is suppressed , while more resonant peaks with large @xmath349 appear , as manifested in figs . 12(a ) , ( b ) and ( d ) . the resonant peaks of the tmr does not almost change with @xmath5 . the @xmath8 dependence of the etc and the tmr for different @xmath173 is presented in figs . 12(c ) and ( d ) . one may observe that a small @xmath173 suppresses the resonant peaks while broadens the peaks of the etc and the tmr . with increasing @xmath173 , the resonant peaks become more and obvious . in accordance with the definition , a larger @xmath173 means that it is more difficult for the electrons with spin down to tunnel through the barrier and then into the leads , while it is easy for the electrons with spin up . so , the resonant peaks of the etc become even sharper at smaller @xmath173 in the antiparallel configuration , while the resonant peaks of the tmr become more obvious . one may note that there are several other peaks of the tmr besides the main resonance and the photonic sidebands as shown in fig . 12(c ) , which are not caused by the additional resonant energy levels of the central region , but by the asymmetric factor of @xmath173 on the spin - dependent transmission . the tunneling process assisted by photons is schematically illustrated in fig . it can be seen that the photonic sidebands provide new channels for the electrons in the left ferromagnet to tunnel through the central spacer and into the right ferromagnet . without the spin - flip scattering , the electrons with spin up ( down ) in the left will be accepted by the spin - up ( -down ) subband in the right . because of the splitting of the energy bands of electrons with different spins , the spin - valve effect will appear . as @xmath8 increases , the photonic sidebands with negative @xmath297 , the main resonant energy level and the photonic sidebands with positive @xmath297 pass through the fermi level of the leads one by one . it is these additional channels from the photonic sidebands to make the etc and the tmr be resonant . the @xmath5 dependence of the etc and the tmr for different @xmath8 is shown in figs . 14(a ) and ( b ) . one may observe that ( i ) the etc oscillates damply with increasing @xmath58 , and a larger @xmath5 may suppress the transmission ; ( ii ) the tmr oscillates with increasing @xmath5 ; ( iii ) the peaks are sharper and more obvious when @xmath8 closes to the main resonance and the photonic sidebands ; ( iv ) when @xmath8 leaves the main resonance energy to a lower value , the first resonant peak of the etc and the tmr moves towards the direction of larger @xmath5 s . these characters are the consequences of the photonic sidebands induced by the ac field . here we should point out that the scattering mechanisms of electrons tunneling process assisted by phonon and by photon are fundamentally different . interestingly enough , however , the features of photon - assisted and optical phonon - assisted tunneling current and conductance are similar , because both tunneling processes are assisted by sidebands induced by photons or optical phonons . in the conventional semiconductor tunnel diodes , the tunneling transitions are dominant by emission of phonon at low temperature ( see ref.@xcite ) , while the present tunneling transitions are contributed by emission or absorption photon quantum . this photon - assisted tunneling in superconductor tunnel junctions was discussed in ref.@xcite . the resonant level is modulated not just by a displacement by emitting or absorbing photons , but is modulated in terms of a probability amplitude which is characterized by the square of the @xmath350th bessel function @xmath351 for each level to be displaced in energy by @xmath352 . these probability amplitudes contain the information about the quantum interference . the tunneling probability of electrons via these virtual energy levels may be modulated by these quantum interferences . in addition , the photon - assisted tunneling can be controlled by tuning an external ac field , but the phonon - assisted tunneling is primarily determined by the material . in the present case , the photon - assisted spin - dependent resonant tunneling is considered and the photon - assisted tmr is observed . this effect may be used to give a tunable tmr , which may be useful for future application . recently , we have noted that a similar phonon - assisted tunneling through a molecular quantum dot was considered in ref.@xcite . we have discussed the spin - dependent transport in a resonant tunneling structure with ferromagnetic multi - terminal under dc and ac fields by means of the nonequilibrium green function technique . the general formulation of the time - dependent spintronic transport in this structure in the presence of ac and dc fields has been established , which might offer a fundamental basis for further discussions on the spin - dependent transport in such spintronic devices in an ac field . first , we have considered the resonant system with two fm terminals in which the magnetizations of the leads are noncollinearly aligned under a dc bias voltage , and have found that for small @xmath173 s where @xmath173 characterizes the asymmetry of the relaxation times of the electrons with different spin in the central region , the tmr is negative , and whose absolute magnitude increases with increasing @xmath139 ( the relative angle between the magnetizations of both leads ) ; when @xmath173 becomes larger , the tmr will be positive and increases with increasing @xmath139 . we have also investigated the tmr as a function of energy , the dc bias voltage and the gate voltage for different polarizations and energy levels of the central scattering region , respectively . the results are diverse , which manifest the effect of the resonant energy level on the tmr and the tunneling current , as discussed in detail in sec.ii . second , we have considered a three - terminal device under a dc bias , which is different from the standard three - terminal device discussed elsewhererefs.@xcite . in the present device , each terminal is applied with a source bias , suggesting that there is net current flowing through every terminal . regardless of the spin - flip scattering , the electrons in up - spin subbands in one terminal will be accepted by the up - spin subbands in another terminal . when the magnetizations of the first and second terminals are antiparallel , the change of @xmath139 which is the relative angle between the magnetizations of the first and third terminals may considerably change the current flowing out of the second terminal . it turns out that tuning @xmath139 the current can be remarkably enhanced , giving rise to the so - called magnetization configuration - induced enhancement of the current or the current ratio @xmath290 . it has been found that the @xmath284 , which is defined in this case by the current change ratio when the configuration of the magnetizations of the first and second terminals is turned from parallel to antiparallel , decreases with increasing @xmath139 . third , as an application of our general formulation we have investigated the time - dependent spintronic transport in a system with two terminals under an ac field . we have uncovered that , when a very small dc bias is applied to the two terminals but an ac field is applied to the central scattering region , the photonic sidebands are formed in the central region . when the sidebands meet with the fermi energy of the terminals , the photon - assisted tunneling will occur . it manifests as the resonant peaks in the etc versus @xmath8 . the tmr exhibits many resonant peaks at the main resonant level and the photonic sidebands . it has been found that the asymmetric factor @xmath173 may lead to additional peaks besides the photon - assisted resonant peaks . we would like to mention that the present study might open a way to control the spin - dependent transport in a spintronic device by applying a time - dependent electrical field . this work is supported in part by the national science foundation of china ( grant nos . 90103023 , 10104015 , 10247002 ) . r. meservey and p.m. tedrow , phys . rep . * 238 * , 173 ( 1994 ) ; m. a. m. gijs and g. e. w. bauer , adv . phys . * 46 * , 285 ( 1997 ) ; j. s. moodera , j. nassar and g. mathon , annu . * 29 * , 381 ( 1999 ) ; s. a. wolf et al , science * 294 , * 1488 ( 2001 ) ; and references therein . the spin accumulation can be estimated by calculating the difference of the diagonal elements of the matrix of the lesser green function in spin space , say , @xmath354_{11}-[g_{d}^{<}(t , t)]_{22}\}$ ] , where the definitions of the lesser green function matrix @xmath355 can be found in the following context . we shall study it elsewhere .

we use the theory of tunnel number knots introduced in to strengthen the tunnel leveling theorem of goda , scharlemann , and thompson . this yields considerable information about bridge numbers of tunnel number knots . in particular , we calculate the minimum bridge number of a knot as a function of the maximum depth invariant of its tunnels . the growth of this value is on the order of , which improves known estimates of the rate of growth of bridge number as a function of the hempel distance of the associated heegaard splitting . we also find the maximum bridge number as a function of the number of cabling constructions needed to produce the tunnel , showing in particular that the maximum bridge number of a knot produced by cabling constructions is the fibonacci number . finally , we examine the special case of the `` middle '' tunnels of torus knots .

0812.1396

the tunnel leveling theorem of h. goda , m. scharlemann , and a. thompson @xcite says that when a tunnel number one knot is in minimal bridge position , any of its tunnel arcs can be slid to lie in a single horizontal level . using the theory of tunnel number @xmath0 knots developed in @xcite , we will prove the _ tunnel leveling addendum . _ roughly speaking , it says that when a tunnel arc is in level position as in the conclusion of the tunnel leveling theorem , the other two knots from the @xmath5-curve which is the union of the knot and its tunnel arc are also ( after trivial repositioning ) in minimal bridge position . its full statement is given near the start of section [ sec : efficient ] . the tunnel leveling addendum gives a great deal of information about bridge numbers of tunnel number @xmath0 knots . some of these applications involve the _ depth _ invariant , which is defined using the theory from @xcite . the depth of a tunnel , @xmath6 , is somewhat similar to the ( hempel ) distance @xmath7 ( see j. johnson @xcite and y. minsky , y. moriah , and s. schleimer @xcite ) , but unlike the distance , the depth is very easy to calculate in terms of the parameter description of tunnels given in @xcite . the two invariants are related by the inequality @xmath8 but the depth can be much larger than the distance . indeed , the `` middle '' tunnels of torus knots that we examine below are easily seen to have distance @xmath9 , but we will see their depths can be arbitrarily large . the depth invariant can be defined geometrically in terms of the cabling constructions of @xcite , but it also has a geometric interpretation in terms of a construction that first appeared in @xcite . that construction , which we call a giant step , is studied in @xcite . there is no upper bound for the bridge number of a knot in terms of the depths of its tunnels , but among our applications of the tunnel leveling addendum is a sharp lower bound : for @xmath10 , the minimum bridge number of a knot having a tunnel of depth @xmath1 is given recursively by @xmath11 , where @xmath12 , @xmath13 , and @xmath14 for @xmath15 . explicitly , @xmath16 and consequently @xmath17 . this improves lemma 2 of @xcite , which is that bridge number grows at least linearly with distance . it also improves proposition 1.11 of @xcite , which implies that bridge number grows at least as fast as @xmath18 . actually , the minimum bridge number theorem can be proven using only the tunnel leveling theorem , which give the lower bounds , and some explicit constructions to realize the minimum values . our upper bound result , however , uses the full strength of the tunnel leveling addendum : write the fibonacci sequence @xmath19 as @xmath20 . the maximum bridge number of a knot having a tunnel produced by @xmath3 cabling constructions , of which the first @xmath21 produce simple or semisimple tunnels , is @xmath22 . ( the terms `` simple '' and `` semisimple '' are recalled in section [ sec : cabling ] . ) for fixed @xmath3 , the largest value for the upper bound in theorem [ thm : max ] occurs when @xmath23 , giving the following absolute maximum : the maximum bridge number of a knot having a tunnel produced by @xmath3 cabling operations is @xmath24 . in fact , this maximum bridge number is achieved by a sequence of torus knot tunnels , as we will see in proposition [ prop : depth - efficienttorusknots ] . in addition to giving general bounds , the tunnel leveling addendum places very strong restrictions on the possible bridge numbers that can occur : suppose that a knot @xmath25 has a tunnel @xmath26 produced by @xmath27 cabling operations , of which the first @xmath21 produce simple or semisimple tunnels . then @xmath28 is one of the @xmath29 values @xmath30 for @xmath31 . here , @xmath32 is the _ fibonacci function _ of @xmath26 , defined in section [ sec : fibonacci_functions ] . it appears almost certain that all @xmath29 possible values in the bridge number set theorem do occur as bridge numbers . as explained in remark [ rem : bridge_number_set ] , this is easy to see for @xmath23 and @xmath33 , but for the general case we have not been able to verify all the necessary examples . we will also examine the interesting case of the `` middle '' tunnels of torus knots . in our paper @xcite , we calculated the invariants of @xcite for all torus knot tunnels . using that information , we will show that torus knot tunnels achieve the minimum rate of growth of bridge numbers in terms of depth , but not the minimum possible values , while they do achieve the maximum possible bridge numbers in terms of the number of cabling constructions . here is an outline of the sections of the paper . the first two sections constitute a concise review of the material from @xcite that we will need for the present applications . section [ sec : ddd ] introduces the distance and depth invariants , and gives a few results that follow quickly from @xcite and work of other authors . section [ sec : tunnel_leveling ] reviews the tunnel leveling theorem , and section [ sec : efficient ] states and proves the tunnel leveling addendum . fibonacci functions are introduced in section [ sec : fibonacci_functions ] , which also contains the more technical results on bridge number , including the bridge number set theorem . the minimum and maximum bridge number theorems are proved in section [ sec : growth ] , and torus knot tunnels are studied in section [ sec : torus_knots ] . finally , most of the results apply to the case of tunnel number @xmath0 links , and in section [ sec : links ] , we briefly discuss these adaptations . let @xmath34 be a genus @xmath9 orientable handlebody , regarded as the standard unknotted handlebody in @xmath35 . for us , a _ disk in h _ means a properly imbedded disk in @xmath34 , _ which is assumed to be nonseparating unless otherwise stated . _ the _ disk complex _ @xmath36 is the @xmath9-dimensional , contractible simplicial complex whose vertices are the isotopy classes of disks in @xmath34 , such that a collection of @xmath37 vertices spans a @xmath38-simplex if and only if they admit a set of pairwise - disjoint representatives . each @xmath0-simplex of @xmath36 is a face of countably many @xmath9-simplices . as suggested by figure [ fig : subdivision ] , @xmath36 grows outward from any of its @xmath9-simplices in a treelike way . in fact , it deformation retracts to the tree @xmath39 seen in figure [ fig : subdivision ] . and the tree @xmath39 . countably many @xmath9-simplices meet along each edge.,width=170 ] each disk @xmath26 in @xmath34 is the cocore disk of a tunnel of the knot @xmath40 which is a core circle of the solid torus obtained by cutting @xmath34 along @xmath26 . on the other hand , each tunnel of a tunnel number 1 knot in @xmath35 determines a collection of disks in @xmath34 as follows . the tunnel is a @xmath0-handle attached to a regular neighborhood of the knot to form an unknotted genus-@xmath9 handlebody . an isotopy carrying this handlebody to @xmath34 carries a cocore @xmath9-disk of that @xmath0-handle to a nonseparating disk in @xmath34 , and carries the tunnel number @xmath0 knot to a core circle of the solid torus obtained by cutting @xmath34 along the image disk in @xmath34 . the indeterminacy of this disk due to the choice of isotopy is the group of isotopy classes of orientation - preserving homeomorphisms of @xmath35 that preserve @xmath34 . this group is called the _ goeritz group @xmath41 . _ work of m. scharlemann @xcite and e. akbas @xcite proves that @xmath41 is finitely presented , and even provides a simple presentation of it . since two disks in @xmath34 determine equivalent tunnels exactly when they differ by an isotopy moving @xmath34 through @xmath35 , _ the collection of all ( equivalence classes of ) tunnels of all tunnel number @xmath0 knots corresponds to the set of orbits of vertices of @xmath36 under @xmath41 . _ so it is natural to examine the quotient complex @xmath42 , which is illustrated in figure [ fig : delta ] . and @xmath43 near the primitive orbits.,width=151 ] through work of the first author @xcite , the action of @xmath41 on @xmath36 is well - understood . a disk @xmath44 in @xmath34 is called _ primitive _ if there is a disk @xmath45 in @xmath46 for which @xmath47 and @xmath48 intersect transversely in one point in @xmath49 . the primitive disks ( regarded as vertices ) span a contractible subcomplex @xmath50 of @xmath36 , called the _ primitive subcomplex_. the action of @xmath41 on @xmath50 is as transitive as possible , indeed the quotient @xmath51 is a single @xmath9-simplex @xmath52 which is the image of any @xmath9-simplex of the first barycentric subdivision of @xmath50 . the vertices of @xmath52 are @xmath53 , the orbit of all primitive disks , @xmath54 , the orbit of all pairs of disjoint primitive disks , and @xmath55 , the orbit of all triples of disjoint primitive disks . on the remainder of @xmath36 , the stabilizers of the action are as small as possible . a @xmath9-simplex which has two primitive vertices and one nonprimitive is identified with some other such simplices , then folded in half and attached to @xmath52 along the edge @xmath56 . the nonprimitive vertices of such @xmath9-simplices are exactly the disks in @xmath36 that are disjoint from some primitive pair , and these are called @xmath57 disks . as tunnels , they are the upper and lower tunnels of @xmath9-bridge knots . the remaining @xmath9-simplices of @xmath36 receive no self - identifications , and descend to portions of @xmath42 that are treelike and are attached to one of the edges @xmath58 where @xmath59 is simple . the tree @xmath39 shown in figure [ fig : subdivision ] is constructed as follows . let @xmath60 be the first barycentric subdivision of @xmath36 . denote by @xmath39 the subcomplex of @xmath60 obtained by removing the open stars of the vertices of @xmath36 . it is a bipartite graph , with `` white '' vertices of valence @xmath61 represented by triples and `` black '' vertices of ( countably ) infinite valence represented by pairs . the valences reflect the fact that moving along an edge from a triple to a pair corresponds to removing one of its three disks , while moving from a pair to a triple corresponds to adding one of infinitely many possible third disks to a pair . the possible disjoint third disks that can be added are called the _ slope disks _ for the pair . the image @xmath62 of @xmath39 in @xmath63 is a tree @xmath43 . the vertices of @xmath63 that are images of vertices of @xmath36 are not in @xmath43 , but their links in @xmath63 are subcomplexes of @xmath43 . these links are known to be infinite trees . for each such vertex @xmath26 of @xmath63 , i. e. each tunnel , there is a unique shortest path in @xmath43 from @xmath55 to _ the vertex in the link of @xmath26 that is closest to @xmath55 . _ this path is called the _ principal path _ of @xmath26 , and this closest vertex is a triple , called the _ principal vertex _ of @xmath26 . the two disks in the principal vertex , other than @xmath26 , are called the _ principal pair _ of @xmath26 . they are exactly the disks called @xmath64 and @xmath65 that play a key role in @xcite . figure [ fig : ppath ] below shows the principal path of a certain tunnel . the white vertices of @xmath43 correspond to unknotted @xmath5-curves in @xmath35 , up to isotopy . for a white vertex gives a triple of nonseparating disks , dual to a @xmath5-curve in @xmath34 in which each arc crosses one of the disks and not the others . these are exactly the unknotted @xmath5-curves , in that a regular neighborhood is isotopic to @xmath34 which is part of a heegaard splitting of @xmath35 . two such @xmath5-curves in @xmath34 are isotopic in @xmath35 exactly when they are equivalent under the goeritz group , so the white vertices of @xmath43 give the isotopy classification . in a sentence , the cabling construction is to `` think of the union of @xmath25 and the tunnel arc as a @xmath5-curve , and rationally tangle the ends of the tunnel arc and one of the arcs of @xmath25 in a neighborhood of the other arc of @xmath25 . '' we sometimes call this `` swap and tangle , '' since one of the arcs in the knot is exchanged for the tunnel arc , then the ends of other arc of the knot and the tunnel arc are connected by a rational tangle . figure [ fig : cabling ] illustrates two cabling constructions , one starting with the trivial knot and obtaining the trefoil , then another starting with the tunnel of the trefoil . more precisely , begin with a triple @xmath66 , where @xmath67 is the principal pair of @xmath26 . choose one of the disks of @xmath67 , say @xmath68 , and a slope disk @xmath69 of the pair @xmath70 , _ other than @xmath71 . _ this is a cabling operation producing the tunnel @xmath69 from @xmath26 . the principal vertex of @xmath69 is @xmath72 . unless otherwise stated , the slope disk @xmath69 is chosen to be nonseparating in @xmath34 . a cabling operation using a separating disk as @xmath69 produces a tunnel number @xmath0 link , and the cabling process can not be continued . this case will be discussed in section [ sec : links ] . theorem 13.2 of @xcite shows that every tunnel of every tunnel number @xmath0 knot can be obtained by a uniquely determined sequence of cabling constructions . a tunnel @xmath59 produced from the tunnel of the trivial knot by a single cabling construction is called a _ simple _ tunnel . as already noted , these are the `` upper and lower '' tunnels of @xmath9-bridge knots . a tunnel is called _ semisimple _ if it is disjoint from a primitive disk , but not from any primitive pair . a @xmath73-knot is a knot that can be put into @xmath0-bridge position with respect to a heegaard torus of @xmath35 . let @xmath25 be a @xmath73-knot , whose heegaard torus splits @xmath35 into two solid tori @xmath74 and @xmath75 . associated to this @xmath73-position are two tunnels obtained as follows . let @xmath76 be an arc in @xmath74 with endpoints in @xmath25 , such that the union of @xmath76 with the arc in @xmath77 bounded by the endpoints of @xmath76 is a core circle of @xmath74 . then @xmath76 determines a tunnel of @xmath25 ; the corresponding tunnel constructed in @xmath75 is the other one . tunnels arising in this way are called @xmath73-tunnels , and are exactly the simple and semisimple tunnels . a tunnel is called _ regular _ if it is not primitive , simple , or semisimple . there is a procedure for assigning rational slopes which record the rational tangle used in a cabling construction . we will not need these slopes in our study of depth , although we will include them , in an inessential way , in our discussion of the torus knot examples in section [ sec : torus_knots ] , and they also appear briefly in remark [ rem : bridge_number_set ] . the slope invariants are usually not needed for working with depth because the depth is completely determined by the second set of invariants associated to a tunnel , the `` binary '' invariants @xmath78 , @xmath79 , which we now define . we have already mentioned that for every tunnel @xmath26 , there is a unique sequence of tunnels @xmath80 , @xmath81 such that @xmath59 is simple and for each @xmath82 , @xmath83 is obtained from @xmath84 by a cabling construction . the cabling that produces @xmath83 retains one arc of the associated knot @xmath85 of @xmath84 , and replaces the other with a tangle , producing @xmath86 . the invariant @xmath87 is @xmath0 exactly when this cabling replaces the arc that was retained by the previous cabling , otherwise @xmath87 is @xmath88 . a tunnel is simple or semisimple if and only if all @xmath89 . the reason is that both conditions characterize cabling sequences in which one of the original primitive disks is retained in every cabling ; this corresponds to the fact that the union of the tunnel arc and one of the arcs of the knot is unknotted . there are two formal definitions of the binary invariants . the first is in terms of the principal path @xmath55 , @xmath54 , @xmath90 , @xmath91 , @xmath92 , @xmath93 , where the @xmath94 are the `` black '' vertices , the @xmath95 are the `` white '' vertices , and @xmath96 : @xmath89 or @xmath97 according to whether or not the unique disk of @xmath98 equals the unique disk of @xmath99 . equivalently , each cabling operation begins with a triple of disks @xmath100 and finishes with @xmath101 . for @xmath102 , put @xmath97 if @xmath103 , and @xmath89 otherwise . figure [ fig : ppath ] shows the principal path of a tunnel with binary invariants @xmath104 . having binary invariants @xmath104 , or equivalently with step sequence `` drrrdrdllldldrr''.,width=188 ] from the viewpoint of a traveler along the principal path , @xmath97 means a change from making right turns ( at the white vertices ) to left turns , or from left turns to right , while @xmath89 means a turn in the same direction as the previous turn . let us say that a _ step _ of the principal path is a portion between successive white vertices . a principal path can then be described as a _ step sequence . _ this is a string of symbols `` l '' , `` r '' , or `` d '' , for `` left '' , `` right '' , and `` down '' as seen from the reader s viewpoint ( as opposed to the `` left '' and `` right '' of a traveler along the path ) . for the example of figure [ fig : ppath ] , the step sequence is `` drrrdrdllldldrr '' . in general , the initial step of a principal path is always `` d '' , and the second step , due to the standard way that we draw the picture , is `` r '' . each subsequent step corresponds to a binary invariant . an `` l '' can only be followed by another `` l '' or a `` d '' , according as the corresponding binary invariant @xmath105 is @xmath88 or @xmath0 , and similarly an `` r '' is followed by another `` r '' or a `` d '' , according as @xmath105 is @xmath88 or @xmath0 . when the previous step is `` d '' , the effect of @xmath105 depends on the step before one that produces the `` d '' . if the `` d '' is in a sequence `` ld '' , then the next step is `` r '' or `` l '' according as @xmath105 is @xmath88 or @xmath0 , while if it is in `` rd '' , then the next step is `` l '' or `` r '' according as @xmath105 is @xmath88 or @xmath0 . functions that translate between the binary sequence and step sequence descriptions are included in the software at @xcite . the main functions there accept either form of input for principal paths . the _ ( hempel ) distance _ @xmath7 is the shortest distance in the curve complex of @xmath49 from @xmath106 to a loop that bounds a disk in @xmath46 ( see j. johnson @xcite and y. minsky , y. moriah , and s. schleimer @xcite ) . it is well - defined since the action of the goeritz group on @xmath49 preserves the set of loops that bound disks in @xmath34 and the set that bound in @xmath46 . a nonseparating disk has distance @xmath0 if and only if it is primitive , since both conditions are equivalent to the condition that cutting @xmath34 along the disk produces an unknotted solid torus . therefore the tunnel of the trivial knot is the only tunnel of distance @xmath0 . a simple or semisimple tunnel has distance @xmath9 , since it is disjoint from a primitive disk . there are , however , regular tunnels of distance @xmath9 . it is an easy observation that the `` middle '' tunnels of torus knots all have distance @xmath9 , and in most cases these are regular . when @xmath107 is a heegaard splitting of the complement of @xmath40 , the _ ( hempel ) distance @xmath108 _ is the minimal distance in the curve complex of @xmath49 between the boundary of a disk in @xmath109 and the boundary of a disk in @xmath46 ( where the disks may be separating ) . clearly , @xmath110 . on the other hand , johnson ( * * lemma 11 ) proved that @xmath111 . [ lem : johnsondistance ] m. scharlemann and m. tomova @xcite proved the following stability result : genus-@xmath112 heegaard splittings of distance more than @xmath113 are isotopic . [ thm : scharlemann - tomova ] using lemma [ lem : johnsondistance ] and theorem [ thm : scharlemann - tomova ] , johnson ( * ? ? ? * corollary 13 ) deduced the following : if @xmath26 is a tunnel of a tunnel number @xmath0 knot @xmath40 and @xmath114 , then @xmath26 is the unique tunnel of @xmath40 . [ thm : uniquetunnel ] theorem 15.2 of @xcite determines all orientation - reversing self - equivalences of tunnels : let @xmath26 be a tunnel of a tunnel number @xmath0 knot or link . suppose that @xmath26 is equivalent to itself by an orientation - reversing equivalence . then @xmath26 is the tunnel of the trivial knot , the trivial link , or the hopf link . [ thm : hopflink ] combining theorems [ thm : uniquetunnel ] and [ thm : hopflink ] gives the following : if @xmath26 is a tunnel of a tunnel number @xmath0 knot , and @xmath114 , then @xmath40 is not amphichiral . [ coro : amphichiral ] for theorem [ thm : hopflink ] shows that an orientation - reversing equivalence from @xmath40 to @xmath40 would produce a second tunnel for @xmath40 . distance also has implications for hyperbolicity : if @xmath40 is a torus knot or a satellite knot , then @xmath115 . consequently , if @xmath116 , then @xmath40 is hyperbolic . [ thm : morimoto - sakuma ] we have already mentioned the fact @xcite that the middle tunnels of torus knots have distance @xmath9 . the other tunnels of torus knots are simple or semisimple , so also have distance @xmath9 . k. morimoto and m. sakuma @xcite found all tunnels of tunnel number @xmath0 satellite knots , showing in particular that they are semisimple . the _ depth _ of @xmath26 is the simplicial distance @xmath6 in the @xmath0-skeleton of @xmath42 from @xmath26 to the primitive vertex @xmath53 . from the definitions , @xmath26 is primitive if and only if @xmath117 , is simple or semisimple if and only if @xmath118 , and is regular if and only if @xmath119 . the inequality @xmath120 mentioned in the introduction is immediate from the definitions . on the other hand , we have already noted that the middle tunnels of torus knots have distance @xmath9 , but we will see in section [ sec : torus_knots ] that their depths can be arbitrarily large . in terms of the step sequence describing the principal path of a tunnel , the depth is simply the number of d s that appear . one can , of course , determine the depth directly from the binary invariants . a maximal block of @xmath0 s in the binary word @xmath121 has the following effect : its first , third , fifth , and so on terms will produce a downward step , increasing the depth , while the other terms correspond to horizontal steps , keeping the same depth . this gives the following simple algorithm to compute @xmath6 from the binary invariants of @xmath26 : 1 . write the binary word @xmath122 as @xmath123 , where @xmath124 and @xmath125 are respectively maximal blocks of ones and zeros ( thus @xmath126 and @xmath127 may have length @xmath88 , while all others have positive length ) . the depth of @xmath26 is @xmath128 , where @xmath129 denotes the least integer greater than or equal to @xmath130 . roughly speaking , the tunnel leveling theorem of goda , scharlemann , and thompson says that a tunnel arc of a tunnel number one knot can be slid so that it lies in a level sphere of some minimal bridge position of the knot . here is the rather technical version of the tunnel leveling theorem that we will need . illustrations of conclusions ( i ) and ( ii ) of the theorem appear in the first drawings of figure [ fig : cc ] and figure [ fig : cc_eyeglass_semisimple ] respectively . let @xmath67 be the principal pair of a tunnel @xmath26 , and let @xmath5 be the @xmath5-curve associated to the principal vertex @xmath131 of @xmath26 . write @xmath132 for the arc dual to @xmath26 , and @xmath133 and @xmath134 for the other two arcs of @xmath5 that are dual to @xmath71 and @xmath68 , so that @xmath135 , @xmath136 , and @xmath137 . then there is a minimal bridge position of @xmath40 for which either : 1 . @xmath132 is slid to an arc in a level sphere , and @xmath132 connects two bridges of @xmath40 . moreover , @xmath138 is isotopic to the original @xmath5 . or , 2 . @xmath132 is slid to an eyeglass in a level sphere . the endpoints of @xmath132 can be slid slightly apart , moving @xmath132 out of the level sphere , producing @xmath138 isotopic to the original @xmath5 , and showing that one of @xmath139 or @xmath140 is a trivial knot , and consequently @xmath26 is simple or semisimple . and furthermore , in the @xmath3-strand trivial tangle above the level sphere : 1 . in case ( i ) , the arcs are parallel to a collection of disjoint arcs in the level sphere , which meet @xmath132 only in its endpoints . 2 . in case ( ii ) , the @xmath141 arcs not meeting @xmath132 are parallel to a collection of disjoint arcs in the level sphere , each meeting the eyeglass in a single point . [ thm : gsttheorem ] by theorem 1.8 of @xcite , we may move @xmath142 , possibly using slide moves of @xmath132 as well as isotopy , so that @xmath40 is in minimal bridge position and @xmath132 either lies on a level sphere and connects two bridges of @xmath40 , or @xmath132 is slid to an `` eyeglass '' . since the leveling process involves sliding the tunnel arc @xmath132 , there is _ a priori _ no reason for the resulting @xmath5-curve to be isotopic to the original @xmath5 . but corollary 3.4 and theorem 3.5 ( combined with lemma 2.9 ) of @xcite show that in ( i ) and ( ii ) , the dual disks to the other two arcs of the @xmath5-curve are the disks called @xmath64 and @xmath65 there . by lemma 14.1 of @xcite , these disks are the principal pair of @xmath26 , that is , @xmath71 and @xmath68 . therefore the resulting @xmath5-curve is isotopic to the original @xmath5 . finally , the description of the trivial tangle above the level sphere in ( iii ) is from theorem 6.1 of @xcite , and in ( iv ) from corollary 6.2 of @xcite , which relies on @xcite . a tunnel arc @xmath132 satisfying conclusion ( i ) of the tunnel leveling theorem is said to be in _ level arc position , _ while for conclusion ( ii ) , _ after sliding the endpoints apart to produce @xmath5 , _ it is in _ eyeglass position . _ if it is in one of these two positions , it may be said to be in _ level position . _ a tunnel arc satisfying all the requirements of level position except that the number of bridges of @xmath40 is not necessarily minimal is said to be in _ weak level position . _ the number of bridges is then called the _ bridge count , _ denoted @xmath143 and dependent , of course , on the choice of weak level position . suppose that @xmath26 is in weak level arc position . the endpoints of @xmath26 cut @xmath40 into two arcs , one dual to @xmath71 and the other dual to @xmath26 . by a simple isotopy , we may assume that one end of the arc dual to @xmath71 leaves the endpoints of the tunnel arc in the upward direction , and the other leaves in the downward direction . for if both leave in the same direction , we can slide an endpoint of the tunnel arc over one of the arches , achieving a level position for which the two ends leave in different directions . we then call this an _ admissible _ weak level arc position . when @xmath26 is in weak level arc position , each of the local maxima of @xmath40 lies in exactly one of @xmath139 or @xmath140 . the numbers that lie in each are called the _ relative bridge counts _ of @xmath139 and @xmath140 for the weak level arc position of @xmath26 , and denoted by @xmath144 and @xmath145 . clearly @xmath146 . if @xmath26 is in weak eyeglass position , with @xmath139 a trivial knot , then we define @xmath147 and @xmath148 . one always has @xmath149 . a first consequence of theorem [ thm : gsttheorem ] is the following . let @xmath26 be a tunnel of a nontrivial knot , and let @xmath67 be the principal pair of @xmath26 . then @xmath150 . if @xmath26 is regular , then @xmath151 . [ lem : bridge_number_inequality ] apply theorem [ thm : gsttheorem ] to @xmath26 . if the tunnel arc is in level arc position , which may be assumed to be admissible , then we have @xmath152 . if the tunnel arc is in eyeglass position , producing , say , @xmath139 trivial , then we have @xmath153 , giving the inequality . if @xmath26 is regular , then the eyeglass configuration can not occur , giving the stronger inequality . in this section , we will prove the following theorem : let @xmath26 be a tunnel with principal vertex @xmath66 . if @xmath26 is not simple , choose notation so that @xmath68 is the tunnel directly preceding @xmath26 in the cabling sequence for @xmath26 . assume that @xmath26 is not the tunnel of the trivial knot or a simple tunnel of a @xmath154 torus knot . then either 1 . all level positions of @xmath26 are level arc positions , and @xmath155 , or 2 . all level positions of @xmath26 are eyeglass positions , @xmath26 is semismiple , and @xmath156 . [ thm : addendum ] the exceptional case of the tunnel leveling addendum is detailed in the next theorem , which is simply a restatement of some results from the work of k. morimoto and m. sakuma on tunnels of @xmath9-bridge knots @xcite : the trefoil knots have unique tunnels , which are simple and can be put into either level arc position or eyeglass position . for the other @xmath154 torus knots , there are two simple tunnels , of which one can be put into both kinds of level position , and the other only into level arc position . [ thm : exceptional_case ] of course the trivial knot has a unique tunnel , which can only be leveled in eyeglass position . it is important to understand the geometric content of the tunnel leveling addendum from the viewpoint of the tunnel leveling theorem [ thm : gsttheorem ] . apart from the exceptional cases , the addendum says that when a tunnel @xmath26 is leveled , giving a positioning of the @xmath5-curve associated to the principal vertex @xmath66 of @xmath26 , then ( possibly after trivial repositioning ) the copies of @xmath139 and @xmath140 in that @xmath5-curve are in minimal bridge position . in this section we will prove the tunnel leveling addendum and theorem [ thm : exceptional_case ] , and in preparation for this we now introduce the technique of efficient cabling . the basic construction is shown in figure [ fig : cc ] , whre notation is selected so that the cabling will replace @xmath71 and retain @xmath68 . we start with @xmath26 in admissible weak level arc position , as shown in the left - hand drawing . there may , of course , be many more bridges , some in @xmath140 and some in @xmath139 . a cabling of some arbitrary slope replaces @xmath71 with a new tunnel @xmath69 ; the rational tangle in @xmath157 created by the cabling is inside a ball represented by the circle in the middle drawing . we may then reposition @xmath157 as in the right - hand drawing of figure [ fig : cc ] , by `` moving the ball up to engulf infinity , '' in such a way that the rectangle in the drawing contains a @xmath158-strand braid . the arc dual to @xmath69 is in weak level arc position , and by a further isotopy , if necessary , we may assume that it is in admissible weak level arc position . the corresponding construction for a tunnel in weak eyeglass position can produce either another semisimple tunnel or a regular tunnel . the resulting tunnel is in weak level arc position , which by isotopy is also assumed to be admissible . efficient cablings for each of the two possibilities are shown in figures [ fig : cc_eyeglass_semisimple ] and [ fig : cc_eyeglass_regular ] , and the constructions should be clear from the discussion of the weak level arc case . the next result details the effect of efficient cabling on bridge counts . the notations @xmath25 and @xmath159 are used to indicate knots obtained from the @xmath5-curves associated to @xmath66 and @xmath72 respectively : suppose that @xmath26 is in weak level arc or weak eyeglass position , and that a cabling operation as in figure [ fig : cc ] , [ fig : cc_eyeglass_semisimple ] , or [ fig : cc_eyeglass_regular ] is performed , producing a new tunnel @xmath69 with principal vertex @xmath160 , and producing a tunnel arc @xmath161 for which @xmath69 is in level arc position . then 1 . 2 . @xmath163 . @xmath164 . [ prop : bridge_counts ] the third equality follows from the first two . the first two are seen by examination of figures [ fig : cc ] , [ fig : cc_eyeglass_semisimple ] , and [ fig : cc_eyeglass_regular ] . for example , let us consider figure [ fig : cc ] . in the leftmost drawing , denote the arcs dual to @xmath71 , @xmath68 , and @xmath26 by @xmath133 , @xmath134 , and @xmath132 respectively . in the rightmost drawing , after the cabling producing @xmath165 has been performed , denote the dual arcs by @xmath166 , @xmath167 , and @xmath168 , where the latter is horizontal . by isotopy we may assume that @xmath69 is also in admissible weak level arc position . the number of bridges that we see in @xmath169 equals the number that appeared in @xmath134 plus the number that appeared in @xmath133 , showing that @xmath163 . the number of bridges in @xmath170 is the number that appeared in @xmath133 , so @xmath162 . the arguments for figures [ fig : cc_eyeglass_semisimple ] and [ fig : cc_eyeglass_regular ] are similar . we are now ready to prove the tunnel leveling addendum and theorem [ thm : exceptional_case ] simultaneously . the tunnel called @xmath26 in the statement of the addendum will be denoted by @xmath69 in our argument . its principal vertex will be written as @xmath72 . if @xmath69 is not simple , then we assume that @xmath26 is the tunnel that precedes @xmath69 in the cabling sequence of @xmath69 , and the principal vertex of @xmath26 will be written as @xmath66 . as in proposition [ prop : bridge_counts ] , we use @xmath25 and @xmath159 to indicate knots obtained from the @xmath5-curves associated to @xmath66 and @xmath72 respectively . note , however , that @xmath140 and @xmath40 are equivalent to @xmath171 and @xmath172 and hence @xmath173 and @xmath174 . we will induct on the length of the cabling sequence of @xmath69 . if the length is @xmath0 , then @xmath69 is an upper or lower tunnel of the @xmath9-bridge knot @xmath175 , so can be put into level arc position . each of @xmath172 and @xmath171 is a trivial knot so has bridge number @xmath0 . therefore we have @xmath176 . so conclusion ( a ) of the tunnel leveling addendum holds for @xmath69 , provided that @xmath69 can not also be put into eyeglass position . the homeomorphism classfication of tunnels of @xmath9-bridge knots is given in table 5.2(b ) of @xcite , and only in the cases called @xmath9 , @xmath61 , and @xmath177 there does there exist a simple tunnel which can also be put into eyeglass position . those cases are defined in lemma 5.1 of @xcite , and upon examination are found to be exactly the @xmath9-bridge torus knots . since this is the excluded case in the tunnel leveling addendum , the addendum holds for tunnels whose cabling sequences have length @xmath0 . closer examination of the tunnel classification in @xcite verifies the precise statement in theorem [ thm : exceptional_case ] , whose proof is now complete . assume now that the length of the cabling sequence of @xmath69 is greater than @xmath0 . put @xmath26 in level position and obtain @xmath69 by efficient cabling as in one of figures [ fig : cc ] , [ fig : cc_eyeglass_semisimple ] , or [ fig : cc_eyeglass_regular ] . suppose first that the resulting weak level arc position for @xmath69 is actually a level arc position . using proposition [ prop : bridge_counts ] and induction , we have @xmath178 . suppose now that the weak level arc position for @xmath69 is not level arc position . then @xmath179 , so lemma [ lem : bridge_number_inequality ] tells us that @xmath69 is semisimple and @xmath180 . but @xmath68 is primitive , since the principal vertex of every semismiple tunnel contains a primitive disk , and @xmath26 is not primitive since @xmath69 is not simple . therefore @xmath181 and @xmath182 . in the latter case , @xmath69 can not be put into level arc position , since then we would have @xmath183 . so either all level positions are level arc positions , or all are eyeglass positions . this completes the proof of the tunnel leveling addendum and theorem [ thm : exceptional_case ] . the next two corollaries are convenient restatements of parts of the tunnel leveling addendum . let @xmath26 be a regular tunnel with principal vertex @xmath66 . then @xmath184 . [ coro : reg_additivity ] let @xmath26 be a semisimple tunnel with principal vertex @xmath185 . then @xmath186 or @xmath187 , according to whether all level positions for @xmath26 are eyeglass positions or all are level arc positions . [ coro : ss_additivity ] let @xmath26 be a tunnel and write the cabling sequence of @xmath26 as @xmath59 , @xmath188 , @xmath189 , @xmath190 , @xmath191 , where @xmath83 is simple or semisimple exactly when @xmath192 . that is , @xmath26 is produced by @xmath3 cablings , the first @xmath21 of which produce depth-@xmath0 tunnels . in particular , @xmath193 when @xmath118 . the principal vertex of @xmath59 has the form @xmath194 , where @xmath53 and @xmath195 are primitive . if we put @xmath196 , the trivial tunnel , then for each @xmath197 , the principal vertex of @xmath198 is of the form @xmath199 for some @xmath200 . if @xmath201 , that is , if @xmath26 is regular , then the first tunnel of depth @xmath9 is @xmath202 , and its principal vertex is @xmath203 . we then define the _ fibonacci function _ @xmath32 of a regular tunnel @xmath26 as follows . to compute @xmath30 , put @xmath204 , @xmath205 , and for @xmath206 , put @xmath207 , where @xmath199 is the principal vertex of @xmath198 . then , put @xmath208 . figure [ fig : bridge_numbers ] shows how to calculate that @xmath209 for a certain depth-@xmath210 tunnel with @xmath211 and @xmath212 . let @xmath26 be a simple or semisimple tunnel produced by @xmath21 cablings . then @xmath213 . [ thm : ss_bridge_num ] induct on @xmath21 , using corollary [ coro : ss_additivity ] . let @xmath26 be a regular tunnel whose cabling sequence contains @xmath21 tunnels of depth @xmath0 . let @xmath214 for @xmath215 . then @xmath216 . [ thm : reg_bridge_num ] induct on the length of the cabling sequence of @xmath26 , using corollary [ coro : reg_additivity ] . suppose that a knot @xmath25 has a tunnel @xmath26 produced by @xmath27 cabling operations , of which the first @xmath21 produce simple or semisimple tunnels . then @xmath28 is one of the @xmath29 values @xmath30 for @xmath31 . [ thm : bridgenumberset ] by theorem [ thm : ss_bridge_num ] , we have @xmath217 , and by corollary [ coro : ss_additivity ] , @xmath218 . the result now follows from theorem [ thm : reg_bridge_num ] . we believe that for every principal path , each of the @xmath29 possible values given in the bridge number set theorem occurs as a bridge number for some knots having a tunnel constructed using the given principal path . this is clear for @xmath23 . in that case , the cabling sequence has only two tunnels @xmath59 and @xmath219 of depth @xmath0 . when @xmath219 is a semisimple tunnel of a @xmath9-bridge knot , @xmath220 , and there are many examples where @xmath221 and @xmath222 , such as semisimple tunnels of torus knots @xcite . so choosing cabling sequences that start with these two types of examples gives tunnels whose knots have bridge numbers @xmath223 and @xmath224 . for @xmath33 , we need to produce the sequences @xmath225 , @xmath226 , @xmath227 , and @xmath228 for @xmath229 . semisimple tunnels of @xmath9-bridge knots give @xmath225 . for @xmath226 , we choose @xmath219 to be a semisimple tunnel of a @xmath9-bridge knot and choose any cabling with slope not of the form @xmath230 to produce @xmath231 ; the results of ( * ? ? ? * section 15 ) then show that @xmath232 can not be @xmath9-bridge . by corollary [ coro : ss_additivity ] , @xmath233 . for @xmath227 , we start by constructing @xmath219 to be an upper tunnel of a @xmath61-bridge torus knot , say the @xmath234 torus knot , as explained in @xcite . the tunnel arc shown in @xcite can be put into eyeglass level position with @xmath235 having three bridges , then a cabling which is geometrically like those of figure 14 of @xcite does not raise bridge number , so @xmath233 as well . finally , for @xmath228 we can just use the upper tunnel of the @xmath236 torus knot , obtained by three cablings as in @xcite . for larger @xmath21 , from upper tunnels of torus knots we obtain the bridge number sequence @xmath237 , and hence realize @xmath238 . and the idea for @xmath226 extends to realized @xmath239 , so @xmath224 occurs as a bridge number . if we start with such an upper tunnel sequence , for which the upper tunnel is in eyeglass position , and then at some point begin using cablings as in figure 14 of @xcite , we obtain all sequences of the form @xmath240 , giving the @xmath241 values @xmath242 . so at least these @xmath243 values in the bridge number set are known to occur . if we follow the latter procedure , but use a complicated tangle for the final cabling , then the sequences @xmath244 should also be obtained , giving the remaining @xmath245 values @xmath246 for @xmath247 . unfortunately we lack a means to prove that the final knot has bridge number @xmath37 rather than @xmath38 . [ rem : bridge_number_set ] a peculiar consequence of theorem [ thm : reg_bridge_num ] is the following : let @xmath26 be a regular tunnel and let @xmath202 be the first tunnel of depth @xmath9 in the cabling sequence of @xmath26 . then @xmath248 is completely determined by the principal path of @xmath26 and the value of @xmath249 . in fact , @xmath250 , [ ( { \operatorname{br}}(k_{\tau_m})+1)/2 ] ) \ .\ ] ] [ coro : first_depth_2 ] by corollary [ coro : reg_additivity ] , @xmath251 . since also @xmath252 and @xmath253 differ by at most @xmath0 , their values must be as in the statement of the corollary . it is not difficult to implement theorems [ thm : reg_bridge_num ] and [ thm : bridgenumberset ] computationally @xcite : depth > fibonacci ( 0011100011100 , 2 , 2 , verbose = true ) f@xmath254tau ( 2 , 2 ) = 182 the iteration sequence is : 2 , 2 , 4 , 6 , 10 , 14 , 18 , 22 , 40 , 62 , 102 , 142 , 182 depth > bridgeset ( 0011100011100 ) [ 182 , 232 , 273 , 323 , 364 , 414 ] using the results of section [ sec : fibonacci_functions ] , we can give some general bounds on bridge number . . ] first we examine lower bounds of bridge number as a function of depth . a bit of experimentation with fibonacci functions shows that the fastest growth of depth relative to bridge number occurs for principal paths whose regular portions ( i. e. the parts starting from @xmath255 and @xmath189 ) are the `` paths of cheapest descent '' seen in figure [ fig : cheapest ] . in that figure , the path on the left is always cheapest descent , and the one on the right is cheapest descent when @xmath256 . any principal path having more than two tunnels at a given depth will produce a larger bridge number , as will any principal path that emerges in the more costly direction out of a downward - pointing @xmath9-simplex . from theorem [ thm : reg_bridge_num ] we now have : let @xmath26 be a regular tunnel of depth @xmath1 , and in the principal path of @xmath26 , let @xmath202 be the first tunnel of depth @xmath9 , with principal vertex @xmath203 . put @xmath257 and @xmath258 . for @xmath27 let @xmath259 be given by the recursion @xmath260 then @xmath261 . [ coro : cheapest_descent ] we can now prove one of our main results . for @xmath10 , the minimum bridge number of a knot having a tunnel of depth @xmath1 is given recursively by @xmath11 , where @xmath12 , @xmath13 , and @xmath14 for @xmath15 . explicitly , @xmath262 and consequently @xmath17 . [ thm : minbridgenum ] the smallest possible values for @xmath252 and @xmath253 in corollary [ coro : cheapest_descent ] are @xmath9 . these occur for any @xmath21 , since there are @xmath9-bridge knot tunnels with arbitrarily long cabling sequences , as seen in ( * ? ? ? * section 15 ) . taking @xmath263 in corollary [ coro : cheapest_descent ] gives a @xmath264 which is a general lower bound for the bridge number of a tunnel at depth @xmath1 , and a little bit of algebra shows that @xmath265 for the recursion in theorem [ thm : minbridgenum ] . as a matrix , the recursion is @xmath266 the eigenvalues of this matrix are @xmath267 , and elementary linear algebra gives the formula @xmath268 . we turn now to upper bounds . there is no upper bound in terms of depth , since there are depth @xmath0 tunnels with arbitrarily large bridge number , such as semisimple tunnels of torus knots @xcite . we can , however , bound the bridge number of @xmath40 in terms of the number of cablings needed to produce @xmath26 . this time , we use the principal path forced by choosing the larger of its two possible sums at every step , shown in figure [ fig : universal ] . write the fibonacci sequence @xmath19 as @xmath20 . the maximum bridge number of a knot having a tunnel produced by @xmath3 cabling constructions , of which the first @xmath21 produce simple or semisimple tunnels , is @xmath22 . [ thm : max ] if @xmath26 is simple , then @xmath269 , @xmath270 and the expression @xmath22 equals @xmath9 . if @xmath26 is semisimple , then @xmath271 and @xmath22 equals @xmath243 , the upper bound given in theorem [ thm : ss_bridge_num ] . so we may assume that @xmath26 is regular . in figure [ fig : universal ] , the top two vertices are @xmath255 and @xmath189 , the last two semisimple tunnels that appear in the cabling sequence of @xmath272 . there are semisimple tunnels @xmath189 produced by @xmath21 cabling constructions which have @xmath273 , such as the semisimple tunnels of the @xmath274 torus knot @xcite . therefore the maximum bridge number is that given by theorem [ thm : reg_bridge_num ] applied to the principal path whose regular portion is shown in figure [ fig : universal ] . using the fact that @xmath275 and @xmath276 , one checks that this value is @xmath22 . the maximum bridge number of a knot having a tunnel produced by @xmath3 cabling operations is @xmath24 . [ coro : max ] for a fixed @xmath3 , the largest upper bound in theorem [ thm : max ] occurs when @xmath23 . proposition [ prop : depth - efficienttorusknots ] below gives an explicit sequence of tunnels of torus knots that achieves the maximum value of corollary [ coro : max ] . the tunnels of torus knots were classified by m. boileau , m. rost , and h. zieschang @xcite and independently by y. moriah @xcite . for a @xmath277 torus knot @xmath278 contained in the standard torus @xmath132 in @xmath35 , the _ middle tunnel _ is represented by an arc in @xmath132 that meets @xmath278 only in its endpoints . there are as many as two other tunnels , which always have depth @xmath0 , but here we focus on the middle tunnels . in this section , we will include some information on slope invariants , for those familiar with them . slopes are not essential to the discussion , and can be ignored if the reader so chooses . for the tunnels of torus knots , the slope and binary invariants were calculated in @xcite . in particular , for the middle tunnels , we have the following theorem , in which @xmath279 and @xmath280 : let @xmath281 and @xmath282 be relatively prime integers with @xmath283 . write @xmath284 as a continued fraction @xmath285 $ ] with all @xmath286 positive and @xmath287 . let @xmath288 put @xmath289 , and for @xmath290 put @xmath291 where the subscripts in the product occur in descending order . then : 1 . the middle tunnel of @xmath292 is produced by @xmath293 cabling constructions whose slopes @xmath294 , @xmath295 , @xmath296 are @xmath297 , \;a_1d_1+b_1c_1,\ ; a_2d_2+b_2c_2,\ ; \ldots,\ ; a_nd_n+b_nc_n\ .\ ] ] 2 . for each @xmath298 , the cabling corresponding to the slope invariant @xmath299 produces the @xmath300 torus knot ; in particular , the first cabling produces the @xmath301 torus knot . the binary invariants of the cabling sequence of this tunnel , for @xmath302 , are given by @xmath303 if @xmath304 and @xmath305 otherwise . [ thm : middle_tunnels_slopes ] note that this enables one to find the invariants of the middle tunnels for all @xmath277 torus knots , since @xmath278 is isotopic to @xmath306 and @xmath307 is equivalent to @xmath278 by an orientation - reversing homeomorphism taking middle tunnel to middle tunnel . such an equivalence negates the slope invariants and does not change the binary invariants . a bit of examination of the binary invariants yields a simple algorithm to find the depth of the middle tunnel of @xmath278 , @xmath308 : 1 . write @xmath284 as a continued fraction @xmath285 $ ] with all @xmath309 positive and @xmath287 . 2 . write the string @xmath310 as @xmath311 , where each @xmath312 is either @xmath313 with @xmath314 , or @xmath309 with @xmath315 . 3 . the depth of the middle tunnel is @xmath316 . this is implemented in the software at @xcite . the @xmath317 torus knot has the smallest bridge number of any torus knot with a depth @xmath158 tunnel.,height=340 ] figure [ fig : cheapest ] shows an initial segment of the principal paths for the tunnels of the @xmath277-torus knots having continued fraction expansions of the form @xmath318 $ ] . notice that this is the path of cheapest descent from figure [ fig : descent ] . the small numbers along the path are the slopes , the letters indicate whether the constructions correspond to multiplication by @xmath319 or by @xmath133 , and the pairs show the @xmath277 for the torus knots determined by the tunnel at each step . the first nontrivial cabling , with @xmath320 $ ] , produces a @xmath321-torus knot , and the second produces a @xmath234-torus knot with bridge number @xmath61 . since we always have @xmath322 , the bridge number is simply the value of @xmath282 . these obey the recursion of corollary [ coro : cheapest_descent ] , starting with @xmath323 and @xmath324 . since the cabling sequence for the middle tunnel @xmath26 of any torus knot contains only one two - bridge knot ( the @xmath301-torus knot produced by the first nontrivial cabling ) , there is no regular torus knot tunnel which has @xmath325 . since the other tunnels of torus knots are semisimple , the maximum depth of any tunnel of a torus knot is the depth of its middle tunnel . therefore each @xmath264 in this sequence gives the minimum bridge number for a torus knot with a tunnel of depth @xmath1 . this gives a version of the minimum bridge number theorem [ thm : minbridgenum ] for torus knot tunnels : for @xmath10 , the minimum bridge number of a torus knot tunnel depth @xmath1 is given recursively by @xmath326 , where @xmath327 , @xmath328 , and @xmath329 for @xmath15 . explicitly , @xmath330 and consequently @xmath331 . we note that @xmath326 is @xmath332 , where @xmath333 is the lower bound in the minimum bridge number theorem [ thm : minbridgenum ] . that is , the minimum bridge number of a torus knot having a tunnel of depth @xmath1 is exactly half the minimum bridge number for all knots having a tunnel of depth @xmath334 , and is approximately @xmath335 times the minimum for all knots having a tunnel of depth @xmath1 . in fact , the middle tunnel any torus knot for which @xmath284 has an expansion @xmath336 $ ] will have a principal path as in the previous argument , since the first term in the continued fraction has no effect on the principal path . middle tunnels for which the expansion is not of this form will have different principal paths , so we can state the following result : the slowest growth of bridge number compared to depth for sequences of middle tunnels of torus knots occurs when the @xmath284 have continued fraction expansions of the form @xmath337 $ ] . [ prop : depth - efficienttorusknots ] by similar considerations , one can obtain the upper bound version . the fastest growth of bridge number of torus knots per number of cablings of the middle tunnels occurs for sequences of tunnels @xmath198 of @xmath278 for which the continued fraction expansions of @xmath284 are of the form @xmath338 $ ] , where there are @xmath38 @xmath0 s . for these tunnels , @xmath339 has bridge number @xmath340 . [ prop : fast - growthtorusknots ] for these tunnels , the terminal part of the corridor is like that shown in figure [ fig : universal ] with @xmath23 . since there are exactly @xmath38 cablings in the cabling sequence of @xmath198 , these tunnels achieve the bridge numbers in the maximum bridge number theorem . as explained in @xcite , our entire theory of tunnel number @xmath0 knot tunnels can be adapted to include tunnels of tunnel number @xmath0 links , simply by adding the separating disks as possible slope disks . the full disk complex @xmath341 is only slightly more complicated than @xmath36 . each separating disk is disjoint from only two other disks , both nonseparating , so the additional vertices appear in @xmath9-simplices attached to @xmath36 along the edge opposite the vertex that is a separating disk . the quotient @xmath342 has only three types of additional @xmath9-simplices : 1 . there is a unique orbit @xmath343 of `` primitive '' separating disks , consisting of separating disks disjoint from a primitive pair , which are exactly the intersections of splitting spheres ( see @xcite ) with @xmath34 . in @xmath342 , @xmath343 is a vertex of a `` half - simplex '' @xmath344 attached to @xmath42 along @xmath345 . it is the unique tunnel of the trivial @xmath9-component link . 2 . simple separating disks lie in half - simplices attached along @xmath346 , just like nonseparating simple disks . 3 . the remaining separating disks lie in @xmath9-simplices attached along edges of @xmath42 spanned by two ( orbits of ) disks , at least one of which is nonprimitive . for the spine , a single `` y '' is added to @xmath43 for each added @xmath9-simplex as in ( 3 ) , and a folded `` y '' for each the half - simplices as in ( 1 ) and ( 2 ) . the link in @xmath347 of a link tunnel is simply the top edges ( or top edge , for the trivial and simple tunnels ) of such a `` y '' ( or folded `` y '' ) . the cabling operation differs only in allowing a separating slope disk , which produces a tunnel of a tunnel number @xmath0 link . the cabling sequence ends with the first separating slope disk . thus the principal paths look exactly like those of the knot case , such as the one in figure [ fig : ppath ] . the only difference is that no further continuation is possible if the final tunnel is the tunnel of a link . for link tunnels , the distance and depth invariants are defined as for knot tunnels . simple tunnels are the upper and lower tunnels of @xmath9-bridge links ( and are the only tunnels of these links , see @xcite , @xcite , or ( * ? ? ? * theorem 16.3 ) ) . depth @xmath0 tunnels are the tunnels of links with one component unknotted . the other component must be a @xmath73-knot , and the link must have torus bridge number @xmath9 ( * ? ? ? * theorem 16.4 ) . lemma [ lem : johnsondistance ] holds when @xmath26 is separating , in fact the argument is an easier version of the argument in @xcite , so theorem [ thm : uniquetunnel ] and corollary [ coro : amphichiral ] hold for links as well as knots . the tunnel leveling addendum extends to tunnels of tunnel number @xmath0 links , since the efficient cabling construction of section [ sec : efficient ] works just as well in the link case . but the statement and proof are very much simpler , since only the level arc case need be considered . using the tunnel leveling addendum for links , the computational results of sections [ sec : fibonacci_functions ] and [ sec : growth ] hold as stated for tunnels of tunnel number @xmath0 links . consequently , the software implementations of @xcite also produce correct results for link tunnels .

we present x - ray spectral analysis of the accreting young star tw hydrae from a 489 ks observation using the _ chandra _ high energy transmission grating . the spectrum provides a rich set of diagnostics for electron temperature , electron density , hydrogen column density , relative elemental abundances and velocities and reveals its source in 3 distinct regions of the stellar atmosphere : the stellar corona , the accretion shock , and a very large extended volume of warm postshock plasma . the presence of , , and emission lines in the spectrum requires coronal structures at mk . lower temperature lines ( e.g. , from , , and ) formed at 2.5 mk appear more consistent with emission from an accretion shock . he - like line ratio diagnostics indicate thatmk and in the shock . these values agree well with standard magnetic accretion models . however , the _ chandra _ observations significantly diverge from current model predictions for the postshock plasma . this gas is expected to cool radiatively , producing as it flows into an increasingly dense stellar atmosphere . surprisingly , indicates , five times lower than in the accretion shock itself , and seven times lower than the model prediction . we estimate that the postshock region producing has roughly 300 times larger volume , and 30 times more emitting mass than the shock itself . apparently , the shocked plasma heats the surrounding stellar atmosphere to soft x - ray emitting temperatures and supplies this material to nearby large magnetic structures which may be closed magnetic loops or open magnetic field leading to mass outflow . our model explains the soft x - ray excess found in many accreting systems as well as the failure to observe high signatures in some stars . such accretion - fed coronae may be ubiquitous in the atmospheres of accreting young stars .

1001.0750

low mass stars in star - forming regions produce strong x - ray emission from coronal magnetic activity , as evidenced by high temperature ( @xmath310 mk ) emission from flares and active regions ( feigelson & montmerle 1999 ; gagn et al . 2004 ; preibisch et al . the role of accretion in the production of x - rays from young stars is less well understood . magnetic structures on stars can extend all the way to the inner accretion disk , as suggested from models of decaying stellar flares ( favata et al . extended x - ray emission from jets also occurs , as observed in dg tau , a system oriented so that the circumstellar disk plane aligns with our line of sight ( gdel et al . soft x - ray emission may arise from shocks at the stellar base of this extended x - ray jet ( gdel et al . 2008 ; schneider & schmitt 2008 ) . x - rays can also be produced in accreting systems when the accreting material accelerates to supersonic velocities and shocks near the stellar surface , heating the gas to a few mk . in the standard model of accreting classical t tauri stars ( ctts ) , the magnetic field of the star connects to the accretion disk near the corotation radius , channeling the accretion stream from the disk to a small area on the star ( calvet & gullbring 1998 ) . x - rays from the shock may be difficult to detect if the shock is formed too deep in the photosphere ( hartmann 1998 ) or if they are absorbed by the stream of preshocked neutral or near - neutral gas . furthermore , the x - ray shock signature may be difficult to distinguish from soft coronal emission without the high resolution spectroscopy necessary to determine temperature and density and identify the emitting regions . shock models predict high electron density @xmath1 ( @xmath8 @xmath6 ) at relatively low electron temperature @xmath0 ( a few mk ) , features that can be tested with x - ray line ratio diagnostics . tw hydrae ( tw hya ) was the first ctts found to show these characteristics of high density and low temperature ( kastner et al . 2002 ; hereafter , k02 ) , based on a 48 ks _ chandra _ high energy transmission grating ( hetg ) spectrum . tw hya is one of the oldest known stars ( @xmath3 10 myr ) still in the ctts ( accreting ) phase . it is uniquely poised in an interesting state of evolution when it will soon stop accreting , lose its disk , and perhaps form planets ( e.g. , calvet et al . 2002 ) . at a distance of only 57 pc , it is also one of the brightest t tauri stars in x - rays . interstellar absorption is minimal , and the observed absorption can be considered intrinsic to the system ( k02 ) . tw hya s circumstellar disk is nearly face - on with an inclination angle of 7@xmath9 ( qi et al . thus the star appears pole - on . key to k02 s argument for accretion is the electron density ( @xmath10 @xmath6 ) determined from he - like emission line ratios , and confirmed by subsequent observations with the _ xmm - newton _ reflection grating spectrometer ( stelzer & schmitt 2004 ) and the _ chandra _ low energy transmission grating ( letg ; raassen 2009 ) . while @xmath1 for some active cool star coronae approaches such high values at high @xmath0 ( @xmath11 mk ) , e.g. , 44 boo , ux ari , and ii peg ( sanz - forcada et al . 2003 ; testa et al . 2004 ) , the high @xmath1 of tw hya is produced at the significantly lower @xmath0 ( @xmath33 mk ) expected for the accretion shock . bp tau ( schmitt et al . 2005 ) , v4046 sag ( gunther et al . 2006 ) , ru lup ( robrade & schmitt 2007 ) , and mp mus ( argiroffi et al . 2007 ) also show high values of the electron density . on the other hand , t tau and ab aur have a low @xmath1 ( gdel et al . 2007a ; robrade & schmitt 2007 ) ; and , the binary hen 3 - 600 appears to lie at an intermediate value of @xmath1 ( huenemoerder et al . 2007 ) . despite large differences in @xmath1 , all of these stars show an excess ratio of soft x - ray emission to hard x - ray emission , manifested by their large / ratios as compared with active main sequence stars ( gdel & telleschi 2007 ; robrade & schmitt 2007 ) . additional soft x - ray diagnostics available with a deep _ chandra _ exposure might be expected to shed light on the accretion process and test the shock models . since accretion is expected to result in higher densities than found in an active stellar corona , it is critical that the density determination be secure . interpretation of the line ratio in terms of high density is not unique , since photoexcitation by ultraviolet radiation mimics the effects of high density . determining the density from is not subject to the same degeneracy , since the relevant ultraviolet radiation , observed with the far ultraviolet explorer ( fuse ; dupree et al . 2005 ) , is too weak to photoexcite its metastable level . the three he - like systems ( , , and ) together can provide independent determinations not only of the electron density , but also of the electron temperature and hydrogen column density . deep spectroscopy can , first , put the accretion interpretation on firm ground and second , allow us to determine the structure of the accretion region itself . we were awarded a _ chandra _ large observing project to establish definitively whether the x - ray emission in tw hya is associated with accretion , using high signal - to - noise ratio spectroscopy with the hetg , and , if so , to determine the emission measure distribution and elemental abundances , search for x - ray signatures of infall , outflow , and turbulence , and constrain the properties of the accretion hot spot . during the _ chandra _ observations we also conducted a ground - based observing campaign , which will be reported elsewhere ( dupree et al . 2010 , in preparation ) . here , we describe the x - ray observations in section 2 and report analysis results in section 3 , where the densities , temperatures , and column densities can be extracted from this rich spectrum . section 3 also presents elemental abundances derived from fits to the spectrum , as well as the measurement of turbulent velocity . section 4 describes spectral predictions from a model of the tw hya shock . in section 5 we discuss evidence for a new type of coronal structure . section 6 gives a summary of results and conclusions . _ chandra _ observed tw hya with the hetg in combination with the acis - s detector array for a total of 489.5 ks intermittently spanning the period between 2007 feb 15 and 2007 mar 3 . the observation consists of 4 segments , with exposure times as follows : 153.3 ks ( obsid 7435 ) , 157.0 ks ( obsid 7437 ) , 158.4 ks ( obsid 7436 ) , and 20.7 ks ( obsid 7438 ) . the summed spectrum has about ten times the exposure time of the spectrum analyzed by k02 . figure 1 shows the total first - order light curve , with flux variations of about a factor two during the observation . assuming a distance of 57 pc , the observed x - ray luminosity @xmath12 , measured using the medium energy grating between 2.0 and 27.5 , is @xmath13 erg s@xmath14 . one clear flare , with total x - ray luminosity of 2.1 @xmath15 erg s@xmath14 , persists for @xmath315 ks , producing a total energy of 4 @xmath16 ergs . the flare is retained in the following analysis since its contribution to line fluxes is negligible ( see section 3.4 ) . we have extracted spectra from the event files , and produced calibration files using the _ chandra _ interactive analysis of observations ( ciao ver . 3.4 ) software package ( fruscione et al . 2006 ) . with response matrices combined from the four observations , _ sherpa_(freeman et al . 2001 ) was used to extract individual emission line fluxes from the co - added observations , maintaining separate first order spectra for the four grating arms ( plus and minus ; meg and heg ) during the fitting procedure . a model for the continuum plus background was found from a global fit to line - free regions and used in addition to the line models . background rejection is high with acis order sorting . thus the background contributes relatively little to this model except at the long wavelength end of the spectrum , where the effective area is low . in the o vii region , for example , the background contains about half as many counts as the continuum . for the relatively weak but important line , background , continuum , and line have about 1 , 2 , and 16 counts , respectively . line models are gaussian functions with negligible or small widths relative to the width of the instrumental line response functions ( @xmath30.012 and 0.023 , for heg and meg , respectively ) . table 1 gives the identified lines and their observed fluxes . for most lines the fit widths are less than half the instrumental width and are not statistically significant . we discuss the widths of the strongest lines in section 3.5 . lrccl & 6.182 & 1.87 @xmath17 0.19 & 163.0 & corona + & 6.648 & 2.72 @xmath17 0.17 & 146.6 & corona + & 6.688 & 0.74 @xmath17 0.14 & 138.3 & corona + & 6.740 & 1.66 @xmath17 0.15 & 151.6 & corona + & 8.421 & 2.24 @xmath17 0.20 & 162.4 & corona + & 9.169 & 2.25 @xmath17 0.22 & 135.4 & shock + & 9.231 & 0.90 @xmath17 0.17 & 133.2 & shock + & 9.314 & 1.27 @xmath17 0.19 & 130.1 & shock + & 9.481 & 1.64 @xmath17 0.20 & 114.5 & corona + & 9.708 & 2.75 @xmath17 0.27 & 86.2 & corona + & 10.239 & 9.23 @xmath17 0.53 & 81.1 & corona + & 10.765 & 3.54 @xmath17 0.43 & 67.9 & shock + & 11.001 & 8.48 @xmath17 0.63 & 67.4 & shock + & 11.544 & 23.7 @xmath17 0.99 & 55.3 & shock + & 11.783 & 3.28 @xmath17 0.50 & 52.3 & corona + & 11.932 & 3.01 @xmath17 0.50 & 49.6 & corona + & 12.134 & 76.1 @xmath17 1.9 & 48.5 & corona + & 12.266 & 4.3 @xmath17 0.6 & 46.6 & mixed + & 13.447 & 177 . @xmath17 4.2 & 22.3 & shock + & 13.553 & 114 . @xmath17 3.3 & 22.7 & shock + & 13.699 & 58.7 @xmath17 2.3 & 25.0 & shock + & 14.208 & 4.59 @xmath17 0.76 & 29.0 & corona + & 14.256 & 1.63 @xmath17 0.53 & 28.5 & corona + & 14.820 & 5.04 @xmath17 0.80 & 23.7 & shock + & 15.014 & 36.5 @xmath17 1.8 & 21.9 & shock + & 15.176 & 10.4 @xmath17 1.2 & 20.8 & shock + & 15.261 & 17.4 @xmath17 1.4 & 20.0 & shock + & 16.006 & 30.6 @xmath17 2.3 & 15.9 & shock + & 16.071 & 5.17 @xmath17 1.0 & 16.3 & corona + & 16.780 & 25.5 @xmath17 3.0 & 12.1 & shock + & 17.051 & 27.9 @xmath17 2.3 & 11.1 & shock + & 17.096 & 26.3 @xmath17 2.3 & 11.2 & shock + & 17.207 & 4.68 @xmath17 1.4 & 10.7 & shock + & 17.396 & 4.68 @xmath17 1.5 & 9.8 & shock + & 17.768 & 3.66 @xmath17 1.6 & 7.7 & shock + & 18.627 & 17.9 @xmath17 1.4 & 7.3 & shock + & 18.969 & 213 . @xmath17 8.4 & 6.6 & shock + & 20.910 & 3.8 @xmath17 2.5 & 3.3 & shock + & 21.601 & 117 . @xmath17 10.0 & 2.8 & postshock + & 21.804 & 72.4 @xmath17 9.1 & 2.7 & postshock + & 22.098 & 15.2 @xmath17 4.4 & 2.1 & postshock + & 24.781 & 68.5 @xmath17 7.6 & 2.6 & shock + all line fluxes are within 3@xmath18 of the values reported by k02 except for @xmath1913.437 , which has slightly more than twice as much flux in our observation . our line fluxes are also within 3@xmath18 of the values reported by stelzer & schmitt ( 2004 ) for lines observed with both instruments , except for @xmath1913.553 , for which our flux is about half . for lines in common with those reported by raassen ( 2009 ) all fluxes agree within 3@xmath18 . with the higher signal - to - noise ratio of our data , new diagnostic information is available . a major goal of this study is the determination of electron density from the he - like ions . we first discuss the diagnostics , and then model the global spectrum . figure 2 shows the high quality in the he - like and line regions . interestingly , the region shows no apparent lines , which can contaminate the diagnostics ( e.g. , ness et al . upper limits on the strongest line at 13.52 imply that contributes 1% or less to the line fluxes . the triplet line region is more complex ( fig . 3 ) , with lower signal - to - noise ratio and apparent line blending between 9.2 and 9.4 . for this region , the fit is constrained such that all known lines are fixed in wavelength relative to the resonance line @xmath199.169 , which is fit separately for each grating arm . lines are gaussian functions with widths set to zero ( i.e. , only the instrumental line response function is used ) . line fluxes of the lyman series lines @xmath20 to @xmath21 are scaled according to their oscillator strengths , since collision strengths are not available in the literature . we note that systematic errors from line blends are not included in the errors given in table 1 . one of the advantages of x - ray spectroscopy with high signal - to - noise characteristics is the availability of numerous line ratio diagnostics to assess conditions in the radiating plasma . in particular , he - like systems provide valuable diagnostic opportunity , as discussed extensively in the literature ( porquet et al . 2001 ; smith et al . 2001 ; and references therein ) . the g - ratio , defined as the flux ratio of the forbidden ( @xmath22 ) plus intercombination ( @xmath23 ) lines to the resonance ( @xmath24 ) line , is sensitive to the electron temperature ( @xmath0 ) , while the r - ratio ( flux ratio of the forbidden to intercombination lines ) is sensitive to @xmath1 . ratios of emission lines in the resonance line series ( transitions from 2p , 3p , etc . to ground , also known as he@xmath25 is the resonance line sometimes labeled @xmath26 or @xmath27 . ] , he@xmath28 , and so on ) are sensitive both to @xmath0 , through their relative boltzmann factors , and to the absorption by an intervening hydrogen column density ( @xmath2 ) through their energy separation . this analysis emphasizes the use of line ratio diagnostics , including those discussed above . we use the astrophysical plasma emission code ( apec ) with the atomic data in atomdb v1.3 ( smith et al . 2001 ) to calculate line emissivities , except for he - like , for which we use the more accurate calculations of chen et al . the model for absorption by gas with cosmic abundances is taken from morrison & mccammon ( 1983 ) . in this section we present the physical conditions of the emitting plasma as determined from the line ratios of specific he - like and other ions , in particular @xmath1 , @xmath0 , and @xmath2 . we use information determined from line ratios to constrain an empirical model of the emission measure distribution ( emd ) , and then allow additional parameters of the model to be constrained by global fits to the spectrum . we give the elemental abundances determined from these fits . finally we present velocity constraints from line centroids and widths . we use g - ratios from multiple ions to determine the electron temperature . figure 4 shows the observed g - ratios of he - like o , ne and mg , overplotted on their respective theoretical functions of @xmath0 . for comparison , the emissivities of their resonance lines peak at @xmath29 2 , 4 , and 6 mk , respectively , as shown in the figure . table 2 gives the derived @xmath0 from these ions . the g - ratio indicates that @xmath30 mk . the @xmath0 determination is highly reliable , given that recent theoretical calculations , which include the resonance contributions to the collisional excitation cross sections , agree with experimental measurements to within 7% ( chen et al . 2006 ; smith et al . 2009a , 2009b ) . on the other hand , the theoretical calculations for and have higher systematic uncertainties , and the observational errors are larger , such that the @xmath0 values derived from and carry less weight ( but are consistent with ) . both and are found far below their temperature of maximum emissivity @xmath31 , supporting an accretion shock scenario for their formation . lcc & [email protected] & 2.4@xmath32 + & [email protected] & 2.5@xmath33 + & [email protected] & 2.1@xmath34 we use r - ratios from multiple ions to determine the electron density . figure 5 shows the observed r - ratios for , , and , overplotted on the theoretical functions of electron density @xmath1 from apec . table 3 presents the @xmath1 values derived from these r - ratios . sensitivity to @xmath0 is negligible near the observed r - ratio values . smith et al . ( 2009 ) show that the apec r - ratio for in the high density range is indistinguishable from the r - ratio calculated by chen et al . the theoretical r - ratios from apec for all three ions are in excellent agreement with porquet et al . ( 2001 ) over the density range reported here . lcc & [email protected] & 5.7@xmath35 + & [email protected] & 3.0@xmath36 + & [email protected] & 5.8@xmath37 from we derive @xmath38 @xmath6 , while gives @xmath39 @xmath6 . the 3@xmath18 range from , between 2.5 and 3.9 @xmath40 @xmath6 , provides a tight constraint . 0.033 , quite close to the value of 0.446 @xmath17 0.124 reported by k02 . the difference of nearly a factor of 2 in the derived @xmath1 appears to be due to the difference in atomic data used , with their models possibly based on the raymond & smith code ( raymond 1988 ) . ] the best - fit value for , using all the lines in the region as described in section 2 , is @xmath41 @xmath6 . for comparison , we have also fit only the 3 lines of mg xi , again with line widths and relative positions fixed , obtaining @xmath42 7 @xmath17 2 @xmath40 @xmath6 , in good agreement with the best - fit value . this comparison suggests that line blending does not unduly affect the result . at low @xmath1 the population of the metastable @xmath43 level builds up since the decay rate is extremely low . at high @xmath1 electron impact excites the metastable level population up to the @xmath44 levels . in principle , low r - ratios may also be produced at low @xmath1 by photoexcitation . stelzer & schmitt ( 2004 ) and k02 dismiss photoexcitation based on the stellar temperature ( @xmath3 4000 k ) , but drake ( 2005 ) suggests that a hot spot above the stellar surface , where the accretion stream shocks , could produce sufficient photoexciting radiation . we have measured the flux at the excitation wavelengths 1647 , 1270 and 1034 for , , and , respectively , from the _ hubble space telescope _ stis ( herczeg et al . 2002 ) and the fuse spectra ( dupree et al . 2005 ) . the intensity at 1034 is an order of magnitude lower than at the longer wavelengths , constraining the black body temperature of the hot spot to be less than 10,000 k , in good agreement with _ international ultraviolet explorer _ analysis showing a blackbody at 7900 k covering about 5% of the stellar surface ( costa et al . 2000 ) . thus the ultraviolet measurements and the r - ratio rule out photoexcitation and support the interpretation of high @xmath1 . together with the temperature derived from the g - ratio , the density supports an accretion origin for . the ratio of @xmath1917.096 to @xmath1917.051 becomes sensitive to @xmath1 above @xmath8 @xmath6 ( mauche et al . . the measured ratio of 1.06 @xmath17 0.13 is at the low @xmath1 limit , ruling out the high density from this line ratio suggested by ness & schmitt ( 2005 ) . the hetg spectrum contains several series of resonance lines that indicate absorption by neutral gas along the line of sight . the most useful of these diagnostics are the and he@xmath25/he@xmath28 ratios because their formation temperatures are independently determined from the g - ratio . table 4 presents the observed line ratios , and their use as diagnostics for @xmath2 . figure 6 shows the theoretical he@xmath25/he@xmath28 ratios as functions of @xmath0 for and , for both unabsorbed and absorbed cases . the observed ratios are overplotted using @xmath0 determined from their g - ratios ( which , as noted above , is more reliable for than for ) . for the best - fit @xmath0 , gives @xmath45 @xmath46 and gives @xmath47 @xmath46 . these two values show a meaningful difference . for at 2.5 mk , the observed ratio is 11@xmath18 below the unabsorbed model , making the presence of an absorbing column quite definitive . the column density from is consistent with the lower limit of @xmath48 @xmath46 determined from the absence of and lines longward of 40 in the _ chandra _ letg spectrum ( raassen 2009 ) . lcrrrr he@xmath49/he@xmath50 & [email protected] & 8.54 & 6.52 & 5.22 & 2.49 + ly@xmath49/ly@xmath50 & [email protected] & 9.63 & 8.07 & 6.82 & 4.55 + he@xmath49/he@xmath50 & [email protected] & 11.14 & 10.07 & 9.23 & 7.34 + ly@xmath49/ly@xmath50 & [email protected] & 12.93 & 11.89 & 10.91 & 8.93 + ly@xmath49/ly@xmath50 & [email protected] & 7.061 & 6.49 & 5.96 & 4.88 ratios of the h - like series lines ( ly@xmath51 , ly@xmath52 , etc . ) are also sensitive to both absorption and temperature , but unlike the he - like ions , h - like ions do not present an independent temperature diagnostic . if the absorption column depends on the ion , as implied by the inconsistency between and , then we need to assume a model @xmath0 to obtain a column density from the h - like line ratios . assuming 2.5 mk , the observed ratio ly@xmath25/ly@xmath28 indicates @xmath53 @xmath46 . this value is consistent within the errors with @xmath2 from and thus a third absorber is not required . for at 2.5 mk , the column density would be larger than the column density for , as shown in table 4 ; however , table 4 also shows that no absorption is required if the lines are formed at 10 mk . we must further consider the effects of temperature on line ratios . we can also estimate the temperature using the charge state balance . table 5 presents the he@xmath25/ly@xmath25 ratios for oxygen and neon . first , the observed ratios are corrected for different column densities as found above , and then @xmath0 is derived from the respective corrected ratios . for oxygen , good consistency is found for the temperatures derived from the g - ratio and the he@xmath51/ ly@xmath51 ratio , given a column density of about 8 @xmath54 @xmath46 . for neon , on the other hand , all the derived temperatures are larger than @xmath0 derived above from the g - ratio . thus it appears that some of the emission originates at higher temperature . using @xmath55 @xmath46 and @xmath56 mk , and the observed he@xmath51 flux , we predict the flux of the ne x ly@xmath51 to be only 18% of the observed value . in this case , the rest of the emission must be formed at higher temperature . if the trend of increasing @xmath2 with charge state continues , then @xmath2 for would be higher than @xmath57 @xmath46 , and the fraction of ne x formed at 2.5 mk would be somewhat larger . lcccc he@xmath49/ ly@xmath49 & [email protected] & 0.70 & 0.85 & 1.67 + he@xmath49/ ly@xmath49 & [email protected] & 2.48 & 2.60 & 3.04 + + identification & + + he@xmath49/ ly@xmath49 & 3.03 & 2.83 & 2.66 & 2.23 + he@xmath49/ ly@xmath49 & 4.16 & 4.08 & 4.01 & 3.87 the emissivity functions for lines of and both peak near 5 mk , suggesting the possibility of comparing their formation temperatures . the ratio of @xmath1915.014 to @xmath1915.261 is sensitive to @xmath0 , due to a blend of an inner shell excitation line of at 15.261 (brown et al . the temperature sensitivity essentially comes from the charge state balance of and . using the reported experimental dependence of the line ratio on the electron beam energy of the electron beam ion trap at lawrence livermore national laboratory ( brown et al . 2001 ) , the observed hetg value gives @xmath58 mk . despite similar peak temperatures , and are not formed together . apparently , the long tail of the ly@xmath51 emissivity function toward high temperature contributes significantly to its emission . the lines are predominantly formed in the accretion shock , while the lines are formed both in the shock and in the hotter corona . one might expect that the ions formed at low temperature ( , , and ) should all experience similar absorption . using @xmath2 derived from , we can derive characteristic emitting @xmath0 for and of 1.4 mk and @xmath59 mk , respectively . these temperatures are far lower than @xmath0 derived from the g - ratios . they are both also far lower than @xmath60 mk derived from the charge state ratio of oxygen represented over the range of @xmath2 found here , as shown in table 5 . we conclude that an increase in @xmath2 with increasing charge state ( presumably temperature ) up through is consistent with all the data . we note that argiroffi et al . ( 2009 ) claim evidence for resonance scattering of the strongest lines in the xmm - rgs spectrum of mp mus , though their rgs spectrum of tw hya is too weak to show the signatures of absorption we find here . we rule out resonance scattering in tw hya using the strong series of . an optical depth @xmath61 of about 0.4 for he@xmath51 would be required to model the observed he@xmath51/he@xmath52 ratio . for a single ion , the optical depth @xmath61 scales as @xmath62 , such that we can then predict the other line ratios . we find then that he@xmath51/he@xmath63 ratio is overpredicted by 50% . since the oscillator strengths decrease rapidly with increasing principal quantum number @xmath64 , only the he@xmath51 line might be affected by resonance scattering . we find instead , that all the lines are affected by absorption , and rule out resonance scattering as the absorption mechanism . we conclude that the absorber is neutral or near neutral , consistent with the preshock accreting gas , as suggested by theoretical studies . for example , lamzin ( 1999 ) considers the absorption by the preshock gas of the x - ray emitting plasma for different geometries and orientations . gregory et al . ( 2007 ) use realistic coronal magnetic fields coupled with a radiative transfer code to calculate the obscuration of x - ray emission by accretion columns , and suggest that this effect can explain the observed low x - ray luminosities of accreting young stars relative to non - accretors . we present here four models for the emission measure ( em ) distribution , where the factor 0.8 accounts for the hydgrogen to electron density ratio and the integral is taken over the volume @xmath65 . the intensity of a spectral feature @xmath66 em @xmath67 where @xmath68 is the emissivity in units of ph @xmath69 s@xmath14 and @xmath70 is the distance to the source . the function @xmath68 depends on @xmath0 , and is given in steps of log [ @xmath0 ( k ) ] = 0.1 in atomdb ( smith et al . 2001 ) . ] and elemental abundances ( table 6 ) . these models serve to illustrate how values derived from the line ratios affect the global fit to the spectrum and to show whether abundance determinations are robust . for all four models , the first - order heg and meg spectra are fit to a set of variable abundance apec models , using _ sherpa _ and applying different constraints to the fitted parameters . all models have acceptable goodness - of - fit values . model a has two @xmath0 components and a single absorber . model a should be appropriate for comparison with other x - ray spectra of cool stars obtained at lower spectral resolution . lccccc @xmath711 ( mk ) & [email protected] & 2.5 & 2.5 & 2.5 + em1 ( @xmath72 @xmath6 ) & [email protected] & [email protected] & [email protected] & [email protected] + @xmath712 ( mk ) & [email protected] & [email protected] & 12.6 & 12.6 + em2 ( @xmath72 @xmath6 ) & [email protected] & [email protected] & [email protected] & [email protected] + @xmath2 ( @xmath73 @xmath46 ) & [email protected] & 1.0 & 1.0 & 1.0 + n ( 8.05 ) & [email protected] & [email protected] & [email protected] & [email protected] + o ( 8.93 ) & [email protected] & [email protected] & [email protected] & [email protected] + ne ( 8.09 ) & [email protected] & [email protected] & [email protected] & [email protected] + mg ( 7.58 ) & [email protected] & [email protected] & [email protected] & [email protected] + si ( 7.55 ) & [email protected] & [email protected] & [email protected] & [email protected] + fe ( 7.67 ) & [email protected] & [email protected] & [email protected] & [email protected] model b is also a two-@xmath0 model , but with constraints imposed from the line ratios . the lower @xmath0 is fixed at 2.5 mk , and the single absorber @xmath2 is fixed at 1.0 @xmath74 @xmath46 ( a rough average of the values obtained from line ratios ) . the @xmath1-sensitive forbidden and intercombination line regions are excluded from the fit . table 6 also gives results from multicomponent models c and d , each with the low @xmath0 and a single @xmath2 fixed , as in model b. assuming that the hotter component is coronal in nature , a broad distribution of @xmath0 is reasonable , although the shape is not well constrained . we broaden the hot component using a gaussian - shaped function centered around 12.5 mk . for model c the entire spectrum is fit simultaneously , with only the forbidden and intercombination line regions excluded as for model b. for model d the emission measures for the two components are fit to line - free regions , identified visually and using apec models . the thermal continuum emission fit to line - free regions is strongly temperature - dependent and thus constrains the emission measure distribution . the abundances are then fit to narrow regions centered on the strong lines . these separate fitting procedures are iterated until the values stop changing . the resulting model d em distribution is shown in figure 7 . the larger statistical error for the low @xmath0 emission measure in model d compared with model c ( table 6 ) stems from the smaller number of bins used in the fits ; however , we expect the systematic errors for model d to be lower , given that we are only using bins whose information content is secure , and thus we choose model d to illustrate the model spectrum . by almost a factor of 2 . we attribute these difficulties primarily to the complexity of the absorption , and secondarily to the few constraints on the shape of the em distribution . ] figure 8 compares the model d spectra predicted by the broad `` coronal '' component and the soft `` accretion '' component with the observed spectra . table 1 lists the origin of each line as accretion or corona based on model d. an additional component with @xmath0 lower than 2.5 mk can be added to both models c and d but is not required , as it does not improve the fit significantly . similar results are obtained with somewhat different choices of width and peak @xmath0 for the hot component . since the line ratios indicate increasing @xmath2 with higher charge state , we tried adding a second absorber to the high @xmath0 gaussian component for models b , c , and d. robust solutions are difficult to obtain , as the absorption , abundances , and normalizations of the emission measures do not independently determine the global x - ray spectrum . we also explored the possibility that the low and high @xmath0 components have different metal abundances but again , without robust results . using @xmath0=2.5 mk , as derived from the line ratio , instead of @xmath0=3.58 mk as found in model a , has important implications for interpreting the formation of some of the emission lines . the ratio of the strong he@xmath51 and ly@xmath51 line emission drives the model a fit to more than 3.0 mk with very little absorption . specifically , only @xmath315% of the ly@xmath25 emission arises from the 2.5 mk component , whereas a 3.58 mk component can produce more than half of the emission . the 2.5 mk component produces essentially all of the , , , and emission and more than half the emission from . the fluxes from these lines formed at 2.5 mk show no increase over their average value during the flare , as shown in figure 1 , indicating that the flare is associated with the hotter corona rather than the accretion shock , and justifying its retention in our analysis . lines from , , ( fig . 1 ) , and are produced by the hotter component and may participate in the flare . while k02 found that was overpredicted by their model by a factor of about 3 , our models b , c , and d have no such problem . in fact , for models c and d , the small flux of the resonance line forces the emission measure to be negligible between @xmath33 and 5 mk ( fig . the relative elemental abundances given in table 6 are reasonably consistent for the different models . the abundances are also similar to those found by k02 , stelzer & schmitt ( 2004 ) , and drake et al . it is important to note that the continuum emission arises primarily from the hot component , except at the longest wavelengths ( fig . thus absolute abundances in the low @xmath0 component for ne , o , fe and other species can not be determined reliably . nevertheless , the extremely large ne / o abundance ratio appears to be a very robust result . differences in the absolute abundances derived here reflect both differences in @xmath0 and the degeneracy between emission measure and @xmath2 . for example , nitrogen and oxygen are formed entirely in the low temperature component and thus the abundance differences among the models reflect differences among the low temperature emission measure and @xmath2 , as does the difference in the neon abundance between models a and b. the difference in the neon abundance between models c and d also reflects the differences in their emission measures at 2.5 mk . weak emission is apparent from the s and ar complexes , but we have not included line fluxes from these elements in table 1 because the individual lines ( in particular the he - like resonance line ) can not not be cleanly determined in the four grating arms . instead we have used the model c global fit to obtain 0.21 @xmath17 0.09 and 1.41 @xmath17 0.49 times solar ( anders & grevesse 1989 ) for the s and ar abundances , respectively . a high ar / o abundance would add weight to drake et al.s ( 2005 ) suggestion that the accretion stream is depleted of grain - forming elements ; however , the high ar abundance is unfortunately not statistically significant . in this section , we discuss velocity constraints from the line measurements . the measurements of the emission line fluxes given in table 1 use gaussian lines in addition to the standard hetg calibration line spread function . table 7 gives the lines with the highest signal - to - noise ratios , for which gaussian widths are measured . lines not listed do not show statistically significant widths , but are consistent with the range of widths found here . for these fits the continuum level is allowed to vary to provide a better local fit . assuming the lines form in the same region , and using a flat continuum flux level from 13.45 to 13.75 , we force the 3 triplet lines to have the same gaussian width to improve the significance of the velocity measurement . figure 2 shows the best fit for the lines . as before , the centroids of the lines measured in each grating arm are fit independently . we thus obtain a velocity of 183 @xmath17 16 km s@xmath14 . the thermal doppler fwhm velocity for neon at 2.5 mk is 75 km s@xmath14 . the error in the calibration is estimated to be about 5% . thus we determine a turbulent velocity of 165 @xmath17 18 km s@xmath14 , including calibration error in quadrature with the 1@xmath18 statistical error . the line profiles are consistent with each other within the errors and suggest a range of velocities from zero up to about the sound speed of the gas ( 185 km s@xmath14 at 2.5 mk ) . the turbulent velocity is well below the preshock gas velocity of @xmath3500 km s@xmath14 . line centroids are consistent with reference wavelengths from atomdb to within the calibration uncertainty on the absolute wavelength scale . lrcccc & 10.239 & 0.0042 @xmath17 0.0030 & 122 @xmath17 87 & 75 & 96 + & 11.544 & 0.0024 @xmath17 0.0024 & 62 @xmath17 62 & 75 & 0 + & 12.134 & 0.0082 @xmath17 0.0009 & 203 @xmath17 23 & 75 & 189 + & 13.447 & 0.0103 @xmath17 0.0010 & 229 @xmath17 22 & 75 & 216 + & 13.553 & 0.0057 @xmath17 0.0014 & 126 @xmath17 32 & 75 & 101 + & 13.699 & 0.0073 @xmath17 0.0019 & 159 @xmath17 41 & 75 & 140 + & 15.014 & 0.0065 @xmath17 0.0040 & 129 @xmath17 80 & 45 & 121 + & 18.969 & 0.0074 @xmath17 0.0026 & 116 @xmath17 41 & 85 & 79 + we summarize here the physical conditions of the x - ray plasma from a consistent analysis of the spectra . the new diagnostic measurements strongly support k02 s argument that the low temperature x - ray component ( @xmath75 2.5 mk ) arises in the accretion shock . while the density we derive from , @xmath76 @xmath6 , is within the range found in active stars at higher @xmath0 ( @xmath3 6 mk ) , it is at least an order of magnitude larger than the few r - ratios reported ( huenemoerder et al . 2001 ; ness et al . 2002 ; osten et al . the @xmath1 derived from the r - ratio is also more than an order of magnitude larger than derived from in other cool stars ( see sanz - forcada et al . 2003 ; ness et al . 2004 ; testa et al . 2004 ) . we also measure the r - ratio and rule out photoexcitation , since the observed r - ratio can only be produced by high @xmath1 , namely @xmath77 @xmath6 . the g - ratios from , , and give similar @xmath78 mk , significantly lower than the @xmath0 of peak emissivity for ( 4 mk ) and ( 6.3 mk ) . the g - ratio from is particularly reliable because of good statistics and accurate atomic data . in our multi-@xmath0 models ( models c and d ) , this low @xmath0 component is isolated and does not connect continuously with the hotter , coronal component . active main - sequence stars tend to have emission measures rising up to 6 mk and above , such that , if biased at all , @xmath0-sensitive line ratios should be biased toward higher rather than lower temperature . thus the @xmath0 measurements also support the accretion shock scenario . more accurate g - ratio models for and are required to establish any @xmath0 differences among these ions . we also measure strong absorption using line ratios from several ions , and require at least two different absorbing column densities , with the higher charge states experiencing more absorption . it seems likely that the gap between the two components of the em distribution is caused by neutral h absorption of the soft x - ray spectrum that is produced by the hot coronal component . the pattern of absorption is reminiscent of the `` two - absorber x - ray sources '' reported by gdel et al . ( 2007b ) , where the coronal component has ten times more absorption than the soft component ; however , unlike tw hya , these sources have high accretion rates and are believed to drive jets . we note that the absorption in tw hya is identified here from line ratio diagnostics , whereas the absorption in gdel et al . ( 2007b ) was determined from ccd spectra . we were not able to determine multiple absorbers using standard global fitting methods , even with the high resolution spectrum presented here , because the continuum spectrum has low signal - to - noise ratio . on the other hand , it seems likely that with low spectral resolution , multiple absorbers could also be easily missed . thus the presence of more than one absorber may be a more universal characteristic than we can presently establish without reliable line ratio diagnostics . all the values of @xmath2 found here are higher than those found using the hydrogen ly@xmath49 profile , for which a conservative upper limit is @xmath79 @xmath46 ( herczeg et al . 2004 ) , indicating that the absorption is not due to the interstellar medium but is intrinsic to the stellar system . our column densities are consistent with the range of values previously reported from rosat and asca ( kastner et al . in fact the two different values from rosat and asca are consistent with our different values and may reflect the complex absorption coupled with the different instrument responses rather than time - variable absorption . as noted also by gdel et al . ( 2007b ) for the highly absorbed coronae of dg tau a , gv tau , dp tau , our x - ray measurements are higher than expected from the optical extinction assuming standard gas - to - dust ratios . for x - ray emission that is highly localized compared with the stellar ly@xmath25 , and preferentially absorbed by the accretion stream directly above it , a difference in @xmath2 derived by the two methods seems entirely reasonable . we can test the hypothesis that some of the measured x - ray emission comes from the accretion shock by constructing a standard one - dimensional ( 1d ) model of the magnetospheric infall , shock heating , and postshock cooling . this model is a simplified version of similar 1d models in the literature ( calvet & gullbring 1998 ; gnther et al . 2007 ) , and is tailored to specific measured properties of tw hya . the magnetospheric accretion is assumed to follow a set of dipole magnetic field lines that thread the accretion disk . we use knigl s ( 1991 ) expression for the inner truncation radius of the disk to determine the distance at which parcels of gas are launched from rest . this expression depends on the mass accretion rate @xmath80 , the surface magnetic field strength @xmath81 , and the stellar mass @xmath82 and radius @xmath83 . for tw hya , we take @xmath84 and @xmath85 ( batalha et al . 2002 ) . as in cranmer ( 2008 ) , we assume a canonical t tauri star magnetic field strength of @xmath86 g. a range of accretion rates for tw hya has been reported , from @xmath87 yr@xmath14 ( muzerolle et al . 2000 ) to @xmath88 yr@xmath14 ( batalha et al . 2002 ) . the low end of this range produces the best agreement with the observations presented here , so the muzerolle et al . accretion rate will be adopted for the cooling - zone models below . with the above values , the knigl ( 1991 ) truncation radius of the disk is found to be approximately @xmath89 , which implies a ballistic free - fall velocity at the stellar surface of @xmath90^{1/2 } \ , \approx 509 \,\ , \mbox{km } \,\ , \mbox{s}^{-1 } \,\ , .\ ] ] since the preshock gas is often assumed to have a sound speed only of order @xmath11 km s@xmath14 ( e.g. , muzerolle et al . 2001 ) , the highly supersonic flow should produce a strong shock at the stellar surface , heating the plasma to a postshock temperature of @xmath91 , where @xmath92 is the boltzmann constant and @xmath93 is the mean atomic mass . for the above parameters , this value is @xmath94 3.4 mk . [ we neglect any preshock heating due to photoionization since the x - ray luminosity is relatively low . see calvet & gullbring ( 1998 ) . ] the preshock mass density @xmath95 in the accretion stream can be estimated via mass flux conservation at the surface , i.e. , @xmath96 where the @xmath97 is the filling factor of the stellar photosphere heated by the accretion stream . the value of @xmath97 was estimated to be 1.1% for the dipole magnetic geometry assumed above ( see also cranmer 2008 ) . assuming complete ionization ( with 10% he by number ) , the preshock electron number density @xmath98 @xmath6 . because the shock is strong , we obtain an immediate postshock density @xmath99 @xmath6 ( i.e. , a factor of four increase across the shock , with the assumed adiabatic index of @xmath100 ) . the postshock velocity @xmath101 is thus approximately 130 km s@xmath14 . as one proceeds deeper through the cooling zone , the density increases and the velocity and temperature decrease , until a depth is reached at which there is no longer any influence from the shock . the `` bottom '' density @xmath102 can be estimated using ram pressure balance between the accretion stream and the unperturbed atmosphere . using equation 2 the pressure of the accretion stream is given by @xmath103 ( e.g. , hartmann et al . 1997 ; calvet & gullbring 1998 ) . assuming the accretion is stopped in the first few scale heights above the photosphere ( at which radiative equilibrium drives the temperature to @xmath104 ) , the pressure of the atmosphere is @xmath105 by equating the two pressures , the density can be found by solving for @xmath106 with @xmath107 k for tw hya ( batalha et al . 2002 ; see also cranmer 2008 ) . converting to electron density , this corresponds to @xmath108 @xmath6 , a factor of 1000 higher than the immediate postshock density . finally , we computed the physical depth and structure of the postshock cooling zone using the analytic model of feldmeier et al . ( 1997 ) , which assumes that the radiative cooling rate @xmath109 has a power - law temperature dependence ( @xmath110 ) . the thickness of the cooling zone is proportional to @xmath111 , and for the case of the tw hya parameters used here , it has a value of 385 km . the spatial dependence of density , temperature , and velocity in the cooling zone is specified by equations ( 8)(11 ) of feldmeier et al . ( 1997 ) , and the shapes of these curves resemble those of gnther et al.(2007)which were computed using a more detailed radiative cooling rate reasonably well ( see fig . note that choosing the stronger mass accretion rate of batalha et al.(2002 ) would have led to a truncation radius closer to the star ( @xmath112 ) , a lower freefall velocity ( @xmath113 km s@xmath14 ) , and thus a lower postshock temperature ( @xmath114 k ) . the densities would all have been higher than their counterparts above by about an order of magnitude , and the thickness of the cooling zone would have been smaller ( 32 km ) . using the temperature and density in the model , we compute the line intensities as functions of distance from the accretion shock . figure 9 shows line intensities through the region for the strongest lines of , , and . we compare the integrated intensities to the observations . assuming the abundances determined from model d , and using @xmath115 @xmath46 , the predicted line intensities for the , and resonance lines are 0.25 , 0.16 , and 0.40 times the observed fluxes , respectively . higher abundances improve the agreement , as do lower column densities . nevertheless , agreement within a factor of 4 seems quite reasonable , given the difficulty of obtaining absolute abundances and the complexity of the absorption . the predicted g - ratios give @xmath29 2.5 and 2.2 mk for and , respectively . the @xmath1 calculated for is @xmath116 @xmath6 , in excellent agreement with the observation ; however , for , the calculated @xmath1 is @xmath117 @xmath6 , seven times larger than observed . the density is clearly discrepant with the standard model . the physical conditions predicted at the shock front are in excellent agreement with the observations ; however , the predictions for the postshock cooling plasma are in stark disagreement with the observations . most problematic is that the electron density derived from is _ lower _ than that for , when the postshock gas should be compressing as it cools and slows and hence the density from should be larger than from . furthermore , @xmath2 determined from is four times _ smaller _ than that determined from , whereas in the standard postshock model should be observed through the same or larger column density of intervening material . in this section , we explore the properties of the postshock cooling plasma based on our new results . using the apec code to calculate emissivities for a relevant grid of @xmath0 and @xmath1 , we construct a set of two - component shock models for comparison with the diagnostic lines from the shock . the two components represent region 1 , the higher @xmath0 region near the shock front , and region 2 , the lower @xmath0 region of postshock plasma . we do not discuss the hot coronal plasma here , but consider the diagnostic lines from ions that are formed below 3 mk , in particular , , , and . each component in the model has four parameters : @xmath0 , @xmath1 , @xmath2 , and volume @xmath65 . the two regions experience different absorption with a factor of 2.5 larger @xmath2 at higher @xmath0 , as required by the resonance series lines . the abundances are taken from the model d fits . the predicted fluxes from the two regions are summed for comparison with the observations . table 8 gives the parameters for the best - fitting model , the predicted fluxes for each of the two regions , and the ratios of predicted to observed line fluxes . for , , and the predictions agree to better than 30% with the observed line fluxes . for the predictions are low by a factor of less than 2 , but the line ratios agree within 30% . the diagnostic line fluxes place tight constraints on the model . lcccc region 1 & @xmath118 & 3.00 & @xmath119 & @xmath120 + region 2 & @xmath121 & 1.75 & @xmath122 & @xmath123 + + & r & i & f & ly@xmath49 + o & 260 . & 166 . & 3.6 & 512 . + ne & 447 . & 260 . & 73.9 & + mg & 3.6 & 1.3 & 1.8 & + + & r & i & f & ly@xmath49 + o & 124 . & 70.8 & 36.7 & 25.1 + ne & 16.1 & 4.8 & 10.1 & + mg & 0.0 & 0.0 & 0.0 & + + & r & i & f & ly@xmath49 + o & 1.00 & 0.69 & 0.53 & 0.69 + ne & 1.31 & 0.85 & 0.75 & + mg & 1.28 & 1.14 & 1.13 & in this simple model , all of the emission from and most of the emission from come from region 1 , which we associate with the shock front . if we assume a cooling length of 200 km , based on the accretion shock model described above ( see fig . 9 ) , and use the derived emitting volume of @xmath124 @xmath125 , we obtain an area filling factor for region 1 of 1.5% of the surface area of the star . region 2 has a larger volume than region 1 by a factor of 300 as determined primarily from . if we assume that the cooling zone is the continuation of region 1 , we again obtain a length of about 200 km ; however , this short length implies a filling factor larger than the stellar surface area and is not plausible . instead , we use the scale height of 0.1 @xmath126 for 1.75 mk plasma to estimate a filling factor of 6.8% for region 2 , a value in good agreement with estimates from the ultraviolet ( costa et al . 2000 ) and optical continuum veiling ( batalha et al . 2002 ) . scaling the mass as @xmath127 implies that the mass of region 2 is 30 times the mass of region 1 . romanova et al . ( 2004 ) have used three - dimensional magnetohydrodynamic ( mhd ) simulations to show that the hot spots formed on the surface by accretion are inhomogeneous , with the lower density and temperature regions filling a larger surface area than the denser central region , seemingly suggestive of region 2 . however , the larger mass of region 2 also implies a higher accretion rate than for region 1 , and thus a larger density , in contradiction to our diagnostics . therefore , we do not interpret region 2 as a second accretion shock with a different accretion rate , but rather as a new accretion - driven phenomena . the impact from the accretion presumably provides the energy to heat additional material in the stellar atmosphere to over 1 mk . very recently , two - dimensional mhd simulations have shown that violent outflows of shock - heated material can propagate from the base of the accretion shock for cases with a high thermal to magnetic pressure ratio @xmath50 ( orlando et al . 2009 ) . one of the key concepts of the standard one - dimensional accretion model suggests a simple picture ( fig . the field line that truncates the disk separates the open magnetic field lines of the polar region from the closed magnetic field loops in the equatorial regions . the region heated by the shock ( our region 2 ) provides a large supply of ionized material available to form the soft x - ray emitting corona presented here , a warm wind ( dupree et al . 2005 ) , and soft x - ray jets ( gdel et al . this corona exists because magnetic loops confine it , but it is fed by the accretion process and is thus unlike stellar coronae in dwarf stars on the main sequence . figure 10 illustrates the shock and its surroundings . this model provides an explanation for the soft x - ray excess discovered in the _ xmm - newton _ extended survey of the taurus molecular cloud ( xest ; gdel et al . 2007c ; telleschi et al . grating spectra show that this excess emission manifests itself in ctts as enhanced relative to as compared with weak - line t tauri ( wtts ) and main - sequence stars ( gdel & telleschi 2007 ; robrade & schmitt 2007 ) . our measurements also demonstrate that the postshock cooling gas can , in some cases , dilute the signature of high @xmath1 and explain why not all accreting systems show high densities . having carefully distinguished between the accretion and corona in section 3 , we reconsider the assumption that these two components arise from different processes . gdel & telleschi ( 2007 ) find that the soft x - ray excess depends both on the level of magnetic activity as measured by the total ( including coronal ) @xmath12 and on the presence of accretion . while we suggest that the role of accretion is to heat and ionize the surrounding stellar atmosphere to create a soft x - ray ( postshock cooling ) plasma , it seems possible that this plasma could be further heated to @xmath128 mk , either by the usual coronal heating ( mhd ) processes , or possibly by the accretion energy . cranmer ( 2009 ) finds that in some systems , accretion can provide the energy for coronal heating to such high temperatures . studies of star forming regions have found that young stars show strong x - ray activity , with energetic flares and extremely high x - ray emitting temperatures ; however , the increase in activity with rotation rate breaks down for the accreting systems , leading to the conjecture that the stellar dynamo might operate differently in these systems ( preibisch et al . 2005 ) . we suggest that the accretion process may play a more prominent role than has generally been believed . while x - ray signatures of the accretion shock itself may not usually dominate the emission , the energy from accretion may in fact contribute to feeding and heating a new type of corona . we have analyzed the deep _ chandra _ hetg spectrum of tw hya to assess the relative contributions of accretion and coronal emission . we summarize the most important of our findings : * the spectrum shows line and continuum emission from @xmath128 mk coronal plasma , which may have a broad emission measure distribution . unlike active main sequence stars , the coronal em distribution abruptly cuts off below @xmath129 mk , probably due to absorption by the accretion stream . * the spectrum also shows emission lines from plasma at @xmath130 mk , as predicted by standard models of the accretion shock . @xmath0 and @xmath1 diagnostics from he - like are in excellent agreement with accretion models for tw hya . * we measure the flux of the forbidden line in tw hya for the first time to better than 3@xmath18 , allowing a determination of electron density . this density is lower than @xmath1 from by a factor of four and lower than @xmath1 from the standard postshock cooling model by a factor of seven , and leads to a new model of an accretion - fed corona . * the high @xmath0 ions suggest a stellar corona . by contrast , line ratio diagnostics from the lower @xmath0 ions , indicate two regions with different densities . in our model , the accretion shock itself has @xmath131 mk and @xmath132 @xmath6 , while the postshock cooling region has @xmath133 mk and @xmath134 @xmath6 . the postshock plasma has 30 times more mass than the shock itself . the surface area filling factor of the shock is 1.5% while the postshock filling factor is approximately 6.8% . * line ratio diagnostics require at least two different absorbers , with higher column density ( @xmath2 ) for higher charge state . absorption by the accretion stream is larger by a factor of 2.5 for the shock than for the postshock region . * a range of models for the em distribution and elemental abundances provides acceptable fits to the global spectrum . while absorption , emission measure , and abundances are not independently determined by such methods , the high ne / o abundance seems to be a robust result . * from line profile analysis of lines we determine a subsonic turbulent velocity of about 165 km s@xmath14 . this high resolution x - ray spectrum of tw hya presents a rich set of diagnostic emission lines for characterizing the physical conditions of the high energy plasma and understanding the dominant physical processes . the observations strongly support the model of an accretion shock producing a few mk gas at high density , as first suggested by kastner et al . ( 2002 ) from an earlier hetg spectrum . the diagnostics also support the role of accretion in producing the soft x - ray excess at previously discovered ( gdel & telleschi 2007 ; robrade & schmitt 2007 ) . this excess emission requires that the accretion shock heat a large volume in the surrounding stellar atmosphere . in tw hya the filling factor at is larger than the filling factor of the shock at , and is consistent with the filling factors of the ultraviolet and optical continua . both open and closed magnetic field lines may emerge from this surrounding area to channel this ionized gas back into the corona . while it is not yet clear whether the energy from the impact of accretion at the star can heat a hot ( 10 mk ) corona , accelerate a highly ionized stellar wind , or drive hot jets , we suggest that accretion - fed coronae may be ubiquitous in accreting young stars . we acknowledge support from nasa to the smithsonian astrophysical observatory ( sao ) under _ chandra _ go7 - 8018x for gjml . nsb and sjw were supported by nasa contract nas8 - 03060 to sao for the chandra x - ray center . src s contribution to this work was supported by the sprague fund of the smithonian institution research endowment , and by nasa grant nng04ge77 g to sao . we thank randall smith for supporting the customized apec code runs . we thank the _ chandra _ mission planning team for their efforts to accommodate the ground - based observing campaign .

the model under consideration is an asymmetric two - dimensional coulomb gas of positively and negatively charged pointlike particles , interacting via a logarithmic potential . this continuous system is stable against collapse of positive - negative pairs of charges for the dimensionless coupling constant ( inverse temperature ) . the mapping of the coulomb gas is made onto the complex bullough - dodd model , and recent results about that integrable 2d field theory are used . the mapping provides the full thermodynamics ( the free energy , the internal energy , the specific heat ) and the large - distance asymptotics of the particle correlation functions , in the whole stability regime of the plasma . the results are checked by a small- expansion and close to the collapse point . the comparison is made with the exactly solvable symmetric version of the model , and some fundamental changes in statistics caused by the charge asymmetry are pointed out . -1 cm -2 cm = cmmib10 scaled1 = cmmib7 scaled1 * key words : * two - dimensional coulomb gas ; exactly solvable models ; bullough - dodd model ; thermodynamics ; pair correlation functions . institute of physics , slovak academy of sciences , dbravsk cesta 9 , 842 28 bratislava , slovakia ; e - mail : [email protected]

cond-mat0210018

a two - dimensional ( 2d ) coulomb plasma is the continuous system of charged particles , plus perhaps a uniformly charged neutralizing background , confined to a plane and interacting via the logarithmic coulomb potential . in this paper , the classical equilibrium statistical mechanics at the ( dimensionless ) inverse temperature @xmath3 is studied . we will restrict ourselves to the simple point - particle coulomb systems : if there are at least two species of charged particles with opposite signs , the coulomb system is stable against the charge collapse for small enough @xmath7 ( in 3d , the collapse point @xmath8 and the stability requires quantum mechanics @xcite ) . going beyond the collapse point needs a short - distance ( e.g. hard core ) regularization of the coulomb interaction . the famous kosterlitz - thouless ( kt ) transition @xcite of infinite order from a high - temperature conducting phase to a low - temperature insulating phase takes place at some @xmath9 . like in other dimensions , the 2d coulomb systems admit the debye - hckel treatment in the high - temperature limit @xmath10 , as was rigorously proven in ref . @xcite . as concerns the bulk thermodynamics , based on the scale invariance of the logarithmic potential , the density derivatives of the free energy , like the pressure , can be obtained exactly in the whole stability range of temperatures @xcite-@xcite . on the other hand , the temperature derivatives of the free energy , like the internal energy or the specific heat , are nontrivial statistical quantities which can not be obtained by simple means and they are known only in special cases ( see below ) . the most studied versions of the coulomb plasma are the one - component plasma ( ocp ) and the symmetric two - component plasma ( tcp ) , or coulomb gas . the 2d ocp of equally , say unit , charged pointlike particles in a uniform neutralizing background is formally related to the fractional quantum hall effect @xcite . no collapse occurs . there are indications that , around @xmath11 , the fluid system undergoes a phase transition to a 2d wigner crystal @xcite . in a more recent paper @xcite , the existence of this transition has been put in doubt . the model is exactly solvable at @xmath12 by mapping onto free fermions , in the bulk @xcite as well as in some inhomogeneous situations ( see review @xcite ) . the symmetric 2d tcp consists of oppositely @xmath13 charged particles , no background is present . the collapse point @xmath14 coincides with the exactly solvable free - fermion point of the equivalent thirring model @xcite-@xcite . at this collapse point , for a fixed fugacity , while the free energy diverges , truncated ursell functions are nonzero and finite . quite recently @xcite , the complete bulk thermodynamics of the symmetric 2d tcp was derived exactly in the whole stability region @xmath15 . the mapping onto a bulk 2d sine - gordon theory with a conformal normalization of the cos - field was made , and recent results about that field theory were applied . subsequently , the surface tension of the same model in contact with an ideal - conductor @xcite and with an ideal - dielectric @xcite rectilinear walls was obtained via a mapping onto integrable 2d boundary sine - gordon theories with dirichlet and neumann boundary conditions , respectively . the large - distance behavior of the bulk charge @xcite and density @xcite correlation functions was derived exactly by exploring the form - factor theory of the equivalent sine - gordon model . hansen and viot @xcite have introduced a general asymmetric 2d coulomb gas which consists of two species of pointlike particles with positive and negative charges of arbitrary strengths @xmath16 @xmath17 . for @xmath18 being an integer , the model describes a plasma of electrons and ions of integer valence . without any loss of generality , we choose @xmath19 where @xmath20 is a positive integer . the model interpolates between the symmetric tcp @xmath21 and , after subtracting the kinetic energy of 2-species , the ocp obtained as the extreme asymmetry case @xmath22 . the boltzmann factor of a pair of positive and negative charges , @xmath23 , is integrable at short distance in 2d space for @xmath24 , so the collapse point is @xmath25 @xcite . in the case of a vanishing but nonzero hard core around particles , the kt phase transition was conjectured to take place at @xmath26 @xcite . highly asymmetric coulomb mixtures in the strong - coupling regime have attracted much attention in the last years @xcite-@xcite due to the phenomenon of overcharging ( charge inversion ) , i.e. the situation when the number of counterions in the vicinity of a macroion is so high that the macro - charge is overcompensated . in this work , we report the exact solution of the @xmath27 asymmetric 2d coulomb gas , with the lowest degree of charge asymmetry @xmath28 and @xmath29 , in the whole stability interval of pointlike particles @xmath2 . the exact solution involves the complete bulk thermodynamics ( the free energy , the internal energy , the specific heat ) and the large - distance behavior of the particle correlation functions . it is obtained by mapping the underlying coulomb gas onto a member of 2d toda field theories @xcite , namely the complex bullough - dodd ( cbd ) model @xcite-@xcite with a conformal normalization of the exponential field . recent results about that integrable field theory , derived by using the thermodynamic bethe ansatz ( thermodynamics ) and the form - factor method ( correlation functions ) , are applied . since the calculations in the cbd model are based on special analyticity assumptions , which are not yet rigorously proven , we check the results for the plasma by a small-@xmath3 expansion , using a renormalized mayer expansion in density @xcite , and close to the collapse @xmath4 point , using an electroneutrality sum rule @xcite combined with an independent - pair conjecture made by hauge and hemmer @xcite . the comparison is made with the symmetric version of the model , and the fundamental changes in statistics due to the charge asymmetry are pointed out . the paper is organized as follows . in section 2 , all relevant aspects of the mapping between the asymmetric @xmath30 2d coulomb gas and the cbd model are presented . the complete thermodynamics of the coulomb system is derived in section 3 . the asymptotic large - distance behavior of the particle correlation functions is obtained within the form - factor method in section 4 . section 5 is a brief recapitulation with some concluding remarks . the results are checked by the small-@xmath3 expansions in appendix a and close to the collapse @xmath4 point in appendix b. we consider the asymmetric tcp made up of two species of pointlike particles with charges @xmath16 @xmath31 given by ( [ 1.1 ] ) . the particles are confined to an infinite 2d space of points @xmath32 , and interact with each other by the pair coulomb potential . the coulomb potential @xmath33 at spatial position @xmath34 , induced by a unit charge at the origin , is given by the 2d poisson equation @xmath35 as follows @xmath36 the length constant @xmath37 , which fixes the zero point of the coulomb potential , will be set for simplicity to unity . the interaction energy @xmath38 of particles @xmath39 reads @xmath40 introducing the microscopic density of particles of species @xmath41 , @xmath42 , the microscopic densities of the total particle number and of the charge are @xmath43 respectively . the energy ( [ 2.3 ] ) can be thus written as @xmath44 where @xmath45 is the ( divergent ) self - energy . we will work in the grand canonical ensemble , with position - dependent fugacities @xmath46 of species @xmath47 . the grand partition function @xmath48 at inverse temperature @xmath3 , considered as a functional of the species fugacities , is defined by @xmath49 & = & \sum_{n_1=0}^{\infty } \sum_{n_2=0}^{\infty } { 1\over n_1 ! } { 1\over n_2 ! } \int \prod_{j=1}^{n_1 } \left [ { \rm d}^2 u_j z_1({\bf u}_j ) \right ] \nonumber \\ & & \times \prod_{j=1}^{n_2 } \left [ { \rm d}^2 v_j z_2({\bf v}_j ) \right ] \exp \left [ - \beta e_{n_1,n_2}(u , v ) \right ] \end{aligned}\ ] ] where @xmath50 denotes the coulomb interaction energy ( [ 2.3 ] ) , resp . ( [ 2.5 ] ) , of @xmath51 particles of type 1 at vector positions @xmath52 and @xmath53 particles of type 2 at positions @xmath54 . the statistical quantity @xmath48 can be expressed in terms of a 2d euclidean theory with the aid of the standard procedure ( see , e.g. , ref . @xcite ) . using the representation ( [ 2.5 ] ) for @xmath38 in @xmath55 and assuming that @xmath56 is the inverse operator of @xmath57 [ see relation ( [ 2.1 ] ) ] , one applies the hubbard - stratonovich transformation @xmath58 = { \int { \cal d}\phi \exp \left [ \int { \rm d}^2 r \left ( { 1\over 16\pi}\phi \delta \phi + { \rm i}\sqrt{\beta/4 } ~ \phi { \hat \rho}\right)\right ] \over \int { \cal d}\phi \exp \left ( \int { \rm d}^2 r { 1\over 16\pi}\phi \delta \phi\right)}\ ] ] where @xmath59 is a real scalar field and @xmath60 denotes the functional integration over this field . the term @xmath61 can be turned into @xmath62 by using integration by parts . the summation over @xmath51 and @xmath53 in ( [ 2.6 ] ) then implies @xmath63 = { \int { \cal d}\phi \exp \left ( - s[z_1,z_2 ] \right ) \over \int { \cal d}\phi \exp \left ( - s[0,0 ] \right)}\ ] ] where the action of the 2d euclidean field theory takes the form [ 2.9 ] @xmath64 & = & \int { \rm d}^2 r \left [ { 1\over 16\pi } ( \nabla \phi)^2 - z_1({\bf r } ) { \rm e}^{{\rm i}b\phi } - z_2({\bf r } ) { \rm e}^{-{\rm i}(b / q)\phi } \right ] \label{2.9a } \\ b^2 & = & \beta/4 \label{2.9b}\end{aligned}\ ] ] here , the fugacities @xmath65 @xmath17 are renormalized by the self - energy terms @xmath66 $ ] . for the homogeneous system with uniform species fugacities , @xmath67 , the action ( [ 2.9 ] ) simplifies to @xmath68\ ] ] note that this action is complex , except in the symmetric case @xmath69 with @xmath70 . the field representation of the multi - particle densities can be obtained from the functional generator @xmath48 , defined by ( [ 2.8 ] ) and ( [ 2.9 ] ) , in a straightforward way . at the one - particle level , the density of particles of type @xmath71 is given by @xmath72 consequently , @xmath73 where @xmath74 denotes the averaging over the action ( [ 2.10 ] ) . note that under the shift of the field variable @xmath75 in ( [ 2.10 ] ) ( which has no effect on @xmath48 as the whole ) , the species fugacities change as @xmath76 and @xmath77 $ ] , and therefore @xmath48 depends only on the combination @xmath78 . according to the definition ( [ 2.11 ] ) , this property of @xmath48 implies the neutrality condition @xmath79 . at the two - particle level , one introduces the two - body densities @xmath80 so that @xmath81 it is useful to introduce also the pair distribution functions @xmath82 and the ( truncated ) pair correlation functions @xmath83 in statistical mechanics , the short - distance behavior of the pair distribution function is dominated by the boltzmann factor of the pair coulomb potential @xcite , @xmath84 ( provided @xmath3 is small enough as explained below ) . the prefactors @xmath85 are expressible as follows @xmath86\ ] ] where @xmath87 is the excess chemical potential of species @xmath41 defined by @xmath88 and @xmath89 with @xmath90 arbitrary represents an extended definition of the excess chemical potential for a `` guest '' particle of charge @xmath90 put into the considered coulomb gas ( [ 1.1 ] ) . to get the stability region of @xmath89 , one has to consider the interaction boltzmann factor of the @xmath90-charge with an opposite charge from the plasma . when @xmath91 ( @xmath92 ) , the boltzmann factor with the opposite @xmath93 ( @xmath94 ) plasma charge at distance @xmath95 , @xmath96 ( @xmath97 ) , is integrable at small @xmath95 if and only if @xmath98 ( @xmath99 ) . @xmath89 is therefore finite if @xmath100 and goes to @xmath101 outside of this stability region . according to the formula ( [ 2.18 ] ) this means that the prefactor @xmath102 remains finite in the whole stability region @xmath103 , while the prefactors @xmath104 and @xmath105 are finite only if @xmath106 . the divergence of @xmath104 and @xmath105 in the middle of the stability region , at the point @xmath107 , is accompanied by a change in the short - distance behavior ( [ 2.17 ] ) of @xmath108 and @xmath109 . the general analysis in ref . @xcite shows that , in the short - distance limit @xmath110 , [ 2.20 ] @xmath111 } , \quad \quad \frac{2 q^2}{2 q - m + 1 } < \beta < \frac{2 q^2}{2 q - m } \label{2.20a } \end{aligned}\ ] ] where @xmath112 is an integer from the interval @xmath113 , and @xmath114 close to the collapse point @xmath115 , one finds [ 2.21 ] @xmath116 the collapse relation ( [ 2.21b ] ) tells us that , for @xmath117 , @xmath118 does not vanish as @xmath110 ( as is intuitively expected since equally charged particles repel each other ) , but diverges . this is related to a paradoxical clustering of counterions in the asymmetric plasma . we now intend to derive the short - distance behavior ( [ 2.17 ] ) , say for @xmath108 , in terms of the 2d field theory ( [ 2.10 ] ) . according to the definitions ( [ 2.6 ] ) , ( [ 2.13 ] ) and ( [ 2.15 ] ) , we have @xmath119 \prod_{j=1}^{n_2 } \left [ { \rm d}^2 v_j z_2({\bf v}_j ) \right ] \nonumber \\ & & \times \exp \left\ { - \beta e_{n_1 - 2,n_2}(u , v ) -\beta \int { \rm d}^2 r '' \left [ v({\bf r}-{\bf r } '' ) + v({\bf r}'-{\bf r } '' ) \right ] { \hat \rho}({\bf r } '' ) \right\}\end{aligned}\ ] ] where @xmath120 denotes the microscopic charge density of @xmath121 particles of type 1 and @xmath53 particles of type 2 . one shifts @xmath122 in the summation over @xmath51 . when @xmath123 , the exponential in ( [ 2.22 ] ) can be written in the form presented on the lhs of eq . ( [ 2.7 ] ) with the substitution @xmath124 . using the hubbard - stratonovich identity ( [ 2.7 ] ) , the previously generated single term @xmath125 now involves two terms : @xmath126 . one integrates over particle coordinates , except the fixed @xmath34-coordinate , and ends up with [ 2.23 ] @xmath127 analogously , one finds that @xmath128 comparing these relations with eqs . ( [ 2.17 ] ) - ( [ 2.19 ] ) , the excess chemical potential of a particle with charge @xmath90 embedded into the considered plasma is related to the one - point expectation of the exponential field as follows @xmath129 note that relations ( [ 2.12 ] ) are special cases of the general formula ( [ 2.24 ] ) . according to the analysis after eq . ( [ 2.19 ] ) and with respect to the relationship ( [ 2.9b ] ) , the one - point expectation @xmath130 is finite if @xmath131 , and goes to @xmath132 otherwise . comparing relations ( [ 2.23 ] ) with eqs . ( [ 2.14 ] ) and ( [ 2.15 ] ) , the short - distance behavior of two - point expectations of the exponential field is given by @xmath133 in quantum field theory , the model ( [ 2.10 ] ) can be regarded as the gaussian conformal field theory perturbed by the operator @xmath134 $ ] . the species fugacities @xmath135 and @xmath136 are renormalized by the ( divergent ) self - energy factors . to give the parameters @xmath135 and @xmath136 a precise meaning , one has to fix the normalization of the adjoint exponential fields . the normalization which corresponds to the short - distance limit of the two - point function ( [ 2.25 ] ) is known as conformal . under the conformal normalization , the diverging self - energy factor disappears from statistical relationships . this makes the bridge between the underlying asymmetric coulomb plasma and the corresponding field theory with the action ( [ 2.10 ] ) , treated within the conformal perturbation theory . note that @xmath90 and @xmath137 should be sufficiently small to ensure that ( [ 2.25 ] ) is the leading short - distance asymptotics . from the point of view of the above equivalence , the crucial is the stability of the two - point expectation asymptotics ( [ 2.25 ] ) for two opposite charges @xmath138 and @xmath139 . in that case , the stability region of @xmath140 is @xmath141 ; for the symmetric coulomb gas @xmath142 , for @xmath27 @xmath143 incidently coincides with @xmath144 and for any finite @xmath145 it holds @xmath146 . there is a large class of integrable 2d field theories , known as the affine toda theories , which are based on the dynkin - diagram classification of simple lie groups ( for a nice review , see ref . @xcite ) . let @xmath147 be a simple lie algebra of rank @xmath95 , @xmath148 a set of simple @xmath95-dimensional roots of the corresponding dynkin diagram and @xmath149 ( the coefficients @xmath150 are called kac labels ) the maximal root . the affine toda theory built on @xmath147 is defined by the action @xmath151\ ] ] where the field @xmath152 consists of @xmath95 real scalar components and @xmath153 is the real coupling constant . the fields in eq . ( [ 3.1 ] ) are normalized so that at @xmath154 @xmath155 for the one - component @xmath156 case of interest , there exist two integrable toda theories . when @xmath157 lie group with @xmath158 and @xmath159 , one obtains the sinh - gordon ( @xmath153 real ) or sine - gordon ( @xmath153 pure imaginary ) models . the sine - gordon model is identified with the action ( [ 2.10 ] ) with @xmath69 and describes the thermodynamics of the symmetric 2d tcp ( see refs . @xcite-@xcite ) . when @xmath160 lie group with @xmath161 and @xmath162 , the action ( [ 3.1 ] ) takes the form @xmath163\ ] ] this theory is known as the bullough - dodd ( bd ) model @xcite . its complex version , obtained via the substitutions @xmath164 and @xmath165 , @xmath166\ ] ] is referred to as the zhiber - mikhailov - shabat model @xcite , or simply the complex bullough - dodd ( cbd ) model . comparing ( [ 3.4 ] ) with ( [ 2.10 ] ) one observes that the cbd model with the short - distance normalization ( [ 2.25 ] ) is the 2d field realization of the @xmath27 asymmetric tcp , i.e. the system of @xmath94 and @xmath167 charged particles . according to ( [ 2.9 ] ) , the coupling constant @xmath153 is related to the inverse temperature by @xmath168 and the parameters @xmath65 @xmath31 represent the renormalized species fugacities . the particle spectrum of the bd model ( [ 3.3 ] ) consists of a single neutral particle of mass @xmath112 . the two - particle @xmath169-matrix was described in ref . the ( dimensionless ) specific grand potential @xmath170 where @xmath48 is given by ( [ 2.8 ] ) and @xmath171 is the volume , was obtained in terms of the particle mass @xmath112 in ref . @xcite following the thermodynamic bethe ansatz technique @xcite . in the same ref . @xcite , the relation between the particle mass @xmath112 and the model parameters @xmath172 was established under the conformal normalization , and an explicit formula for the mean value of the exponential field @xmath173 was suggested by exploring a reflection relationship between the bd model and the 2d liouville theory @xcite . the derivation of the results was based on special analyticity assumptions , so their verification by various checks is needed . the spectrum of the cbd model ( [ 3.4 ] ) exhibits an extremely complicated hierarchy of particles @xcite . the fundamental particle is a three - component kink . the kinks generate one - component bound states ( breathers ) and higher kinks , these higher kinks generate new breathers and new kinks , etc . the important simplifying fact is that , in the whole stability interval of interest @xmath174 @xmath175 , the lightest particle , the 1-breather , corresponds to the analytic continuation of the only particle in the spectrum of the bd model . since the lightest particle dominates in the thermodynamic limit @xmath176 , we can apply all results of ref . @xcite presented in the above paragraph , with the substitutions @xmath164 and @xmath165 , also to the cbd model . in particular , the specific grand potential ( [ 3.5 ] ) takes the form [ 3.6 ] @xmath177 and the mass of the lightest 1-breather reads @xmath178^{(1+\xi)/6 } \left [ \frac{2 z_2 \pi \gamma(1-b^2/4)}{\gamma(b^2/4 ) } \right]^{(1+\xi)/3}\ ] ] note that at the collapse point @xmath179 @xmath180 the particle mass @xmath181 . the expected divergence of @xmath182 ( [ 3.6 ] ) at @xmath179 is therefore caused by @xmath183 , while the prefactor to @xmath183 remains finite . such behavior is opposite to that observed in the symmetric 2d tcp @xcite at the corresponding collapse point where the mass of the lightest sine - gordon breather , @xmath184 , is finite and @xmath182 diverges due to the prefactor to @xmath185 . the expectation value of the exponential field is given by ( see also @xcite ) [ 3.8 ] @xmath186^{\frac{2a}{3b } } \left [ \frac{m \gamma(1-\xi/3 ) \gamma((1+\xi)/3)}{2^{2/3 } \sqrt{3 } \gamma(1/3 ) } \right]^{2a^2-ab } \nonumber \\ & & \times \exp \left\ { \int_0^{\infty } \frac{{\rm d}t}{t } \left [ \frac{\sinh((2-b^2)t ) \psi(t , a)}{\sinh(3(2-b^2)t ) \sinh(2t)\sinh(b^2 t ) } - 2 a^2 { \rm e}^{-2 t } \right ] \right\ } \label{3.8a}\end{aligned}\ ] ] where @xmath187 \label{3.8b}\end{aligned}\ ] ] the integral in ( [ 3.8a ] ) is well defined if @xmath188 taking into account ( [ 2.12 ] ) , the charge - neutrality condition in the considered coulomb gas , @xmath189 , results in the equality @xmath190 using the integral formula for the logarithm of the gamma function @xcite @xmath191 , \quad \quad { \rm re}(z ) > 0\ ] ] it can be readily verified that the suggested formula ( [ 3.8 ] ) is consistent with the neutrality condition ( [ 3.10 ] ) . we are now ready to derive the basic density - fugacity relationship for the asymmetric @xmath30 coulomb gas . the total particle number density is @xmath192 where the auxiliary parameter @xmath193 , introduced in ( [ 3.6b ] ) , is expressible in the inverse temperature @xmath194 as follows @xmath195 after some algebra , eqs . ( [ 3.6 ] ) and ( [ 3.7 ] ) give the explicit density - fugacity relationship @xmath196^{1-\beta/8 } \nonumber \\ & & \times \left [ \frac{\pi \gamma(1-\beta/4)}{\gamma(\beta/4 ) } \right]^{1/3 } \left [ \frac{2 \pi \gamma(1-\beta/16)}{\gamma(\beta/16 ) } \right]^{2/3}\end{aligned}\ ] ] the length constant @xmath37 in ( [ 2.2 ] ) was set to unity . this can be shown to imply that the fugacity product @xmath197 in neutral configurations of ( [ 2.6 ] ) has dimension @xmath198^{-6(1-\beta/8)}$ ] , and so eq . ( [ 3.14 ] ) is dimensionally correct . the small-@xmath3 expansion of the rhs of ( [ 3.14 ] ) reads @xmath199 \frac{\beta}{8 } + \left [ 3 \psi'(2/3 ) - 2 \pi^2 \right ] \frac{\beta^2}{1728 } + o(\beta^3 ) \right\}\ ] ] where @xmath200 is the euler number and @xmath201/{\rm d}x$ ] is the psi function . the series representation of the first derivative of @xmath202 reads @xmath203 the series expansion ( [ 3.15 ] ) is checked up to the indicated order in appendix a.1 by using a bond - renormalized mayer expansion in density @xcite . for a fixed fugacity product @xmath197 , relation ( [ 3.14 ] ) implies the expected collapse singularity of the density @xmath204 : @xmath205 ^ 2 \left [ \frac{\gamma(3/4)}{\gamma(1/4 ) } \right]^{4/3 } \frac{1}{(1-\beta/4)^{2/3 } } \quad \quad { \rm as}\ \beta \to 4 ^ -\ ] ] this singularity is reproduced indirectly in appendix b.1 by applying a ( slightly modified ) perfect - screening sum rule , which is another important check of the basic result ( [ 3.14 ] ) . to obtain the complete thermodynamics of the asymmetric @xmath30 coulomb gas , we pass from the grandcanonical to the canonical ensemble via the legendre transformation @xmath206 where [ 3.19 ] @xmath207 @xmath208 , @xmath209 with @xmath210 being the total particle number , and @xmath211 is the chemical potential of species @xmath41 . the dimensionless specific free energy , @xmath212 , is thence given by @xmath213 with the aid of the density - fugacity relationship ( [ 3.14 ] ) , one has explicitly @xmath214 \nonumber \\ & & - \frac{1}{3 } \ln \left [ \frac{\gamma(1-\beta/4)}{\gamma(\beta/4 ) } \right ] - \frac{2}{3 } \ln \left [ \frac{2\gamma(1-\beta/16)}{\gamma(\beta/16 ) } \right]\end{aligned}\ ] ] the ( excess ) internal energy per particle , @xmath215 , and the ( excess ) specific heat at constant volume per particle , @xmath216 , are determined by elementary thermodynamics as follows [ 3.23 ] @xmath217 the specific heat is independent of the particle number density @xmath204 , which is a specificity of the 2d pointlike coulomb gases . the expansion of @xmath218 near the collapse point @xmath4 results into the laurent series @xmath219 the leading two terms in ( [ 3.24 ] ) are reproduced indirectly in appendix b.2 by applying an independent - pair conjecture of hauge and hemmer @xcite . in a 2d integrable field theory with particle spectrum @xmath220 , correlation functions of local operators @xmath221 ( @xmath222 is a free parameter ) are expressible as infinite convergent series over multi - particle intermediate states ( see e.g. ref . @xcite ) . for the truncated two - point correlation functions @xmath223 the series reads @xmath224 here , @xmath225 is the rapidity which parametrizes the energy @xmath38 and the momentum @xmath226 of a particle @xmath227 of mass @xmath228 as follows @xmath229 and the normalization constants in the form factors @xmath230 depend on the specific form of the operator @xmath221 . in what follows , we will consider @xmath221 to be an exponential field : @xmath231 in the case of the bd model with the action ( [ 3.3 ] ) , and @xmath232 in the case of the cbd model with the action ( [ 3.4 ] ) . the form - factor representation ( [ 4.2 ] ) is particularly useful in the large - distance limit @xmath233 . the dominant contribution in this asymptotic limit comes from a one - particle intermediate state with the minimum value of the particle mass @xmath112 , at the point of the vanishing rapidity @xmath234 . the consequent exponential decay @xmath235 is multiplied by a slower ( inverse power law ) decaying function , whose particular form depends on the one - particle form factors . for the bd model ( [ 3.3 ] ) with the only particle @xmath236 in the spectrum , the one - particle form factors @xmath237 and @xmath238 are presented for the exponential field ( [ 4.4a ] ) in refs . the transition to the cbd model ( [ 3.4 ] ) is straightforward since , as was already mentioned in section 3 , the analytic continuation of the single particle in the bd spectrum corresponds to the lightest particle of interest in the cbd model , the 1-breather with mass @xmath112 given by ( [ 3.7 ] ) . in particular , after the substitutions @xmath239 and @xmath164 in the one - particle form - factor formulae for the bd model @xcite , the lightest - particle form factor of the exponential field ( [ 4.4b ] ) in the cbd model is given by [ 4.5 ] @xmath240 where @xmath241^{1/2 } \nonumber \\ & & \times \exp \left\ { - 2 \int_0^{\infty } \frac{{\rm d}t}{t } \frac{\cosh(t/6 ) \sinh(t\xi/3 ) \sinh(t(1+\xi)/3)}{\sinh t \cosh(t/2 ) } \right\ } \label{4.5b}\end{aligned}\ ] ] according to formulae ( [ 2.12])-([2.16 ] ) adapted to the considered asymmetric @xmath30 2d coulomb gas , the pair correlation function @xmath242 of species @xmath41 and @xmath243 with the corresponding charges @xmath16 and @xmath244 is expressible in terms of the averages over the equivalent cbd model as follows @xmath245 using the form - factor representation ( [ 4.2 ] ) for @xmath246 , the leading contribution to @xmath242 in the limit @xmath247 is determined by the lightest particle in the cbd model , the 1-breather with mass @xmath112 given by ( [ 3.7 ] ) and the form factor @xmath248 given by ( [ 4.5 ] ) . consequently , @xmath249 since @xmath250 at asymptotically large @xmath95 ( @xmath251 is the modified bessel function of second kind ) , we finally arrive at @xmath252 here , @xmath253 and the mass @xmath112 ( [ 3.7 ] ) is expressible in terms of the inverse debye length for the considered coulomb gas @xmath254 by combining eqs . ( [ 3.6 ] ) and ( [ 3.12 ] ) , @xmath255^{1/2}\ ] ] we recall that @xmath256 . in the high - temperature limit , the parameters @xmath257 ( [ 4.10 ] ) and @xmath112 ( [ 4.12 ] ) of the large - distance asymptotics ( [ 4.9 ] ) have the following small-@xmath3 expansions @xmath258 \right\ } + o(\beta^3 ) \label{4.13a } \\ m & = & \kappa \left [ 1 + \frac{\pi \beta}{48 \sqrt{3 } } + o(\beta^2 ) \right ] \label{4.13b}\end{aligned}\ ] ] these small-@xmath3 expansions are checked up to the indicated order in appendix a.2 by using the renormalized mayer expansion . the leading - order terms @xmath259 and @xmath260 correspond to the debye - hckel approximation , the first corrections to this approximation are implied by the renormalized meeron graph . let us introduce the charge density and particle number density combinations of the pair correlation functions @xmath261 respectively . it is evident that within the debye - hckel approximation the two - particle correlations are determined at large distances by the charge - charge correlation function @xmath262 , while @xmath263 is identically equal to zero . taking into account the first correction in @xmath257 ( [ 4.13a ] ) , the strong division between @xmath264 and @xmath265 disappears : both of them decay at large @xmath95 exponentially with the same correlation length @xmath266 , only the corresponding @xmath3-dependent prefactors differ from one another . the prefactor is of the form @xmath267 in the case of the symmetric 2d tcp @xcite , and so at any @xmath3 in the stability regime the large - distance asymptotics of the two - particle correlations are determined exclusively by @xmath262 ( @xmath263 is related to the heavier 2-breather and therefore goes to zero faster then @xmath262 ) . the asymmetry in the particle charges thus causes a fundamental change in the relative large - distance behavior of the charge and density correlation functions . since @xmath268 at the collapse point @xmath4 , the particle mass @xmath112 ( [ 4.12 ] ) diverges , and @xmath269 given by ( [ 4.9 ] ) reduces trivially to zero . on the other hand , the mass of the lightest particle is finite ( for a fixed @xmath270 ) at the collapse point for the symmetric 2d tcp @xcite , and the corresponding ursell functions @xmath271 have a nontrivial large - distance dependence . also from this point of view the asymmetry in the particle charges has a relevant influence on the large - distance characteristics of particle correlation functions . in this paper , we have solved exactly the 2d coulomb gas of pointlike charged particles , with the charge asymmetry @xmath28 and @xmath29 , via the equivalent 2d cbd field theory . the previous calculations in the cbd model were based on special analyticity conjectures . this is why we check our results for the asymmetric coulomb gas , in the small-@xmath3 limit and close to the collapse @xmath4 point . the small-@xmath3 series expansions are generated using the renormalized mayer expansion in density , which is a simplified coulomb version of the feynman diagrammatic perturbation technique in 2d field theories . on the other hand , the important check of the results close to the opposite collapse point , based on screening properties of the coulomb plasma , is original and has no counterpart in the field theory . the asymmetry in the strength of the particle charges brings two fundamental modifications in the statistical behavior of the 2d coulomb gas in comparison with its symmetric version . firstly , the large - distance exponential decays of the charge and density correlation functions are characterized by the same correlation length ( in the symmetric coulomb gas @xcite , the charge correlation length is twice larger than the density one ) . such behavior does not occur in the debye - hckel treatment of the model . secondly , the mass of the lightest particle in the spectrum of the cbd model , which determines the thermodynamics and the asymptotics of the particle correlations , diverges at the collapse point @xmath4 . as a consequence , the truncated particle distributions are trivially equal to zero at @xmath4 . this is in contrast to the symmetric 2d coulomb gas with finite and nonzero particle distributions at its collapse free - fermion point . the other cases of the asymmetric coulomb gases do not belong to the family of integrable 2d toda field theories . on the other hand , the 2d ocp , which corresponds to the extreme charge - asymmetric case , is exactly solvable at its free - fermion point @xcite . the 2d ocp has a field representation @xcite which resembles the one of the 2d complex liouville model with a kind of `` background '' charge . although this theory does not belong to the toda theories , it is integrable at least at the aforementioned free - fermion point . it might be therefore useful to explore its integrability properties at an arbitrary temperature , first in the classical limit and subsequently at the quantum level . let us consider a general fluid composed of distinct species of particles @xmath272 with the corresponding position - dependent densities @xmath273 . the particles @xmath274 and @xmath275 interact through the pair potential @xmath276 , where vector position @xmath277 is represented simply by @xmath274 . the technique of the bond - renormalization in the ordinary mayer expansion in density ( for details , see refs . @xcite ) is based on an expansion of each mayer function in the inverse temperature @xmath3 , and on a consequent series resummation of two - coordinated field circles . the renormalized @xmath278-bonds are given by @xmath279{$1,\sigma_1 $ } \text(32,0)[]{$2,\sigma_2 $ } \text(17,23)[]{$k$ } \end{picture } } \ \ \ = \ \ \ { \begin{picture}(32,20)(0,7 ) \dashline(0,10)(32,10){5 } \bcirc(0,10){2.5 } \bcirc(32,10){2.5 } \text(0,0)[]{$1,\sigma_1 $ } \text(32,0)[]{$2,\sigma_2 $ } \text(17,23)[]{$-\beta v$ } \end{picture}}\ \ + \ \ { \begin{picture}(64,20)(0,7 ) \dashline(0,10)(32,10){5 } \dashline(32,10)(64,10){5 } \bcirc(0,10){2.5 } \bcirc(64,10){2.5 } \vertex(32,10){2.2 } \text(0,0)[]{$1,\sigma_1 $ } \text(64,0)[]{$2,\sigma_2 $ } \end{picture}}\ \ + \ \ { \begin{picture}(96,20)(0,7 ) \dashline(0,10)(32,10){5 } \dashline(32,10)(64,10){5 } \dashline(64,10)(96,10){5 } \bcirc(0,10){2.5 } \bcirc(96,10){2.5 } \vertex(32,10){2.2 } \vertex(64,10){2.2 } \text(0,0)[]{$1,\sigma_1 $ } \text(96,0)[]{$2,\sigma_2 $ } \end{picture}}\ \ + \ldots\ ] ] or , algebraically , @xmath280 \nonumber \\ & & + \sum_{\sigma_3 } \int { \rm d}3~ [ -\beta v(1,\sigma_1 \vert 3,\sigma_3)]~ n(3,\sigma_3)~ k(3,\sigma_3 \vert 2,\sigma_2)\end{aligned}\ ] ] it is straightforward to verify by variation of ( [ a.1 ] ) that it holds @xmath281 the excess helmholtz free energy @xmath282 is expressible in the renormalized format as follows [ a.3 ] @xmath283 = \ \ \begin{picture}(50,20)(0,7 ) \dashline(0,10)(40,10){5 } \vertex(0,10){2.2 } \vertex(40,10){2.2 } \end{picture } + \ \ d^{(0)}[n ] + \sum_{s=1}^{\infty } d^{(s)}[n]\ ] ] where @xmath284 0.7 cm is the sum of all ring diagrams which can not undertake the bond - renormalization procedure and @xmath285{$d^{(1)}$ } \end{picture } \begin{picture}(60,40)(0,7 ) \photonarc(20,-10)(28,45,135){1}{9 } \photonarc(20,30)(28,225,315){1}{9 } \photonarc(20,6)(20,15,165){1}{11 } \photonarc(20,14)(20,195,345){1}{11 } \vertex(0,10){2 } \vertex(40,10){2 } \text(20,-25)[]{$d^{(2)}$ } \end{picture } \begin{picture}(60,40)(0,16 ) \photonarc(32,6)(32,115,170){1}{7.5 } \photonarc(-12,40)(32,295,355){1}{7.5 } \photonarc(9,5)(32,10,70){1}{7.5 } \photonarc(52,40)(32,185,250){1}{7.5 } \photon(0,10)(40,10){1}{8 } \vertex(0,10){2 } \vertex(40,10){2 } \vertex(20,35){2 } \text(20,-15)[]{$d^{(3)}$ } \end{picture } \setscale{0.9 } \begin{picture}(60,40)(0,13 ) \photon(0,0)(0,40){1}{7 } \photon(40,0)(40,40){1}{7 } \vertex(0,0){2 } \vertex(40,0){2 } \vertex(0,40){2 } \vertex(40,40){2 } \photonarc(20,-20)(28,45,135){1}{8 } \photonarc(20,20)(28,225,315){1}{8 } \photonarc(20,20)(28,45,135){1}{8 } \photonarc(20,60)(28,225,315){1}{8 } \text(20,-19)[]{$d^{(4)}$ } \end{picture } \begin{picture}(60,40)(0,12 ) \photon(0,0)(40,0){1}{7 } \photon(0,0)(0,40){1}{7 } \photon(0,40)(40,40){1}{7 } \photon(40,0)(40,40){1}{7 } \photon(0,0)(40,40){1}{8 } \photon(0,40)(40,0){1}{8 } \vertex(0,0){2 } \vertex(40,0){2 } \vertex(0,40){2 } \vertex(40,40){2 } \text(20,-19)[]{$d^{(5)}$ } \end{picture } \setscale{1 } \ { \rm etc.}\ ] ] 1.5 cm are all remaining completely renormalized graphs . the free energy is the generating functional for one - body , two - body , etc . densities . the density - fugacity relationship is generated via [ a.4 ] @xmath286 & = & \frac{\delta ( -\beta f^{\rm ex})}{\delta n(1,\sigma_1 ) } \nonumber \\ & = & \ \ \begin{picture}(50,20)(0,7 ) \dashline(0,10)(40,10){5 } \bcirc(0,10){2.2 } \vertex(40,10){2.2 } \text(0,0)[]{$1,\sigma_1 $ } \end{picture } + \ \ d^{(0)}(1,\sigma_1 ) + \sum_{s=1}^{\infty } d^{(s)}(1,\sigma_1 ) \label{a.4a}\end{aligned}\ ] ] where @xmath287 can be readily obtained in the form @xmath288 \big\vert_{\sigma_2=\sigma_1}\ ] ] and @xmath289 the direct correlation function is generated via [ a.5 ] @xmath290{$1,\sigma_1 $ } \text(43,0)[]{$2,\sigma_2 $ } \end{picture } \ \ c^{(0)}(1,\sigma_1\vert 2,\sigma_2 ) + \sum_{s=1}^{\infty } c^{(s)}(1,\sigma_1\vert 2,\sigma_2 ) \label{a.5a}\end{aligned}\ ] ] where @xmath291 corresponds to the meeron graph , @xmath292{$1,\sigma_1 $ } \text(43,0)[]{$2,\sigma_2 $ } \end{picture } = \frac{1}{2 ! } k^2(1,\sigma_1\vert 2,\sigma_2)\ ] ] and @xmath293 the pair correlation function @xmath294 , defined by relations ( [ 2.13 ] ) - ( [ 2.16 ] ) , is related to @xmath295 via the ornstein - zernike ( oz ) equation @xmath296 notice that the functional derivatives with respect to the density field generate root circles not only at obvious field - circle positions , but also on @xmath278 bonds according to formula ( [ 4.2 ] ) , causing their right @xmath278-@xmath278 division . let us return to the 2d asymmetric coulomb gas with two kinds of particles @xmath47 of charges @xmath297 , interacting via the logarithmic interaction [ a.7 ] @xmath298 we consider the infinite - volume limit , characterized by homogeneous densities @xmath299 constrained by the neutrality condition @xmath300 , so that @xmath301 with @xmath204 being the total particle density . two - body functions are both isotropic and translationally invariant , @xmath302 , @xmath303 . it follows from eq . ( [ a.1 ] ) that the renormalized bonds exhibit the same charge - dependence as the interaction under consideration ( [ a.7 ] ) , [ a.9 ] @xmath304 here , @xmath251 is the modified bessel function of second kind and @xmath305^{1/2 } = \sqrt{\pi \beta n}\ ] ] is the inverse debye length . the small-@xmath306 expansion of the modified bessel function @xmath307 reads @xcite @xmath308 where @xmath309 . thus , in the uniform regime with the interaction ( [ a.7 ] ) and the renormalized interaction ( [ a.9 ] ) , eq . ( [ a.4b ] ) yields @xmath310 , \quad \quad \sigma = 1 , 2\ ] ] the first term on the rhs of ( [ a.4a ] ) disappears due to the charge neutrality . consequently , @xmath311 + \frac{\partial}{\partial n_{\sigma } } \sum_{s=1}^{\infty } \frac{d^{(s)}}{v } , \quad \quad \sigma = 1 , 2\ ] ] where @xmath176 is the volume of the system . for the dimensionless quantity of interest @xmath312 , ( [ a.13 ] ) gives @xmath313 \frac{\beta}{8 } - \left ( \frac{1}{3 } \frac{\partial}{\partial n_1 } + \frac{2}{3 } \frac{\partial}{\partial n_2 } \right ) \sum_{s=1}^{\infty } \frac{d^{(s)}}{v } \right\}\ ] ] when all @xmath314 are set to zero , we have the debye - hckel approximation valid in the @xmath315 limit and , indeed , ( [ a.14 ] ) then reproduces the leading term of the exact @xmath3-expansion ( [ 3.15 ] ) . the scaling form of the renormalized interaction bonds ( [ a.9 ] ) permits us to perform the @xmath3-classification of @xmath316 . let the given completely renormalized diagram @xmath317 be composed of @xmath318 skeleton vertices and @xmath319 bonds . every dressed bond @xmath278 brings the factor @xmath320 and enforces the substitution @xmath321 which manifests itself as the factor @xmath322 for each field - circle integration @xmath323 . since there are @xmath324 independent field - circle integrations in @xmath317 we conclude that @xmath325 in the sketch ( [ a.3c ] ) , the only diagram which contributes to the @xmath326 order is @xmath327 , the next diagrams @xmath328 and @xmath329 constitute the complete set of contributions to the @xmath330 order , etc . we are interested in the lowest correction to the debye - hckel limit , so only the diagram @xmath327 has to be analyzed . the contribution of this diagram is expressible as @xmath331 ^ 3 \nonumber \\ & = & - \frac{1}{2 ! 3 ! } \beta^2 \frac{(n_1-n_2/8)^2}{n_1+n_2/4 } \int \frac{{\rm d}^2 r}{2\pi } k_0 ^ 3({\bf r})\end{aligned}\ ] ] to evaluate the integral in ( [ a.16 ] ) , we can make use of the 2d fourier components of @xmath332 and @xmath333 ( see appendix of ref . @xcite ) to derive @xmath334}{(1+k^2)\sqrt{1+(k/2)^2}}\ ] ] after the substitution @xmath335 , one gets @xmath336\end{aligned}\ ] ] where @xmath337 is given by ( [ 3.16 ] ) . using the readily derivable relationships @xcite [ a.19 ] @xmath338 \label{a.19a } \\ \psi'(x ) & = & \frac{1}{9 } \left [ \psi'\left(\frac{x}{3}\right ) + \psi'\left(\frac{x+1}{3}\right ) + \psi'\left(\frac{x+2}{3}\right ) \right ] \label{a.19b } \\ \psi'(x ) & = & \psi'(x+1 ) + \frac{1}{x^2 } \label{a.19c}\end{aligned}\ ] ] valid for any @xmath306 , one finds @xmath339\ ] ] the consideration of ( [ a.20 ] ) in ( [ a.16 ] ) leads to @xmath340 \frac{\beta^2}{1728}\ ] ] where ( [ a.8 ] ) was taken into account . finally , inserting ( [ a.21 ] ) into ( [ a.14 ] ) , the term of the order @xmath326 is reproduced exactly in the exponential of the small-@xmath3 expansion ( [ 3.15 ] ) , generated from the exact density - fugacity relationship ( [ 3.14 ] ) . in the fourier picture , the oz equation ( [ a.6 ] ) reads @xmath341 it follows from ( [ a.5 ] ) that , at the lowest order in @xmath3 ( debye - hckel approximation ) , [ a.23 ] @xmath342 consequently , from ( [ a.22 ] ) , [ a.24 ] @xmath343 and the asymptotic form of @xmath344 is @xmath345 reproducing thus the expected result ( [ 4.9 ] ) with @xmath346 ( [ 4.13a ] ) and @xmath347 $ ] ( [ 4.13b ] ) . with the aid of eq . ( [ a.5 ] ) , the direct correlation function is determined up to the @xmath326 order as follows @xmath348 taking @xmath349 as the unit of length , one has in the fourier space @xmath350}{k~ \sqrt{1+(k/2)^2}}\ ] ] the oz equation ( [ a.22 ] ) implies [ a.28 ] @xmath351 \right\ } \label{a.28a } \\ { \hat h}_{12}(k ) & = & \frac{1}{d } \left\ { \frac{\beta}{2 } + \frac{\beta^2 k}{4 \sqrt{4+k^2 } } \ln \left [ \frac{k+\sqrt{4+k^2}}{2 } \right ] \right\ } \label{a.28b } \\ { \hat h}_{22}(k ) & = & \frac{1}{d } \left\ { -\frac{\beta}{4 } + \frac{\beta^2(6 + k^2)}{16 k \sqrt{4+k^2 } } \ln \left [ \frac{k+\sqrt{4+k^2}}{2 } \right ] \right\ } \label{a.28c } \end{aligned}\ ] ] where the denominator is @xmath352\ ] ] the asymptotic behavior of @xmath344 is governed by the poles of @xmath353 closest to the real axis . when @xmath10 , there poles are at @xmath354 ( or @xmath355 ) . in close analogy with ref . @xcite , to find the @xmath3-correction of these poles , it is sufficient to expand the @xmath3-dependent part of the denominator @xmath356 around say @xmath357 up to the first order in @xmath358 : @xmath359 \nonumber \\ & = & ( k^2 + m^2 ) \left [ 1 - \beta \left ( \frac{1}{24 } + \frac{\pi}{9\sqrt{3 } } \right ) + \cdots \right]\end{aligned}\ ] ] where @xmath360 the prefactors @xmath257 , defined by @xmath361 are deducible from ( [ a.28 ] ) ( with numerators evaluated at @xmath362 ) and ( [ a.30 ] ) , [ a.33 ] @xmath363 the large - distance behavior of @xmath344 is of type ( [ 4.9 ] ) : the prefactors @xmath257 ( [ a.33 ] ) and the parameter @xmath112 ( [ a.31 ] ) have the small-@xmath3 expansions ( [ 4.13a ] ) and ( [ 4.13b ] ) , respectively , confirming thus the predictions of the form - factor method . based on the analysis in section 2 applied to our specific @xmath27 case , close to the collapse point @xmath4 @xmath364 , the two - body densities behave like [ b.1 ] @xmath365 in the short - distance limit @xmath110 . the average in ( [ b.1a ] ) is taken over the cbd action ( [ 3.4 ] ) with @xmath366 . the one - body and two - body densities in a general coulomb system satisfy the electroneutrality sum rule @xcite @xmath367 where @xmath41 numerates the charged species . for our asymmetric @xmath30 2d coulomb gas , one has in particular [ b.3 ] @xmath368 it follows from ( [ 3.17 ] ) that , for fixed @xmath135 and @xmath136 , the density @xmath204 exhibits the singularity of type @xmath369 as @xmath370 . this singularity originates in ( [ b.3 ] ) as a result of the short - distance integration over the two - body densities @xmath371 ; neither @xmath372 nor @xmath373 give a diverging contribution as @xmath374 after being integrated out over short distances @xmath95 . the problem with eqs . ( [ b.3 ] ) is that they imply the dependences of @xmath204 on the relevant ( diverging ) integral @xmath375 with two different prefactors . we have to admit that the collapse mechanism alters the form of the sum rules ( [ b.3 ] ) . the particles can be divided into three basic groups : one third of particles is of type 1 ( charge @xmath376 and density @xmath377 ) , one third of particles is of type 2 ( charge @xmath29 and density @xmath378 ) and the remaining third of particles of type @xmath379 ( charge @xmath380 and density @xmath381 ) is excluded from the collapse phenomenon ( they only feel the charge @xmath382 of each collapsed pair of 1,2-particles and do not enter into diverging screening integrals ) . under this assumption , close to @xmath4 , the electroneutrality sum rules ( [ b.2 ] ) are modified as follows @xmath383 where @xmath384 and @xmath385 . the couple of eqs . ( [ b.4 ] ) is consistent and implies the only singular relationship @xmath386 we now put the short - distance expansion of @xmath387 , eq . ( [ b.1a ] ) , into the integral on the rhs of ( [ b.5 ] ) cut at some finite @xmath388 ( @xmath389 is a length over which the coulomb interaction is screened ) and obtain @xmath390 to evaluate the mean value of the exponential field , we apply the conjectured formula ( [ 3.8 ] ) for @xmath391 , @xmath366 and @xmath392 . consequently , @xmath393^{1/3 } \nonumber \\ & & \times \exp \left\ { \int_0^{\infty } \frac{{\rm d}t}{2 t } \left [ \frac{1}{\cosh t ( 2\cosh t-1 ) } - { \rm e}^{-2 t } \right ] \right\}\end{aligned}\ ] ] where we have applied @xmath394 . it is simple to show by using the integral representation of the logarithm of the gamma function ( [ 3.11 ] ) that @xmath395 \right\ } = \frac{1}{(2\pi)^2 } \left ( \frac{2}{3 } \right)^{1/2 } \frac{\gamma(3/4)}{\gamma(1/4 ) } \left [ \gamma\left ( \frac{1}{6 } \right ) \gamma\left ( \frac{1}{3}\right ) \right]^2\ ] ] eqs . ( [ b.6 ] ) - ( [ b.8 ] ) reproduce exactly the collapse singularity ( [ 3.17 ] ) deduced from the exact density - fugacity relationship ( [ 3.14 ] ) . although the thermodynamics of the system close to the collapse point is derivable directly from the collapse @xmath396 relationship discussed in the previous subsection , there exists another simpler derivation in the spirit of an independent - pair approximation by hauge and hemmer @xcite . as was already mentioned , close to the collapse point , from the total number of @xmath397 particles , @xmath398 particles of type 1 and @xmath398 particles of type 2 form pairs . their statistical weights contribute dominantly to the configuration integral @xmath399^{n/3 } = \left [ \frac{\pi}{1-\beta/4 } l^{2(1-\beta/4 ) } \right]^{n/3 } \quad \quad { \rm as}\ \beta \to 4 ^ -\ ] ] thence @xmath400 in full agreement with the laurent series ( [ 3.24 ] ) . i am indebted to bernard jancovici for careful reading of the manuscript and useful comments . the support by grant vega 2/7174/20 is acknowledged .

dialogue systems rely on a careful reinforcement learning design : the learning algorithm and its state space representation . in lack of more rigorous knowledge , the designer resorts to its practical experience to choose the best option . in order to automate and to improve the performance of the aforementioned process , this article formalises the problem of online off - policy reinforcement learning algorithm selection . a meta - algorithm is given for input a portfolio constituted of several off - policy reinforcement learning algorithms . it then determines at the beginning of each new trajectory , which algorithm in the portfolio is in control of the behaviour during the full next trajectory , in order to maximise the return . the article presents a novel meta - algorithm , called epochal stochastic bandit algorithm selection ( esbas ) . its principle is to freeze the policy updates at each epoch , and to leave a rebooted stochastic bandit in charge of the algorithm selection . under some assumptions , a thorough theoretical analysis demonstrates its near - optimality considering the structural sampling budget limitations . then , esbas is put to the test in a set of experiments with various portfolios , on a negotiation dialogue game . the results show the practical benefits of the algorithm selection for dialogue systems , in most cases even outperforming the best algorithm in the portfolio , even when the aforementioned assumptions are transgressed .

1701.08810

reinforcement learning ( rl ) @xcite is a machine learning framework intending to optimise the behaviour of an agent interacting with an unknown environment . for the most practical problems , trajectory collection is costly and sample efficiency is the main key performance indicator this is for instance the case for dialogue systems @xcite and robotics @xcite . as a consequence , when applying rl to a new problem , one must carefully choose in advance a model , an optimisation technique , and their parameters in order to learn an adequate behaviour given the limited sample set at hand . in particular , for 20 years , research has developed and applied rl algorithms for spoken dialogue systems , involving a large range of dialogue models and algorithms to optimise them . just to cite a few algorithms : monte carlo @xcite , @xmath0-learning @xcite , sarsa @xcite , mvdp algorithms @xcite , kalman temporal difference @xcite , fitted-@xmath0 iteration @xcite , gaussian process rl @xcite , and more recently deep rl @xcite . additionally , most of them require the setting of hyper parameters and a state space representation . when applying these research results to a new problem , these choices may dramatically affect the speed of convergence and therefore , the dialogue system performance . facing the complexity of choice , rl and dialogue expertise is not sufficient . confronted to the cost of data , the popular _ trial and error _ approach shows its limits . _ algorithm selection _ @xcite is a framework for comparing several algorithms on a given _ problem instance_. the algorithms are tested on several problem instances , and hopefully , the algorithm selection learns from those experiences which algorithm should be the most efficient given a new problem instance . in our setting , only one problem instance is considered , but several experiments are led to determine the fittest algorithm to deal with it . thus , we developed an _ online _ learning version @xcite of algorithm selection . it consists in testing several algorithms on the task and in selecting the best one at a given time . indeed , it is important to notice that , as new data is collected , the algorithms improve their performance and that an algorithm might be the worst at a short - term horizon , but the best at a longer - term horizon . in order to avoid confusion , throughout the whole article , the algorithm selector is called a _ meta - algorithm _ , and the set of algorithms available to the meta - algorithm is called a _ portfolio_. defined as an online learning problem , our algorithm selection task has for objective to minimise the expected regret . at each online algorithm selection , only the selected algorithm is experienced . since the algorithms learn from their experience , it implies a requirement for a fair budget allocation between the algorithms , so that they can be equitably evaluated and compared . budget fairness is in direct contradiction with the expected regret minimisation objective . in order to circumvent this , the reinforcement algorithms in the portfolio are assumed to be _ off - policy _ , meaning that they can learn from experiences generated from an arbitrary non - stationary behavioural policy . section [ sec : rlmodel ] provides a unifying view of reinforcement learning algorithms , that allows information sharing between all algorithms of the portfolio , whatever their decision processes , their state representations , and their optimisation techniques . then , section [ sec : asrl ] formalises the problem of online selection of off - policy reinforcement learning algorithms . it introduces three definitions of pseudo - regret and states three assumptions related to experience sharing and budget fairness among algorithms . beyond the sample efficiency issues , the online algorithm selection approach addresses furthermore four distinct practical problems for spoken dialogue systems and online rl - based systems more generally . first , it enables a systematic benchmark of models and algorithms for a better understanding of their strengths and weaknesses . second , it improves robustness against implementation bugs : if an algorithm fails to terminate , or converges to an aberrant policy , it will be dismissed and others will be selected instead . third , convergence guarantees and empirical efficiency may be united by covering the empirically efficient algorithms with slower algorithms that have convergence guarantees . fourth and last , it enables staggered learning : shallow models converge the fastest and consequently control the policy in the early stages , while deep models discover the best solution later and control the policy in late stages . afterwards , section [ sec : amabma ] presents the epochal stochastic bandit algorithm selection ( esbas ) , a novel meta - algorithm addressing the online off - policy reinforcement learning algorithm selection problem . its principle is to divide the time - scale into epochs of exponential length inside which the algorithms are not allowed to update their policies . during each epoch , the algorithms have therefore a constant policy and a stochastic multi - armed bandit can be in charge of the algorithm selection with strong pseudo - regret theoretical guaranties . a thorough theoretical analysis provides for esbas upper bounds of the pseudo - regrets defined in section [ sec : asrl ] under the assumptions stated in the same section . = [ rectangle , draw , text width=8em , text centered , rounded corners , minimum height=4em ] = [ draw , -latex ] ( agent ) agent ; ( environment ) stochastic environment ; ( agent.0 ) ++ ( 4em,0em ) |- node [ near start]@xmath1 ( environment.0 ) ; ( environment.190 ) ++ ( -6em,0em ) |- node [ near start ] @xmath2 ( agent.170 ) ; ( environment.170 ) ++ ( -4.25em,0em ) |- node [ near start , right ] @xmath3 ( agent.190 ) ; next , section [ sec : experesults ] experiments esbas on a simulated dialogue task , and presents the experimental results , which demonstrate the practical benefits of esbas : in most cases outperforming the best algorithm in the portfolio , even though its primary goal was just to be almost as good as it . finally , sections [ sec : related ] and [ sec : conclusion ] conclude the article with respectively related works and prospective ideas of improvement . the goal of this section is to enable information sharing between algorithms , even though they are considered as black boxes . we propose to share their trajectories expressed in a universal format : the _ interaction process_. reinforcement learning ( rl ) consists in learning through trial and error to control an agent behaviour in a stochastic environment . more formally , at each time step @xmath4 , the agent performs an action @xmath5 , and then perceives from its environment a signal @xmath6 called observation , and receives a reward @xmath7 . figure [ fig : rl ] illustrates the rl framework . this interaction process is not markovian : the agent may have an internal memory . in this article , the reward function is assumed to be bounded between @xmath8 and @xmath9 , and we define the rl problem as episodic . let us introduce two time scales with different notations . first , let us define _ meta - time _ as the time scale for algorithm selection : at one meta - time @xmath10 corresponds a meta - algorithm decision , _ i.e. _ the choice of an algorithm and the generation of a full episode controlled with the policy determined by the chosen algorithm . its realisation is called a _ trajectory_. second , _ rl - time _ is defined as the time scale inside a trajectory , at one rl - time @xmath11 corresponds one triplet composed of an observation , an action , and a reward . let @xmath12 denote the space of trajectories . a _ trajectory _ @xmath13 collected at meta - time @xmath10 is formalised as a sequence of ( observation , action , reward ) triplets : @xmath14 where @xmath15 is the length of trajectory @xmath16 . the objective is , given a discount factor @xmath17 , to generate trajectories with high discounted cumulative reward , also called _ return _ , and noted @xmath18 : @xmath19 since @xmath20 and @xmath21 , the return is bounded : @xmath22 the _ trajectory set _ at meta - time @xmath23 is denoted by : @xmath24 a sub - trajectory of @xmath16 until rl - time @xmath11 is called the _ history _ at rl - time @xmath11 and written @xmath25 with @xmath26 . the history records what happened in episode @xmath16 until rl - time @xmath11 : @xmath27 the goal of each reinforcement learning algorithm @xmath28 is to find a policy @xmath29 which yields optimal expected returns . such an algorithm @xmath28 is viewed as a black box that takes as an input a trajectory set @xmath30 , where @xmath31 is the ensemble of trajectory sets of undetermined size : @xmath32 , and that outputs a policy @xmath33 . consequently , a reinforcement learning algorithm is formalised as follows : @xmath34 such a high level definition of the rl algorithms allows to share trajectories between algorithms : a trajectory as a sequence of observations , actions , and rewards can be interpreted by any algorithm in its own decision process and state representation . for instance , rl algorithms classically rely on an mdp defined on a state space representation @xmath35 thanks to a projection @xmath36 : @xmath37 the state representation may be built dynamically as the trajectories are collected . many algorithms doing so can be found in the literature , for instance @xcite . then , @xmath28 may learn its policy @xmath38 from the trajectories projected on its state space representation and saved as a transition set : a transition is defined as a quadruplet @xmath39 , with state @xmath40 , action @xmath41 , reward @xmath42 , and next state @xmath43 . off - policy reinforcement learning optimisation techniques compatible with this approach are numerous in the literature @xcite : @xmath0-learning @xcite , fitted-@xmath0 iteration @xcite , kalman temporal difference @xcite , etc . another option would be to perform end - to - end reinforcement learning @xcite . as well , any post - treatment of the state set , any alternative decision process model , such as pomdps @xcite , and any off - policy technique for control optimisation may be used . the algorithms are defined here as black boxes and the considered meta - algorithms will be indifferent to how the algorithms compute their policies , granted they satisfy the assumptions made in the following section . algorithm selection @xcite for combinatorial search @xcite consists in deciding which experiments should be carried out , given a problem instance and a fixed amount of computational resources : generally speaking computer - time , memory resources , time and/or money . algorithms are considered efficient if they consume little resource . this approach , often compared and applied to a scheduling problem @xcite , experienced a lot of success for instance in the sat competition @xcite . algorithm selection applied to machine learning , also called _ meta - learning _ , is mostly dedicated to error minimisation given a corpus of limited size . indeed , these algorithms do not deliver _ in fine _ the same answer . in practice , algorithm selection can be applied to arbitrary performance metrics and modelled in the same framework . in the classical batch setting , the notations of the machine learning algorithm selection problem are described in @xcite as follows : * @xmath44 is the space of problem instances ; * @xmath45 is the _ portfolio _ , _ i.e. _ the collection of available algorithms ; * @xmath46 is the _ objective function _ , _ i.e. _ a performance metrics enabling to rate an algorithm on a given instance ; * @xmath47 are the features characterising the properties of problem instances . the principle consists in collecting problem instances and in solving them with the algorithms in the portfolio @xmath45 . the measures @xmath48 provide evaluations of the algorithms on those instances . then , the aggregation of their features @xmath49 with the measures @xmath48 constitutes a training set . finally , any supervised learning techniques can be used to learn an optimised mapping between the instances and the algorithms . nevertheless , in our case , @xmath44 is not large enough to learn an efficient model , it might even be a singleton . consequently , it is not possible to regress a general knowledge from a parametrisation @xmath49 . this is the reason why the online learning approach is tackled in this article : different algorithms are experienced and evaluated during the data collection . since it boils down to a classical exploration / exploitation trade - off , multi - armed bandit @xcite have been used for combinatorial search algorithm selection @xcite and evolutionary algorithm meta - learning @xcite . in the online setting , the algorithm selection problem for off - policy reinforcement learning is new and we define it as follows : * @xmath30 is the _ trajectory set _ ; * @xmath50 is the _ portfolio _ ; * @xmath51 is the _ objective function _ defined in equation [ eq : return ] . pseudo - code [ alg : algorithmselectionproblem ] formalises the online algorithm selection setting . meta - algorithm _ is defined as a function from a trajectory set to the selection of an algorithm : @xmath52 the meta - algorithm is queried at each meta - time @xmath53 , with input @xmath54 , and it ouputs algorithm @xmath55 controlling with its policy @xmath56 the generation of the trajectory @xmath57 in the stochastic environment . let @xmath58 be a condensed notation for the expected return of policy @xmath59 that was learnt from trajectory set @xmath60 by algorithm @xmath28 : @xmath61.\label{eq : expmu}\ ] ] the final goal is to optimise the cumulative expected return . it is the expectation of the sum of rewards obtained after a run of @xmath23 trajectories : _ = _ = 1^t _ ^-1 ( |)_^(),[eq : cumulativeexpectedreturn1 ] _ = _ = 1^t_,[eq : cumulativeexpectedreturn2 ] _ = _ . [ eq : cumulativeexpectedreturn3 ] equations [ eq : cumulativeexpectedreturn1 ] , [ eq : cumulativeexpectedreturn2 ] and [ eq : cumulativeexpectedreturn3 ] transform the cumulative expected return into two nested expectations . the outside expectation @xmath62 assumes the algorithm selection fixed and averages over the trajectory set stochastic collection and the corresponding algorithms policies , which may also rely on a stochastic process . the inside expectation @xmath63 assumes the policy fixed and averages the evaluation over its possible trajectories in the stochastic environment . equation [ eq : cumulativeexpectedreturn1 ] transforms the expectation into its probabilistic equivalent , @xmath64 denoting the probability density of generating trajectory set @xmath60 conditionally to the meta - algorithm @xmath65 . equation [ eq : cumulativeexpectedreturn2 ] transforms back the probability into a local expectation , and finally equation [ eq : cumulativeexpectedreturn3 ] simply applies the commutativity between the sum and the expectation . _ nota bene _ : there are three levels of decision : meta - algorithm @xmath65 selects an algorithm @xmath28 that computes a policy @xmath66 that in turn controls the actions . we focus in this paper on the meta - algorithm level . in order to evaluate the meta - algorithms , let us formulate two additional notations . first , the _ optimal expected return _ @xmath67 is defined as the highest expected return achievable by a policy of an algorithm in portfolio @xmath45 : @xmath68 second , for every algorithm @xmath28 in the portfolio , let us define @xmath69 as its _ canonical meta - algorithm _ , _ i.e. _ the meta - algorithm that always selects algorithm @xmath28 : @xmath70 , @xmath71 . the _ absolute pseudo - regret _ @xmath72 defines the regret as the loss for not having controlled the trajectory with an optimal policy . defnabsreg [ def : abs ] the absolute pseudo - regret @xmath72 compares the meta - algorithm s expected return with the optimal expected return : @xmath73 . \label{eq : absoluteregret}\ ] ] the absolute pseudo - regret @xmath72 is a well - founded pseudo - regret definition . however , it is worth noting that an optimal meta - algorithm will not yield a null regret because a large part of the absolute pseudo - regret is caused by the sub - optimality of the algorithm policies when the trajectory set is still of limited size . indeed , the absolute pseudo - regret considers the regret for not selecting an optimal policy : it takes into account both the pseudo - regret of not selecting the best algorithm and the pseudo - regret of the algorithms for not finding an optimal policy . since the meta - algorithm does not interfere with the training of policies , it can not account for the pseudo - regret related to the latter . in order to have a pseudo - regret that is relative to the learning ability of the best algorithm and that better accounts for the efficiency of the algorithm selection task , we introduce the notion of _ relative pseudo - regret_. defnrelreg [ def : rel ] the relative pseudo - regret @xmath74 compares the @xmath65 meta - algorithm s expected return with the expected return of the best canonical meta - algorithm : @xmath75 - \mathbb{e}_\sigma\left[\sum_{\tau=1}^t\mathbb{e}\mu_{\mathcal{d}_{\tau-1}^\sigma}^{\sigma(\tau ) } \right ] . \label{eq : relregret}\ ] ] it is direct from equations [ eq : absoluteregret ] and [ eq : relregret ] that the relative pseudo - regret can be expressed in function of absolute pseudo - regrets of the meta - algorithm @xmath65 and the canonical meta - algorithms @xmath69 : @xmath76 since one shallow algorithm might be faster in early stages and a deeper one more effective later , a good meta - algorithm may achieve a negative relative pseudo - regret , which is ill - defined as a pseudo - regret definition . still , the relative pseudo - regret @xmath74 is useful as an empirical evaluation metric . a large relative pseudo - regret shows that the meta - algorithm failed to consistently select the best algorithm(s ) in the portfolio . a small , null , or even negative relative pseudo - regret demonstrates that using a meta - algorithm is a guarantee for selecting the algorithm that is the most adapted to the problem . the theoretical analysis is hindered by the fact that algorithm selection , not only directly influences the return distribution , but also the trajectory set distribution and therefore the policies learnt by algorithms for next trajectories , which will indirectly affect the future expected returns . in order to allow policy comparison , based on relation on trajectory sets they are derived from , our analysis relies on three assumptions whose legitimacy is discussed in this section and further developed under the practical aspects in section [ sec : transgressions ] . assmorass the algorithms produce better policies with a larger trajectory set on average , whatever the algorithm that controlled the additional trajectory : @xmath77}.\ ] ] [ ass : monotony ] assumption [ ass : monotony ] states that algorithms are off - policy learners and that additional data can not lead to performance degradation on average . an algorithm that is not off - policy could be biased by a specific behavioural policy and would therefore transgress this assumption . assordass if an algorithm produces a better policy with one trajectory set than with another , then it remains the same , on average , after collecting an additional trajectory from any algorithm : @xmath78 \leq \mathbb{e}_{\alpha'}\left[\mathbb{e}\mu_{\mathcal{d}'\cup\varepsilon^{\alpha'}}^{\alpha}\right ] . \end{array}\ ] ] [ ass : compatibility ] assumption [ ass : compatibility ] states that a performance relation between two policies learnt by a given algorithm from two trajectory sets is preserved on average after adding another trajectory , whatever the behavioural policy used to generate it . from these two assumptions , theorem [ th : notworse ] provides an upper bound in order of magnitude in function of the worst algorithm in the portfolio . it is verified for any algorithm selection @xmath65 : thmnotthm the absolute pseudo - regret is bounded by the worst algorithm absolute pseudo - regret in order of magnitude : @xmath79 [ th : notworse ] see the appendix . contrarily to what the name of theorem [ th : notworse ] suggests , a meta - algorithm might be worse than the worst algorithm ( similarly , it can be better than the best algorithm ) , but not in order of magnitude . its proof is rather complex for such an intuitive and loose result because , in order to control all the possible outcomes , one needs to translate the selections of algorithm @xmath28 with meta - algorithm @xmath65 into the canonical meta - algorithm @xmath69 s view , in order to be comparable with it . this translation is not obvious when the meta - algorithm @xmath65 and the algorithms it selects act tricky . see the proof for an example . the _ fairness of budget distribution _ has been formalised in @xcite . it is the property stating that every algorithm in the portfolio has as much resources as the others , in terms of computational time and data . it is an issue in most online algorithm selection problems , since the algorithm that has been the most selected has the most data , and therefore must be the most advanced one . a way to circumvent this issue is to select them equally , but , in an online setting , the goal of algorithm selection is precisely to select the best algorithm as often as possible . in short , exploration and evaluation require to be fair and exploitation implies to be unfair . our answer is to require that all algorithms in the portfolio are learning _ off - policy _ , _ i.e. _ without bias induced by the behavioural policy used in the learning dataset . by assuming that all algorithms learn off - policy , we allow _ information sharing _ @xcite between algorithms . they share the trajectories they generate . as a consequence , we can assume that every algorithm , the least or the most selected ones , will learn from the same trajectory set . therefore , the control unbalance does not directly lead to unfairness in algorithms performances : all algorithms learn equally from all trajectories . however , unbalance might still remain in the exploration strategy if , for instance , an algorithm takes more benefit from the exploration it has chosen than the one chosen by another algorithm . in this article , we speculate that this chosen - exploration effect is negligible . more formally , in this article , for analysis purposes , the algorithm selection is assumed to be absolutely fair regardless the exploration unfairness we just discussed about . this is expressed by assumption [ ass : fairlearning ] . assfairass if one trajectory set is better than another for one given algorithm , it is the same for other algorithms . @xmath80 [ ass : fairlearning ] in practical problems , assumptions 2 and 3 are defeated , but empirical results in section [ sec : experesults ] demonstrate that the esbas algorithm presented in section [ sec : amabma ] is robust to the assumption transgressions . an intuitive way to solve the algorithm selection problem is to consider algorithms as arms in a multi - armed bandit setting . the bandit meta - algorithm selects the algorithm controlling the next trajectory @xmath81 and the trajectory return @xmath82 constitutes the reward of the bandit . however , a stochastic bandit can not be _ directly _ used because the algorithms performances vary and improve with time @xcite . adversarial multi - arm bandits are designed for non - stationary environments @xcite , but the exploitation the structure of our algorithm selection problem makes it possible to obtain pseudo - regrets of order of magnitude lower than @xmath83{t})$ ] . alternatively , one might consider using policy search methods @xcite , but they rely on a state space representation in order to apply the policy gradient . and in our case , neither policies do share any , nor the meta - algorithm does have at disposal any other than the intractable histories @xmath25 defined in equation [ eq : histories ] . to solve the off - policy rl algorithm selection problem , we propose a novel meta - algorithm called epochal stochastic bandit algorithm selection ( esbas ) . because of the non - stationarity induced by the algorithm learning , the stochastic bandit can not directly select algorithms . instead , the stochastic bandit can choose fixed policies . to comply to this constraint , the meta - time scale is divided into epochs inside which the algorithms policies can not be updated : the algorithms optimise their policies only when epochs start , in such a way that the policies are constant inside each epoch . as a consequence and since the returns are bounded in equation [ eq : returnbounds ] , at each new epoch , the problem can be cast into an independent stochastic @xmath84-armed bandit @xmath85 , with @xmath86 . the esbas meta - algorithm is formally sketched in pseudo - code [ alg : epochalgorithmselectionalgorithm ] embedding ucb1 @xcite as the stochastic @xmath84-armed bandit @xmath85 . the meta - algorithm takes as an input the set of algorithms in the portfolio . meta - time scale is fragmented into epochs of exponential size . the @xmath87^th^ epoch lasts @xmath88 meta - time steps , so that , at meta - time @xmath89 , epoch @xmath87 starts . at the beginning of each epoch , the esbas meta - algorithm asks each algorithm in the portfolio to update their current policy . inside an epoch , the policy is never updated anymore . at the beginning of each epoch , a new @xmath85 instance is reset and run . during the whole epoch , @xmath85 decides at each meta - time step which algorithm is in control of the next trajectory . it has to be noted that esbas simply intends to minimise the regret for not choosing the best algorithm at a given meta - time @xmath10 . it is somehow short - sighted : it does not intend to optimise its algorithms learning . the following _ short - sighted pseudo - regret _ definition captures the function esbas intends to minimise . defnssreg [ def : ss ] the short - sighted pseudo - regret @xmath90 is the difference between the immediate best expected return algorithm and the one selected : @xmath91 . \label{eq : shortsightedregret}\ ] ] the short - sighted pseudo - regret optimality depends on the meta - algorithm itself . for instance , a poor deterministic algorithm might be optimal at meta - time @xmath10 but yield no new information , implying the same situation at meta - time @xmath92 , and so on . thus , a meta - algorithm that exclusively selects the deterministic algorithm would achieve a short - sighted pseudo - regret equal to 0 , but selecting other algorithms are , in the long run , more efficient . theorem [ th : shortsightedregret ] expresses in order of magnitude an upper bound for the short - sighted pseudo - regret of esbas . but first , let define the gaps : @xmath93 it is the difference of expected return between the best algorithm during epoch @xmath87 and algorithm @xmath28 . the smallest non null gap at epoch @xmath87 is noted : @xmath94 if @xmath95 does not exist , _ i.e. _ if there is no non - null gap , the problem is trivial and the regret is null . upper bounds on short - sighted and absolute pseudo - regrets are derived . thmssthm if the stochastic multi - armed bandit @xmath85 guarantees a regret of order of magnitude @xmath96 , then : @xmath97 [ th : shortsightedregret ] see the appendix . several upper bounds in order of magnitude on @xmath98 can directly be deduced from theorem [ th : shortsightedregret ] , depending on an order of magnitude of @xmath95 . table [ tab : bounds ] reports some of them for a two - fold portfolio . it must be read by line . according to the first column : the order of magnitude of @xmath95 , the esbas short - sighted pseudo - regret bounds are displayed in the second column . one should notice that the first two bounds are obtained by summing the gaps . this means that esbas is unable to recognise the better algorithm , but the pseudo - regrets may still be small . on the contrary , the two last bounds are deduced from theorem [ th : shortsightedregret ] . in this case , the algorithm selection is useful . the worse algorithm @xmath99 is , the easier algorithm selection gets , and the lower the upper bounds are . [ c?c ] @xmath95 & @xmath100 + @xmath101 & @xmath102 + @xmath103 , @xmath104 & @xmath105 + @xmath103 , @xmath106 & @xmath107 + @xmath108 & @xmath109 + about esbas ability to select the best algorithm : if @xmath110{t})$ ] , then the meta - algorithm will not be able to distinguish the two best algorithms . still , we have the guarantee that pseudo - regret @xmath111{t})$ ] . the impossibility to determine which is the better algorithm is interpreted in @xcite as a budget issue . the meta - time necessary to distinguish arms that are @xmath95 apart with an arbitrary confidence interval takes @xmath112 meta - time steps . as a consequence , if @xmath110{t})$ ] , then @xmath113 . however , the budget , _ i.e. _ the length of epoch @xmath87 starting at meta - time @xmath114 , equals @xmath23 . in fact , even more straightforwardly , the stochastic bandit problem is known to be @xmath115 @xcite , which highlights the limit of distinguishability at @xmath116 . spare the @xmath117 factor in table [ tab : bounds ] bounds , which comes from the fact the meta - algorithm starts over a novel bandit problem at each new epoch , esbas faces the same hard limit . as equation [ eq : ] recalls , the absolute pseudo - regret can be decomposed between the absolute pseudo - regret of the canonical meta - algorithm of the best algorithm and the relative pseudo - regret , which is the regret for not running the best algorithm alone . the relative pseudo regret can in turn be upper bounded by a decomposition of the selection regret : the regret for not always selecting the best algorithm , and potentially not learning as fast , and the short - sighted regret : the regret for not gaining the returns granted by the best algorithm . these two successive decomposition lead to theorem [ th : absolute ] that provides an upper bound of the absolute pseudo - regret in function of the canonical meta - algorithm of the best algorithm , and the short - sighted pseudo - regret , which order magnitude is known to be bounded thanks to theorem [ th : shortsightedregret ] . thmabsthm if the stochastic multi - armed bandit @xmath85 guarantees that the best arm has been selected in the @xmath23 first episodes at least @xmath118 times , with high probability @xmath119 , then : @xmath120 where algorithm selection @xmath121 selects exclusively algorithm @xmath122 . [ th : absolute ] see the appendix . [ c?c|c ] & + & @xmath123 & @xmath124 + @xmath101 & @xmath102 & + @c@@xmath103 , + & @xmath105 & @xmath105 + @c@@xmath103 , + & @xmath107 & @c@@xmath125 , + + @xmath107 , + + @xmath108 & @xmath109 & @xmath125 + table [ tab : absbounds ] reports an overview of the absolute pseudo - regret bounds in order of magnitude of a two - fold portfolio in function of the asymptotic behaviour of the gap @xmath95 and the absolute pseudo - regret of the meta - algorithm of the the best algorithm @xmath126 , obtained with theorems [ th : notworse ] , [ th : shortsightedregret ] and [ th : absolute ] . table [ tab : bounds ] is interpreted by line . according the order of magnitude of @xmath95 in the first column , the second and third columns display the esbas absolute pseudo - regret bounds cross depending on the order of magnitude of @xmath126 . several remarks on table [ tab : absbounds ] can be made . firstly , like in table [ tab : bounds ] , theorem [ th : notworse ] is applied when esbas is unable to distinguish the better algorithm , and theorem [ th : shortsightedregret ] are applied when esbas algorithm selection is useful : the worse algorithm @xmath99 is , the easier algorithm selection gets , and the lower the upper bounds . secondly , @xmath123 implies that @xmath127 . thirdly and lastly , _ in practice _ , the second best algorithm absolute pseudo - regret @xmath128 is of the same order of magnitude than the sum of @xmath95 : @xmath129 . for this reason , in the last column , the first bound is greyed out , and @xmath130 is assumed in the other bounds . it is worth noting that upper bounds expressed in order of magnitude are all inferior to @xmath131{t})$ ] , the upper bounds of the adversarial multi - arm bandit . _ nota bene _ : the theoretical results presented in table [ tab : absbounds ] are verified if the stochastic multi - armed bandit @xmath85 satisfies both conditions stated in theorems [ th : shortsightedregret ] and [ th : absolute ] . successive and median elimination @xcite and upper confidence bound @xcite under some conditions @xcite are examples of appropriate @xmath85 . esbas algorithm for off - policy reinforcement learning algorithm selection can be and was meant to be applied to reinforcement learning in dialogue systems . thus , its practical efficiency is illustrated on a dialogue negotiation game @xcite that involves two players : the system @xmath132 and a user @xmath133 . their goal is to reach an agreement . @xmath134 options are considered , and at each new dialogue , for each option @xmath135 , players have a private uniformly drawn cost @xmath136 $ ] to agree on it . each player is considered fully empathetic to the other one . as a result , if the players come to an agreement , the system s immediate reward at the end of the dialogue is : @xmath137 where @xmath138 is the last state reached by player @xmath132 at the end of the dialogue , and @xmath135 is the agreed option ; if the players fail to agree , the final immediate reward is : @xmath139 and finally , if one player misunderstands and agrees on a wrong option , the system gets the cost of selecting option @xmath135 without the reward of successfully reaching an agreement : @xmath140 players act each one in turn , starting randomly by one or the other . they have four possible actions : * refprop@xmath141 : the player makes a proposition : option @xmath135 . if there was any option previously proposed by the other player , the player refuses it . * askrepeat : the player asks the other player to repeat its proposition . * accept@xmath141 : the player accepts option @xmath135 that was understood to be proposed by the other player . this act ends the dialogue either way : whether the understood proposition was the right one ( equation [ eq : rsuccess ] ) or not ( equation [ eq : rfailure ] ) . * enddial : the player does not want to negotiate anymore and ends the dialogue with a null reward ( equation [ eq : rgiveup ] ) . understanding through speech recognition of system @xmath132 is assumed to be noisy : with a sentence error rate of probability @xmath142 , an error is made , and the system understands a random option instead of the one that was actually pronounced . in order to reflect human - machine dialogue asymmetry , the simulated user always understands what the system says : @xmath143 . we adopt the way @xcite generates speech recognition confidence scores : @xmath144 if the player understood the right option @xmath145 , otherwise @xmath146 . the system , and therefore the portfolio algorithms , have their action set restrained to these five non parametric actions : refinsist @xmath147 refprop@xmath148 , @xmath149 being the option lastly proposed by the system ; refnewprop @xmath147 refprop@xmath141 , @xmath135 being the preferred one after @xmath149 , askrepeat , accept@xmath147 accept@xmath141 , @xmath135 being the last understood option proposition and enddial . all learning algorithms are using fitted-@xmath0 iteration @xcite , with a linear parametrisation and an @xmath150-greedy exploration : @xmath151 , @xmath87 being the epoch number . six algorithms differing by their state space representation @xmath152 are considered : * _ simple _ : state space representation of four features : the constant feature @xmath153 , the last recognition score feature @xmath154 , the difference between the cost of the proposed option and the next best option @xmath155 , and finally an rl - time feature @xmath156 . @xmath157 . * _ fast _ : @xmath158 . * _ simple-2 _ : state space representation of ten second order polynomials of _ simple _ features . @xmath159 , @xmath154 , @xmath155 , @xmath160 , @xmath161 , @xmath162 , @xmath163 , @xmath164 , @xmath165 , @xmath166 . * _ fast-2 _ : state space representation of six second order polynomials of _ fast _ features . @xmath159 , @xmath154 , @xmath155 , @xmath161 , @xmath162 , @xmath167 . * _ n-@xmath168-\{simple / fast / simple-2/fast-2 } _ : versions of previous algorithms with @xmath168 additional features of noise , randomly drawn from the uniform distribution in @xmath169 $ ] . * _ constant-@xmath48 _ : the algorithm follows a deterministic policy of average performance @xmath48 without exploration nor learning . those constant policies are generated with _ simple-2 _ learning from a predefined batch of limited size . + + [ fig : simplevssquare ] in all our experiments , esbas has been run with ucb parameter @xmath170 . we consider 12 epochs . the first and second epochs last @xmath171 meta - time steps , then their lengths double at each new epoch , for a total of 40,920 meta - time steps and as many trajectories . @xmath172 is set to @xmath173 . the algorithms and esbas are playing with a stationary user simulator built through imitation learning from real - human data . all the results are averaged over 1000 runs . the performance figures plot the curves of algorithms individual performance @xmath69 against the esbas portfolio control @xmath174 in function of the epoch ( the scale is therefore logarithmic in meta - time ) . the performance is the average return of the reinforcement learning problem defined in equation [ eq : return ] : it equals @xmath175 in the negotiation game , with @xmath176 value defined by equations [ eq : rsuccess ] , [ eq : rgiveup ] , and [ eq : rfailure ] . the ratio figures plot the average algorithm selection proportions of esbas at each epoch . sampled relative pseudo - regrets are also provided in table [ tab : recap ] , as well as the gain for not having chosen the worst algorithm in the portfolio . relative pseudo - regrets have a 95% confidence interval about @xmath177 , which is equivalent to @xmath178 per trajectory . three experience results are presented in this subsection . 1.81 _ simple-2 _ + _ fast-2 _ & 35 & -181 + _ simple _ + _ n-1-simple-2 _ & -73 & -131 + _ simple _ + _ n-1-simple _ & 3 & -2 + _ simple-2 _ + _ n-1-simple-2 _ & -12 & -38 + _ all-4 _ + _ constant-1.10 _ & 21 & -2032 + _ all-4 _ + _ constant-1.11 _ & -21 & -1414 + _ all-4 _ + _ constant-1.13 _ & -10 & -561 + _ all-4 _ & -28 & -275 + _ all-2-simple _ + _ constant-1.08 _ & -41 & -2734 + _ all-2-simple _ + _ constant-1.11 _ & -40 & -2013 + _ all-2-simple _ + _ constant-1.13 _ & -123 & -799 + _ all-2-simple _ & -90 & -121 + _ fast _ + _ simple-2 _ & -39 & -256 + _ simple-2 _ + _ constant-1.01 _ & 169 & -5361 + _ simple-2 _ + _ constant-1.11 _ & 53 & -1380 + _ simple-2 _ + _ constant-1.11 _ & 57 & -1288 + _ simple _ + _ constant-1.08 _ & 54 & -2622 + _ simple _ + _ constant-1.10 _ & 88 & -1565 + _ simple _ + _ constant-1.14 _ & -6 & -297 + _ all-4 _ + _ all-4-n-1 _ + _ constant-1.09 _ & 25 & -2308 + _ all-4 _ + _ all-4-n-1 _ + _ constant-1.11 _ & 20 & -1324 + _ all-4 _ + _ all-4-n-1 _ + _ constant-1.14 _ & -16 & -348 + _ all-4 _ + _ all-4-n-1 _ & -10 & -142 + _ all-2-simple _ + _ all-2-n-1-simple _ & -80 & -181 + 4*_n-2-simple _ & -20 & -20 + 4*_n-3-simple _ & -13 & -13 + 8*_n-1-simple-2 _ & -22 & -22 + _ simple-2 _ + _ constant-0.97 _ ( no reset ) & 113 & -7131 + _ simple-2 _ + _ constant-1.05 _ ( no reset ) & 23 & -3756 + _ simple-2 _ + _ constant-1.09 _ ( no reset ) & -19 & -2170 + _ simple-2 _ + _ constant-1.13 _ ( no reset ) & -16 & -703 + _ simple-2 _ + _ constant-1.14 _ ( no reset ) & -125 & -319 + 1.81.5 figures [ fig : simplevssquare1 ] and [ fig : simplevssquare2 ] plot the typical curves obtained with esbas selecting from a portfolio of two learning algorithms . on figure [ fig : simplevssquare1 ] , the esbas curve tends to reach more or less the best algorithm in each point as expected . surprisingly , figure [ fig : simplevssquare2 ] reveals that the algorithm selection ratios are not very strong in favour of one or another at any time . indeed , the variance in trajectory set collection makes _ simple _ better on some runs until the end . esbas proves to be efficient at selecting the best algorithm for each run and unexpectedly obtains a negative relative pseudo - regret of -90 . more generally , table [ tab : recap ] reveals that most of such two - fold portfolios with learning algorithms actually induced a strongly negative relative pseudo - regret . figures [ fig : squarevsconstant1 ] and [ fig : squarevsconstant2 ] plot the typical curves obtained with esbas selecting from a portfolio constituted of a learning algorithm and an algorithm with a deterministic and stationary policy . esbas succeeds in remaining close to the best algorithm at each epoch . one can also observe a nice property : esbas even dominates both algorithm curves at some point , because the constant algorithm helps the learning algorithm to explore close to a reasonable policy . however , when the deterministic strategy is not so good , the reset of the stochastic bandits is harmful . as a result , learner - constant portfolios may yield quite strong relative pseudo - regrets : @xmath179 in figure [ fig : squarevsconstant1 ] . however , when the constant algorithm expected return is over 1.13 , slightly negative relative pseudo - regrets may still be obtained . subsection [ sec : noarmreset ] offers a straightforward improvement of esbas when one or several algorithm are known to be constant . esbas also performs well on larger portfolios of 8 learners ( see figure [ fig:8learners1 ] ) with negative relative pseudo - regrets : @xmath180 ( and @xmath181 against the worst algorithm ) , even if the algorithms are , on average , almost selected uniformly as figure [ fig:8learners2 ] reveals . esbas offers some staggered learning , but more importantly , early bad policy accidents in learners are avoided . the same kind of results are obtained with 4-learner portfolios . if we add a constant algorithm to these larger portfolios , esbas behaviour is generally better than with the constant vs learner two - fold portfolios . we interpret esbas s success at reliably outperforming the best algorithm in the portfolio as the result of the four following potential added values : * calibrated learning : esbas selects the algorithm that is the most fitted with the data size . this property allows for instance to use shallow algorithms when having only a few data and deep algorithms once collected a lot . * diversified policies : esbas computes and experiments several policies . those diversified policies generate trajectories that are less redundant , and therefore more informational . as a result , the policies trained on these trajectories should be more efficient . * robustness : if one algorithm learns a terrible policy , it will soon be left apart until the next policy update . this property prevents the agent from repeating again and again the same blatant mistakes . * run adaptation : obviously , there has to be an algorithm that is the best on average for one given task at one given meta - time . but depending on the variance in the trajectory collection , it is not necessarily the best one for each run . the esbas meta - algorithm tries and selects the algorithm that is the best at each run . all these properties are inherited by algorithm selection similarity with ensemble learning @xcite . simply , instead of a vote amongst the algorithms to decide the control of the next transition @xcite , esbas selects the best performing algorithm . in order to look deeper into the variance control effect of algorithm selection , in a similar fashion to ensemble learning , we tested two portfolios : four times the same algorithm _ n-2-simple _ , and four times the same algorithm _ n-3-simple_. the results show that they both outperform the _ simple _ algorithm baseline , but only slightly ( respectively @xmath182 and @xmath183 ) . our interpretation is that , in order to control variance , adding randomness is not as good as changing hypotheses , _ i.e. _ state space representations . esbas s worst results concern small portfolios of algorithms with constant policies . these ones do not improve over time and the full reset of the @xmath84-multi armed bandit urges esbas to explore again and again the same underachieving algorithm . one easy way to circumvent this drawback is to use the knowledge that these constant algorithms do not change and prevent their arm from resetting . by operating this way , when the learning algorithm(s ) start(s ) outperforming the constant one , esbas simply neither exploits nor explores the constant algorithm anymore . figure [ fig : no - arm - reset ] displays the learning curve in the no - arm - reset configuration for the constant algorithm . one can notice that esbas s learning curve follows perfectly the learning algorithm s learning curve when this one outperforms the constant algorithm and achieves a strong negative relative pseudo - regret of -125 . still , when the constant algorithm does not perform as well as in figure [ fig : no - arm - reset ] , another harmful phenomenon happens : the constant algorithm overrides the natural exploration of the learning algorithm in the early stages , and when the learning algorithm finally outperforms the constant algorithm , its exploration parameter is already low . this can be observed in experiments with constant algorithm of expected return inferior to 1 , as reported in table [ tab : recap ] . several results show that , in practice , the assumptions are transgressed . firstly , assumption [ ass : compatibility ] , which states that more initial samples would necessarily help further learning convergence , is violated when the @xmath184-greedy exploration parameter decreases with meta - time and not with the number of times this algorithm has been selected . indeed , this is the main reason of the remaining mitigated results obtained in subsection [ sec : noarmreset ] : instead of exploring in early stages , the agent selects the constant algorithm which results in generating over and over similar fair but non optimal trajectories . finally , the learning algorithm might learn slower because of @xmath184 being decayed without having explored . secondly , we also observe that assumption [ ass : fairlearning ] is transgressed . indeed , it states that if a trajectory set is better than another for a given algorithm , then it s the same for the other algorithms . this assumption does not prevent calibrated learning , but it prevents the run adaptation property introduced in subsection [ sec : reasons ] that states that an algorithm might be the best on some run and another one on other runs . still , this assumption infringement does not seem to harm the experimental results . it even seems to help in general . thirdly , off - policy reinforcement learning algorithms exist , but in practice , we use state space representations that distort their off - policy property @xcite . however , experiments do not reveal any obvious bias related to the off / on - policiness of the trajectory set the algorithms train on . and finally , let us recall here the unfairness of the exploration chosen by algorithms that has already been noticed in subsection [ sec : assumptions ] and that also transgresses assumption [ ass : fairlearning ] . nevertheless , experiments did not raise any particular bias on this matter . related to algorithm selection for rl , @xcite consists in using meta - learning to tune a fixed reinforcement algorithm in order to fit observed animal behaviour , which is a very different problem to ours . in @xcite , the reinforcement learning algorithm selection problem is solved with a portfolio composed of online rl algorithms . in those articles , the core problem is to balance the budget allocated to the sampling of old policies through a _ lag _ function , and budget allocated to the exploitation of the up - to - date algorithms . their solution to the problem is thus independent from the reinforcement learning structure and has indeed been applied to a noisy optimisation solver selection @xcite . the main limitation from these works relies on the fact that _ on - policy _ algorithms were used , which prevents them from sharing trajectories among algorithms . meta - learning specifically for the eligibility trace parameter has also been studied in @xcite . a recent work @xcite studies the learning process of reinforcement learning algorithms and selects the best one for learning faster on a new task . this approach assumes several problem instances and is more related to the batch algorithm selection ( see section [ sec : batch ] ) . as for rl can also be related to ensemble rl . @xcite uses combinations of a set of rl algorithms to build its online control such as policy voting or value function averaging . this approach shows good results when all the algorithms are efficient , but not when some of them are underachieving . hence , no convergence bound has been proven with this family of meta - algorithms . horde @xcite and multi - objective ensemble rl @xcite are algorithms for hierarchical rl and do not directly compare with as . regarding policy selection , esbas advantageously compares with the rl with policy advice s regret bounds of @xmath185 on static policies @xcite . in this article , we tackle the problem of selecting online off - policy reinforcement learning algorithms . the problem is formalised as follows : from a fixed portfolio of algorithms , a meta - algorithm learns which one performs the best on the task at hand . fairness of algorithm evaluation is granted by the fact that the rl algorithms learn off - policy . esbas , a novel meta - algorithm , is proposed . its principle is to divide the meta - time scale into epochs . algorithms are allowed to update their policies only at the start each epoch . as the policies are constant inside each epoch , the problem can be cast into a stochastic multi - armed bandit . an implementation with ucb1 is detailed and a theoretical analysis leads to upper bounds on the short - sighted regret . the negotiation dialogue experiments show strong results : not only esbas succeeds in consistently selecting the best algorithm , but it also demonstrates its ability to perform staggered learning and to take advantage of the ensemble structure . the only mitigated results are obtained with algorithms that do not learn over time . a straightforward improvement of the esbas meta - algorithm is then proposed and its gain is observed on the task . as for next steps , we plan to work on an algorithm inspired by the _ lag _ principle introduced in @xcite , and apply it to our off - policy rl setting . 1.81 [ cols="<,<,<",options="header " , ] & section [ sec : algos ] + _ n-1-fast-2 _ & f@xmath0i with @xmath186 & section [ sec : algos ] + _ constant-@xmath48 _ & non - learning algorithm with average performance @xmath48 & section [ sec : algos ] + _ @xmath168 _ & number of noisy features added to the feature set & section [ sec : algos ] + 1.81.5 [ tab : gloss2 ] from definition [ def : abs ] : ^_abs(t ) = t^*_- _ , ^_abs(t ) = t^*_- _ _ , ^_abs(t ) = _ _ , where @xmath187 is the subset of @xmath60 with all the trajectories generated with algorithm @xmath28 , where @xmath188 is the index of the @xmath189^th^ trajectory generated with algorithm @xmath28 , and where @xmath190 is the cardinality of finite set @xmath191 . by convention , let us state that @xmath192 if @xmath193 . then : ^_abs(t ) = _ _ i=1^t _ . to conclude , let us prove by mathematical induction the following inequality : _ _ ^ is true by vacuity for @xmath194 : both left and right terms equal @xmath195 . now let us assume the property true for @xmath189 and prove it for @xmath196 : _ = _ , _ = _ , _ = _ . if @xmath197 , by applying mathematical induction assumption , then by applying assumption [ ass : compatibility ] and finally by applying assumption [ ass : monotony ] recursively , we infer that : _ _ ^ , _ _ ^ , _ _ ^ , _ _ ^. if @xmath198 , the same inequality is straightforwardly obtained , since , by convention @xmath199 , and since , by definition @xmath200 . the mathematical induction proof is complete . this result leads to the following inequalities : ^_abs(t ) _ _ i=1^t _ ^ , ^_abs(t ) _ ^^_abs(t ) , ^_abs(t ) k_^^_abs(t ) , which leads directly to the result : , _ abs^(t)(_^k _ abs^^k(t ) ) . _ this proof may seem to the reader rather complex for such an intuitive and loose result but algorithm selection @xmath65 and the algorithms it selects may act tricky . for instance selecting algorithm @xmath28 only when the collected trajectory sets contains misleading examples ( _ i.e. _ with worse expected return than with an empty trajectory set ) implies that the following unintuitive inequality is always true : @xmath201 . in order to control all the possible outcomes , one needs to translate the selections of algorithm @xmath28 into @xmath69 s view . _ by simplification of notation , @xmath202 . from definition [ def : ss ] : ^^esbas_ss(t ) = _ ^esbas , ^^esbas_ss(t ) = _ ^esbas , ^^esbas_ss(t ) _ ^esbas , ^^esbas_ss(t ) _ = 0^_2(t ) _ ss^^esbas ( ) , [ eq : th2:epochregret ] where @xmath203 is the epoch of meta - time @xmath10 . a bound on short - sighted pseudo - regret @xmath204 for each epoch @xmath87 can then be obtained by the stochastic bandit @xmath85 regret bounds in @xmath205 : _ ss^^esbas ( ) = _ ^esbas , _ ss^^esbas ( ) ( ) , _ ss^^esbas ( ) ( ) , _ 10 , _ ss^^esbas ( ) , where = _ and where_^= \ { ll + & if _ ^= _ _ ^ , + _ _ ^-_^ & otherwise . . since we are interested in the order of magnitude , we can once again only consider the upper bound of @xmath206 : _ ( ) , ( _ ) , _ 20 , , where the second best algorithm at epoch @xmath87 such that @xmath207 is noted @xmath208 . injected in equation [ eq : th2:epochregret ] , it becomes : _ ss^^esbas(t ) _ 1_2_=0^_2(t ) , [ eq : deltaminbound ] which proves the result . if @xmath209 , then @xmath210 , where @xmath211 . @xmath212 means that only one algorithm @xmath213 converges to the optimal asymptotic performance @xmath214 and that @xmath215 such that @xmath216 , @xmath217 , such that @xmath218 , @xmath219 . in this case , the following bound can be deduced from equation [ eq : deltaminbound ] : _ ss^^esbas(t ) _ 4 + _ = _ 1^(t ) , _ ss^^esbas(t ) _ 4 + [ eq : cor3:result ] , where @xmath220 is a constant equal to the short - sighted pseudo - regret before epoch @xmath221 : _ 4 = _ ss^^esbas(2^_1 - 1 ) equation [ eq : cor3:result ] directly leads to the corollary . if @xmath222 , then @xmath223 . if @xmath95 decreases slower than polynomially in epochs , which implies decreasing polylogarithmically in meta - time , _ i.e. _ @xmath224 , @xmath225 , such that @xmath226 , @xmath227 , then , from equation [ eq : deltaminbound ] : _ ss^^esbas(t ) _ 6 + _ = _ 2^(t ) , _ ss^^esbas(t ) _ 6 + _ = _ 2^(t ) ^m^+1 , _ ss^^esbas(t ) ^m^+2(t ) , [ eq : cor4:result ] where @xmath228 is a constant equal to the short - sighted pseudo - regret before epoch @xmath229 : _ 6 = _ ss^^esbas(2^_2 - 1 ) . equation [ eq : cor4:result ] directly leads to the corollary . if @xmath230 , then @xmath231 . if @xmath95 decreases slower than a fractional power of meta - time @xmath23 , then @xmath232 , @xmath233 , @xmath234 , such that @xmath235 , @xmath236 , and therefore , from equation [ eq : deltaminbound ] : _ ss^^esbas(t ) _ 8 + _ = _ 3^(t ) , _ ss^^esbas(t ) _ 8 + _ = _ 3^(t ) , _ ss^^esbas(t ) _ 8 + _ = _ 3^(t ) ( 2^c^)^ , [ eq : sumxex ] where @xmath237 is a constant equal to the short - sighted pseudo - regret before epoch @xmath238 : _ 8 = _ ss^^esbas(2^_3 - 1 ) . the sum in equation [ eq : sumxex ] is solved as follows : _ i = i_0^n ix^i = x_i = i_0^n ix^i-1 , _ i = i_0^n ix^i = x_i = i_0^n , _ i = i_0^n ix^i = x , _ i = i_0^n ix^i = x , _ i = i_0^n ix^i = ( ( x-1)nx^n - x^n -(x-1)i_0x^i_0 - 1 + x^i_0 ) . this result , injected in equation [ eq : sumxex ] , induces that @xmath239 , @xmath240 , @xmath241 : _ ss^^esbas(t ) _ 8 + ( t ) 2^c^(t ) , _ ss^^esbas(t ) _ 8 + t^c^ ( t ) , which proves the corollary . the esbas absolute pseudo - regret is written with the following notation simplifications : @xmath242 and @xmath243 : note that @xmath121 is the optimal constant algorithm selection at horizon @xmath23 , but it is not necessarily the optimal algorithm selection : there might exist , and there probably exists a non constant algorithm selection yielding a smaller pseudo - regret . the esbas absolute pseudo - regret @xmath244 can be decomposed into the pseudo - regret for not having followed the optimal constant algorithm selection @xmath121 and the pseudo - regret for not having selected the algorithm with the highest return , _ i.e. _ between the pseudo - regret on the trajectory and the pseudo - regret on the immediate optimal return : where @xmath245 is the expected return of policy @xmath246 , learnt by algorithm @xmath213 on trajectory set @xmath247 , which is the trajectory subset of @xmath54 obtained by removing all trajectories that were not generated with algorithm @xmath213 . on the one side , assumption [ ass : fairlearning ] of fairness states that one algorithm learns as fast as any another over any history . the asymptotically optimal algorithm(s ) when @xmath248 is(are ) therefore the same one(s ) whatever the the algorithm selection is . on the other side , let @xmath249 denote the probability , that at time @xmath10 , the following inequality is true : with probability @xmath250 , inequality [ eq : subsize ] is not guaranteed and nothing can be inferred about @xmath245 , except it is bounded under by @xmath251 . let @xmath252 be the subset of @xmath253 such that @xmath254 . then , @xmath250 can be expressed as follows : let consider @xmath255 the set of all sets @xmath60 such that @xmath256 and such that last trajectory in @xmath60 was generated by @xmath28 . since esbas , with @xmath85 , a stochastic bandit with regret in @xmath257 , guarantees that all algorithms will eventually be selected an infinity of times , we know that : we recall here that the stochastic bandit algorithm @xmath85 was assumed to guarantee to try the best algorithm @xmath213 at least @xmath259 times with high probability @xmath260 and @xmath261 . now , we show that at any time , the longest stochastic bandit run ( _ i.e. _ the epoch that experienced the biggest number of pulls ) lasts at least @xmath262 : at epoch @xmath263 , the meta - time spent on epochs before @xmath264 is equal to @xmath265 ; the meta - time spent on epoch @xmath266 is equal to @xmath267 ; the meta - time spent on epoch @xmath87 is either below @xmath267 , in which case , the meta - time spent on epoch @xmath266 is higher than @xmath268 , or the meta - time spent on epoch @xmath87 is over @xmath267 and therefore higher than @xmath268 . thus , esbas is guaranteed to try the best algorithm @xmath213 at least @xmath269 times with high probability @xmath249 and @xmath270 . as a result :

most biological data are multidimensional , posing a major challenge to human comprehension and computational analysis . principal component analysis is the most popular approach to rendering two- or three - dimensional representations of the major trends in such multidimensional data . the problem of multidimensionality is acute in the rapidly growing area of phylogenomics . evolutionary relationships are represented by phylogenetic trees , and very typically a phylogenomic analysis results in a collection of such trees , one for each gene in the analysis . principal component analysis offers a means of quantifying variation and summarizing a collection of phylogenies by dimensional reduction . however , the space of all possible phylogenies on a fixed set of species does not form a euclidean vector space , so principal component analysis must be reformulated in the geometry of tree - space , which is a cat(0 ) geodesic metric space . previous work has focused on construction of the first principal component , or principal geodesic . here we propose a geometric object which represents a-th order principal component : the locus of the weighted frchet mean of points in tree - space , where the weights vary over the standard-dimensional simplex . we establish basic properties of these objects , in particular that locally they generically have dimension , and we propose an efficient algorithm for projection onto these surfaces . combined with a stochastic optimization algorithm , this projection algorithm gives a procedure for constructing a principal component of arbitrary order in tree - space . simulation studies confirm these algorithms perform well , and they are applied to data sets of apicomplexa gene trees and the african coelacanth genome . the results enable visualizations of slices of tree - space , revealing structure within these complex data sets .

1609.03045

one of the great opportunities offered by modern genomics is that phylogenetics applied on a genomic scale ( phylogenomics ) should be especially powerful for elucidating gene and genome evolution , relationships among species and populations , and processes of speciation and molecular evolution . however , a well - recognized hurdle is the sheer volume of genomic data that can now be generated relatively cheaply and quickly , but for which analytical tools are lagging . there is a major need to explore new approaches to undertake comparative genomic and phylogenomic studies much more rapidly and robustly than existing tools allow . principal component analysis is the most popular approach for reducing the dimension of multidimensional data sets . the problem of multidimensionality is acute in the rapidly growing area of phylogenomics , which can provide insight into relationships and evolutionary patterns of a diversity of organisms , from humans , plants and animals , to microbes and viruses . data sets consisting of collections of phylogenetic trees are challenging to analyse , due both to high dimensionality and the complexity of the space containing the data . multivariate statistical procedures such as outlier detection @xcite , clustering @xcite and multidimensional scaling @xcite have previously been applied to such data sets . however , principal component analysis is perhaps the most useful multivariate statistical tool for exploring high - dimensional data sets , due to its flexibility and its attractive properties . for example , @xcite and @xcite recently showed that principal component analysis automatically projects to the subspace where the global solution of k - means clustering lies , and so facilitates k - means clustering to find near - optimal solutions . although principal component analysis for data in @xmath2 can be defined in several different ways , the following description is natural for reformulating the procedure in tree - space . suppose we have data @xmath3 where @xmath4 for @xmath5 . for any set of @xmath1 points @xmath6 we can define @xmath7 so that @xmath8 is the hyperplane in @xmath2 containing @xmath9 . the orthogonal @xmath10 distance of any point @xmath11 from @xmath8 is denoted @xmath12 , and the sum of squared projected distances of the data @xmath13 onto the hyperplane is defined by @xmath14 then the @xmath0-th order principal component @xmath15 corresponds to a choice of @xmath16 which _ minimizes _ this sum . in @xmath2 , @xmath17 is the sample mean , @xmath18 is the line through the sample mean which minimizes the sum of squared projected distances , and so on for @xmath19 . although it is not explicit in the definition above , in @xmath2 the principal components are nested : @xmath20 this description of principal component analysis relies heavily on the vector space properties of @xmath2 : @xmath8 is defined as a linear combination of vectors and the procedure uses orthogonal projection . however , the space of phylogenetic trees with @xmath21 leaves is not an euclidean vector space . it follows that we can not directly apply classical principal component analysis to a data set consisting of phylogenetic trees . the set @xmath22 of all phylogenetic trees with @xmath21 leaves labelled @xmath23 is a so - called cat(0 ) space @xcite . this means that @xmath22 is a metric space with a unique geodesic ( shortest length path ) between any pair of points , with the geodesic computable in @xmath24 steps @xcite . amongst other properties , projection onto closed sets is well - defined in cat(0 ) spaces . the analogue of the zero - th order principal component is given by the _ frchet mean _ of the data @xmath25 . the frchet mean is a statistic which characterizes the central tendency of a distribution in arbitrary metric spaces . for any metric space @xmath26 equipped with metric @xmath27 , the frchet population mean , @xmath28 , with respect to distribution @xmath29 is defined by @xmath30 the discrete analogue , the weighted frchet mean of a sample @xmath31 with respect to a weight vector @xmath32 , is @xmath33 where the weights @xmath34 satisfy @xmath35 for @xmath36 . in any cat(0 ) space , @xmath37 is a well - defined unique point given data @xmath13 and weight vector @xmath32 . the definition of the zero - th order principal component @xmath17 in @xmath2 given above coincides with the definition of the frchet sample mean with weights @xmath38 in any cat(0 ) space . several algorithms for computing the frchet sample mean in @xmath22 have been developed @xcite and we review these later in section [ sec : fmalgorithms ] as they play an important role in our methodology . methods for constructing a principal geodesic in tree - space , an analogue of @xmath39 as defined above , have recently been developed . in @xcite , the approach involved firing geodesics from some mean tree . for each candidate geodesic @xmath40 , the sum of squared projected distances @xmath41 was computed and a greedy algorithm was used to adjust @xmath40 in order to mimimize @xmath41 . the geodesics considered were infinitely long , but these have the disadvantage that in some cases many such geodesics fit the data equally well . subsequent approaches therefore considered finitely long geodesic segments @xcite . the geodesic segment between two points @xmath42 is analogous to @xmath8 in equation @xmath43 with @xmath44 , except that the weights @xmath45 must constrained to be positive under the analogy . @xcite constrained the ends of the geodesic to be points in the sample @xmath13 and sought the corresponding geodesic @xmath40 which mimimizes @xmath41 , whereas @xcite did not restrict the geodesic and used a stochastic optimization algorithm to perform the minimization . in this paper we address two fundamental questions : ( i ) which geometric object most naturally plays the role of a @xmath0-th order principal component in tree - space ; and ( ii ) given such an object , how can we efficiently project data points onto the object ? our proposed solution is to replace the definition of @xmath46 given in equation @xmath43 with the locus of the weighted frchet mean of points @xmath9 in tree - space . specifically , suppose @xmath47 and define @xmath48 by @xmath49 where @xmath50 is the @xmath0-dimensional simplex of probability vectors @xmath51 and @xmath52 is the frchet mean of the points in set @xmath16 with weights @xmath53 . we call @xmath8 the _ locus of the frchet mean _ of @xmath16 . our choice of notation is intended to emphasize the analogy between the definition of @xmath8 in tree - space and the corresponding definition for @xmath2 in equation @xmath43 . the locus of the frchet mean is a type of _ minimal surface _ , as the following physical analogy suggests : imagine connecting a point @xmath54 to points @xmath55 by @xmath1 pieces of elastic . when the point @xmath56 is free to move , it will move under the action of the elastic into an equilibrium position in tree - space . we can imagine how this equilibrium point changes as the stiffness in the pieces of elastic is varied , which corresponds to varying @xmath57 . as the equilibrium point moves around it scans out a surface in tree - space . in euclidean space the locus of the frchet mean of some collection of points is a subset of a hyperplane . in tree - space , as we will show , the hyper - surface can be curved . surfaces of this kind have recently been studied by @xcite in the context of riemannian manifolds and other geodesic metric spaces . we discuss the relationship of the present paper to that work in section [ sec : discussion ] . our main theoretical results are as follows . first , when @xmath58 we derive a set of local implicit equations for @xmath8 . these allow us to derive conditions for @xmath8 to be locally flat , and also enable us to construct explicit realizations of @xmath8 in certain interesting cases . secondly , using the implicit equations we show the locus of the frchet mean @xmath8 in @xmath22 is locally @xmath0 dimensional for generic @xmath16 , and thus forms a suitable candidate for a @xmath0-th order principal component . third , we present an algorithm for projection onto @xmath8 which relies only on the cat(0 ) properties of @xmath22 . we demonstrate accuracy of the projection algorithm via a simulation study . the remainder of the paper is organized in the following way . section [ review ] reviews the fundamental concepts and results in phylogenetic tree - space : the construction of geodesics , calculation of the frchet mean and convex hulls . section [ sec : lfm ] studies the geometric features of the locus of the frchet mean , in particular establishing its dimension with theorem [ thm : lfm_dimension ] . we draw the reader s attention to section [ sec : lfm_example ] which contains an explicit example which illustrates geodesics in tree - space , convex hulls , and details of the calculation of dimension . section [ sec : projection_and_pca ] presents the algorithm for projecting sets of phylogenetic tree data onto the locus of the frchet mean of @xmath1 fixed trees and describes the algorithm used to fit these objects to the data . this section also contains a simulation study to test the effectiveness of the algorithms . in section [ sec : results ] we apply these methods to two data sets . the first is a set of gene trees coming from fish and tetrapods intended to investigate the relationship of coelacanth and lungfish to the tetrapods , and the second is a set of gene trees from apicomplexa . we make some concluding remarks in section [ sec : discussion ] . throughout the paper , the @xmath59-dimensional euclidean vector space is denoted by @xmath2 . the non - negative and positive orthants in @xmath60 are denoted by @xmath61 and @xmath62 , respectively . for any vectors @xmath63 , @xmath64 denotes the euclidean norm of @xmath65 , and @xmath66 denotes the euclidean inner product . a _ phylogenetic tree _ with the leaf set @xmath67 is an undirected weighted acyclic graph with @xmath21 degree-@xmath68 vertices labelled @xmath23 , and with no degree-@xmath69 verices . we consider rooted trees , and the root is the leaf labelled `` 0 '' . each such tree contains @xmath21 _ pendant _ edges , which connect to the leaves , and up to @xmath70 _ internal edges_. the maximum number of internal edges is achieved when the tree is binary , in which case all vertices have degree @xmath71 other than the leaves , in which case the tree is called _ fully resolved_. if a tree contains fewer edges then it is called _ unresolved _ and there must be at least one vertex with degree @xmath72 . each edge in a phylogeny is assigned a strictly positive weight ( also called the edge _ length _ ) . given a tree @xmath73 , the set of edges of @xmath65 is denoted @xmath74 , and the weight associated to @xmath75 is denoted @xmath76 . it is convenient to define @xmath76 to be zero whenever @xmath77 is not contained in @xmath65 . tree - space @xmath22 is the set of all phylogenetic trees with leaf set @xmath78 @xcite . tree - space can be embedded in @xmath79 for @xmath80 in the following way . if we cut any edge @xmath75 then the tree @xmath65 splits into two disconnected pieces . this determines a _ split _ @xmath81 of the leaf set @xmath78 , where @xmath82 and @xmath83 . by convention we choose @xmath84 to be the set containing the root 0 , and so there are @xmath85 possible splits of @xmath78 . the collection of splits represented by a tree @xmath65 is called the _ topology _ of @xmath65 . since edges and splits are equivalent , we use the notation @xmath74 to also represent the set of splits in @xmath65 . by choosing some arbitrary ordering of the set of all splits , each tree @xmath73 can be represented as a vector in @xmath79 with up to @xmath86 positive entries given by the edge weights of @xmath65 , and zeros for each split which is not contained in @xmath65 . however , an arbitrary choice of vector will not necessarily represent a tree : for example the splits @xmath87 and @xmath88 can not both be contained in the same tree so any vector for which these splits both have strictly positive value does not represent a tree . two splits @xmath89 are _ compatible _ if one of the four sets @xmath90 , @xmath91 , @xmath92 , @xmath93 is empty , in which case there is at least one tree containing both splits . any collection of pairwise compatible splits determines a valid tree topology ( * ? ? ? * theorem 3.1.4 ) . the embedding into euclidean space reveals the combinatorial structure of @xmath22 . every tree @xmath73 contains @xmath94 pendant edges and so @xmath22 is the product of @xmath95 and a space corresponding to the internal edges . it is therefore convenient to ignore the pendant edges , and consider the corresponding embedding of tree - space into @xmath96 . given any tree topology @xmath97 containing @xmath59 internal edges , the set of trees with topology @xmath97 corresponds to a subset @xmath98 which is isomorphic to @xmath99 . ( the isomorphism is with respect to the local euclidean structure . ) each such region is called the _ orthant _ for topology @xmath97 . the boundary of @xmath100 in @xmath101 corresponds to trees obtained by removing one or more internal edges from @xmath97 . equivalently , the trees on the boundary can be obtained by taking a tree @xmath65 in @xmath100 and continuously shrinking one or more internal edges down to length zero . thus , for a fully - resolved topology @xmath97 , the codimension-@xmath68 boundaries of @xmath100 correspond to trees containing @xmath102 internal edges , and in general each codimension-@xmath0 boundary corresponds to trees containing @xmath103 internal edges , for @xmath104 . there are @xmath105 possible fully resolved rooted tree topologies and so @xmath22 is built from @xmath105 orthants isomorphic to @xmath106 together with the boundaries of these orthants which correspond to trees which are not fully resolved . orthants are glued together at their boundaries since a given unresolved tree containing @xmath59 internal edges can be obtained by removing edges from several different trees containing @xmath107 edges . orthants corresponding to fully - resolved topologies are glued at their codimension-@xmath68 boundaries in a relatively simple way . if a single internal edge in a tree with fully - resolved topology @xmath97 is contracted to length zero and removed from the tree , the result is a vertex of degree @xmath108 . there are @xmath71 possible ways to add in an additional edge to give another fully resolved topology ( including the original edge which was removed ) so each codimension-@xmath68 face of @xmath100 is glued to two other such orthants . trees containing no internal edges are called _ star - trees _ : the point @xmath109 corresponds to the set of star - trees and is contained in the boundary of every orthant @xmath100 . the topology of @xmath22 is taken to be that induced by the embedding into euclidean space . geodesics are constructed by considering continous paths in @xmath22 which are euclidean straight - line segments in each orthant . the length of a path is the sum of the euclidean segment lengths . as shown in @xcite , the shortest such path or _ geodesic _ between two points @xmath110 is unique , and it will be denoted @xmath111 . the distance @xmath112 is defined to be the length of @xmath111 and this defines the metric @xmath113 on @xmath22 . by definition , @xmath112 incorporates information about both the topologies and edge lengths of @xmath65 and @xmath56 . given two points @xmath114 in the same orthant @xmath111 is simply the euclidean line segment between @xmath114 , whereas when @xmath114 are in different orthants @xmath111 consists of a series of straight line segments traversing orthants corresponding to different topologies . @xcite proved that @xmath22 is a cat(0 ) space , and so it has several additional geometrical properties @xcite . @xcite established a polynomial time @xmath115 algorithm to compute the geodesic between any two trees in @xmath22 . the details of their algorithm are not important for the present application , but we do require some notation for the form of the geodesics it constructs . given @xmath110 let @xmath116 be the set of splits in @xmath117 which are compatible with every split in @xmath74 and every split in @xmath118 . adopting notation from @xcite , the geodesic @xmath111 is characterized by disjoint sets of internal splits @xmath119 where @xmath120 is an integer which depends on @xmath114 . these sets of splits determine the order in which edges are removed and added as the geodesic is traversed . the union @xmath121 for @xmath122 is @xmath123 and similarly for tree @xmath56 . we let @xmath124 be the ordered list of sets @xmath125 and similarly define @xmath126 . the _ support _ of @xmath111 is defined to be the triple @xmath127 . it characterizes the sequence of orthants the geodesic traverses . for any set @xmath128 we adopt the notation @xmath129 and similarly for subsets of @xmath118 . @xcite showed that @xmath130 where @xmath131 is the @xmath132-dimensional vector whose @xmath133-th element is @xmath134 and similarly for @xmath135 the @xmath133-th element is @xmath136 . vectors @xmath137 and @xmath138 have dimension @xmath139 , and respectively contain the edge lengths @xmath76 and @xmath140 for @xmath141 . it follows from equation @xmath142 that @xmath143 where @xmath144 is the sum of squared edge lengths in @xmath65 and similarly for @xmath56 . the following definition characterizes certain geodesics which behave rather like euclidean straight lines . [ def : simple ] suppose @xmath110 are fully resolved . the geodesic @xmath111 is called _ simple _ if each set @xmath145 and @xmath146 contains exactly one element for @xmath147 . equivalently , @xmath111 is simple if and only if at most one edge length contracts to zero at a time as the geodesic is traversed . the following definition determines the set of trees @xmath56 such that the geodesics @xmath111 to a fixed point @xmath65 all share the same support . fix some point @xmath73 and an orthant @xmath100 corresponding to a fully - resolved topology @xmath97 . given any valid support @xmath148 , the set @xmath149 is called a _ support region_. @xcite considered very similar subsets of @xmath22 and established their properties . given @xmath65 and @xmath97 there are only finitely many possible supports . we will use the fact that the union over the set of possible supports @xmath150 is dense in @xmath100 , where @xmath151 denotes the interior of each support region . the boundaries between the support regions are continuous codimension-@xmath68 surfaces within each orthant . several algorithms for computing the unweighted or weighted frchet mean of a sample in @xmath22 have been developed @xcite . these algorithms have the following general structure . let the sample of trees be @xmath152 . at the @xmath153-th iteration there is an estimate @xmath154 of the frchet mean of @xmath13 . to find the next estimate , @xmath155 , a data point @xmath156 is selected , either deterministically or stochastically depending on the particular algorithm . the geodesic @xmath157 is constructed , and @xmath155 is taken to be the point a certain proportion of the distance along the geodesic . this proportion can depend on the weights when the weighted frchet mean is estimated . in each case , some form of convergence of the sequence @xmath158 to the frchet mean of @xmath13 can be proved , independent of the initial estimate @xmath159 . our methodology does not make direct use of these algorithms . however , our proposed algorithm for projecting data onto the locus of the frchet mean is adapted from the algorithm of @xcite ( see section [ sec : projection ] ) , and so we present sturm s algorithm here . the algorithm computes the frchet mean of @xmath160 using weights @xmath161 . by definition , the frchet mean is invariant under positive scaling of the weights , so we can take @xmath162 . sturm s algorithm proceeds in the following way . [ alg : sturm ] sturm s algorithm for the weighted frchet mean . fix an initial estimate @xmath159 and set @xmath163 . + * repeat * : + 1 . sample @xmath164 such that @xmath165 . + 2 . construct @xmath166 . let @xmath155 be the point a proportion @xmath167 along @xmath166 where @xmath168 . + 4 . set @xmath169 . + * until * the sequence @xmath170 converges . convergence can be tested in various ways , for example repeating until a specified number of consecutive estimates @xmath154 all lie within distance @xmath171 of each other . sturm proved that the points @xmath154 converge in probability to the frchet mean of the distribution defined by sampling @xmath160 according to probabilities @xmath172 . the deterministic algorithm of @xcite for computing the weighted frchet mean is similar to sturm s algorithm , except the data points are used cyclically , as opposed to being randomly sampled as the algorithm progresses , and the weighting is instead taken into account in the definition of the proportions @xmath167 . we use the algorithm of @xcite for computing the frchet mean in order to test our projection algorithm , and this procedure is also described in section [ sec : projection ] . @xcite suggested that the convex hull of @xmath1 points in @xmath22 might be a suitable geometrical object to represent a @xmath0-th order principal component . a set @xmath173 is _ convex _ if and only if for all points @xmath174 the geodesic @xmath111 is also contained in @xmath175 . the _ convex hull _ of a set of points is the smallest convex set containing those points . any geodesic segment is the convex hull of its end - points , and using the convex hull of @xmath71 points to represent a second order principal component is a natural generalization of the idea of a principal geodesic . convexity is also a desirable property when performing projections , as occurs in a principal component analysis . however , convex hulls in tree - space do not have the correct dimension . examples for which the convex hull of @xmath71 points is @xmath71 dimensional can readily be constructed @xcite . it is demonstrated in @xcite that the dimension of a convex hull of @xmath71 points in @xmath22 can be arbitrarily high as @xmath94 increases . more generally , convex hulls in tree - space are difficult to characterize geometrically with several fundamental questions unanswered . these issues make convex hulls less appealing as geometrical objects to represent principal components , and at this point we turn our attention to the locus of the frchet mean . however we demonstrate the relationship between the locus of the frchet mean and convex hull for an explicit configuration of @xmath71 points @xmath176 later in section [ sec : lfm_example ] . throughout this section we work with @xmath1 points @xmath55 and let @xmath58 . as in the introduction , we define @xmath177 by @xmath178 the locus of the frchet mean of @xmath16 , denoted @xmath48 , is @xmath179 here we establish some basic properties of @xmath8 , while the next section presents a more detailed analysis of @xmath8 within orthant interiors . first , the map @xmath28 is continuous and so @xmath8 is compact since it is the continuous image of a compact set . continuity of @xmath28 can be proved using the deterministic algorithm for calculating the weighted frchet mean given by @xcite : the output of the algorithm depends continuously on the inputs @xmath16 and @xmath53 . ( we do not give a detailed proof for reasons of brevity . ) secondly , the points @xmath180 are contained in @xmath8 , since @xmath181 where @xmath182 denotes the @xmath153-th standard basis vector in @xmath50 . similarly each geodesic @xmath183 is contained in @xmath8 , by taking @xmath53 to be a convex combination of @xmath182 and @xmath184 . by the same argument , @xmath8 contains @xmath185 where @xmath186 is any non - empty subset of @xmath16 . in euclidean space the convex hull of @xmath1 points coincides with the locus of the frchet mean of the points . however , this is not the case in tree - space , though @xmath8 is contained in the closure of the convex hull of @xmath16 . this follows because any point in @xmath8 can be approximated arbitrarily closely by performing a finite number of steps of the algorithm of @xcite ( see section [ sec : fmalgorithms ] ) . provided the algorithm is initialized with one of the points @xmath180 , each of these steps remains within the convex hull , and so the limit point is contained in the closure of the convex hull . it is important to note that @xmath8 is itself generally _ not _ convex . as a consequence , there might not be a unique closest point on @xmath8 to any given point @xmath187 , although the minimum distance of @xmath187 from @xmath8 is well - defined . by using @xmath8 as a principal component we have therefore lost the desirable property of uniqueness of projection . frchet means in tree - space exhibit a property called _ stickiness _ @xcite . this essentially means that for fixed @xmath16 the map @xmath188 can fail to be injective . specifically , depending on the points in @xmath16 , there might exist open sets in @xmath50 which all map to the same point in tree - space . this has implications when we project data points onto @xmath8 : given a data point @xmath187 , the value of @xmath53 which minimizes @xmath189 might be non - unique , even if there is a unique closest point @xmath190 to @xmath187 . the algebraic form of tree - space geodesics described in section [ sec : bhv ] can be used to derive implicit equations for the edge lengths of trees lying on the locus of the frchet mean @xmath8 , and these equations are fundamental to establishing the dimension of @xmath8 . for fixed @xmath191 consider the objective function @xmath192 defined by @xmath193 suppose we fix an orthant @xmath100 for a fully resolved topology @xmath97 . let @xmath194 have edge lengths @xmath195 where @xmath196 for @xmath197 . @xcite showed that functions of the form @xmath198 are continuously differentiable on @xmath100 with respect to the edge lengths @xmath199 . in order to minimize @xmath200 we additionally assume @xmath65 lies in a set @xmath201 for some choice of supports @xmath202 . we call sets of this form _ mutual support regions _ with respect to @xmath180 . they are dense in @xmath100 using the properties of support regions given in section [ sec : bhv ] . each mutual support region is essentially a piece of tree - space for which the combinatorics of the geodesics to @xmath180 do not vary as a reference point moves around the region . an example of a decomposition of orthants into mutual support regions is given in section [ sec : lfm_example ] . under this assumption on @xmath65 , we can write down the algebraic form of @xmath198 using equation @xmath203 to give @xmath204 if the point @xmath205 lies on the locus of the frchet mean @xmath8 then @xmath206 for all @xmath133 , and so we want to evaluate these derivatives to obtain implicit equations relating the edge lengths @xmath199 to the vector @xmath53 . let @xmath56 be any of the trees @xmath180 . by definition @xmath207 since @xmath199 is the length of split @xmath184 and so the derivative of @xmath208 is just a constant . the term @xmath209 has a more general functional dependence on @xmath199 . by definition @xmath210 for any edge @xmath211 this expression does not depend on @xmath199 so the derivative is zero . when @xmath212 only the first term in brackets will depend on @xmath199 . since the sets @xmath213 are disjoint it must be the case that @xmath184 is contained in exactly one set and we define @xmath214 to be the index @xmath215 of that set when @xmath216 . then @xmath217 in the case that @xmath218 contains only @xmath184 and no other splits , we have @xmath219 so the expression becomes @xmath220 which is a constant . substituting these expressions into equation @xmath221 gives @xmath222 where @xmath223 if @xmath224 and is zero otherwise . we define @xmath225 by @xmath226 this function is continuously differentiable with respect to the edge lengths for all @xmath65 lying within the interior of mutual support regions . on the boundary between mutual support regions @xmath227 is continuous but may not be differentiable . in section [ sec : lfm_dimension ] we show the matrix of second derivatives of @xmath200 is positive definite on each mutual support region , and so every solution to @xmath228 is a minimum . it follows that @xmath8 is locally the solution to @xmath229 . the following lemma establishes conditions for @xmath8 to be a hyperplane within the mutual support region @xmath230 . [ lem : planar ] if the supports @xmath202 are such that the geodesics @xmath231 are simple for all @xmath232 ( in the sense of definition [ def : simple ] ) then @xmath8 is a hyperplane of dimension @xmath0 or lower in @xmath233 . if all the geodesics @xmath231 are simple for @xmath205 then each set @xmath234 contains exactly one split . then equation @xmath235 becomes @xmath236 for some constants @xmath237 . solving @xmath229 gives each edge length @xmath199 as a linear combination of @xmath238 , which establishes the result . generically , @xmath8 is therefore locally a @xmath0-dimensional hyperplane , but the dimension may be lower . further discussion about this point is given in section [ sec : lfm_dimension ] . we aim to prove that @xmath8 has dimension @xmath0 in each mutual support region . the strategy is to first show that the matrix of second derivatives of @xmath200 , or equivalently the matrix with elements @xmath239 with @xmath227 defined by equation @xmath240 , is positive definite . calculation of the dimension of @xmath8 follows by applying the implicit function theorem . the matrix with elements @xmath239 is positive definite for all @xmath65 in the mutual support region @xmath26 . using equation @xmath235 we have @xmath241 where the matrix @xmath242 has elements @xmath243 we start by assuming @xmath244 for all @xmath245 ( so that @xmath246 for all @xmath245 ) , and drop this assumption later . then @xmath247 where @xmath248 is a @xmath249 dimensional matrix with @xmath250 whenever @xmath251 is contained in @xmath252 and zero otherwise . equivalently @xmath248 indicates whether splits @xmath184 and @xmath251 are simultaneously contracted to zero on @xmath231 . we will show that the matrices @xmath242 are positive semi - definite . for any vector @xmath253 we have @xmath254 where @xmath255 and @xmath256 . now fix a single set of splits @xmath234 and let @xmath257 denote the indices of splits in this set . if we restrict the right - hand side of the last equation to indices @xmath258 we obtain @xmath259 the terms @xmath260 adopt the same value for all @xmath258 , and similarly for @xmath261 , so they are independent of the summation index in the last expression . also for @xmath258 @xmath262 then @xmath263 the cauchy - schwartz inequality shows the right - hand side is @xmath264 since @xmath265 are constant over this range of @xmath133 . it follows that the right - hand side of equation @xmath266 is @xmath264 , so each matrix @xmath242 is positive semi - definite . if we drop the assumption that @xmath244 for all @xmath245 , this introduces rows and columns of zeros into each matrix @xmath242 . however , the matrices @xmath242 must therefore remain positive semi - definite , and this establishes the calim in the statement of the proof . [ thm : lfm_dimension ] within the mutual support region @xmath26 , the locus of the frchet mean @xmath8 is a submanifold of dimension @xmath0 or lower . for generic selections of the points @xmath267 the dimension is @xmath0 . application of the implicit function theorem to the map @xmath227 when @xmath205 establishes that there is a locally - defined function @xmath268 such that @xmath269 and that the locus @xmath270 is a @xmath0-dimensional submanifold of @xmath271 . in fact , the image @xmath272 will be @xmath0-dimensional when the derivative of @xmath227 with respect to @xmath53 , @xmath273 , has rank @xmath0 which is the case for generic arrangements of points @xmath180 in tree - space . this is analogous to considering the hyperplane containing @xmath1 given points in euclidean space : generically the hyperplane has dimension @xmath0 but the dimension can be lower . for the example in section [ sec : lfm_example ] . weights for internal edges are shown . right : coordinates of the trees @xmath274 under the identification with orthants in @xmath275 . the @xmath276 axis points out of the page . the geodesics between @xmath274 are shown : @xmath277 ` kinks ' around the origin . the dashed line is between points @xmath278 and @xmath279 on @xmath280 and @xmath281 respectively . [ fig : example],title="fig : " ] for the example in section [ sec : lfm_example ] . weights for internal edges are shown . right : coordinates of the trees @xmath274 under the identification with orthants in @xmath275 . the @xmath276 axis points out of the page . the geodesics between @xmath274 are shown : @xmath277 ` kinks ' around the origin . the dashed line is between points @xmath278 and @xmath279 on @xmath280 and @xmath281 respectively . [ fig : example],title="fig : " ] in this section we construct an explicit example of the locus of the frchet mean for three points in @xmath282 . this example helps demonstrate the nature of geodesics in tree - space , the derivation of the impicit equations for @xmath8 , the relationship with the convex hull and other geometrical features . we start by fixing @xmath274 to have the topologies and edge lengths shown in figure [ fig : example ] . we will ignore the pendant edge lengths , and so the orthants containing these trees can be identified with three orthants in @xmath275 equipped with standard coordinates @xmath283 . there are five splits contained in these trees ( other than the pendant splits ) : they will be denoted @xmath284 , @xmath285 , @xmath286 , @xmath287 , @xmath288 by neglecting the complements in @xmath67 . we then use the notation @xmath289 to denote the length associated to split @xmath284 in tree @xmath65 , for example . under the identification with @xmath275 we have @xmath290 and @xmath291 . figure [ fig : example ] shows the location of trees @xmath274 under this identification . the orthant @xmath292 does not correspond to a valid tree topology as @xmath287 is not compatible with @xmath288 . at each codimension-@xmath68 face between the orthants shown there is in fact a third orthant in @xmath282 glued at the same boundary , but these do not play a role in this example . figure [ fig : example ] shows that the geodesics @xmath280 and @xmath281 are straight line segments under the identification with @xmath275 , while the geodesic @xmath277 ` kinks ' at a codimension-@xmath69 face . this behaviour is typical of geodesics in @xmath22 : they are straight line segents within each orthant but they can contain kinks at the voundaries between orthants . figure [ fig : example ] also shows how the convex hull of @xmath274 has dimension @xmath71 . the dashed line shows the geodesic between points @xmath278 and @xmath279 on @xmath280 and @xmath281 respectively . the convex hull therefore contains the points @xmath293 and @xmath294 and so there are @xmath108 points which are not coplanar within each orthant of the convex hull . and the points @xmath274 . left : when @xmath65 is contained in the orange region , none of the geodesics @xmath231 hit codimension-@xmath69 orthant faces , and so lemma [ lem : planar ] shows @xmath8 is planar . the same applies to the two yellow mutual support regions . right : when @xmath65 is contained one of the green regions then @xmath295 is not simple ( it hits a codimension-@xmath69 boundary ) and so @xmath8 is not planar . [ fig : lfm_decomp],title="fig : " ] and the points @xmath274 . left : when @xmath65 is contained in the orange region , none of the geodesics @xmath231 hit codimension-@xmath69 orthant faces , and so lemma [ lem : planar ] shows @xmath8 is planar . the same applies to the two yellow mutual support regions . right : when @xmath65 is contained one of the green regions then @xmath295 is not simple ( it hits a codimension-@xmath69 boundary ) and so @xmath8 is not planar . [ fig : lfm_decomp],title="fig : " ] figure [ fig : lfm_decomp ] shows the decomposition of the orthants into mutual support regions for @xmath274 . there are five regions in total , and the geodesics @xmath231 are simple for all @xmath296 when @xmath65 is contained in three of the regions . lemma [ lem : planar ] shows that @xmath8 is therefore planar in those regions with equation @xmath297 we can also explicitly calculate equations for @xmath8 in the mutual support region contained in @xmath298 and shown on the right in figure [ fig : lfm_decomp ] . for @xmath65 contained in this region , the squared distances to the vertices are @xmath299 where @xmath65 has coordinates @xmath283 . these can be used to write down an equation for @xmath300 , and then equation @xmath235 becomes @xmath301 then @xmath302 can be solved to give @xmath303 whenever @xmath304 , where @xmath305 . the resulting surface is show in figure [ fig : lfm_3d ] . the figure shows how @xmath8 forms a non - convex @xmath69-dimensional surface which is contained within the convex hull . in order to use the surface @xmath8 as a principal component , we need to be able to project data onto @xmath8 . let @xmath306 denote a data point and fix @xmath307 . a projection of @xmath187 onto @xmath8 is a point which minimizes @xmath308 . this point might not be unique as @xmath8 is not convex . a naive algorithm to find a projection is to perform exhaustive search , as follows . exhaustive search to project @xmath187 onto @xmath8 . [ alg : exhaustive ] 1 . construct a lattice of points @xmath309 . for @xmath310 this is a triangular lattice . 2 . for each point @xmath311 use a standard algorithm to compute @xmath52 . ( see section [ sec : fmalgorithms ] . ) find @xmath311 which minimizes @xmath312 . we implemented this algorithm for @xmath310 and used the algorithm of @xcite at step 2 to compute frchet means . algorithm [ alg : exhaustive ] is computationally very expensive , since the resolution of the lattice @xmath313 needs to be quite high in order to obtain accurate results . consequently we only use the exhaustive search algorithm in what follows as a benchmark in order to assess other methods . as an alternative to exhuastive search , we would like a more efficient algorithm which is defined entirely in terms of the geodesic geometry , since any reliance on local differentiable structure is likely to be problematic at orthant boundaries . we propose algorithm [ alg : projection ] , which we call the _ geometric projection algorithm_. geometric projection algorithm to project @xmath187 onto @xmath8 . [ alg : projection ] fix an initial estimate @xmath159 of the projection of @xmath187 , let @xmath314 and set @xmath163 . + * repeat * : + 1 . construct @xmath315 for @xmath316 . + 2 . for @xmath317 let @xmath318 be the point a proportion @xmath168 along @xmath315 . + 3 . find @xmath319 which minimizes @xmath320 . + 4 . set @xmath321 and let @xmath322 where @xmath323 is the @xmath324-th + standard basis vector in @xmath50 . + 5 . set @xmath169 . + * until * the sequence @xmath170 converges . the algorithm is a modification of sturm s algorithm for computing the frchet mean of @xmath16 ( algorithm [ alg : sturm ] . ) at each step of sturm s algorithm , one of the points @xmath318 is used as the new estimate @xmath155 , and the point @xmath318 is sampled according to a fixed probability vector @xmath53 . here , the new estimate for the projection , @xmath155 , is again chosen from @xmath325 but instead is selected to greedily minimize the distance from @xmath187 . the vector @xmath57 estimates the weight vector associated to the projected point : at iteration @xmath153 , @xmath326 is a vector with integer entries which counts the number of times the algorithm has moved the estimate of the projection towards each vertex in @xmath16 . the computational cost of the algorithm is similar to that for computing a single frchet mean using the sturm algorithm . for @xmath310 the initial point @xmath159 is sampled uniformly from the perimeter of @xmath8 . convergence is tested as follows : at iteration @xmath153 it is determined whether @xmath327 for all @xmath328 where @xmath329 and @xmath59 are fixed . if that is the case then the algorithm terminates . the output from the algorithm after @xmath330 iterations is an estimate @xmath331 of the projection of @xmath187 and a vector @xmath57 . the geometric projection algorithm is presented here without a proof of convergence and without further theoretical study of its properties . instead we rely on a simulation study in the next section to assess the effectiveness of the algorithm . we ran a set of simulations designed to demonstrate that , specifically in the case that @xmath310 , algorithm [ alg : projection ] converges to a tree on @xmath8 which minimizes @xmath308 . for each iteration of the simulation , a random species tree @xmath332 with @xmath333 taxa was generated under the @xcite coalescent . three trees @xmath274 and a fourth test tree @xmath187 were then generated under a coalescent model constrained to be contained within the tree @xmath332 , and thus corresponded to gene trees coming from the underlying species tree @xmath332 . ( see @xcite for more information about the relationship between species trees and gene trees . ) the dendropy library @xcite was used to generate these trees . the test tree @xmath187 was then projected onto @xmath8 for @xmath334 using the exhaustive search algorithm and the geometric projection algorithm . all calculations were carried out ignoring pendant edges . this particular simulation scheme was chosen in order to generate a variety of different geometrical configurations for the points @xmath335 , as well as being biologically reasonable . if the trees were sampled with topologies chosen independenly uniformly at random , for example , the simulation procedure would only have explored instances of @xmath8 with widely differing vertices . the results obtained from the two algorithms were compared in two ways . first , the distances from the data tree to the trees obtained with the two algorithms were computed and checked to ensure that the projection algorithm obtained a distance less than or equal to the exhaustive search . second , the distance between the tree from geometric projection and tree from exhaustive search was checked to ensure that the two trees were close together . for the second check we considered any distance greater than 1% of the total internal length of the data tree to be a failure . in a run of 10,000 iterations of this procedure , 95.65% of the iterations passed the two tests . however , even the set of failing iterations produced a projection result which were quite close to the exhaustive search result . among the 435 failing iterations , the perpendicular distance for the projection was an average of 3.7% greater than the perpendicular distance of the exhaustive search , and the distance between the two results was an average of 4.7% of the total internal length of the data tree . we believe that the failing results are attributable to the projection algorithm becoming trapped in local minima of the perpendicular distance . starting the algorithm from several locations and comparing the results would help to mitigate this problem . however , for the present purpose of fitting higher - order principal components to a collection of data trees , we believe these small deviations from the exhaustive search solution are an acceptable trade for the great increase in computational speed obtained . given data @xmath336 , our objective is to find @xmath307 which minimizes the sum of squared projected distances @xmath337 . from this point on in the paper , we restrict to the case @xmath310 . the geometric projection algorithm is used to compute @xmath337 given @xmath16 , at least approximately , and so we must now consider how to search over the possible configurations of the vertices @xmath16 . we adopt a stochastic optimization approach , algorithm [ alg : optimization ] below , which is similar to that for fitting principal geodesics in @xcite . we assume we have available a set of proposals @xmath338 , each of which is a map from @xmath22 to the set of distributions on @xmath22 . in particular , given any tree @xmath65 , each @xmath339 is asuumed to be a distribution on @xmath22 from which we can easily sample . stochastic optimization algorithm to fit @xmath8 to @xmath13 . [ alg : optimization ] fix an initial set @xmath334 and compute @xmath337 . + * repeat * : + * for * @xmath296 : + * for * @xmath340 : + 1 . sample a tree @xmath32 from @xmath341 . let @xmath342 be the set @xmath16 but with @xmath32 replacing @xmath343 . + 3 . compute @xmath344 using the geometric projection algorithm . + 4 . if @xmath345 set @xmath346 . + * until * convergence . the optimization algorithm attempts to minimize @xmath337 by stochastically varying one point @xmath347 at a time using the proposals @xmath348 . the algorithm is greedy : whenever a configuration @xmath342 improves upon the current configuration @xmath16 we replace @xmath16 with @xmath342 . convergence is assessed by considering the relative change in @xmath337 over a certain fixed number of iterations . if this is less than some proportion then the algorithm terminates . we used three different types of proposal . the first samples a tree uniformly at random with replacement from the data set @xmath13 . the second type is a refinement of this : given a tree @xmath65 it similarly samples a tree @xmath187 uniformly at random with replacement from the data set @xmath13 . then the geodesic @xmath349 is computed , and a beta distribution is used to sample a tree some proportion of the distance along @xmath349 . the third type of proposal is a random walk starting from @xmath65 , as described in @xcite . the random walk proposals can have different numbers of steps and step - sizes . the algorithm is not guaranteed to find a global optimum , and it can become stuck in local minima . it is therefore necesary to run the algorithm several times with different starting points for each data set , and then compare the results from each run . two statistics can be used to summarize the fit of @xmath8 to a data set @xmath13 : the sum of squared projected distances @xmath337 and a non - euclidean proportion of variance statistic , denoted @xmath350 . if the projection of each data point @xmath187 onto @xmath8 is denoted @xmath351 and @xmath352 denotes the frchet mean of @xmath353 , then @xmath354 the denominator in this expression varies with @xmath8 since pythagoras theorem does not hold in tree - space . unlike @xmath337 , the @xmath350 statistic is quite sensitive to small changes in @xmath16 , but it can be interpreted broadly as the proportion of variance explained by @xmath8 . in order to assess the performance of the algorithm we performed a small simulation study . eight data sets of 100 trees containing @xmath355 taxa were generated in the following way . for each data set a tree topology was sampled from a coalescent process , and each edge length was sampled from a gamma distribution with shape @xmath356 and rate @xmath357 , to give a tree @xmath358 . two trees @xmath359 were then obtained by applying random topological operations to @xmath358 . in four of the data sets @xmath359 were obtained by performing nearest neighbour interchange operations , while in the other four data sets sub - tree prune and regraft operations were used . then , to construct each data set given @xmath360 , @xmath361 points were sampled from a dirichlet distribution on @xmath362 with parameter @xmath363 and the corresponding points on @xmath185 were found using the bak algorithm . each point was then perturbed by using a random walk , so that each data set resembled a cloud of points around the surface @xmath185 . the step - size of the random walk was tuned to produce data sets classified as having either low or high dispersion . table [ tab : opt_sim ] summarizes the data sets used and the simulation results . it shows the sum of squared projected distances @xmath364 ( the ` true ' sum of squared distances ) and the fitted value @xmath337 , as well as the non - euclidean @xmath350 statistic . the exhaustive projection algorithm was used to compute @xmath364 while the geometric projection algorithm was used for @xmath337 . from the table it can be seen that the algorithm performs well in every scenario . .simulations to assess the stochastic optimization algorithm . the left column describes number and type of topological operation used to obtain @xmath359 from @xmath358 for each data set . the abbreviation nni stands for nearest neighbour interchange and spr stands for sub - tree prune and regraft . for each scenario , two data sets were generated by perturbing points on @xmath185 via random walks , with low and high dispersions respectively . the table shows the value of @xmath337 for the fitted principal component and in brackets the reference value @xmath364 . similarly the non - euclidean @xmath350 statistic is shown with the reference value in brackets . [ tab : opt_sim ] [ cols="^,^,^,^,^ " , ] figure [ fig : api_pc2 ] illustrates the second principal component , with the simplex shaded according to the corresponding tree topology on @xmath8 . three separate runs of algorithm [ alg : optimization ] converged to give similar results . the summary statistics for the second principal component are : sum of squared projected distances @xmath365 , @xmath350 statistic @xmath366 . while these summary statistics were consistent between runs , the set of topologies displayed on @xmath8 was subject to more variation , so figure [ fig : api_pc2 ] is a representative choice , although topologies 1,4 and 6 were present in all runs . the results show how the second principal component is able to tease more from the data than the variation in overall scale captured by the principal geodesic . topology 4 is congruent with the generally accepted phylogeny of taxa within the apicomplexa and is a resolution of the frchet mean tree : _ theileria annulata _ ( ta ) and _ babesia bovis _ ( bb ) group together ; the two plasmodium species ( pf and pv ) group together ; _ cryptosporidium parvum _ is the deepest rooting apicomplexan ; and pv , pf , ta , bb are monophyletic ( they are all hemosporidians or blood parasites ) . the figure shows that the second principal component corresponds to variation in topology consisting of nearest neighbour interchange operations which transform topology 4 into topologies 1 and 6 . none of the projected trees have topology 5 , although this is the topology of one of the vertices of @xmath8 . this topology appears to be present in order for @xmath8 to be positioned in such a way as to capture the other topologies . topology 2 shows evidence of stickiness ( see section [ sec : props ] ) : although the topology is unresolved , so that the coloured triangle lies in a codimension-@xmath68 region of tree - space , it occupies non - zero area on the simplex . as for the lungfish , the exhaustive and geometric projection algorithms were compared on the surface @xmath8 produced by algorithm [ alg : optimization ] . the distances between the projected points obtained with the two algorithms were very small compared to the distances of the data points from @xmath8 ( the sum of squared distances between pairs of projected points was 3.91e-4 ) . this paper presents three main innovations : ( i ) use of the locus of the frchet mean @xmath8 as an analogue of a principal component in tree - space , ( ii ) proof that @xmath8 has the desired dimension , and ( iii ) the geometric projection algorithm for projecting data onto @xmath8 . the locus of the frchet mean was first proposed as a geometric object for principal component analysis in tree - space by @xcite , though in @xcite pennec has made a similar proposal for analogues of principal component analysis in riemannian manifolds and other geodesic metric spaces . the barycentric subspaces of pennec correspond exactly to the surfaces @xmath8 considered in this paper . pennec s methodology , however , is principally based in the context of a riemannian manifold rather than in tree - space , though he points out the potential for generalization . there are substantial differences between pennec s analogue of principal component analysis on riemannian manifolds ( barycentric subspace analysis ) and the methodology presented in this paper . in particular , a key aim of barycentric subspace analysis is to produce _ nested _ principal components , in the sense of equation @xmath367 , while we do not make that restriction here . for example , if we consider a surface @xmath8 for @xmath310 then the only geodesics which are obviously contained in @xmath8 are the edges , and it is unappealing to restrict the principal geodesic to be one of these . our analysis has been restricted to data sets with relatively few taxa and to the construction of the first and second principal components . the algorithms presented in this paper scale linearly with respect to the number of data points @xmath368 , but run in polynomial time with respect to the number of taxa @xmath94 . however , by partitioning the data set for the geometric projection algorithm , parallel computer architectures can be employed and the speed - up is approximately proportional to the number of processors used . while the geometric projection algorithm runs relatively quickly , the calculations involved in searching for the optimal set of vertices @xmath16 can be very substantial . the experimental data sets in section [ sec : results ] took between 1 and 3 days to analyse , running on 4 processors each . for higher order components with @xmath369 , this computational burden will increase , and it is likely that finding a global minimum for @xmath337 will be more difficult . the figures in section [ sec : results ] show the potential for creating visualizations of the data which reveal meaningful biological structure . the pattern of projected points obtained for the experimental data sets we considered were very similar to results obtained via multidimensional scaling . however , multidimensional scaling is not capable of revealing the features of the data set which cause the observed variation . more information could be included in the graphical representation of our results , such as the distance of the data points from their projections , information about the principal geodesic , and the proximity of points to orthant boundaries . we presented the geometric projection algorithm without a proof of convergence , and used simulation to assess its accuracy . the algorithm is attractive in that it is defined entirely in terms of the geodesic structure on tree - space , and so it could be used on any geodesic metric space , including riemannian manifolds . the algorithm clearly deserves further investigation , and we aim to study its properties in a future publication . the authors thank d. howe from university of kentucky for useful comments on the analysis of apicomplexa data set . mitchell s. abrahamsen , thomas j. templeton , shinichiro enomoto , juan e. abrahante , guan zhu , cheryl a. lancto , mingqi deng , chang liu , giovanni widmer , saul tzipori , gregory a. buck , ping xu , alan t. bankier , paul h. dear , bernard a. konfortov , helen f. spriggs , lakshminarayan iyer , vivek anantharaman , l. aravind , and vivek kapur . complete genome sequence of the apicomplexan , cryptosporidium parvum . , 304:441445 , 2004 . amit bahl , brian brunk , jonathan crabtree , martin j. fraunholz , bindu gajria , gregory r. grant , hagai ginsburg , dinesh gupta , jessica c. kissinger , philip labo , li li , matthew d. mailman , arthur j. milgram , david s. pearson , david s. roos , jonathan schug , christian j. stoeckert , and patricia whetzel . plasmodb : the plasmodium genome resource . a database integrating experimental and computational data . , 31:212215 , 2003 . kelly a. brayton , audrey o. t. lau , david r. herndon , linda hannick , lowell s. kappmeyer , shawn j. berens , shelby l. bidwell , wendy c. brown , jonathan crabtree , doug fadrosh , tamara feldblum , heather a. forberger , brian j. haas , jeanne m. howell , hoda khouri , hean koo , david j. mann , junzo norimine , ian t. paulsen , diana radune , qinghu ren , roger k. smith jr . , carlos e. suarez , owen white , jennifer r. wortman , donald p. knowles jr.1 , terry f. mcelwain , and vishvanath m. nene . genome sequence of babesia bovis and comparative analysis of apicomplexan hemoprotozoa . , 3:e148 , 2007 . jonathan a eisen , robert s coyne , martin wu , dongying wu , mathangi thiagarajan , jennifer r wortman , jonathan h badger , qinghu ren , paolo amedeo , kristie m jones , luke j tallon , arthur l delcher , steven l salzberg , joana c silva , brian j haas , william h majoros , maryam farzad , jane m carlton , roger k smith , jr . , jyoti garg , ronald e pearlman , kathleen m karrer , lei sun , gerard manning , nels c elde , aaron p turkewitz , david j asai , david e wilkes , yufeng wang , hong cai , kathleen collins , b. andrew stewart , suzanne r lee , katarzyna wilamowska , zasha weinberg , walter l ruzzo , dorota wloga , jacek gaertig , joseph frankel , che - chia tsao , martin a gorovsky , patrick j keeling , ross f waller , nicola j patron , j. michael cherry , nicholas a stover , cynthia j krieger , christina del toro , hilary f ryder , sondra c williamson , rebecca a barbeau , eileen p hamilton , and eduardo orias . macronuclear genome sequence of the ciliate _ tetrahymena thermophila _ , a model eukaryote . , 4:16201642 , 2006 . a. feragen , m. owen , j. petersen , m.m.w . wille , l.h . thomsen , a. dirksen , and de bruijne m. tree - space statistics and approximations for large - scale analysis of anatomical trees . in _ 23rd biennial international conference on information processing in medical imaging ( ipmi ) _ , 2013 . bindu gajria , amit bahl , john brestelli , jennifer dommer , steve fischer , xin gao , mark heiges , john iodice , jessica c. kissinger , aaron j. mackey , deborah f. pinney , david s. roos , christian j. stoeckert , haiming wang , and brian p. brunk . toxodb : an integrated toxoplasma gondii database resource . , 36:d553d556 , 2008 . malcolm j. gardner , neil hall , eula fung , owen white , matthew berriman , richard w. hyman , jane m. carlton , arnab pain , karen e. nelson , sharen bowman , ian t. paulsen , keith james , jonathan a. eisen , kim rutherford , steven l. salzberg , alister craig , sue kyes , man - suen chan , vishvanath nene , shamira j. shallom , bernard suh , jeremy peterson , sam angiuoli , mihaela pertea , jonathan allen , jeremy selengut , daniel haft , michael w. mather , akhil b. vaidya , david m. a. martin , alan h. fairlamb , martin j. fraunholz , david s. roos , stuart a. ralph , geoffrey i. mcfadden , leda m. cummings , g. mani subramanian , chris mungall , j. craig venter , daniel j. carucci , stephen l. hoffman , chris newbold , ronald w. davis , claire m. fraser , and bart barrell . genome sequence of the human malaria parasite plasmodium falciparum . , 419:498511 , 2002 . mark heiges , haiming wang , edward robinson , cristina aurrecoechea , xin gao , nivedita kaluskar , philippa rhodes , sammy wang , cong - zhou he , yanqi su , john miller , eileen kraemer , and jessica c. kissinger . cryptodb : a cryptosporidium bioinformatics resource update . , 34:d419d422 , 2006 . christiane hertz - fowler , chris s. peacock , valerie wood , martin aslett , arnaud kerhornou , paul mooney , adrian tivey , matthew berriman , neil hall , kim rutherford , julian parkhill , alasdair c. ivens , marie - adele rajandream , and bart barrell . genedb : a resource for prokaryotic and eukaryotic organisms . , 32:d339d343 , 2004 .

we consider models of random groups in which the typical group is of intermediate rank ( in particular , it is not hyperbolic ) . these models are parallel to m. gromov s well - known constructions and include for example a density model " for groups of intermediate rank . the main novelty is the higher rank nature of the random groups . they are randomizations of certain families of lattices in algebraic groups ( of rank 2 ) over local fields . this paper introduces models of random groups `` of higher rank '' . the construction , basic properties , and applications are detailed in [ s - basic idea ] to [ propertyt ] below , which we now summarize . the construction ( see [ s - basic idea ] ) is rather general . if is a group which acts properly on a simply connected complex of dimension 2 with compact , and is a subgroup of `` very large '' finite index , then one can choose at random a family of-orbits of 2-cells inside . then let denote the universal cover of . the random group is the group of transformations of the galois covering this construction leads to several distinct models of random groups including a density model " , following m. gromov . the initial structural data ( , , ... ) for the model is called the _ deterministic data_. the basic properties of depend on the deterministic data . an idea of groups `` of intermediate rank '' was introduced in in particular to address the following question , where is cat(0 ) and is compact : ? ( this is the periodic flat plane problem " which has been formulated in many places , see for an early reference . ) since the assumption is equivalent to being non hyperbolic , the new models are relevant to the study of this question . we will see that in some cases ( depending on the deterministic data , the density parameter , etc . ) the answer is positive generically " , but that the precise relation between the two conditions `` '' and `` '' remains mysterious even for random groups associated with lattices in . before turning to these models let us discuss briefly gromov s original construction of random groups and the density model introduced in ( see also , or ) . let be a hyperbolic group in the sense of gromov , and take successive quotients , say where is a finite set of additional relations . as explained in the set can `` in general '' be chosen so that : * stays hyperbolic * is injective on larger and larger balls property ( ii ) ensures the existence of an infinite limit group , while the attribute in general " accounts for the oversupply of choices in the construction ; its precise meaning depends on the size and the nature of . for example , if is torsion free and the sets consist of a single relation which is a higher and higher " power of the element in a list exhausting , then the limit is a finitely generated infinite torsion group . here ( i ) and ( ii ) become geometric assertions relying on , and the construction offers almost total freedom . ( gromov s construction is related to the burnside problem the existence of infinite torsion groups was established by golod , and the existence of infinite groups of finite exponent by adian and novikov , and by olshanskii using the small cancellation theory . ) a prominent feature is the _ genericity of hyperbolic groups _ , as put forward in and illustrated by ( i ) above . m. gromov has since invented several models of random groups and constructed many exotic infinite groups using them . we are interested here in his so - called _ density model _ , which studies one - step " random quotients for very large " random sets of very long " relations . if one starts with a free group on generators and let denote the density parameter , then the random group in the density model is a quotient of the form , where is a set of words chosen uniformly independently at random in the sphere of radius in . gromov shows that if , then the resulting random group is hyperbolic with overwhelming probability as , while if it is trivial ( meaning 1 or ) with overwhelming probability . in this model , small cancellations occur for . ( one can also start here with a non elementary hyperbolic group and take random quotients by elements in the spheres , ; the same phase transition hyperbolic " and " is then valid provided that is torsion free . ) an earlier model of gromov called the few relator model " studies the situation where is bounded . we refer to for a survey of these groups . the groups of intermediate rank constructed in can be put on a `` rank interpolation line '' : ( we emphasis again that this only has a schematic value : as discussed in already , the phenomenon of rank interpolation is not unidimensional . ) the two extreme cases in this picture are the hyperbolic groups ( rank 1 , or more generally the groups with isolated flats , of rank ) and the lattices in nonarchimedean groups ( rank 2 ) . the value 1.94 refers to the bowtie group introduced in and further studied in . the present paper constructs many groups whose rank is arbitrarily close to 2 : if the deterministic data arises from a nonarchimedean lie group of rank 2 , then the `` rank '' of the random group ( for example , the local rank in the sense of ( * ? ? ? * definition 4.5 ) ) is as close to 2 as desired . we now formulate our main result in the special case of the density model with deterministic data the cartwright steger lattices in and their congruence subgroups . the techniques and constructions involved in the proof of this result apply in more general situations , and we will state and establish more general statements along the text . in fact , most of the assertions in theorem [ intrdensity ] , with the notable exception of the fact that " , will be proved under less restrictive assumptions . the cartwright steger lattices are uniform lattices associated with the ring{\hookrightarrow}{{\bf{f}}}_q(y)$ ] . their congruence subgroups correspond to ideals . the groups act transitively on the vertices of the bruhat tits building of . below ; the random groups discussed in theorem [ intrdensity ] have as deterministic data the lattices associated with the cartwright steger lattices of rank 2 . [ intrdensity ] let be prime power . fix two sequences and where $ ] is a monic irreducible polynomial prime to and and is an integer . if the density parameter satisfies then the random group in the density model of parameter with ( cartwright steger ) deterministic data satisfies with overwhelming probability , where the congruence subgroups are associated with the ideal generated by in $ ] . if in addition for large enough , then with overwhelming probability whenever ( which can be made as close to 1 as desired , independently of ) . furthermore , if ( which can be made as close to 1 as desired , independently of ) then acts freely uniformly on a space of dimension 2 with the geodesic extension property , and if then has kazhdan s property t with overwhelming probability . in addition , if is an arbitrary real number given in advance , then there exists such that if , then has kazhdan s property t with overwhelming probability for every . if on the other hand then does nt have property fa . finally , if ( and is arbitrary ) , then has the extension rigidity property of ( namely , it remembers " the building it comes from ) . finally , if and are real numbers given in advance , then there exists such that if , then has local rank ( in the sense of ) uniformly with overwhelming probability for every . the paper is structured as follows . an analog of gromov s few relator model is studied in [ s - pfp i ] in relation with the periodic flat plane problem . the density model is studied from [ s - critical density ] onwards , where we introduce critical densities " for various properties of groups in this model . in our framework , the density parameter regulates the size of the random subset . for example , we have a critical density for kazhdan s property t , for property fa , etc . , and most importantly the critical density for the property that . the critical densities depends on the deterministic data . in [ s - phase transition ] we discuss various analogs of gromov phase transition theorem in the density model , while [ sect : density ] is devoted to estimating for nonarchimedean lattices ( in positive characteristic ) . finally [ propertyt ] derives other properties of the random group ( for example property t ) and its intermediate rank " behavior . the reason why property t arises only for sufficiently large residue fields " is the same as in garland s paper , namely that the spectral gap is large enough only for large enough . we conclude with a question , the issue of which seems hard to predict at this stage : [ q - rank 2 ] is there of local rank such that for every ? here is a cat(0 ) of dimension 2 and freely with compact and the local rank is defined in . the question can be considered both in the general case , or when the order is bounded . see ( * ? ? ? * question 0.2 ) for a related question ( some aspects of the local to global problem " implicit in question [ q - rank 2 ] are also discussed in ) and also ( * ? ? ? * problem 3 ) . * acknowledgements . * we are grateful to the referees for helpful comments on the text .

1308.2315

let @xmath8 be a discrete group acting freely simplicially on a 2-complex @xmath6 with @xmath10 compact , and @xmath94 be a family of finite index subgroups of @xmath8 with @xmath95\to \infty$ ] , @xmath39 . the random group defined below is a randomization " of the following ( deterministic ) sequence of covering maps : @xmath96_{{\gamma}}\ar[rd]_{{\gamma}_1 } \ar[rrd]_{{\gamma}_2}\ar[rrrd]_{{\gamma}_3}&&\\ & { x/{\gamma}_1}\ar[ld]\hspace{.1 cm } & x/{\gamma}_2\ar[lld ] \hspace{.3 cm } & x/{\gamma}_3\ar[llld]\hspace{.3cm}\cdots\\ { x/{\gamma}}}\ ] ] associated with @xmath6 and @xmath97 . ( the family may be nested @xmath98 and correspond to a tower @xmath99 of compact spaces . ) the randomization is achieved by inserting a random topological noise " to the spaces @xmath100 ( where @xmath32 is very large ) which is detected by the fundamental group . we call @xmath101 the _ deterministic data_. the topological perturbation is implemented as follows . for each @xmath102 , remove a family of 2-cells in @xmath100 ( equivalently , a family of @xmath103-orbits of 2-cells in the _ fixed _ space @xmath6 ) at random ( with respect to a probability scheme for removing 2-cells , for example , bernoulli ) . the universal cover @xmath104 of the resulting ( random ) space @xmath105 has a ( random ) group @xmath106 of deck transformations : @xmath107^{\tilde { \gamma}_p}&\\ x/{\gamma}_p & x_p/{\gamma}_p\ar[l]}\ ] ] as described in the following overall diagram : @xmath108\ar[dd]^{\tilde { \gamma}_p}&\cdots\\ x\ar[dd]_{{\gamma}}\ar[rd]_{{\gamma}_1 } \ar[rrd]^{{\gamma}_p}&&\\ & { x/{\gamma}_1}\ar[ld ] \hspace{.5cm}\cdots&x/{\gamma}_p\ar[lld ] & x_p/{\gamma}_p\ar[l]\ar[llld]&\cdots\\ { x/{\gamma}}}\ ] ] we study the properties of @xmath109 when the order of approximation @xmath32 is very large . the construction provides a random group @xmath106 , a random space @xmath104 , a random action @xmath110 , and a random compact space @xmath111 . at this stage , the model will be precisely defined as long as the random scheme removing 2-cells is specified ; several options are possible . we study here analogs of m. gromov s well known models for adding relators at random to a finitely generated group ( more precisely , analogs of the few relator model " and the density model " of @xcite ) . fix a deterministic data @xmath101 ( different choices for the triples @xmath101 give rise to different models of random groups and lead to a priori distinct random objects ) . the set @xmath112-orbits of faces in @xmath113 ( whose element are called _ equivariant faces _ of @xmath6 with respect to @xmath103 , or sometimes _ equivariant chambers _ when we have in mind a bruhat tits building ) is finite , and in many interesting cases it is rapidly growing . it plays the role of the set @xmath114 all words ( or reduced words ) of length @xmath32 ( or at most @xmath32 ) in @xmath8 in gromov s models , from which the relations are picked up at random and added to the given group @xmath8 ( e.g. the free group @xmath115 ) . for a finite subset @xmath116 of @xmath117 , we set * @xmath118 * @xmath119 and @xmath120 be the corresponding universal cover * @xmath121 be the galois group of the covering map @xmath122 the random group is defined by : [ def - lattice models ] fix for every @xmath102 a process @xmath123 for selecting random subsets of elements in @xmath117 . the _ random group of order @xmath32 _ in the @xmath124-model is the group @xmath121 associated by the construction above to the @xmath123-generic subset @xmath125 . we say that a property @xmath126 occurs _ with overwhelming probability _ in this model if the probability that the random group @xmath121 of order @xmath32 satisfies @xmath126 converges to 1 as @xmath39 . \1 ) in all cases considered below , the random process @xmath127 is _ universal _ in that it does not depend on the deterministic data @xmath101 . more precisely , a predetermined process @xmath128 is chosen for selecting a random finite subset in an arbitrary finite set , and this process @xmath128 is applied recursively to the terms of the sequence @xmath129 . \2 ) most of the results of the present paper extend easily to the case of _ proper _ actions on _ cell _ complexes with compact quotient . we also note that these models of random group only take into account the profinite information " contained in the deterministic data . we study two special cases of definition [ def - lattice models ] , the bounded model " and the density model " . in the first model , a uniformly bounded number of elements of @xmath117 is chosen at random : [ fewmisschambers ] fix an integer parameter @xmath130 . the _ bounded model _ over @xmath101 is the @xmath124-model associated with the process @xmath131 choose @xmath132 chambers in @xmath117 , uniformly and independently at random " . this corresponds to gromov s few - relator model " . in the second model , the number of chosen chambers in @xmath117 is unbounded and rapidly growing . [ fewmisschambers ] fix a real parameter @xmath133 ( the density ) . the _ density model _ over @xmath101 is the @xmath124-model associated with the process @xmath131 choose @xmath134 chambers in @xmath117 , uniformly and independently at random " . this corresponds to the gromov density model " . as in gromov models , the bounded model can be seen as a manifestation of the density model with @xmath135 " ( see also section [ s - critical density ] ) . in a sense , the new models can be thought of as mirror images " of the gromov models : rather than starting with a group with a large supply of quotients , for example non abelian free groups , and gradually adding relations at random , we typically start ( see below ) with lattices in some algebraic group of rank 2 , which have as many relations as is conceivable " for an infinite group ( in particular they are just infinite up to centre ) , and remove them at random . it is unclear how to randomize ( say , residually finite ) discrete groups of higher cohomological dimension ; for example , the above construction provides a precise meaning for the expression @xmath8 is a random extension of a lattice in @xmath136 " for @xmath54 , where @xmath137 is a local field what about @xmath138 ? let @xmath101 be a deterministic data , and let @xmath102 , @xmath125 , @xmath139 , @xmath121 , @xmath140 ( following the notation of section [ s - basic idea ] ) be fixed throughout the section . observe that the sequence of covering spaces @xmath141 provides an exact sequence ( non - split in general ) @xmath142 of discrete groups . [ l - cohomological dim ] if @xmath6 is contractible , then : 1 . the homology groups @xmath143 vanish for @xmath144 , and the covering space @xmath145 is contractible . 2 . the group @xmath140 is a free group on countably many generators . the group @xmath121 is a finitely presented group of geometric dimension 2 . if @xmath146 , then @xmath121 is a strict extension of @xmath8 . the first part of the first assertion is clear and the second part classically follows from the first ( @xmath145 is weakly homotopy equivalent to a point , and therefore contractible ) . the universal coefficient theorem @xmath147 where the module @xmath148 is free and thus projective , shows that the second integral cohomology @xmath149 vanishes . thus , @xmath140 is a free group . this follows from the stallings swan theorem that a group of cohomological dimension 1 is free . ( note that the group @xmath140 does not act naturally " on a tree in general . ) let us prove ( 3 ) . since @xmath8 acts freely on @xmath6 contractible , @xmath10 is an eilenberg maclane space , @xmath150 , and it follows from ( 1 ) that @xmath151 is a @xmath152 and in particular @xmath121 is finitely presented of geometric dimension 2 . ( if the action @xmath93 is only assumed to be proper , then the groups @xmath121 are of proper ( bredon ) geometric dimension 2 . ) assume that @xmath146 . let @xmath153 be the boundary of an element of @xmath154 . if @xmath155 is an isomorphism , then @xmath153 is a boundary in @xmath139 , and therefore is an equator in a 2-sphere of @xmath6 . this contradicts the fact that @xmath139 is aspherical , so @xmath156 , and by equivariance , @xmath140 is infinitely generated . [ r - 10 ] if @xmath8 is of higher cohomological dimension , then @xmath140 is not necessarily free . the density model is interesting to study in this situation , and seems to be working properly only in high density regimes " ( this to compensate , especially in the case of @xmath157 , @xmath158 , for the very strong ambient rigidity properties of the deterministic data ) . the following flat plane correspondence " assertion is useful in connection with gromov s periodic flat plane problem . [ l - lifting lemma ] assume that @xmath6 is a cat(0 ) space and that @xmath93 is isometric . for every flat @xmath159 , there exists a flat @xmath160 such that the restriction of @xmath161 to @xmath162 is an isometry onto @xmath163 . furthermore , if there exists a subgroup @xmath164 , @xmath165 , such that @xmath166 is a compact torus , then there is a corresponding a subgroup @xmath167 , @xmath168 , such that @xmath169 is a compact torus , and the following diagram commutes : @xmath170&\\ { { \mathbb z}}^2\ar[r]\ar[ur ] & \lambda\subset { \gamma}_p}$ ] conversely , if @xmath160 is a flat in @xmath145 , then @xmath171 is a flat in @xmath139 . if in addition there exists a subgroup @xmath167 , @xmath168 , such that @xmath169 is a compact torus , then its projection @xmath172 in @xmath8 is a subgroup isomorphic to @xmath173 , such that @xmath166 is a compact torus . since @xmath174 is a locally isometric covering map and @xmath175 is contractible , the map @xmath176 admits a unique isometric lifting @xmath177 through any point @xmath178 such that @xmath179 , such that the following diagram commute : @xmath180^{\pi_a}&\\ { { \mathbb r}}^2\ar[r]_{j}\ar[ur]^{\tilde j } & x_a}\ ] ] the map @xmath181 , being a local isometry , restricts to an isometry from @xmath162 onto @xmath163 . let @xmath182 be a subgroup isomorphic to @xmath173 such that @xmath166 is a compact torus . let @xmath183 , and let @xmath184 be any lift of @xmath183 . by equivariance of @xmath185 , we have @xmath186 . choose @xmath187 such that @xmath188 . then @xmath189 is an isometric lifting of @xmath163 which contains @xmath190 . uniqueness of liftings implies that @xmath191 . let @xmath192 be a generating set of @xmath172 and let @xmath193 be such that @xmath194 and @xmath195 . since @xmath196 and @xmath197 is isometric , the elements @xmath198 and @xmath199 of @xmath121 commute and therefore generate an ( infinite torsion - free ) abelian group @xmath200 . if @xmath201 , let @xmath202 be a generator , and let @xmath203 be its image then @xmath204 and @xmath205 , contradicting @xmath165 . therefore @xmath206 is conjugate to a discrete action of @xmath173 on @xmath175 , hence @xmath207 is compact . conversely , let @xmath163 be the image of @xmath162 under @xmath185 . since @xmath208 is a locally isometric covering map , @xmath163 is either a torus , a cylinder , or a flat plane . the first two possibilities are incompatible with the fact that @xmath209 , where @xmath6 is a cat(0 ) space of dimension 2 . in particular , we see that if @xmath210 is a subgroup isomorphic to @xmath173 acting on @xmath162 with @xmath169 compact , then @xmath211 . but this implies that its image @xmath172 in @xmath8 is isomorphic to @xmath173 , acting freely on @xmath163 with @xmath166 compact . [ l - hyperbolic ] assume that @xmath6 is a cat(0 ) space and that @xmath8 acts isometrically . assume that @xmath212 . 1 . if @xmath121 is gromov hyperbolic , then @xmath213 is gromov hyperbolic . if @xmath213 contains a subgroup isomorphic to @xmath173 , then @xmath121 contains a subgroup isomorphic to @xmath173 . \(1 ) by invariance under quasiisometry , it is enough to prove that @xmath214 is hyperbolic . if it is not hyperbolic then , being a cat(0 ) space , it contains a flat plane @xmath215 by the flat plane theorem . since the image of @xmath163 in @xmath216 under the covering map @xmath217 is a flat plane , the space @xmath139 contains a flat plane , and therefore , @xmath145 contains a flat plane by lemma [ l - lifting lemma ] . therefore , @xmath145 is not hyperbolic and @xmath121 is not a hyperbolic group , contrary to assumption . thus @xmath213 is hyperbolic . \(2 ) since @xmath218 , the map @xmath219 lifts to a map @xmath220 let @xmath221 , @xmath222 , and let @xmath223 be the pull - back of @xmath224 in @xmath225 . let also @xmath226 be a subgroup isomorphic to @xmath173 , and let @xmath227 be the corresponding periodic flat . we apply lemma [ l - lifting lemma ] to the isometric action @xmath228 , and the subset @xmath223 of @xmath225 . then @xmath229 coincide with the image of @xmath214 in @xmath225 and @xmath230 , @xmath231 . it follows that @xmath232 is a flat in @xmath233 and that the projection @xmath172 of @xmath234 in @xmath235 is a subgroup isomorphic to @xmath173 , such that @xmath166 is compact . this shows that @xmath121 contains a copy of @xmath173 . the same is true for the isolated flats property : [ l - isol flats ] assume that @xmath6 is a cat(0 ) space and that @xmath8 acts isometrically . if @xmath6 has the isolated flats property , then @xmath145 has the isolated flats property . we consider the bounded model first . recall that a countable group @xmath8 is called _ virtually indicable _ if some finite index subgroup of @xmath8 admits an infinite abelian quotient . the main theorem in this section is a result that is significantly more general than theorem [ intrdensity ] but is restricted to the bounded model . it produces groups which are _ infinitesimal perturbations _ of the deterministic data . [ l -notvirtind ] assume that the deterministic data @xmath101 satisfies : * @xmath6 is a cat(0 ) space and @xmath8 acts isometrically * @xmath236 , and @xmath95 \to_p \infty$ ] * @xmath8 is not virtually indicable * @xmath13 then the random group in the bounded @xmath101-model contains a copy of @xmath173 with overwhelming probability . let @xmath172 be a subgroup of @xmath8 isomorphic to @xmath173 . let @xmath237 be a flat plane on which @xmath172 acts freely with @xmath238 a compact torus ( which exists by the flat torus theorem ) . consider the subgroup @xmath239 and let @xmath240 be the set of @xmath241-orbits of chambers in @xmath163 . since @xmath242\leq [ { \gamma}:{\gamma}_p]\ ] ] the set @xmath240 is finite . we denote by @xmath243 the image of @xmath240 into @xmath117 under the map which associates to a @xmath244-orbit in @xmath163 its corresponding @xmath103-orbit in @xmath6 . let @xmath245 be the random group associated with a random subset @xmath246 in the bounded model ( so @xmath247 is a fixed integer ) . we will show that @xmath248 { } 0 $ ] . assume towards a constradiction that the limit is not zero . then there exist @xmath249 and a sequence @xmath250 of integers such that , for all @xmath251 , the random set @xmath252 in @xmath253 contains , with probability at least @xmath249 , an equivariant chamber @xmath223 such that @xmath254 : @xmath255 a fortiori , @xmath256 thus , for all @xmath251 , @xmath257 since the condition @xmath254 is equivalent to the fact that @xmath258 , we obtain @xmath259 consider the groups @xmath260 and @xmath261 , which act freely on @xmath117 and @xmath240 respectively . it follows that there exists a constant @xmath262 , which depends only on @xmath27 , @xmath132 , the number of @xmath8-orbit of chambers in @xmath6 and the number of @xmath172-orbits of chambers in @xmath163 , such that @xmath263\leq \kappa.\ ] ] hence , for every @xmath264 , the subgroup @xmath265 of @xmath8 is ( by the isomorphism theorems ) of index @xmath266\leq \kappa.\ ] ] as @xmath8 is finitely generated ( for it acts freely on @xmath6 with @xmath10 compact ) , the family @xmath267 of subgroups of index at most @xmath262 is finite . a finitely generated group @xmath8 is not virtually indicable if and only if every finite index subgroup @xmath268 of @xmath8 has finite abelianization @xmath269 $ ] . since @xmath270 we have @xmath271|<\gamma\ ] ] for some fixed @xmath272 , not depending on @xmath264 . on the other hand , since @xmath165 , @xmath273\ ] ] and as @xmath274 { } \infty\ ] ] we have @xmath271|\xrightarrow[p_i\to \infty ] { } \infty.\ ] ] which gives the desired contradiction . thus , @xmath275 { } 0.\ ] ] we now apply lemma [ l - lifting lemma ] , whose assumptions are satisfied with overwhelming probability . this shows that @xmath276 with overwhelming probability . theorem [ t- stable random ] below can be expressed by saying that the periodic flat plane problem is stable under randomization in the bounded model " , under suitable assumptions . more precisely : we say that a property @xmath126 is _ stable under randomization _ in the @xmath124-model if the random group has property @xmath126 with overwhelming probability provided that the deterministic data @xmath101 has property @xmath126 . [ t- stable random ] let @xmath101 be a deterministic data , where @xmath8 acts isometrically on @xmath6 cat(0 ) , and @xmath236 is a sequence of normal subgroups such that @xmath95\to \infty$ ] . if @xmath8 is not virtually indicable , then the periodic flat plane alternative is stable under randomization in the bounded @xmath101-model . assume that @xmath8 satisfies the periodic flat plane alternative and let us prove that the random group does . if @xmath8 contains @xmath173 , then lemma [ l -notvirtind ] shows more : every copy of @xmath173 eventually survives in the random group . if @xmath8 does not contain @xmath173 then it is hyperbolic , and lemma [ l - hyperbolic ] shows that the random group is hyperbolic as well . according to the flat closing conjecture " ( @xmath277 the hypothesis that the periodic flat plane problem always has a positive answer ) , one expects that virtual indicability does nt play a role regarding the existence of a periodic flat plane in the bounded model . the conjecture implies that for an arbitrary deterministic data @xmath101 , where @xmath6 is cat(0 ) and @xmath8 acts isometrically , the periodic flat plane alternative is stable under randomization in the bounded @xmath101-model . this can be checked easily in a few situations , including spaces with isolated flats , using lemma [ l - isol flats ] . let us conclude this section with a statement which ( in the case of the bounded model ) weakens the assumption on the deterministic data for the @xmath13 assertion in theorem [ intrdensity ] . let @xmath72 be a non - archimedean local field with discrete valuation , and @xmath278 be an algebraic group of rank 2 over @xmath72 . let @xmath6 be the associated bruhat - tits building , @xmath8 be a uniform lattice in @xmath278 acting freely on @xmath6 , and @xmath279 be a sequence of finite index normal subgroups , and let us take @xmath101 as deterministic data . [ c- random building with chambers missing bounded model ] if @xmath8 is the random group in the bounded @xmath101-model , then @xmath13 with overwhelming probability . note that such a determinitic data can be constructed . indeed , let @xmath0 be any uniform lattice in @xmath278 ( see ( * ? ? ? ix.3 ) ; if the characteristic of @xmath72 is zero , then all lattices are all uniform ) . then , by a well - known result of selberg , @xmath8 has a torsion - free subgroup of finite index @xmath8 , and since @xmath278 over @xmath72 is a linear group , malcev s theorem shows that @xmath8 residually finite . now @xmath103 may be taken to be any sequence of finite index subgroups of @xmath8 , for example a sequence of normal subgroups with trivial intersection . being an algebraic group of rank 2 , @xmath278 satisfies kazhdan s property t ( see ( * ? ? ? * chap . 1 , theorem 1.6.1 ) ) and so does the lattice @xmath8 . a fortiori , @xmath8 is not virtually indicable and therefore lemma [ l -notvirtind ] applies . the present section defines two relevant critical densities for studying the periodic flat plane problem in the density model : @xmath280 ( the critical density for being hyperbolic ) and @xmath88 ( the critical density for containing @xmath173 ) . existence follows from the `` lift / quotient stability in the density model '' of the corresponding group property ( compare proposition [ p - hyp - z2 critical densities ] ) . we call a group property @xmath126 _ lift stable _ ( resp . _ factor stable _ or _ quotient stable _ ) , if for every surjective morphism @xmath281 where @xmath0 ( resp . @xmath8 ) has property @xmath126 , then @xmath8 ( resp . @xmath0 ) has property p. it is clear that * @xmath126 lift stable @xmath282 @xmath283 factor stable * @xmath126 lift stable & @xmath126 factor stable @xmath284 @xmath126 trivial . ( in this respect factor and lift stable properties are `` half - trivial '' properties . ) the following result explains the origin , for lift / factor stable group properties , of transition thresholds observed in the density models . [ p -phase transition ] let @xmath126 be a lift stable ( resp . @xmath285 be a factor stable ) group property and let @xmath286 be a deterministic data . there exists a critical parameter @xmath287 ( resp . @xmath288 ) such that the random group @xmath8 in the density model of parameter @xmath27 over @xmath286 satisfies the following : * @xmath8 has property @xmath126 if @xmath289 ( resp . @xmath8 has property @xmath285 if @xmath290 ) * @xmath8 does nt have property @xmath126 if @xmath291 ( resp . @xmath8 does nt have property @xmath285 if @xmath292 ) ( with overwhelming probability ) . in particular , @xmath293 and @xmath294 respectively leading to a `` threshold '' : @xmath295{random_lattices_threshold.pdf}\ ] ] _ `` half - trivial '' properties have one threshold _ it seems that in known cases @xmath296 . for @xmath126 such that @xmath297 it could be informative to determine the true behaviour of the probabilities on the interval between @xmath298 and @xmath299 . the proposition shows the existence of critical densities for * kazhdan s property t ( factor stable ) , say @xmath86 , * serre s property fa ( factor stable ) , say @xmath87 * largeness , which is the existence of a finite index subgroup of @xmath8 surjecting to a non abelian free group ( lift stable ) , say @xmath300 * virtual indicability ( lift stable ) , say @xmath301 where @xmath302 observe that if @xmath303 where both @xmath304 are factor stable / both are lift stable or if @xmath305 where @xmath126 is factor stable and @xmath285 is lift stable , then @xmath306 . the proof of proposition [ p -phase transition ] will follow from an assertion that the random group at density @xmath307 is a quotient of the random group at a higher density @xmath308 " , which we make precise in lemma [ l -phase transition ] . the existence of transition thresholds does not generalize to properties which are not lift stable or factor stable . for example , a property like @xmath309 admits a priori two thresholds . in general , the above result can be extended to any finite conjunction of properties which are lift stable or factor stable , resulting in a finite number of phase transitions . we prove lemma [ l -phase transition ] in a more general model , called the @xmath310-model " , which extends both the bounded model ( where @xmath310 is bounded ) and the density model of parameter @xmath27 for any @xmath249 ( where @xmath311 ) : let @xmath101 be a deterministic data and @xmath312 be an increasing sequence . the _ @xmath310-model _ over @xmath101 is the @xmath124-model of random groups in which @xmath123 selects @xmath313 chambers in @xmath117 , uniformly and independently at random . given @xmath314 , we say that the @xmath310-model is dominated by the @xmath315-model if @xmath316 for every @xmath317 sufficiently large . zero density models , in which say @xmath318 for every @xmath249 ( this includes the bounded model ) , can be used to provide variations on the model density @xmath249 , for example : @xmath319 . the bounded model is dominated by the density model of parameter @xmath27 , which is dominated by the density model of parameter @xmath320 . l -phase transition ] makes explicit the fact that if @xmath310 dominates @xmath315 , then the random group in the @xmath310-density model is a quotient of the random group in the @xmath315-density model " . we prove the case of lift stable properties only , since the corresponding statement for factor stable properties is analogous . [ l -phase transition ] let @xmath126 be a lift stable group property and @xmath286 be a deterministic data . if the @xmath310-model on @xmath286 is dominated by the @xmath315-model on @xmath286 , then the probability that the random group has property p in the @xmath310-model is dominated by the probability that the random group has property @xmath126 in the @xmath315-model . in particular , if @xmath126 occurs with overwhelming probability in the @xmath310-model , then it does so in the @xmath315-model , establishing prop . [ p -phase transition ] . if @xmath126 is lift stable , then @xmath321 therefore if @xmath322 denote the random subset of @xmath117 in the @xmath310-model and @xmath323 denote the random subset of @xmath117 in the @xmath315-model , then : @xmath324 this proves the assertion . however , a property like `` hyperbolicity '' is not factor stable nor lift stable , even though gromov establishes sharp phase transition in his density model . the following weakening of lift / factor - stability is consistent with the density model and allows for more general properties . [ d - liftquotient stable density ] a group property @xmath126 is said to be _ lift stable ( resp . factor stable or quotient stable ) in the density model _ if the random group at density @xmath27 satisfies property @xmath126 with overwhelming probability whenever the random group at density @xmath325 ( resp . @xmath320 ) satisfies property @xmath126 with overwhelming probability . thus , we have ( tautologically ) a critical density @xmath326 for every lift stable property @xmath126 in the density model , and similarly for factor stable properties . furthermore @xmath327 if @xmath126 is lift stable and @xmath283 factor stable . [ p - hyp - z2 critical densities ] let @xmath286 be a deterministic data , where @xmath6 is a cat(0 ) space and @xmath8 acts isometrically . then : 1 . gromov - hyperbolicity is lift stable in the density model over @xmath286 . therefore , there exists a critical density , denoted @xmath328 , for the random group in the density model to be hyperbolic , and a critical density , denoted @xmath329 , for the random universal cover to contain @xmath175 . the existence of subgroups isomorphic to @xmath173 is factor stable in the density model . therefore , there exists a critical density , denoted @xmath330 , for the random group in the density model to contain a subgroup isomorphic to @xmath173 , and a critical density , denoted @xmath331 , for the random group to contain no subgroup isomorphic to @xmath173 . by the periodic flat plane theorem , @xmath332 given a deterministic data @xmath286 where @xmath6 is a cat(0 ) space and @xmath8 acts properly uniformly isometrically , is it true that @xmath333 the flat closing conjecture implies that the answer is positive . \(1 ) by lemma [ l - hyperbolic ] ( a ) , we have : @xmath334 this shows that the random group in the @xmath315-model is hyperbolic provided that the random in the @xmath310-model is hyperbolic , for any @xmath315 dominating @xmath310 , which applies in particular to the density model . this shows the existence of @xmath335 and similarly of @xmath280 . ( 2 ) the same proof applies ( using now lemma [ l - hyperbolic ] ( b ) ) , with a reverse inequality . namely we have : @xmath336 this shows that the random group in the @xmath310-model contains @xmath173 provided that the random in the @xmath315-model contains @xmath173 , for any @xmath315 dominating @xmath310 . this shows the existence of @xmath88 and similarly of @xmath337 . the above discussion of `` lift stable '' and `` factor stable '' properties also applies to the classical gromov models and shows the existence of critical densities , for example in the density model . recall that there exist morphisms between the random group of different models at the _ same _ density , for example when comparing the usual model with the triangular model , see ( * ? ? ? * i.3.g ) . the above results are of a different nature in that they concern morphisms between random groups at distinct density regimes in a given model . the random group @xmath8 in gromov s density model exhibits the following well - known phase transition at density @xmath338 ( see @xcite and theorem 11 in @xcite ) : 1 . if @xmath33 , then with overwhelming probability @xmath8 is infinite hyperbolic , torsion free and of geometric dimension 2 . 2 . if @xmath35 , then with overwhelming probability @xmath8 is trivial ( either @xmath339 or @xmath340 ) . in other words , @xmath341 . the analogous result in our setting takes the following form . the proof , as in gromov s case , is an illustration of the box principle . [ p- r separated ] let @xmath286 be a deterministic data , and @xmath342 be fixed . then the random space @xmath139 in the density model over @xmath101 exhibits a phase transition with overwhelming probability : 1 . if @xmath33 , then the random set @xmath343 is @xmath26-separated ( that is , the simplicial distance between any two elements of @xmath154 is at least @xmath26 ) . 2 . if @xmath35 , then there exists at least @xmath26 pairs of adjacent elements of @xmath154 . \(1 ) let @xmath344 be a sequence of independent uniformly distributed random variables with values in @xmath117 and for @xmath345 denote by @xmath346 the event @xmath347 we have @xmath348 given the event @xmath349 , namely , that the faces @xmath350 have been chosen and any @xmath351 of them have diameter at least @xmath26 , we obtain , letting @xmath352 be the number of faces in @xmath6 whose union with @xmath353 of the faces @xmath354 form a set of diameter at most @xmath26 , @xmath355 let @xmath356 be the the number of faces in the @xmath26 neighbourhood of a single chamber of @xmath6 ( this does nt depend on the given chamber ) . then @xmath357 so that @xmath358 using @xmath359 for all @xmath360 , we get @xmath361 therefore if @xmath362 , where @xmath33 , then @xmath363 is arbitrarily close to 1 . this concludes the proof of ( 1 ) . the proof of ( 2 ) is of a similar nature . then the density model exhibits several extra phase transitions as the parameter @xmath27 increases . the transitions depend on the local structure of the initial space @xmath6 : [ d12 ] let @xmath286 be a deterministic data , and @xmath342 , @xmath364 be fixed . then the random space @xmath139 in the density model over @xmath286 admits a phase transition at density @xmath365 . namely , the following holds with overwhelming probability : 1 . if @xmath366 , every ball @xmath367 of radius @xmath26 in @xmath368 contains to at most @xmath72 elements of @xmath154 . if @xmath369 , there exists an @xmath26-separated set @xmath370 of cardinal at least @xmath371 , such that every edge @xmath372 is included in exactly @xmath373 elements of @xmath154 . \(1 ) let @xmath344 be a sequence of independent uniformly distributed random variables with values in @xmath117 . for a ball @xmath374 of radius @xmath26 , we write @xmath375 to mean that both @xmath376 and @xmath377 are included in @xmath367 . let @xmath374 be a ball of radius @xmath26 and fix @xmath378 . since @xmath379 where @xmath380 ( which is finite as @xmath10 is compact ) , we obtain @xmath381 therefore if @xmath362 , where @xmath382 , then with overwhelming probability every ball in @xmath383 of radius at most @xmath26 contains at most @xmath72 elements of the random subset @xmath154 of @xmath117 , at density @xmath27 . \(2 ) let @xmath344 be a sequence of independent uniformly distributed random variables with values in @xmath117 . let @xmath384 be the set of all @xmath72-tuples of elements of @xmath117 whose faces are mutually adjacent to a same edge . for every @xmath385 we write @xmath386 for the event @xmath387 and let @xmath388 be the characteristic function of @xmath386 . we will estimate the expectation of the random variable @xmath389 whenever @xmath390 $ ] . fix @xmath72 and @xmath391 . write @xmath392 we have @xmath393 by inclusion - exclusion , where @xmath394 and @xmath395 therefore , @xmath396 substituting @xmath134 for @xmath264 we deduce @xmath397 since for @xmath32 large enough , @xmath398 whenever @xmath399 and @xmath400 we have @xmath401 and therefore , @xmath402 now , @xmath403 so @xmath404 showing that @xmath405 whenever @xmath406 . since is @xmath6 uniformly locally bounded , this shows that we can find at least @xmath371 edges @xmath407 in @xmath6 ( for any @xmath371 fixed in advance ) , having at least @xmath72 chambers adjacent to @xmath408 removed . in addition , since the number of such edges grows to infinity , we may suppose that the pairwise distance between these edges is at least @xmath26 from each other in @xmath6 ( for any @xmath26 fixed in advance ) , and since the chambers are chosen independently at random , we may also suppose that exactly @xmath409 distinct chambers will be removed on @xmath407 . [ p - d1 ] let @xmath286 be a deterministic data and let @xmath410 . then the random group @xmath121 in the density model over @xmath101 satisfies the following with overwhelming probability : 1 . if @xmath411 , then @xmath121 acts freely on a simply connected space @xmath6 without boundary with @xmath412 compact . 2 . if @xmath413 , then @xmath121 splits off a free factor isomorphic to a free group @xmath414 on @xmath371 generators : @xmath415 where @xmath235 is a finitely presented group , and @xmath416 . in particular , @xmath417 where @xmath87 is the critical density for serre s property fa ( existence of fixed point for actions on trees ) . the first assertion follows directly from lemma [ d12 ] by contracting the free faces , so we prove ( 2 ) . since @xmath10 is compact , there exists a constant @xmath272 such that , if @xmath418 denotes the lowest order of an edge in @xmath6 , then @xmath419 satisfies @xmath420 for every @xmath421 . let @xmath422 and let @xmath423 be the set of all @xmath72-tuples of elements of @xmath117 whose faces are mutually adjacent to a same edge @xmath424 . the proof of lemma [ d12 ] , with the same notation except for @xmath423 replacing @xmath384 , shows that @xmath425 and since now @xmath426 we obtain @xmath427 . therefore there exists , with overwhelming probability , an @xmath26-separated set @xmath428 of cardinal @xmath371 , such that every edge @xmath372 is included in exactly @xmath429 elements of @xmath125 . let @xmath430 be the quotient space @xmath431 , and let @xmath432 be a maximal tree of the 1-skeleton of @xmath433 , and choose a root @xmath434 . let @xmath435 be the topological space obtained from @xmath433 by collapsing @xmath432 to @xmath436 , which is homotopy equivalent to @xmath433 . the elements of @xmath437 , being face free edges in @xmath438 form a wedge sum @xmath439 of @xmath371 edges in the collapse @xmath440 of @xmath441 in @xmath435 , and @xmath440 is the wedge sum of @xmath439 and a compact complex @xmath442 . the van kampen theorem shows that @xmath443 splits as @xmath444 where @xmath445 and @xmath446 is finitely presented . in particular we have @xmath447 proving the proposition . we now prove the implication @xmath448 in theorem [ intrdensity ] and similar assertions over local rings . in theorem [ intrdensity ] the fixed global field ( denoted @xmath72 ) is @xmath449 and the uniform lattice @xmath8 in the deterministic data @xmath101 is the cartwright steger lattice @xcite which is of type @xmath450 over @xmath451 . the presentation of @xmath8 by lubotzky , samuels and vishne in @xcite is especially useful , where the congruence subgroups @xmath452 are described explicitly to illustrate the ramanujan property of their quotients . we first review briefly the construction for notational purposes , referring to @xcite for complete details . let @xmath453 be the central simple algebra of degree 3 over @xmath72 defined by @xmath454 with relations @xmath455 and @xmath456 , where @xmath457 is a generator of @xmath458 and @xmath459 is a basis for @xmath460 over @xmath461 . associated with @xmath453 are algebraic groups over @xmath72 defined by @xmath462 and @xmath463 . for a valuation @xmath464 on @xmath72 we let @xmath465 and say that @xmath466 splits whenever @xmath467 . let @xmath468 consisting of the degree valuation @xmath469 on @xmath72 , and the valuation @xmath470 associated with the prime @xmath471 , namely , @xmath472 where the polynomials @xmath473 are prime to @xmath474 . then by ( * ? ? ? 3.1 ) the algebra @xmath466 splits for all valuations @xmath475 on @xmath72 , while it is a division algebra for @xmath476 . let @xmath477 and @xmath478 $ ] . since @xmath479 splits , we have @xmath480 , so @xmath481 embeds as a discrete subgroup of @xmath45 . since @xmath482 and @xmath483 is compact for @xmath476 , then strong approximation ( cf . * section 4 ) ) shows that @xmath481 is a cocompact lattice in @xmath45 . the cartwright steger lattice ( * ? ? ? * section 2 ) is the subgroup @xmath8 of @xmath481 defined as follows . we note that , as stated above , the lattice @xmath481 is well defined only up to commensurability . a strict definition of @xmath481 depends upon fixing a embedding @xmath484 into a linear group over @xmath72 , which is chosen here to be @xmath485 ( see ( * ? ? ? 3.3 ) ) , so that @xmath486 . then @xmath8 consists of matrices of @xmath481 whose reduction modulo @xmath487 are upper triangular with @xmath488 identity blocks on the diagonal . another description of @xmath8 from ( * ? ? ? * section 4 ) is as follows . let @xmath489 be the subring of @xmath72 given by @xmath48 $ ] , and @xmath490 be the @xmath489-algebra @xmath491 having the same defining relations as @xmath453 , so that @xmath453 appears as the algebra of central fractions of @xmath490 , namely , @xmath492 . we have @xmath493 ( ( * ? ? ? let @xmath494 and for @xmath495 , set @xmath496 . let @xmath497 be the subgroup of @xmath498 generated by the elements @xmath499 . then @xmath8 is the quotient of @xmath497 modulo @xmath500 . write @xmath6 for the bruhat tits building associated to @xmath45 . the vertices of @xmath6 are @xmath501$]-lattices in @xmath502 and incidence is given by flags . the action of @xmath481 on @xmath6 ( via its embedding in @xmath45 ) is transitive on vertices . the group @xmath8 is a normal subgroup of @xmath481 of finite index ( see ( * ? ? ? * theorem 2.6 ) ) which also acts transitively on the vertices of @xmath6 ( in fact , the reduced norm of @xmath503 is @xmath504 , which coincide with @xmath60 up to the invertible element @xmath471 of @xmath501 $ ] ) . the congruence subgroups of @xmath8 are defined as follows . let @xmath505 be an ideal in @xmath489 . the quotient map @xmath506 induces an epimorphism @xmath507 , which itself induces a group morphism @xmath508 . denote the kernel by @xmath509 and its quotient modulo @xmath500 by @xmath510 . then we set @xmath511 , and let @xmath512 be the quotient group . let @xmath59 $ ] be a monic irreducible polynomial prime to @xmath60 and @xmath61 , @xmath62 be an integer , and @xmath66 . the deterministic data for theorem [ intrdensity ] is @xmath513 where : * @xmath6 is the bruhat - tits building of @xmath45 * @xmath8 is the cartwright - steger lattice in @xmath45 * @xmath514 are the arithmetic lattices associated with @xmath515 . we suppose @xmath516 . the assertion that @xmath13 for these random groups is related to the elementary algebraic structure of @xmath517 and @xmath14 over fields and local rings . as usual for a local ring @xmath518 , @xmath519 is the group of units of @xmath520 , @xmath521 is the subgroup of @xmath519 of matrices with determinant 1 , and @xmath522 , @xmath523 are the projective versions . we start with a lemma . let @xmath524 be a finite local principal ring of prime characteristic with residue field @xmath525 , uniformizer @xmath526 and length @xmath527 ( the smallest integer @xmath528 such that @xmath529 ) . [ lemdens1 ] the order of the group @xmath530 generated by two commuting elements @xmath531 is at most @xmath532 where @xmath533 is the logarithm relative to @xmath56 , and @xmath534 is the upper integer value . in the reduction modulo @xmath351 @xmath535 an element @xmath536 such that @xmath537 has a representative of the form @xmath538 for some @xmath539 , @xmath540 and @xmath541 . the frobenius map gives @xmath542 therefore @xmath543 . thus , if @xmath153 and @xmath544 are two commuting elements in @xmath522 , then the kernel @xmath137 in the exact sequence @xmath545 is an abelian group of order at most @xmath546 . we now estimate @xmath547 . assume without loss of generality that @xmath548 . let @xmath549 be representatives of @xmath550 in @xmath519 , and let @xmath551 $ ] be the characteristic polynomial of @xmath552 . assume that @xmath553 does nt have a root in @xmath518 . then @xmath553 is irreducible over @xmath518 ( since a factorization would split off a degree one factor ) and therefore equal to its minimal polynomial . it follows that the rational canonical form of @xmath554 has only one block associated with @xmath553 , and @xmath554 has a cyclic vector @xmath555 . given @xmath556 , there exists @xmath557 $ ] such that @xmath558 if @xmath559 for some @xmath539 , then @xmath560 so @xmath561 thus , @xmath562 in the basis @xmath563 . taking determinants the equality @xmath564 shows that @xmath565 is a cube root of 1 . thus every element of @xmath566 belong to the subspace @xmath567\right\}\ ] ] of @xmath568 , possibly after multiplying by an element of the form @xmath569 @xmath539 a cube root of 1 . since the dimension of @xmath570\right\}$ ] over @xmath518 is 3 , the order of @xmath566 is at most @xmath571 . a similar argument works more generally if @xmath553 equals the minimal polynomial . otherwise , @xmath553 has a root @xmath572 in @xmath518 and we may assume up to conjugacy that @xmath554 has the form @xmath573 where @xmath574 and @xmath575 . write @xmath576 @xmath577 , @xmath578 , and @xmath579 . since @xmath580 for some @xmath539 , we have @xmath581 so @xmath582 and thus @xmath583 , @xmath584 and @xmath585 if @xmath586 , then the argument of the previous shows that the order of @xmath587 is at most @xmath588 , so the order of @xmath589 is at most @xmath590 . otherwise , @xmath572 is an eigenvalue of @xmath310 and it is easy to see that the order of @xmath589 is at most @xmath590 in this case too , which proves the lemma . the local ring @xmath591/\langle f_p(y)^{s_p } \rangle\ ] ] has residue field @xmath592/\langle f_p(y)\rangle.\ ] ] by theorem 6.6 in @xcite @xmath593 let @xmath172 be any subgroup of @xmath8 isomorphic to @xmath173 . lemma [ lemdens1 ] shows that the order of the image of @xmath172 in @xmath594 is at most @xmath595 where @xmath596 is the order of the residue field . denote by @xmath597 the flat associated to @xmath172 in @xmath6 and let @xmath117 be the set of @xmath598-orbits of chambers in @xmath6 . the set @xmath240 of @xmath599 orbits of chambers in @xmath163 is finite , and there is a constant @xmath223 ( namely , @xmath600the number of @xmath172 orbits of chambers in @xmath163 ) such that @xmath601 let @xmath602 be the image of @xmath240 in @xmath117 . consider a sequence @xmath344 of independent identically distributed random variable with values in @xmath117 . the probability @xmath363 of the event : @xmath603 can be estimated below by @xmath604 therefore @xmath605 since @xmath594 acts freely on @xmath117 with @xmath606 fixed , @xmath607 for some constant @xmath608 . let us compute @xmath609 . the reduction map @xmath610 induces a surjective map @xmath611 since @xmath612 commutes with reduction modulo @xmath613 , and since an element is invertible in @xmath614 if and only if its image in @xmath615 is nonzero ( as @xmath614 is local ) . therefore any pull - back of a matrix of @xmath616 in @xmath617 is a matrix in @xmath618 . the kernel of @xmath526 consists of matrices of the form @xmath619 where all coefficient of @xmath620 belong to the ideal @xmath613 . therefore @xmath621 as @xmath622 is surjective and @xmath623 if and only if @xmath572 is not a multiple of @xmath624 , we have that @xmath625 this gives @xmath626 . therefore , @xmath627 thus , since @xmath628 when @xmath629 , we obtain : * if @xmath630 for large @xmath317 , then @xmath631 is cyclic , @xmath632 ( for large @xmath317 ) and thus @xmath633 if @xmath634 that is , @xmath63 * if @xmath635 for large @xmath317 , then @xmath631 is not cyclic in general , but using the rough estimate @xmath636 ( or @xmath637 if the order satisfy @xmath638 , @xmath639 ) , we have that @xmath633 if @xmath640 that is @xmath641 * if @xmath642 does nt satisfy ( i ) or ( ii ) , we separate the cases @xmath643 and @xmath630 and obtain the desired conclusion when @xmath644 as @xmath645 for @xmath646 . this shows that @xmath633 . therefore , lemma [ l - lifting lemma ] applies with overwhelming probability and we have that @xmath13 with overwhelming probability . we have proved the stronger result that every embedding @xmath647 eventually lifts to an embedding @xmath13 with overwhelming probability . recall that according to mostow and prasad , the set of embeddings @xmath647 is dense @xcite . are these random groups residually finite ? ( for comparison , note that there exist non residually finite @xcite central extensions of the form @xmath648 . ) what is the value of @xmath649 ? what is the value of @xmath650 ? if @xmath6 is a bruhat tits building of rank 2 ( possibly exotic ) and @xmath651 is an @xmath489-separated subset of chambers , does @xmath652 for @xmath489 large enough ? it is proved in @xcite that the congruence quotients of @xmath82 by @xmath653 are ramanujan complexes . this fact , whose proof relies on the jacquet - langlands correspondence " ( in positive characteristic ) , is not used in the present paper . ( nevertheless , the random quotients @xmath10 as in theorem [ intrdensity ] are ramanujan complexes up to @xmath654 " , at least if the density is not too large . ) the main input from algebraic groups used in the proof is strong approximation , and so it is most likely that results in the spirit of theorem [ intrdensity ] can also be established for other algebraic groups ( e.g. @xmath655 ) and over other fields ( e.g. characteristic 0 ) , the first step being to construct explicit families of lattices . we also note that the case of positive characteristic is easier to handle from an algorithmic perspective ( the construction of @xmath624 can be done in polynomial time using the shoup algorithm ) . the critical density @xmath656 for kazhdan s property t can be estimated using the so - called spectral criterion for property t " , which is local . this is done by expliciting ( spectral ) bounds on the local geometric perturbations , following @xcite and @xcite . [ t - garland ] let @xmath657 be finite simplicial complex of dimension @xmath371 in which every simplex belongs to an @xmath371-cell , and let @xmath658 . assume that for every vertex @xmath659 , the @xmath26-th reduced cohomology group @xmath660 [ i.e. @xmath661 if @xmath662 and @xmath663 of the link @xmath664 of @xmath657 vanishes , and denote by @xmath665 the smallest nonzero eigenvalue of @xmath666 acting on @xmath667 . set @xmath668 . if @xmath669 then @xmath670 for every finite dimensional unitary representation @xmath671 of @xmath672 . the statement in @xcite deals only with those @xmath657 whose universal coverings are bruhat - tits buildings , but borel points out ( * ? ? ? * theorem 1.5 ) that garland s argument is general . ( while we are mostly interested in bruhat - tits buildings here , it is the general case that is useful for us . ) borel also notes in 2.2 that the assumption that @xmath657 is finite can be replaced by uniformly locally finite " , provided that the conclusion refers to the vanishing of @xmath676 ( @xmath677-summable cochains ) . it was later shown that the family of admissible @xmath671 with vanishing @xmath678 could be substantially enlarged , in particular , the linearity assumption on @xmath671 can be disposed of ( in the smooth category @xcite ) , as can the finite dimensionality assumption on @xmath671 ( see @xcite ) . the latter leads to the conclusion that @xmath670 for every unitary representation @xmath671 of @xmath672 which , in the case @xmath674 , is equivalent to @xmath8 having kazhdan s property t by a well known result of delorme and guichardet ( see @xcite ) . garland studies in particular the 2-dimensional case , exploiting the computation of the exact value of @xmath679 for tits buildings ( associated with bn - pairs , the computation appears in @xcite and ) to derive the vanishing of @xmath680 when the order is large enough " . ( as garland observes , his cohomology vanishing results for lattices with finite dimensional unitary targets were in fact already known in degree 1 as they were covered by kazhdan s work . ) we now explain why property t holds for random groups when the order is large enough " ( in particular when @xmath681 if @xmath33 , and the deterministic data is of type @xmath450 , and for @xmath682 if @xmath27 is only assumed to be bounded away from 1 ) . [ l - large order ] for every @xmath249 and every integer @xmath683 , there exists a constant @xmath684 such that if @xmath685 is a graph of order @xmath686 with @xmath687 , and @xmath688 is a subgraph with @xmath689 , then @xmath690 . indeed , by the cheeger inequality ( for graphs ) , it is enough to prove the corresponding statement for the cheeger constant , then the numerator of the two isoperimetric ratios differ by a constant that is washed out as @xmath691 . for a subset @xmath692 , let @xmath693 , where @xmath694 is the number of edges with extremities belonging to both @xmath154 and @xmath695 , and @xmath696 . the cheeger constant of @xmath685 is @xmath697 . let @xmath154 be a subset of @xmath698 such that @xmath699 . by the cheeger inequality , @xmath700 since @xmath686 , we obtain @xmath701 . thus @xmath702 using again the cheeger inequality . let @xmath6 be a thick irreducible euclidean building of rank 2 . the links of @xmath6 are spherical buildings and there exist integers @xmath703 such that the degree of an edge in @xmath6 is either @xmath704 or @xmath705 . we call @xmath706 the order of @xmath6 . in the @xmath450 case , @xmath707 . [ t - t ] for every @xmath83 , there exists @xmath76 such that if @xmath78 and if in the deterministic data @xmath101 , the space @xmath6 is a thick classical ( i.e. associated with an algebraic group over a local field ) irreducible bruhat tits building of rank 2 and order @xmath708 , then the random group at density @xmath27 with deterministic data @xmath101 has kazhdan s property t with overwhelming probability . thus @xmath709 as @xmath710 . the groups appearing in the above result ( and in corollary [ c- random building with chambers missing bounded model ] ) are buildings with chambers missing in the sense of @xcite . some ( weak ) buildings with chambers missing have the opposite " haagerup property , see @xcite . we also note that random groups in the density model admit an infinite quotient with property t , _ independently of the order _ , as soon as we are given a deterministic data with property t. let @xmath72 be large enough so that @xmath711 . let @xmath712 be given . by lemma [ l - large order ] , we can find @xmath713 such that if @xmath714 then @xmath715 . higman @xcite and garland @xcite show that if @xmath716 is a spherical building of order @xmath56 , then @xmath717 converges to 1 as @xmath691 . let @xmath76 be larger than @xmath718 and be such that @xmath719 for any spherical building @xmath720 of order @xmath77 . if in the deterministic data @xmath101 , the space @xmath6 is a thick classical irreducible building of rank 2 such that @xmath721 , and if @xmath78 , then by proposition [ d12 ] , the links of @xmath6 contain at most @xmath72 chambers missing at every vertex , with overwhelming probability . therefore , the random perforated building @xmath722 at density @xmath27 satisfies @xmath723 with overwhelming probability . since the random universal cover @xmath724 is locally isomorphic to @xmath725 , theorem [ t - garland ] applies . thus the random group has property t ( and the spectral gap is arbitrarily close to 1 ) with overwhelming probability . it is natural to wonder if the bound on the order can be estimated . it seems difficult to give a general answer , but we can do it at least when the density is @xmath726 ( or in the model with few chambers missing ) . in fact , one can compute the _ exact _ value of @xmath727 , in the spirit of feit and higman s paper @xcite . let @xmath733 be an extension into an @xmath728 building of order @xmath56 ( see @xcite ) and let @xmath126 be the projective plane corresponding to @xmath734 . we choose the following basis for the space of functions on the vertex set of @xmath685 : @xmath735 where @xmath736 ( resp . @xmath737 ) corresponds to the point @xmath32 ( resp . the line @xmath738 ) of @xmath126 associated to the missing chamber , while @xmath739 ( resp . @xmath740 ) correspond to an enumeration of the points of @xmath738 distinct from @xmath32 ( resp . the lines adjacent to @xmath32 distinct from @xmath738 ) in @xmath126 . @xmath741 where @xmath154 is the normalized adjacency matrix , namely @xmath742 , where @xmath743 is the usual ( bipartite ) adjacency matrix ( we recall that @xmath744 when @xmath745 is adjacent to @xmath746 ) , and @xmath747 is the diagonal matrix having @xmath748 down the diagonal , except on the first entry which is @xmath56 . denote @xmath749 . a computation shows that @xmath750 where @xmath751 is the @xmath752 matrix with @xmath753 on the diagonal , and @xmath754 elsewhere , while @xmath755 is the @xmath756 matrix with @xmath757 on the diagonal and @xmath758 elsewhere . set @xmath759 we have @xmath760 whose kernel is of codimension 3 . consider the vectors @xmath761 a direct computation shows that @xmath762 thus the problem reduces to computing the spectrum of the @xmath488 matrix @xmath763 the characteristic polynomial is @xmath764 where @xmath765 is an obvious root . the two other roots are @xmath766 and @xmath767 therefore , the eigenvalue of @xmath768 we are interested in is @xmath769 a similar computation can be worked out when two chambers are missing ( arguing according to the respective positions of the chambers in an apartment ) but the value of @xmath770 when @xmath685 is a spherical building ( of dimension 1 ) might be more challenging to find . it would also be interesting to make similar estimates with more general ( e.g. smooth ) targets @xmath771 and deduce the corresponding fixed points theorem for the random groups following @xcite . the results in @xcite on buildings with chambers missing and @xcite on property rd can be applied to our random groups in the @xmath450 case . this shows in particular that the random group is _ rigid _ in the sense that it remembers the building that it comes from ( see ( * ? ? ? * section 5 ) ) . furthermore , by corollary 6 in @xcite , the random group satisfies the baum connes conjecture without coefficients . 1 . the random group @xmath772 at density @xmath773 admits unique extension @xmath774 ( in the sense of ( * ? ? ? * section 1 ) ) into a euclidean building . 2 . the random group @xmath8 at arbitrary density satisfy the baum connes conjecture without coefficients . the first assertion is a consequence of proposition [ p- r separated ] and ( * ? ? ? * theorem 5.11 ) . it is unknown if @xmath8 satisfies the baum connes conjecture with coefficients . the conjecture without coefficients is also unknown for the other coxeter types ( in the irreducible case ) but would follow from the `` interpolation of property rd to intermediate rank situations '' ( see @xcite ) , that is , between hyperbolic groups and ( uniform ) higher rank lattices of the corresponding type , provided that the latter groups satisfy property rd ( which is conjectured by a. valette ) . gromov , m. geometric group theory . proceedings of the symposium held at sussex university , sussex , july 1991 . edited by graham a. niblo and martin a. roller . london mathematical society lecture note series , 182 . cambridge university press , cambridge , 1993 . ollivier , yann a january 2005 invitation to random groups . ensaios matemticos [ mathematical surveys ] , 10 . sociedade brasileira de matematica , rio de janeiro , 2005 . ( see also january 2010 random groups updates " ) tits , jacques spheres of radius @xmath780 in triangle buildings . i. finite geometries , buildings , and related topics ( pingree park , co , 1988 ) , 1728 , oxford sci . publ . , oxford univ . press , new york , 1990 .

this paper proposes the ricci flow equation as a general geometric model for various reaction diffusion systems ( and related dissipative solitons ) in mathematical biology . more precisely , we propose a hypothesis that any kind of reaction diffusion processes in biology , chemistry and physics can be modelled by the combined geometric diffusion system . in order to demonstrate the validity of this hypothesis , we review a number of popular reaction diffusion systems and try to show that they can all be subsumed by the presented geometric framework of the ricci flow . + * keywords : * ricci flow , bio reaction diffusion , dissipative solitons and breathers

0806.2194

parabolic _ reaction diffusion _ systems are abundant in mathematical biology . they are mathematical models that describe how the concentration of one or more substances distributed in space changes under the influence of two processes : local chemical reactions in which the substances are converted into each other , and diffusion which causes the substances to spread out in space . more formally , they are expressed as semi linear parabolic partial differential equations ( pdes , see e.g. , @xcite ) . the evolution of the state vector @xmath0 describing the concentration of the different reagents is determined by anisotropic diffusion as well as local reactions : @xmath1where each component of the state vector @xmath0 represents the concentration of one substance , @xmath2 is the standard laplacian operator , @xmath3 is a symmetric positive definite matrix of diffusion coefficients ( which are proportional to the velocity of the diffusing particles ) and @xmath4 accounts for all local reactions . the solutions of reaction diffusion equations display a wide range of behaviors , including the formation of travelling waves and other self organized patterns like _ dissipative solitons _ ( dss ) . on the other hand , the _ ricci flow equation _ ( or , the parabolic einstein equation ) , introduced by r. hamilton in 1982 @xcite , is the nonlinear heat like evolution equation topologically a sphere if every closed curve in @xmath5 can be shrunk continuously to a point ? in other words , poincar conjectured : a simply - connected compact 3manifold is diffeomorphic to the 3sphere @xmath6 ( see e.g. , @xcite ) . ] @xmath7for a time dependent riemannian metric @xmath8 on a smooth real @xmath9manifold @xmath5 with the ricci curvature tensor @xmath10 . ) was chosen by hamilton for much the same reason that a. einstein introduced the ricci tensor into his gravitation field equation , @xmath11where @xmath12 is the energy momentum tensor . einstein needed a symmetric 2index tensor which arises naturally from the metric tensor @xmath13 and its first and second partial derivatives . the ricci tensor @xmath14 is essentially the only possibility . in gravitation theory and cosmology , the ricci tensor has the volume decreasing effect ( i.e. , convergence of neighboring geodesics , see @xcite ) . ] this equation roughly says that we can deform any metric on a 2surface or @xmath9manifold by the negative of its curvature ; after _ normalization _ ( see figure [ ricci ] ) , the final state of such deformation will be a metric with constant curvature . however , this is not true in general since , in addition to the presence of singularities , the limits could be ricci solitons ( see below ) . the factor of 2 in ( [ rf ] ) is more or less arbitrary , but the negative sign is essential to insure a kind of global _ volume exponential decay_,]for a positive function @xmath15 . we can get some insight into its solution from the simple exponential decay ode , @xmath16(where @xmath17 is the observed quantity with its initial value @xmath18 and @xmath19 is a positive decay constant ) , as well as the corresponding @xmath20th order rate equation ( where @xmath21 is an integer),@xmath22 ] since the ricci flow equation ( [ rf ] ) is a kind of nonlinear geometric generalization of the standard linear _ _ heat equation _ _ @xmath23like the heat equation ( [ h1 ] ) , the ricci flow equation ( [ rf ] ) is well behaved in forward time and acts as a kind of smoothing operator ( but is usually impossible to solve in backward time ) . if some parts of a solid object are hot and others are cold , then , under the heat equation , heat will flow from hot to cold , so that the object gradually attains a uniform temperature . to some extent the ricci flow behaves similarly , so that the ricci curvature ` tries ' to become more uniform @xcite , thus resembling a monotonic _ entropy growth _ , of the metric @xmath13 is viewed as the leading quantity of the system and plays the role of a probability density , in perelman s entropy the leading quantity describing the system is the metric @xmath13 itself . hamilton established the monotonicity of his entropy along the volume - normalized ricci flow on the 2sphere @xmath24 surface . perelman established the monotonicity of his entropy along the ricci flow in all dimensions @xcite . ] @xmath25 , which is due to the positive definiteness of the metric @xmath26 , and naturally implying the _ arrow of time _ @xcite . in a suitable local coordinate system , the ricci flow equation ( [ rf ] ) has a nonlinear heat type form , as follows . at any time @xmath27 , we can choose local harmonic coordinates so that the coordinate functions are locally defined harmonic functions in the metric @xmath28 . then the ricci flow takes the general form ( see e.g. , @xcite ) @xmath29where @xmath30 is the _ laplace beltrami operator _ ( [ lb ] ) and @xmath31 is a lower order term quadratic in @xmath32 and its first order partial derivatives @xmath33 . from the analysis of nonlinear heat pdes , one obtains existence and uniqueness of forward time solutions to the ricci flow on some time interval , starting at any smooth initial metric @xmath34 . the quadratic ricci flow equation ( [ rh ] ) is our geometric framework for general bio reaction diffusion systems , so that the spatio temporal pde ( [ re - dif3d ] ) corresponds to the quadratic ricci flow pde @xmath35 with : * the metric @xmath36 on an @xmath9manifold @xmath5 corresponding to the @xmath9dimensional ( or @xmath9component , or @xmath9phase ) concentration @xmath37 ; * the laplace beltrami differential operator @xmath30 , as defined on @xmath38functions on an @xmath9manifold @xmath5 , with respect to the riemannian metric @xmath13 , by @xmath39 @xmath40 ; and * the quadratic @xmath9dimensional ricci term , @xmath31 , corresponding to the @xmath9dimensional bio reaction term , @xmath41 . as a simple example of the ricci flow equations ( [ rf])([rh ] ) , consider a round spherical boundary @xmath42 of the 3ball radius @xmath43 . the metric tensor on @xmath42 takes the form @xmath44where @xmath45 is the metric for a unit sphere , while the ricci tensor @xmath46is independent of @xmath43 . the ricci flow equation on @xmath42 reduces to @xmath47 thus the boundary sphere @xmath42 collapses to a point in finite time ( see milnor ) . more generally , the geometrization conjecture @xcite holds for any 3manifold @xmath5 ( see below ) . suppose that we start with a compact initial 3manifold @xmath48 whose ricci tensor @xmath14 is everywhere positive definite . then , as @xmath48 shrinks to a point under the ricci flow ( [ rf ] ) , it becomes rounder and rounder . if we rescale the metric @xmath13 on @xmath48 so that the volume of @xmath48 remains constant , then @xmath48 converges towards another compact 3manifold @xmath49 of constant positive curvature ( see ham82 ) . in case of even more general @xmath50manifolds ( outside the class of positive ricci curvature metrics ) , the situation is much more complicated , as various singularities may arise . one way in which singularities may arise during the ricci flow is that a spherical boundary @xmath51 of an @xmath50manifold @xmath5 may collapse to a point in finite time . such collapses can be eliminated by performing a kind of ` geometric surgery ' on the 3manifold @xmath5 , that is a sophisticated sequence of cutting and pasting without accumulation of time errors ) with surgery is well understood , then one would be able to recognize the topological structure of the initial manifold . thus hamilton s program , when carried out successfully , would lead to a proof of the poincar conjecture and thurston s geometrization conjecture yau . ] ( see @xcite ) . after a finite number of such surgeries , each component either : ( i ) converges towards a 3manifold of constant positive ricci curvature which shrinks to a point in finite time , or possibly ( ii ) converges towards an @xmath52 which shrinks to a circle @xmath53 in finite time , or ( iii ) admits a ` thin thick ' decomposition of thurston . therefore , one can choose the surgery parameters so that there is a well defined ricci flow with surgery , that exists for all time perel2 . in this paper we use the evolving @xmath9dimensional geometric machinery of the volume decaying and entropy growing ricci flow @xmath28 , given by equations ( [ rf])([rh ] ) , for modelling various biological reaction diffusion systems and dissipative solitons , defined by special cases of the general spatio temporal model ( [ re - dif3d ] ) . in case of ideal mixtures , the driving force for the general diffusion @xmath54 ( re - dif3d ) is the concentration gradient @xmath55 , or the gradient of the chemical potential @xmath56 of each species @xmath57 , giving the _ diffusion flux _ by the first fick s law,@xmath58assuming the diffusion coefficients @xmath3 to be a constant , the second fick s law gives the linear parabolic heat equation,@xmath59while , in case of variable diffusion coefficients @xmath60 we get ( slightly ) more general parabolic _ diffusion equation_,@xmath61 which is still analogous to the ` linear ' part of the quadratic ricci flow equation ( [ rh ] ) , @xmath62 due to general ` diffusion properties ' of the laplace beltrami operator @xmath30 . the @xmath9dimensional diffusion coefficient @xmath63 at different temperatures @xmath64 can be approximated by the arrhenius exponential decay relation,@xmath65where @xmath66 is the maximum possible diffusion coefficient ( at infinite temperature @xmath64 ) , @xmath67 is the activation energy for diffusion ( i.e. , the energy that must be overcome in order for a chemical reaction to occur ) and @xmath68 is the gas constant . using the first fick s first law ( [ fick1 ] ) , the diffusion equation ( [ difeq ] ) can be derived in a straightforward way from the _ continuity equation _ , which states that a change in density in any part of the system is due to inflow and outflow of material into and out of that part of the system ( effectively , no material is created or destroyed),@xmath69where @xmath70 is the flux of the diffusing material . the most important special case of ( [ heat1 ] ) is at a steady state , when the concentrations @xmath71 do not change in time , giving the _ laplace s equation _ , @xmath72for harmonic functions @xmath73 . the stochastic version of the deterministic heat equation ( [ heat1 ] ) , connected with the study of brownian motion , is @xmath74 , where @xmath75 is the laplacian on @xmath76 more generally , a brownian motion on an @xmath9manifold @xmath5 is given by one - half of the laplace beltrami operator @xmath77 ( [ lb ] ) . ] is the _ fokker planck equation _ ( see e.g. , @xcite ) , @xmath78 + \partial _ { x^{i}x^{j}}\left [ d_{ij}^{2}(x^{i})f\right ] , \label{fokker}\]](@xmath79 @xmath80 ) , where where @xmath81 is the drift vector and @xmath82 the diffusion tensor ( which results from the presence of the stochastic force ) . the fokker planck equation ( [ fokker ] ) is used for computing the probability densities of stochastic differential equations.]where @xmath83 is the state of an @xmath9dimensional stochastic system at time @xmath27 and @xmath84 is the standard @xmath85d wiener process . if the initial distribution is @xmath86 , then the probability density of the state is given by the fokker planck equation ( [ fokker ] ) with the drift and diffusion terms , @xmath87 ] also , notice that the real valued heat equation ( [ heat1 ] ) is formally similar to the complex valued _ schrdinger equation _ ( see e.g. , quleap ) , @xmath88where @xmath89 is the wave function of the particle , @xmath90 and @xmath91 is planck s constant divided by @xmath92 in the remainder of this section , we will review a number of particular bio reaction diffusion systems , which are likely to be subsumed by the quadratic ricci flow model ( [ rh ] ) . the simplest bio re - action diffusion pde concerning the concentration @xmath93 of a single substance in one spatial dimension , @xmath94is also referred to as the kolmogorov piscounov ( kpp ) equation . if the reaction term vanishes , then the equation represents a pure diffusion process described by the heat equation . in particular , the choice @xmath95yields _ fisher s equation _ that was originally used to describe the spreading of biological populations.]where @xmath96 represents the population density , @xmath97 is a spatially homogenous diffusion constant , @xmath98 is a spatially inhomogeneous growth term , and @xmath99 is a competition term ( @xmath100 is a competition rate and @xmath101 is the microscopic length scale at which two particles will compete with one another ) . one simple form of inhomogeneity considered in these works is a ` square well ' potential @xmath102 which consists of a uniform space with negative growth rate ( termed the ` desert ' ) , in which a single region of positive growth rate ( an ` oasis ' ) is placed . this model has proven to be applicable to experiments with bacteria populations in adverse environments @xcite . ] the one component kpp equation ( [ re - dif1d ] ) can also be written in the variational ( gradient ) form @xmath103and therefore describes a permanent decrease ( a kind of exponential decay ) of the system s _ free energy _ functional @xmath104 dx,\]]where @xmath105 is the potential such that @xmath106 the swift hohenberg ( sh ) equation , no - ted for its pattern forming behavior , is the decaying reaction diffusion pde , @xmath107 given by the variational ( gradient ) equation ( [ var ] ) with the free energy functional @xmath108 dxdy,\ ] ] where @xmath109 is given by ( [ ru ] ) , while @xmath110 is a 2dimensional region in which ( bio)chemical pattern formation occurs . the time derivative of the free energy @xmath111 is given by @xmath112 \partial _ { t}u\,dxdy,\ ] ] and , since the expression in square brackets is equal to the negative right hand side of ( [ sh ] ) , we have @xmath113 therefore , the free energy @xmath111 is the _ lyapunov functional _ that may only decrease as it evolves along its trajectory in some phase space . if @xmath111 has no minima , then when the horizontal scale of the liquid container is large compared to the instability wavelength , a propagating front will be observed ( e.g. , in chemically reacting flames ) . in this case , @xmath111 will decrease continuously until the front approaches the boundary of the medium . an alternative possibility is realized when @xmath111 has one or several minima , each corresponding to a local equilibrium state in time . in this case the so called multi stability is possible . therefore , the limit behavior of gradient systems of the form of ( [ ru ] ) is characterized by either a steady attractor or propagating fronts @xcite . one of the most popular models in the pattern formation theory is the complex ginzburg landau equation ( see e.g. , @xcite ) , @xmath114 where @xmath115 is the complex wave amplitude , @xmath116 is the super - criticality parameter , while @xmath117 and @xmath118 measure linear and nonlinear dispersion ( the dependence of the frequency of the waves on the wave - number ) , respectively . the equation ( [ cgl ] ) describes a vast array of phenomena including nonlinear waves , second - order phase transitions , rayleigh bnard convection and superconductivity . the equation describes the evolution of amplitudes of unstable modes for any process exhibiting a hopf bifurcation , for which a continuous spectrum of unstable wave numbers is taken into account . it can be viewed as a highly general normal form for a large class of bifurcations and nonlinear wave phenomena in spatially extended systems . ) , which describes strongly resonant multi frequency forcing of the form @xmath119 was recently proposed in @xcite by considering the analogous center manifold reduction of the extended dynamical system in which the forcing amplitudes @xmath120 , @xmath121 , and @xmath122 are considered as dynamical variables that vary on the slow time scale @xmath27 . under time translations @xmath123 , they transform as @xmath124 . to cubic order in @xmath115 the most general equation that is equivariant under @xmath125 is then given by @xmath126 where @xmath127 , @xmath128 , @xmath129 , @xmath130 . the forcing terms @xmath131 satisfy decoupled evolution equations on their own . in the simplest case this evolution expresses a de - tuning @xmath132 of the forcing @xmath131 from the respective resonance and the @xmath131 satisfy @xmath133 in general , the de - tuning introduces time dependence into ( [ eq : cgle_early ] ) . ] in particular , if we put @xmath134 in ( [ cgl ] ) , we get the real , or dissipative , ginzburg landau equation , @xmath135which is a gradient equation : @xmath136 with the free energy functional @xmath137 dxdy.\]]using the fact that @xmath138solutions of ( [ dgl ] ) at @xmath139 are either stationary field distributions satisfying , for @xmath140 the equation @xmath141of fronts whose propagation is accompanied by a decrease of the functional @xmath142 . the functional must reach its minimum at stable stationary solutions of ( [ da ] ) . the dynamical system from which the temporal evolution of neural activation fields is generated is constrained by the postulate that localized peaks of activation are stable objects , or formally , _ fixed point attractors_. such a field dynamics has the generic form @xcite @xmath143where @xmath144 is the activation field defined over the metric dimension @xmath145 and time @xmath146 the first three terms define an input driven regime , in which attractor solutions have the form @xmath147 the _ rate of relaxation _ is determined by the time scale parameter @xmath148 . the interaction stabilizes localized peaks of activation against decay by local excitatory interaction and against diffusion by global inhibitory interaction . in amari s formulation @xcite the conceptual model ( [ dft1 ] ) is specified as a _ continuous model for neural activity in cortical structures _ , @xmath149where @xmath150 is a constant resting level , @xmath151 is spatially and temporally variable input function , @xmath152 is an interaction kernel and @xmath153 is a sigmoidal nonlinear threshold function . the interaction term collects input from all those field sites @xmath154 at which activation is sufficiently large . the interaction kernel determines if inputs from those sites are positive , driving up activation ( excitatory ) , or negative , driving down activation ( inhibitory ) . excitatory input from nearby location and inhibitory input from all field locations generically stabilizes localized peaks of activation . for this class of dynamics , detailed analytical results provide a framework for the inverse dynamics task facing the modeler , determining a dynamical system that has the appropriate attractor solutions @xcite.[multiblock footnote omitted ] two component systems allow for a much larger range of possible phenomena than their one component counterparts . an important idea that was first proposed by a. turing is that a state that is stable in the local system should become unstable in the presence of diffusion @xcite . this idea seems counter intuitive at first glance as diffusion is commonly associated with a stabilizing effect . however , the linear stability analysis shows that when linearizing the general two component system@xmath155and perturbing the system against plane waves@xmath156close to a stationary homogeneous solution one finds @xcite @xmath157turing s idea can only be realized in four equivalence classes of systems characterized by the signs of the jacobian @xmath158 of the reaction function . in particular , if a finite wave vector k is supposed to be the most unstable one , the jacobian must have the signs@xmath159this class of systems is named activator@xmath160inhibitor system after its first representative : close to the ground state , one component stimulates the production of both components while the other one inhibits their growth . its most prominent representative is the fitzhugh nagumo equation ( [ fn ] ) . classical model of an autocatalytic chemical reaction is prigogine s brusselator ( see e.g. , @xcite ) @xmath161which describe the spatio temporal evolution of the intermediate components @xmath162 and @xmath163 with diffusion coefficients @xmath164 and @xmath165 , while reactions @xmath166describe the concentration of the original substances @xmath117 and @xmath167 for which the final products @xmath168 and @xmath169 are constant when all four reaction rates @xmath170 equal unity . a discretized ( temporal only ) version of the brusselator pde ( brus ) reads @xmath171 the brusselator displays oscillatory behavior in the species @xmath162 and @xmath172 when reverse reactions are neglected and the concentrations of @xmath173 and @xmath174 are kept constant . turbulence of scroll waves is a kind of spatio temporal chaos that exists in 3dimensional excitable media . cardiac tissue and the belousov zhabotinsky reaction are examples of such media . in cardiac tissue , chaotic behavior is believed to underlie fibrillation which , without intervention , precedes cardiac death . fast computer simulation of waves in excitable media have been often performed using the 2component barkley model of excitable media @xcite , @xmath175where @xmath176 is a small parameter @xmath177 characterising mutual time scales of the fast @xmath162 and slow @xmath172 variables , and @xmath178 and @xmath179 specify the kinetic properties of the system . parameter @xmath179 determines the excitation threshold and thus controls the excitability of the medium . the term @xmath180 represents an ` extra transmembrane current ' . suppression of the turbulence using stimulation of two different types , ` modulation of excitability ' and ` extra transmembrane current ' was performed in @xcite , using the barkley model ( [ bark ] ) . with cardiac defibrillation in mind , the authors used a single pulse as well as repetitive extra current with both constant and feedback controlled frequency . they show that turbulence can be terminated using either a resonant modulation of excitability or a resonant extra current . the turbulence is terminated with much higher probability using a resonant frequency perturbation than a non - resonant one . suppression of the turbulence using a resonant frequency is up to fifty times faster than using a non - resonant frequency , in both the modulation of excitability and the extra current modes . they also demonstrate that resonant perturbation requires strength one order of magnitude lower than that of a single pulse , which is currently used in clinical practice to terminate cardiac fibrillation . spontaneous pattern formation in initially almost homogeneous systems is common in both organic and inorganic systems . the gierer meinhardt model @xcite is a reaction diffusion system of the activator inhibitor type that appears to account for many important types of pattern formation and morphogenesis observed in biology , chemistry and physics . the model describes the concentration of a short range autocatalytic substance , the activator , that regulates the production of its long range antagonist , the inhibitor . it is given as a 2component nonlinear pde system , @xmath181where @xmath178 is a short range autocatalytic substance , i.e. , _ activator _ , and @xmath182 is its long range antagonist , i.e. , _ inhibitor_. @xmath183 and @xmath184 describe respectively the changes of activator and inhibitor concentrations per second , @xmath185 and @xmath186 are the corresponding decay rates , while @xmath187 and @xmath188 are the corresponding diffusion coefficients . @xmath189 is a positive constant . @xmath190 is a small activator independent production rate of the activator and is required to initiate the activator autocatalysis at very low activator concentration , e.g. , in the case of regeneration . a low baseline production of the inhibitor , @xmath191 , leads to a stable non patterned steady state ; the system can be asleep until an external trigger occurs by an elevation of the activator concentration above a threshold @xcite . an important example of bio reaction diffusion systems , frequently used in neurodynamics , is the 2component fitzhugh nagumo activator inhibitor system fitzhugh , nagumo ( see also @xcite ) @xmath192with @xmath193 , which describes how an action potential travels through a nerve , @xmath164 and @xmath165 are diffusion coefficients , @xmath194and @xmath195 are time characteristics , while @xmath196 and @xmath197 are positive constants . in matrix form , system ( [ fn ] ) reads @xmath198when an activator inhibitor system undergoes a change of parameters , one may pass from conditions under which a homogeneous ground state is stable to conditions under which it is linearly unstable . the corresponding bifurcation may be either a _ hopf bifurcation _ to a globally oscillating homogeneous state with a dominant wave number @xmath199 or a _ turing bifurcation _ to a globally patterned state with a dominant finite wave number . the latter in two spatial dimensions typically leads to stripe or hexagonal patterns . in particular , for the fitzhugh nagumo system ( [ fn ] ) , the neutral stability curves marking the boundary of the linearly stable region for the turing and hopf bifurcation are given by @xmath200if the bifurcation is subcritical , often localized structures ( i.e. , dissipative solitons ) can be observed in the hysteretic region where the pattern coexists with the ground state . other frequently encountered structures comprise pulse trains , spiral waves and target patterns . the reduced ( temporal ) non dimensional fitzhugh nagumo equations read : @xmath201where @xmath202 is essentially the threshold value , @xmath179 and @xmath203 are positive constants and @xmath204 is the applied current . the drift field for this model is given by @xmath205 as can be seen from ( [ fhn2 ] ) the null cline of the deterministic dynamics of this equations is the line @xmath206 . by substitution on the r.h.s of equation ( [ fhn1 ] ) we find the following equation for steady states : @xmath207 when this system is in a noisy environment , in the limit of weak noise , we can approximate the dynamics of the fluctuations by the _ langevin equation _ @xcite @xmath208 that is , the fluctuations run along the line @xmath206 . in particular , parameters in the _ fitzhugh nagumo neuron _ model tacanody , compmind @xmath209can be tuned so that the model describes spiking dynamics of many resonator neurons . since one needs to simulate the shape of each spike , the time step in the model must be relatively small , e.g. , @xmath210 . since the model is a 2dimensional system of odes , without a reset , it can not exhibit autonomous chaotic dynamics or bursting . adding noise to this , or some other 2dimensional models , allows for stochastic bursting . classical belousov zhabotinsky ( bz ) reaction is a family of oscillating chemical reactions . during these reactions , transition metal ions catalyze oxidation of various , usually organic , reductants by bromic acid in acidic water solution . most bz reactions are homogeneous . the bz reaction makes it possible to observe development of complex patterns in time and space by naked eye on a very convenient human time scale of dozens of seconds and space scale of several millimeters . the bz reaction can generate up to several thousand oscillatory cycles in a closed system , which permits studying chemical waves and patterns without constant replenishment of reactants @xcite . consider the water in oil micro emulsion bz reaction @xcite @xmath211 \frac{v - q_{0}}{v+q_{0}}+\frac{1}{% \varepsilon _ { 0}}\left [ \frac{1-mz}{1-mz+\varepsilon _ { 1}}\right ] v - v^{2 } , \notag \\ \partial _ { t}z & = & d_{z}\delta z - z+v\left [ \frac{1-mz}{1-mz+\varepsilon _ { 1}}% \right ] , \label{bz1}\end{aligned}\]]where @xmath212 are dimensionless concentrations of activator hbro@xmath213 and oxidized catalyst @xmath214^{3+}$ ] respectively ; @xmath165 and @xmath215 are dimensionless diffusion coefficients of activator and catalyst ; @xmath216 and @xmath217 are parameters of the standard tyson model @xcite ; @xmath218 represents the photoinduced production of inhibitor , and @xmath219 represents the strength of oxidized state of the catalyst with @xmath220 . this reaction was shown experimentally and numerically to admit localized spot patterns that persist for long time @xcite . we can rescale the variables in ( [ bz1 ] ) as @xcite @xmath221 in the new variables , after dropping the hats , we obtain the non dimensional 2component bz reaction @xmath222where @xmath223 \frac{v - q}{v+q}+\left [ \frac{w}{1+\alpha w% } \right ] v - v^{2},\qquad g(v , w)=1-\left [ \frac{w}{1+\alpha w}\right ] v,\]]with the non - dimensional constants given by @xmath224 the oregonator model is based on the so called fkn mechanism @xcite , which provided the first successful explanation of the chemical oscillations that occur in the experimental belousov zhabotinsky reaction . it is is composed of five coupled elementary chemical stoichiometries . during the last two decades , the oregonator model has been modified in many ways by inclusion of additional chemical reaction steps or by changing the rate constants . if we denote the concentration of the species s by [ s ] , then we define : @xmath225bro@xmath226 , @xmath227h@xmath228 , @xmath229hbro@xmath213 ] , @xmath231br@xmath232 , @xmath233ce@xmath234 . the original oregonator model was described by the following three coupled nonlinear pdes , @xmath235where @xmath236 is a stoichiometric factor @xcite , @xmath237 are rate constants , while @xmath238 , @xmath239 , and @xmath240 are the diffusion constants of the species hbro@xmath213 , br@xmath241 , and ce@xmath242 respectively ( for dilute solutions , the diffusion matrix is diagonal ) . for a thorough discussion of the chemistry on which the oregonator is based , the reader is referred to @xcite . the oregonator temporal mass action dynamics is a well stirred , homogeneous system of odes given by @xmath243which are typically scaled as @xcite@xmath244the basic chemistry of the bz oscillations involves jumps between high and low hbro2 ( @xmath245 ) states , which is reflected in the relaxation oscillator nature of the oregonator . this fundamental bistability may be stabilized in a flow reactor ( cstr ) with reactants and br@xmath241 in the feed stream . hysteresis between the two states is observed both experimentally and in the oregonator . quasiperiodicity and chaos also are observed in cstr and can be modeled by the oregonator @xcite . our last reaction diffusion system is a general model of multi phase tumor growth , in the form of nonlinear parabolic pde , as reviewed recently in @xcite @xmath246 ( @xmath247 ) , where for phase @xmath248 , @xmath249 is the volume fraction ( @xmath250 ) , @xmath251 is the random motility or diffusion , @xmath252 is the chemical and phase dependent production , and @xmath253 is the chemical and phase dependent degradation / death , and @xmath254 is the cell velocity defined by the constitutive equation @xmath255 where @xmath256 is a positive constant describing the viscous like properties of tumor cells and @xmath257 is the spheroid internal pressure . in particular , the multi phase equation ( [ multiph ] ) splits into two heat like mass conservation pdes @xcite , @xmath258 where @xmath259 and @xmath260 are the tissue cell / matrix and fluid volume fractions , respectively , @xmath261 and @xmath262 are the cell / matrix and the fluid velocities ( both defined by their constitutive equations of the form of ( [ constit ] ) ) , @xmath263 is the rate of production of solid phase tumor tissue and @xmath264 is the creation / degradation of the fluid phase . conservation of matter in the tissue , @xmath265 , implies that @xmath266(@xmath267 + @xmath268 ) = @xmath269 . the assumption that the tumor may be described by two phases only implies that the new cell / matrix phase is formed from the fluid phase and vice versa , so that @xmath270 . the detailed biochemistry of tumor growth can be coupled into the model above through the growth term @xmath263 , with equations added for nutrient diffusion , see @xcite and references therein . the multi phase tumor growth model ( [ multiph ] ) has been derived from the classical transport / mass conservation equations for different chemical species @xcite , @xmath271here @xmath272 are the concentrations of the chemical species , subindex @xmath178 for oxygen , @xmath179 for glucose , @xmath168 for lactate ion , @xmath169 for carbon dioxide , @xmath273 for bicarbonate ion , @xmath236 for chloride ion , and @xmath32 for hydrogen ion concentration ; @xmath274 is the net rate of consumption / production of the chemical species both by tumor cells and due to the chemical reactions with other species ; and @xmath275 is the flux of each of the chemical species inside the tumor spheroid , given ( in the simplest case of uncharged molecules of glucose , @xmath276 and @xmath277 ) by fick s law , @xmath278where @xmath251 are ( positive ) constant diffusion coefficients . in case of charged molecules of ionic species , the flux @xmath275 contains also the ( negative ) gradient of the volume fractions @xmath279 . there are three distinct stages to cancer development : avascular , vascular , and metastatic researchers often concentrate their efforts on answering specific oupc related questions on each of these stages @xcite . in particular , as some tumor cell lines grown in vitro form spherical aggregates , the relative cheapness and ease of in vitro experiments in comparison to animal experiments has made 3d _ multicellular tumor spheroids _ ( mts , see figure 6 in @xcite ) very popular in vitro model system of avascular tumors @xcite . they are used to study how local micro - environments affect cellular growth / decay , viability , and therapeutic response @xcite . mts provide , allowing strictly controlled nutritional and mechanical conditions , excellent experimental patterns to test the validity of the proposed mathematical models of tumor growth / decay @xcite . in this section we will derive the geometric formalism associated with the quadratic ricci flow equation ( [ rh ] ) , as a general framework for all presented bio reaction diffusion systems . geometry and topology of smooth surfaces are related by the _ gauss bonnet formula _ for a closed surface @xmath280 ( see , e.g. , ganeshsprbig , ganeshadg ) @xmath281 where @xmath282 is the area element of a metric @xmath32 on @xmath280 , @xmath283 is the gaussian curvature , @xmath284 is the euler characteristic of @xmath280 and @xmath285 is its _ genus _ , or number of handles , of @xmath286 . every closed surface @xmath280 admits a metric of constant gaussian curvature @xmath287 , or @xmath288 and so is uniformized by elliptic , euclidean , or hyperbolic geometry , which respectively have @xmath289 ( sphere ) , @xmath290 ( torus ) and @xmath291 ( torus with several holes ) . the integral ( [ gb ] ) is a _ topological invariant _ of the surface @xmath280 , always equal to 2 for all topological spheres @xmath42 ( that is , for all closed surfaces without holes that can be continuously deformed from the geometric sphere ) and always equal to 0 for the topological torus @xmath292 ( i.e. , for all closed surfaces with one hole or handle ) . topological framework for the ricci flow ( [ rf ] ) is thurston s _ geometrization conjecture _ @xcite , which states that the interior of any compact 3manifold can be split in an essentially unique way by disjoint embedded 2spheres @xmath24 and tori @xmath293 into pieces and each piece admits one of 8 geometric structures ( including ( i ) the 3sphere @xmath294 with constant curvature @xmath295 ; ( ii ) the euclidean 3space @xmath296 with constant curvature 0 and ( iii ) the hyperbolic 3space @xmath297 with constant curvature @xmath288).[multiblock footnote omitted ] the geometrization conjecture ( which has the poincar conjecture as a special case ) would give us a link between the geometry and topology of 3manifolds , analogous in spirit to the case of 2surfaces . in higher dimensions , the gaussian curvature @xmath283 corresponds to the riemann curvature tensor @xmath298 on a smooth @xmath9manifold @xmath5 , which is in local coordinates on @xmath5 denoted by its @xmath299components @xmath300 , or its @xmath301components @xmath302 ( see appendix , as well as e.g. , ganeshsprbig , ganeshadg ) . the trace ( or , contraction ) of @xmath298 , in @xmath299case using the inverse metric tensor @xmath303 , is the ricci tensor @xmath304 , the contracted curvature tensor , which is in a local coordinate system @xmath305 defined in an open set @xmath306 , given by @xmath307(using einstein s summation convention ) , while the _ scalar curvature _ is now given by the second contraction of @xmath298 as @xmath308 in general , the ricci flow @xmath309 is a one parameter family of riemannian metrics on a compact @xmath9manifold @xmath5 governed by the equation ( [ rf ] ) , which has a unique solution for a short time for an arbitrary smooth metric @xmath310 on @xmath5 @xcite . if @xmath311 at any local point @xmath312 on @xmath5 , then the ricci flow ( [ rf ] ) contracts the metric @xmath313 near @xmath314 , to the future , while if @xmath315 , then the flow ( [ rf ] ) expands @xmath309 near @xmath314 . the solution metric @xmath309 of the ricci flow equation ( [ rf ] ) shrinks in positive ricci curvature direction while it expands in the negative ricci curvature direction , because of the minus sign in the front of the ricci tensor @xmath316 . in particular , on a 2sphere @xmath24 , any metric of positive gaussian curvature will shrink to a point in finite time . at a general point , there will be directions of positive and negative ricci curvature along which the metric will locally contract or expand ( see @xcite ) . also , if a simply connected compact 3manifold @xmath5 has a riemannian metric @xmath13 with positive ricci curvature then it is diffeomorphic to the 3sphere @xmath317 @xcite . more generally speaking , the ricci flow deforms manifolds with positive ricci curvature to a point which can be renormalized to the 3sphere . all three riemannian curvatures ( @xmath318 and @xmath298 ) , as well as the associated volume forms , _ evolve _ during the ricci flow ( [ rf ] ) . in general , the ricci flow evolution equation ( [ rf ] ) for the metric tensor @xmath13 implies the reaction diffusion type evolution equation for the riemann curvature tensor @xmath298 on an @xmath9manifold @xmath5 , @xmath319where @xmath320 is a quadratic expression of the riemann @xmath9curvatures , corresponding to the @xmath9component bio chemical reaction , while the term @xmath321 corresponds to the @xmath9component diffusion . from the general @xmath9curvature evolution ( [ rmflow ] ) we have two important particular cases:[multiblock footnote omitted ] \1 . the evolution equation for the ricci curvature tensor @xmath322 on a 3manifold @xmath5 , @xmath323where @xmath324 is a quadratic expression of the ricci 3curvatures , corresponding to the 3component bio chemical reaction , while the term @xmath325 corresponds to the 3diffusion . the evolution equation for the scalar curvature @xmath326 , @xmath327 in which the term @xmath328 corresponds to the 2component bio chemical reaction , while the term @xmath329 corresponds to the 2component diffusion . by the _ maximum principle _ ( see subsection [ rentr ] ) , the minimum of the scalar curvature @xmath326 is non decreasing along the flow @xmath28 , both on @xmath5 and on its boundary @xmath330 ( see @xcite ) . let us now see in detail how various geometric quantities evolve given the _ short time solution _ of the ricci flow equation ( [ rf ] ) on an arbitrary @xmath9manifold @xmath5 . for this , we need first to calculate the _ variation formulas _ for the christoffel symbols and curvature tensors on @xmath5 ; then the corresponding evolution equations will naturally follow ( see ham82,caochow , chowknopf ) . if @xmath331 is a 1parameter family of metrics on @xmath5 with @xmath332 then the variation of the christoffel symbols @xmath333 on @xmath5 is given by @xmath334(where @xmath335 is the covariant derivative with respect to the riemannian connection ) from which follows the evolution of the christoffel symbols @xmath336 under the ricci flow @xmath28 on @xmath5 given by ( [ rf]),@xmath337 from ( [ vc ] ) we calculate the variation of the ricci tensor @xmath14 on @xmath338 as @xmath339and the variation of the scalar curvature @xmath326 on @xmath5 by @xmath340where @xmath341 is the trace of @xmath342 . if an @xmath9manifold @xmath5 is oriented , then the _ volume _ @xmath9form on @xmath5 is given , in a positively oriented local coordinate system @xmath343 , by @xcite @xmath344if @xmath345 then @xmath346 the evolution of the volume @xmath9form @xmath347 under the ricci flow @xmath28 on @xmath348 is given by the exponential decay / growth relation with the scalar curvature @xmath326 as the ( variable ) rate parameter , @xmath349 which gives an exponential decay for @xmath350 ( elliptic geometry ) and exponential growth for @xmath351 ( hyperbolic geometry ) for any small constant @xmath178 ( scalar curvature must be bounded away from zero ) . the elementary volume evolution ( [ dtmu ] ) implies the integral form of the exponential relation for the total @xmath9volume @xmath352 which again gives an _ exponential decay _ for elliptic @xmath353 and _ exponential growth _ for hyperbolic @xmath354 . this is a crucial point for the _ tumor suppression _ by the body : the immune system needs to keep the elliptic geometry of the mts , evolving by ( [ multiph ] ) by all possible means . in the healthy organism this normally happens , because the initial mts started as a spherical shape with @xmath353 . the immune system just needs to keep the mts in the spherical / elliptic shape and prevent any hyperbolic distortions of @xmath354 . thus , it will naturally have an exponential decay and vanish . since the @xmath9volume is not constant and sometimes we would like to prevent the solution from shrinking to an @xmath9point on @xmath5 ( elliptic case ) or expanding to an infinity ( hyperbolic case ) , we can also consider the _ normalized ricci flow _ on @xmath5 ( see figure [ ricci ] as well as ref . caochow ) : @xmath355is the average scalar curvature on @xmath5 . we then have the @xmath9__volume conservation law : _ _ @xmath356 to study the _ long time existence _ of the normalized ricci flow ( nrf ) on an arbitrary @xmath9manifold @xmath5 , it is important to know what kind of curvature conditions are preserved under the equation . in general , the ricci flow @xmath357 on @xmath5 , as defined by the fundamental relation ( [ rf ] ) , tends to preserve some kind of positivity of curvatures . for example , positive scalar curvature @xmath326 ( i.e. , elliptic geometry ) is preserved both on @xmath5 and on its boundary @xmath330 in any dimension . this follows from applying the maximum principle to the evolution equation ( [ dtr ] ) for scalar curvature @xmath326 both on @xmath5 and on @xmath330 . similarly , positive ricci curvature is preserved under the ricci flow on a 3manifold @xmath5 . this is a special feature of dimension 3 and is related to the fact that the riemann curvature tensor may be recovered algebraically from the ricci tensor and the metric on 3manifolds @xcite . in particular , we have the following result for 2surfaces ( see @xcite ) : let @xmath358 be a closed 2surface , which is a boundary of a compact 3manifold @xmath5 . then for any initial 2metric @xmath34 on @xmath330 , the solution to the normalized ricci flow ( [ nrf ] ) on @xmath330 exists for all time . in other words , the normalized ricci flow in 2d always converges . moreover , ( i ) if the euler characteristic of @xmath330 is non positive , then the solution metric @xmath28 on @xmath330 converges to a constant curvature metric as @xmath359 ; and ( ii ) if the scalar curvature @xmath326 of the initial metric @xmath34 is positive , then the solution metric @xmath360 on @xmath330 converges to a positive constant curvature metric as @xmath361 ( for surfaces with non positive euler characteristic , the proof is based primarily on maximum principle estimates for the scalar curvature . ) applying to the tumor evolution ( [ multiph ] ) , the normalized ricci flow ( [ nrf ] ) of the mts will make it completely round with a geometric sphere shell , which is ideal for surgical removal . this is our second option for the mts growth / decay control . if we can not force it to exponential decay , then we must try to normalize into a round spherical shell which is suitable for surgical removal . the negative flow of the total @xmath9volume @xmath362 represents the _ einstein hilbert functional _ ( see @xcite ) @xmath363if we put @xmath345 we have@xmath364 so , the critical points of @xmath365 satisfy _ einstein s equation _ @xmath366 in the vacuum . the gradient flow of @xmath365 on an @xmath9manifold @xmath5 , @xmath367is the ricci flow ( [ rf ] ) plus @xmath368 . thus , einstein metrics are the fixed points of the normalized ricci flow.manifolds are metrics with constant ricci curvature . however , along the way , the deformation will encounter singularities . the major question , resolved by perelman , was how to find a way to describe all possible singularities . ] let @xmath75 denote the laplacian acting on functions on a closed @xmath9manifold @xmath5 , which is in local coordinates @xmath343 given by@xmath369for any smooth function @xmath236 on @xmath5 we have @xcite@xmath370 from which it follows that if we have @xmath371 using this laplacian @xmath75 , we can write the linear heat equation on @xmath5 as @xmath372 where @xmath162 is the temperature . in particular , the laplacian acting on functions with respect to @xmath28 will be denoted by @xmath373 . if @xmath374 is a solution to the ricci flow equation ( rf ) , then we have@xmath375 the evolution equation ( [ dtr ] ) for the scalar curvature @xmath326 under the ricci flow ( [ rf ] ) follows from ( [ vr ] ) . using equation ( [ cb ] ) from appendix , we have : @xmath376 showing again that the scalar curvature @xmath326 satisfies a heat type equation with a quadratic nonlinearity both on a 3manifold @xmath5 and on its boundary 2surface @xmath330 . next we will find the exact form of the evolution equation ( [ rcflow ] ) for the ricci tensor @xmath322 under the ricci flow @xmath28 given by ( [ rf ] ) on any @xmath50manifold @xmath5 . ( note that in higher dimensions , the appropriate formula of huge complexity would involve the whole riemann curvature tensor @xmath298 . ) given a variation @xmath377 , from ( [ vric ] ) we get @xmath378where @xmath379 denotes the so called lichnerowicz laplacian ( which depends on @xmath298 , see @xcite ) . since@xmath380by ( [ cb ] ) ( after some algebra ) we get that under the ricci flow ( [ rf ] ) the evolution equation for the ricci tensor @xmath322 on a 3manifold @xmath5 is @xmath381so , just as in case of the evolution ( [ dtr ] ) of the scalar curvature @xmath382 ( both on a 3manifold @xmath5 and on its 2boundary @xmath383 ) , we get a heat type evolution equation with a quadratic nonlinearity for @xmath384 , which means that positive ricci curvature ( @xmath385 ) of elliptic 3geometry is preserved under the ricci flow @xmath386 on @xmath5 . more generally , we have the following result for 3manifolds ( see ham82 ) : let @xmath387 be a compact riemannian @xmath50manifold with positive ricci curvature @xmath322 . then there exists a unique solution to the normalized ricci flow @xmath28 on @xmath5 with @xmath388 for all time and the metrics @xmath28 converge exponentially fast to a constant positive sectional curvature metric @xmath389 on @xmath5 . in particular , @xmath5 is diffeomorphic to a 3sphere @xmath317 . ( as a consequence , such a @xmath50manifold @xmath5 is necessarily diffeomorphic to a quotient of the @xmath50sphere by a finite group of isometries . it follows that given any homotopy @xmath50sphere , if one can show that it admits a metric with positive ricci curvature , then the poincar conjecture would follow @xcite . ) in addition , compact and closed @xmath50manifolds which admit a non - singular solution can also be decomposed into geometric pieces @xcite . from the geometric evolution equations reviewed in this subsection , we see that both short time and long time geometric solution can always be found for 2compo - nent bio reaction diffusion equations , as they correspond to evolution of the scalar 2curvature @xmath326 . regarding the 3component bio reaction diffusion equations , corresponding to evolution of the ricci 3curvature @xmath390 we can always find the short time geometric solution , while the long time solution exists only under some additional ( compactnes and/or closure ) conditions . finally , in case of @xmath9component bio reaction diffusion equations , corresponding to evolution of the riemann @xmath9curvature @xmath391 only short time geometric solution is possible . an important class of bio reaction diffusion systems are _ dissipative solitons _ ( dss ) , which are stable solitary localized structures that arise in nonlinear spatially extended dissipative systems due to mechanisms of _ self organization_. they can be considered as an extension of the classical soliton concept in conservative systems . apart from aspects similar to the behavior of classical particles like the formation of bound states , dss exhibit entirely nonclassical behavior e.g. , scattering , generation and annihilation all without the constraints of energy or momentum conservation . the excitation of internal degrees of freedom may result in a dynamically stabilized intrinsic speed , or periodic oscillations of the shape . in particular , stationary dss are generated by production of material in the center of the dss , diffusive transport into the tails and depletion of material in the tails . a propagating pulse arises from production in the leading and depletion in the trailing end . among other effects , one finds periodic oscillations of dss , the so called ` breathing ' dissipative solitons @xcite . dss in many different systems show universal particle like properties . to understand and describe the latter , one may try to derive ` particle equations ' for slowly varying order parameters like position , velocity or amplitude of the dss by adiabatically eliminating all fast variables in the field description . this technique is known from linear systems , however mathematical problems arise from the nonlinear models due to a coupling of fast and slow modes @xcite . similar to low dimensional dynamic systems , for supercritical bifurcations of stationary dss one finds characteristic normal forms essentially depending on the symmetries of the system ; e.g. , for a transition from a symmetric stationary to an intrinsically propagating ds one finds the pitchfork normal form for the ds velocity @xmath392 @xcite , @xmath393where @xmath394 represents the bifurcation parameter and @xmath395 the bifurcation point . for a bifurcation to a ` breathing ' ds , one finds the hopf normal form @xcite @xmath396for the amplitude @xmath115 of the oscillation . note that the above problems do not arise for classical solitons as inverse scattering theory yields complete analytical solutions @xcite . closely related to dissipative solitons are the so called _ breathers _ , solitonic structures given by localized periodic solutions of some nonlinear soliton pdes , including the exactly solvable sine gordon equation of the ( 1 + 1)d sine gordon equation @xmath397@xmath398which , for @xmath399 , is periodic in time @xmath27 and decays exponentially when moving away from @xmath400 . ] and the focusing nonlinear schrdinger equation.]with a breather solution of the form : @xmath401which gives breathers periodic in space @xmath314 and approaching the uniform value @xmath178 when moving away from the focus time @xmath402 . ] a metric @xmath309 evolving by the ricci flow @xmath28 given by ( [ rf ] ) on any 3manifold @xmath5 is called a _ ricci breather _ , if for some @xmath403 and @xmath404 the metrics @xmath405 and @xmath406 differ only by a diffeomorphism ; the cases @xmath407 correspond to steady , shrinking and expanding breathers , respectively . trivial breathers on @xmath5 , for which the metrics @xmath408 and @xmath409 differ only by diffeomorphism and scaling for each pair of @xmath410 and @xmath411 , are called _ ricci solitons_. thus , if one considers ricci flow as a dynamical system on the space of riemannian metrics modulo diffeomorphism and scaling , then breathers and solitons correspond to periodic orbits and fixed points respectively . at each time the ricci soliton metric satisfies on @xmath5 an equation of the form perel1 @xmath412where @xmath168 is a number and @xmath413 is a 1form ; in particular , when @xmath414 for some function @xmath178 on @xmath415 we get a gradient ricci soliton . an important example of a gradient shrinking soliton is the _ gaussian soliton _ , for which the metric @xmath13 is just the euclidean metric on @xmath296 , @xmath416 and @xmath417 . given a @xmath418 function @xmath419 on a riemannian @xmath50manifold @xmath5 , its laplacian is defined in local coordinates @xmath420 to be @xmath421where @xmath422 is the _ covariant derivative _ ( levi civita connection , see appendix ) . we say that a @xmath418 function @xmath423 where @xmath424,$ ] is a solution to the heat equation if ( [ h1 ] ) holds . one of the most important properties satisfied by the heat equation is the _ maximum principle _ , which says that for any smooth solution to the heat equation , whatever point - wise bounds hold at @xmath402 also hold for @xmath425 @xcite . more precisely , we can state : let @xmath426 be a @xmath418 solution to the heat equation ( [ h1 ] ) on a complete riemannian @xmath50manifold @xmath5 . if @xmath427 for all @xmath428 for some constants @xmath429 then @xmath430 for all @xmath431 and @xmath432 this property exhibits the averaging behavior of the heat equation ( [ h1 ] ) on @xmath5 . now , consider perelman s _ entropy functional _ @xcite on a 3manifold @xmath433 @xmath434for a riemannian metric @xmath13 and a ( temperature - like ) scalar function @xmath236 ( which satisfies the backward heat equation ) on a closed 3manifold @xmath5 , where @xmath347 is the volume 3form ( [ dmu ] ) . during the ricci flow ( [ rf ] ) , @xmath435 evolves on @xmath5 as @xmath436now , define @xmath437 where infimum is taken over all smooth @xmath438 satisfying @xmath439@xmath440 is the lowest eigenvalue of the operator @xmath441 then the entropy evolution formula ( [ df ] ) implies that @xmath442 is nondecreasing in @xmath443 and moreover , if @xmath444 then for @xmath445 $ ] we have @xmath446 for @xmath236 which minimizes @xmath435 on @xmath338 @xcite . thus a steady breather on @xmath5 is necessarily a steady soliton . if we define the conjugate _ heat operator _ on @xmath5 as @xmath447then we have the _ _ conjugate heat equation__[multiblock footnote omitted ] @xcite @xmath448 the entropy functional ( [ f ] ) is nondecreasing under the coupled _ ricci diffusion flow _ on @xmath5 ( see @xcite ) @xmath449where the second equation ensures @xmath450 to be preserved by the ricci flow @xmath28 on @xmath5 . if we define @xmath451 , then the right hand equation in ( [ conheat ] ) is equivalent to the generic scalar field @xmath452evolution equation on @xmath5 , @xmath453which instead preserves ( [ edm ] ) . the coupled ricci diffusion flow ( [ conheat ] ) , or equivalently , the dual system @xmath454 is our _ global decay _ model for a general @xmath9dimensional bio reaction diffusion process , including both geometric and bio chemical multi phase evolution . the sole _ hypothesis of this paper _ is that any kind of reaction diffusion processes in biology , chemistry and physics is subsumed by the geometric diffusion system ( [ conheat ] ) , or the dual system ( [ dual ] ) . recall that proper differentiation of vector and tensor fields on a smooth riemannian @xmath9manifold is performed using the _ levi civita covariant derivative _ ( see , e.g. , @xcite ) . formally , let @xmath5 be a riemannian @xmath9manifold with the tangent bundle @xmath455 and a local coordinate system @xmath305 defined in an open set @xmath456 . the covariant derivative operator , @xmath457 , is the unique linear map such that for any vector fields @xmath458 constant @xmath168 , and function @xmath236 the following properties are valid : @xmath459,\qquad \text{(torsion free property)}\end{aligned}\]]where @xmath460 $ ] is the lie bracket of @xmath245 and @xmath461 ( see , e.g. , @xcite ) . in local coordinates , the metric @xmath32 is defined for any orthonormal basis @xmath462 in @xmath306 by @xmath463 then the affine _ levi civita connection _ is defined on @xmath5 by @xmath464are the ( second - order ) _ christoffel symbols_. now , using the covariant derivative operator @xmath465 we can define the _ riemann curvature _ @xmath301tensor @xmath298 by ( see , e.g. , @xcite ) @xmath466}z.\]]@xmath298 measures the curvature of the manifold by expressing how noncommutative covariant differentiation is . the @xmath301components @xmath467 of @xmath298 are defined in @xmath306 by @xmath468also , the riemann @xmath299tensor @xmath469 is defined as the @xmath470based inner product on @xmath5 , @xmath471 the @xmath475 _ ricci tensor _ @xmath322 is the trace of the riemann @xmath301tensor @xmath298 , @xmath476 its components @xmath477 are given in @xmath306 by the contraction @xcite @xmath478 being a symmetric second order tensor , @xmath322 has @xmath479 independent components on an @xmath9manifold @xmath5 . in particular , on a 3manifold , it has 6 components , and on a 2surface it has only the following 3 components : @xmath480 which are all proportional to the corresponding coordinates of the metric tensor , @xmath481 j.j . tyson , a quantitative account of oscillations , bistability and travelling waves in the belousov - zhabotinskii reaction , in oscillation and travelling waves in chemical systems , eds . field , r. j. , burger , m. , wiley - intersc . , new york , ( 1985 ) .

quian quiroga et al . [ nature * 435 * , 1102 ( 2005 ) ] have recently discovered neurons that appear to have the characteristics of grandmother ( gm ) cells . here we quantitatively assess the compatibility of their data with the gm - cell hypothesis . we show that , contrary to the general impression , a gm - cell representation can be information - theoretically efficient , but that it must be accompanied by cells giving a distributed coding of the input . we present a general method to deduce the sparsity distribution of the whole neuronal population from a sample , and use it to show there are two populations of cells : a distributed - code population of less than about 5% of the cells , and a much more sparsely responding population of putative gm cells . with an allowance for the number of undetected silent cells , we find that the putative gm cells can code for or more categories , sufficient for them to be classic gm cells , or to be gm - like cells coding for memories . we quantify the strong biases against detection of gm cells , and show consistency of our results with previous measurements that find only distributed coding . we discuss the consequences for the architecture of neural systems and synaptic connectivity , and for the statistics of neural firing . and neural representations ; grandmother cells ; distributed coding ;

q-bio0603014

critical to understanding how information is processed in the brain is the form of the neural coding that underlies the storage and recall of memories . is there a local , or gnostic @xcite , code colloquially called a grandmother - cell ( gm - cell ) representation in which the firing of a single neuron ( or group of neurons ) exclusively codes recognition of a particular object , person or memory ? or is the code much more distributed ? although it is generally accepted [ e.g. , @xcite ] that gm representations are not used in reality , experiments @xcite often find localist responses by individual neurons . most dramatically , @xcite have recently found many neurons in humans that , within the limits of the measurements , behave like classic gm cells . in this paper , we therefore quantitatively re - examine the viability of gm - cell representations , with the outcome that we refute the standard quantitative arguments against them , both theoretical and phenomenological . the information - theoretic argument is that gm representations need far too many neurons for the information coded @xcite . we show that this argument fails when one examines the information storage capacity of the synapses rather than the representational capacity of neurons for input stimuli . the standard efficiency argument applies only to the input representation , needed to represent any of the myriad possible stimuli . for storage , a gm representation can be optimally efficient . the phenomenological argument is that gm cells should fire in response to a much smaller fraction of stimuli than has been deduced from measurements of neural responses @xcite . a gm cell can be regarded as a categorizer , and the data appear to imply that any apparent gm cell responds to many categories of stimuli rather than to one category . however , our information theoretic argument shows that associated with any gm - cell population , with its ultra - low sparsity , is a more conventional population with a much higher sparsity . this two - population property , always a part of the gm - cell idea @xcite , was not allowed for in older analyses , including that @xcite by the group responsible for the new data @xcite . we devise a very general method of analyzing neural systems with multiple sparsities , and apply it to the data of @xcite . it enables us to quantify the biases against experimental detection of gm - like cells , most of which simply appear as unreported silent cells , and whose estimated numbers @xcite may be a factor of 30 more than the reported cells . we find that the two - population property holds , and that less than @xmath1 of _ detected _ cells are in the distributed - code population : the vast majority of the cells can be gm - like . then we find that the number of categories coded by the gm - like cells can be @xmath0 or more . uncertainties are minor relative to the orders of magnitude involved . the data of @xcite therefore appear in strong quantitative agreement with the gm - cell hypothesis . the biases against detecting gm cells are enough to allow consistency with previous @xcite measurements and analyses that use a single - population model and that quantitatively argued against gm cells . we will examine other arguments against gm representations in the discussion section . inappropriate or excessively rigid definitions can exclude biologically interesting cases . for example , @xcite define a local representation as one where `` all the information that a particular stimulus or event occurred is provided by the activity of one of the neurons '' . the word `` all '' appears to exclude a system where a local representation codes the result of testing a distributed representation of a stimulus against remembered items . not all the available information is in the firing of the neurons in the local output representation . therefore in this section we present our definitions , and explain important features and consequences of the definitions needed for later sections . we define a set of cells to form a local or gm - like system when nodes of the system can be divided into groups of one or more nodes , and , to a good approximation , each group corresponds to one particular meaningful and distinct property of the stimulus input to the system . each group we call a gm group . typically we treat the properties corresponding to different gm groups as being mutually exclusive . we will usually identify the nodes with actual neurons , so that measurement in one of a gm group s cells of firing above a suitable threshold is strong evidence that the stimulus is associated with the corresponding property . but it is also possible that the nodes could be , for example , part of a dendritic tree . then the correspondence between neural firing and local coding might not be direct . in any case we can treat the system as a categorization system . in contrast , a distributed representation is formed by a set of cells where the categorization can only be determined from the activity of multiple cells / nodes and where the patterns of activity overlap between distinct properties , even when the properties themselves are mutually exclusive . [ [ classic - gm - cells - and - generalizations ] ] classic gm cells and generalizations + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the classic example of gm system is a facial identification system , where the firing of a particular gm group mediates the `` unitary perception '' @xcite of a retinal image as corresponding to a particular individual person . a characteristic property of a _ classic _ gm system is therefore that firing is exclusive between different gm groups i.e. , only one gm group is active at a time . this corresponds to the fact that the individuals associated with the different stimuli are themselves completely distinct . of course , there will be situations where the gm group firing is not completely exclusive ; for example , if a particular stimulus is ambiguous , or if the picture of a face of one person is artificially morphed into that of another person . our definition , however , was worded to allow certain natural generalizations from the case of classic gm cells . in particular , we will apply the terminology to declarative memories in general ( episodes , facts , etc ) . thus we could have a gm cell or gm group corresponding to each episodic memory . in this case it is evidently of practical importance for a memory to be recalled from a stimulus containing a few components of the original memory . since the same components could be part of other memories , the pattern of recognition firing need not be exclusive between memories . let us regard these patterns as priming the recall of the memories . full conscious recall of one particular memory requires some extra cues and modulation . with a gm - like memory system , non - exclusive priming recall would be at a relatively low level of firing above some threshold , with full recall involving exclusive firing at a much higher level . in this case the exclusivity is not between the actual firing of different gm groups , but between the concepts corresponding to the groups . of course , the situation is a little more complicated for episodic memory , since episodes are happenings along a continuum in time . a memory cell for an episode corresponds to a small range in time . the exclusivity between different gm groups is between well - separated episodes , of which there are evidently a very large number . another natural generalization of the gm - cell concept is to local coding for output , with strong experimental evidence in the work of @xcite . [ [ high-level-gm-systems-v.low-level-local-coding ] ] high - level gm systems v. low - level local coding + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we choose to restrict our use of the `` gm - cell '' terminology to the higher levels of neural processing . for lower levels , we will use the broader term `` local coding '' ; by this we mean coding by a set of cells each of which is responsive to some patch of stimulus space ( with fall off at the edges naturally ) . when the relevant properties of the stimulus are in a low - dimensional space as for color or position in an environment the collection of patches can give complete coverage . but when the relevant stimulus space is high dimensional , one can only expect local coverage of a minute fraction of possible stimuli for example , to correspond to linguistic phonemes out of all possible auditory stimuli . there is considerable evidence that can be interpreted as supporting some kind of local coding below the highest perceptual levels . the key question for us is whether local coding is also used at the highest levels . [ [ internal - stimuli ] ] internal stimuli + + + + + + + + + + + + + + + + input to a memory system need not correspond to actual physical events external to the brain . some input can be completely generated internally , as with an author planning a novel . memories of people and episodes in the novel have the same neural status as memories of real people and events . one big computational advantage of a gm - like system is the simplicity of information flow , fig . [ fig : gm - struct ] , so that the representation is easy to construct and manipulate e.g. , @xcite . new memories are formed with new or unallocated neurons , so that they interfere minimally with old memories @xcite . the input to gm cells is from a distributed representation of the stimuli . output to downstream neurons using the categorization can be simply taken from a single gm cell . above - threshold firing of a gm - cell models a property of the cause of the stimulus . thus when firing is exclusive between different gm groups , this corresponds to exclusivity between the modeled properties of the associated external stimuli , like the identity of a person . it is therefore tempting to use exclusivity of firing between different gm groups as the primary measurable criterion for characterizing gm - like systems . but natural generalizations to declarative memory motivate us to relax this criterion . for example , with episodic memories , a stimulus may cause a response in nodes for those episodes having important commonalities with the stimulus . in addition , there can be multiple categorization systems , and there is no requirement of exclusivity between different systems . this is particularly clear at low levels in the processing hierarchy , where we could have separate local representations , for example , of the color and shape of an object . a collection of local representations of such relatively low - level features then forms a distributed representation of the whole object , suitable for input to the next level of hierarchical processing . note that we prefer to use the terminology `` gm - like system '' rather than `` gm representation '' to emphasize two aspects : the first is that if the gm - cell idea applies in its classic sense of recognition of individual people , it is likely to apply much more generally to all declarative memories . the second is that we wish to treat the firing of a gm cell as coding the result of a _ recognition computation _ from a stimulus . but the use of the word `` representation '' for a gm - like system would carry the connotation of representing the stimulus , which is not generally appropriate . if nothing else , local coding typically dramatically fails to cover the stimulus space . for example , consider a distributed input representation on 100 binary neurons , a small number compared with real sensory systems . there are @xmath2 distinct stimuli , many orders of magnitude larger than the total number of neurons in any brain . a practical local representation can only apply to a minute fraction of stimuli , presumably ones that are especially salient . a local representation can only provide full coverage for a stimulus space of very low dimension , like that for color . practical experiments only involve a limited number of stimuli and cells , and definitely do not give the detailed synaptic information that determines all possible causes of a cell s firing . thus it is non - trivial to distinguish a gm system from a distributed representation , when the distributed representation is sparse , and when we allow natural generalizations of the gm - cell idea . we illustrate the issues by comparing two very different computational memory models , one by @xcite , and one by @xcite ( bmw ) . [ [ hopfields - model ] ] hopfield s model + + + + + + + + + + + + + + + + in hopfield s particular example , the input stimulus concerns properties of people , and the input representation is carried by a set of 1000 binary neurons . these are divided into 50 sets of 20 neurons . the 20 neurons in each set give a local representation of 20 possible values of a property of the stimulus . for example , to input the name of a person , one of a set of 20 name - coding neurons would be active . binary synapses connect every neuron to every other neuron . each stored memory is considered as the set of 50 property values for a particular person , and is coded in the state of the synapses . the synaptic strengths are set by a hebbian - like rule on presentation of stimuli . recall of a memory is caused by a stimulus that consists of partial data about an individual , i.e. , values for a subset of the 50 properties . memory retrieval results in completion of a partial stimulus to the full set of properties for the corresponding individual . there is no corresponding gm cell ; the model is a fully distributed memory system . [ [ bmw - model ] ] bmw model + + + + + + + + + the basic bmw model is the simplest possible form of a gm system : a feed - forward perceptron with one intermediate memory - cell layer , fig.[fig : bmw ] . the memory cells are arranged to respond in a gm - cell style to recognize inputs that correspond to stored memories . the model gains its power by applying it to the situation that the input forms a sparse binary representation of stimuli . this would typically be obtained from the top of a processing hierarchy , as in fig.[fig : gm - struct ] . improvements in the model can be made , for example , by adding inhibitory interneurons to enforce a winner - take - all action in the memory - cell layer , but the simplest form of the model is sufficiently robust for our illustration . to solve the same pattern - completion task as hopfield s model , we make the output cells identical to the input cells , thereby specializing from a heteroassociative memory system to a homoassociative system . memories are created by presenting a full stimulus to the system , and arranging for some unallocated memory neuron @xmath3 to match its synaptic strengths to the stimulus . the chosen neuron changes to a state of being allocated , with correspondingly greatly reduced synaptic plasticity . let @xmath4 denote the prototype pattern for memory neuron @xmath3 . then the strengths of the neuron s input synapses are set to a constant times the input pattern : @xmath5 , and similarly for the output . the constant @xmath6 can be scaled out of all of our formulae , but its use assists in relating the formulae to properties of real neurons . when another stimulus @xmath7 is presented , the response of neuron @xmath3 is obtained from a thresholded sum over its inputs : @xmath8 this formulation is for artificial analog neurons , but it is readily generalized to realistic spiking neurons , and the model is robust enough that minor variations do not affect the principles governing its performance . we now review how standard properties @xcite of sparse binary codes show that the system implements the pattern completion task if the threshold is set suitably . let a stimulus @xmath7 be presented ; it may be some full pattern , in which @xmath1 of the input neurons are active , or it may be a partial pattern . the input to neuron @xmath3 is @xmath9 , which is @xmath6 times the overlap between @xmath7 and the prototype pattern for memory @xmath3 , i.e. , it measures the number of common on - bits . if the new stimulus is unrelated to memory @xmath3 , then its on - bits are random relative to those for memory @xmath3 . thus for a full pattern the typical overlap is @xmath1 of the total number of on - bits in a full patten , i.e. , @xmath10 , and less for a partial pattern . we therefore set the threshold intermediate between the input for the full prototype pattern , i.e. , @xmath11 , and the typical input for a full unrelated pattern , i.e. , @xmath12 . then unrelated patterns almost never cause neuron @xmath3 to fire . thus with sparse input patterns we have the well - known automatic orthogonalization against unrelated stimuli . but now consider an input corresponding to a subset of the features in memory @xmath3 s prototype pattern . for example , a quarter of the features could have been detected . in that case , the input to neuron @xmath3 would be @xmath13 . hence with an appropriate threshold setting , we get firing of neuron @xmath3 in response to a subset of the features in its prototype pattern . this then causes firing of the whole set of output neurons that correspond to the prototype , i.e. , we have pattern completion . because the input to a memory cell is much bigger for patterns related to its prototype than for unrelated patterns , the operation of the system is robust against changes in the exact equations describing its operation . [ [ measurements ] ] measurements + + + + + + + + + + + + to mimic a biological experiment , one could measure the responses of a sample of model neurons to a sample of stimuli . in hopfield s model if there are fewer than about 10 or 20 stimuli , the firing of many of the individual neurons would correspond to single people . thus the distributed nature of the memory representation would not be immediately apparent . but it is not necessary to enlarge the stimulus set to test this . data from a sample of neurons and stimuli give the fraction of stimuli to which neurons respond . then by using knowledge of the capacity of the system , which in this case is several hundred memories , one can statistically extrapolate from the sample to show that each neuron is responsive to multiple unrelated stimuli . as for the bmw model , the input / output cells would have the same kind response characteristics as all the cells of the hopfield model , responding to @xmath1 of the stimuli . but there are also gm memory cells , which respond much more rarely . if the sample data include at least a few responsive gm neurons , one can detect the different properties of the input and gm cells , and thereby distinguish the system from the hopfield model . in this paper , we will construct a method to extrapolate in general from data with a sample of cells and stimuli to the whole system . the method enables us to compute which of the following properties of a set of putative gm cells is consistent with data : ( a ) the cells actually could code for single properties . ( b ) each cell is expected to respond to multiple _ unrelated _ properties , after an extrapolation to the full set of possible stimuli . one confounding issue is that if one detects a response from a gm - like cell it is easy to misidentify the property corresponding to the cell s firing . if a cell only responds , within limited data , to a particular individual person , the cell could indeed be a classic gm cell corresponding to that person . but it could also be , for example , a gm cell for an episodic memory that contains the person . in that case it would respond to stimuli containing other components of the episode . the distinguishing feature of the most general kind of gm - like cell , but one that can be hard to test , is that apparently different stimuli that cause it to respond are in fact related . in contrast , with a purely distributed representation at the highest level , there is no high - level relation between , for example , the multiple people causing a particular cell to respond . the only relation is an identity of lower - level features in the input representation . the locality is at the feature level , not at the high level . we now quantify the information requirements for a memory or categorization system . first we distinguish the activity state and the storage state . the activity state is the pattern of firing of the neurons , and the storage state @xcite is the pattern of synaptic connectivity and strengths . furthermore , within the activity state of a memory system , we distinguish an input representation and a recognition representation . the input representation is of the current input stimulus , like a visual scene , while the recognition representation concerns which stored pattern ( if any ) corresponds to the current input . the immediate input to a memory system must be able to represent the relevant features of any possible stimulus , and not just those previously encountered stimuli for which there is a memory trace . here the standard arguments for distributed representations apply unambiguously , and a gm representation is not possible . the argument is simply that @xmath14 neurons code for at most @xmath14 exclusive properties in a gm system , but that they code for exponentially more with a distributed representation , for example @xmath15 with a simple binary code . in general the immediate input to a memory or categorization system is not the raw sensory input but a highly processed representation of those high - level features that are relevant for the system s particular task . for example , input for face recognition could involve neurons coding for the presence , shape and position of eyes , nose and mouth , etc ; individual features could well be coded locally , but the collection of feature representations would always form a distributed representation . each neuron has a range of distinguishable firing rates , so that the raw information capacity in the activity of @xmath14 neurons is a few times @xmath14 bits . but robustness requires a certain amount of redundancy , and firing is often sparse , both of which reduce the information per neuron . measurements @xcite show that in hippocampal neurons about @xmath16 bits of independent information are coded per neuron and suggest that a few hundred neurons suffice for coding possible faces . for example , 300 neurons code about 100 bits , sufficient for about @xmath17 different categories of face , an entirely satisfactory number . note that once an input representation is sufficiently small , pure representation efficiency is not a dominant consideration . issues of processing speed , metabolic efficiency , and algorithmic robustness can be more important . for example , sparse distributed representations appear to be favored @xcite , since they can give weak interference between memories during synaptic plasticity . the information capacity in the storage state , i.e. , in the synapses , was estimated by @xcite and by @xcite . there are two contributions , from the synaptic topology and from the synaptic strengths , giving a total of 5 to 10 bits of raw storage capacity per synapse . we divide by 10 , to provide a plausible allowance for redundancy . thus we need about one synapse for each bit of storage information . this calculation uses only very basic physical information about synapses and neural processing , so it is certainly accurate at the order of magnitude level . since measures of information in units of bits are independent of the physical implementation , the numbers are directly compared with those for ordinary digital computers , and experience with data processing and storage can be used to derive minimum numbers of synapses for a task . a human brain has around @xmath18 neurons and @xmath19 synapses per neuron . so its synapses store about @xmath20 bits , i.e. , about @xmath21{gbyte}$ ] , up to a factor of 10 or so . suppose the system has a repertoire of @xmath22 stored memories . each is an arbitrary association of a category to stimulus features . so we attribute to each memory @xmath23 bits of association information , which we term its semantics . this includes both input and output information . ( e.g. , facial structure and name for a person , and , in fact , _ all remembered information about the person_. ) it is important to include output semantics in the associations , since they are what allow the retrieval of a particular memory to cause memory - specific behavior . these are quite arbitrary associations , for example of a linguistic name to a specific person , and without such output associations there is no purpose to storing a memory . the output associations do need to be stored , and they therefore contribute to the calculation of the minimum size of the system just like the input associations . the total association information is @xmath24 bits , so that approximately this number of synapses is needed . for example , with a repertoire of @xmath25 and an average of @xmath26 bits of information per item , we need about @xmath27 synapses . the measured number of synapses per neuron is around @xmath28 , thereby implicating about @xmath29 neurons , i.e. , a number proportional to the size of the system s repertoire ( and the information per memory trace ) . hence if the system uses @xmath22 gm cells for a recognition representation , this is at most a constant factor beyond the neurons needed to carry the storage synapses . moreover , if @xmath23 is larger than about @xmath28 , there is not even any overhead at all in using gm cells . we can characterize this by saying that if the memories are semantically rich , then a gm strategy for the recognition output can be optimally efficient , as in the model of @xcite . in contrast , if one ignored the storage requirement , one would assert that @xmath14 recognition neurons can code exponentially more recognized categories , e.g. , @xmath15 . it is important that in quantifying the information the term `` bit '' is used in the strict information theoretic sense . this means that if each memory were coded as an actual bit pattern , each of the @xmath30 possible patterns would be equiprobable . thus , on a computer , the bit count refers to an optimally compressed representation . however , only certain features of the input are relevant for tasks like face identification : a minimalist line drawing often suffices for unambiguous identification . so in computer terms the association bits are with respect to a representation that is both lossy and compressed . in the opposite direction , it can be difficult to perform computations on optimally compressed representations , and it is also difficult to measure accurately the probabilities of occurrences of different kinds of stimuli , since the number of possible stimuli far exceeds the number actually experienced . moreover redundancy ( in the information theoretic sense of using more than the minimum necessary number of bits ) is useful in giving robustness to a system . for all these reasons , we must expect the physical capacity of a system needed to code memories to be a substantial factor large than a minimal physical implementation of @xmath24 bits . nevertheless this measure is important in quantifying information in an implementation - independent way . it also enables us to estimate the information storage requirements by examining implementations of related tasks on a digital computer . well - known examples of simple line drawings and pictures of artificially low resolution show that the information to identify faces could be quite modest , if a suitable representation is used . but the synaptic size of the system also depends on the remaining association information , treated as output . this can be much larger in size , effectively amounting to a biography of the individual concerned in each memory . a number of objections to our bounds and ideas for evading them have been proposed , which we now answer . the general answer is simply that the information theoretic bound represents an absolute physical limit that it is impossible to exceed . all that is required is that we count the bits of information in the strictly correct information theoretic sense , and that we have identified the correct physical location of memory in the synapses . [ [ counting - memories ] ] counting memories + + + + + + + + + + + + + + + + + does not the idea of quantifying memories as discrete items that can be counted carry the implication that we use a gm system ? are there not difficulties in counting memories in distributed memory systems ? in fact the classical kinds of distributed memory , e.g. , @xcite , are regarded as storing patterns , which can be counted . what we do have in mind is declarative , or explicit memory , the kind considered as prototypically hippocampal . here it is reasonably clear what is meant by a single memory : a picture , a scene , or the meaning of a word ; all of these are discrete . even with episodic memory , where there is a continuous time variable , we can observe that there is a correlation time within a continuous series of events . as regards storage requirements , we simply identify a single memory with the happenings within a correlation time , which is evidently of the order of seconds or minutes . of course , only a small fraction of these are stored in long term memory . we will not require great precision here . in the contrasting case of implicit procedural memory , it is much less obvious what should be defined as a single memory item . but we are not concerned with this case . [ [ multiplexing ] ] multiplexing + + + + + + + + + + + + could not one gain by allowing a neuron to respond to multiple different stimuli ? could not a single face - identification cell respond to either george bush , jennifer aniston or john hopfield , for example ? this multiplexing is just going in the direction of a distributed representation . our argument so far does not rule that out ; all it says is that this does not provide a way of reducing the synapse count . it is our statistical argument in later sections that enables us to estimate the degree of multiplexing . now if a neuron responds to multiple categories , then there is interference at the neural level between different memory traces . an unambiguous categorization then requires the use of the firing information from more than one neuron . in the case of lightweight memories , i.e. , with substantially less than 10000 bits of associations per memory , multiplexing of memories can indeed reduce the neuron count , and is allowed by our general argument . but with richer memories , there is no gain . [ [ representation-v.memory ] ] representation v. memory + + + + + + + + + + + + + + + + + + + + + + + + the memories we have discussed are typified by hippocampal memories . consider instead visual area v1 ; the number @xmath31 of distinguishable activation patterns is exponential in the number @xmath14 of neurons , e.g. , @xmath32 . an outside observer could identify different faces from the different activation patterns . why should we not regard this as memory , if we are to regard observation of ( much simpler ) activation patterns in gm cells as identifying faces , and as in fact part of a memory system ? the difference is not in the outside observations , but in the use the organism itself makes of the information . a memory is not useful unless it produces some consequence when recalled . what we mean , for example , is that seeing john hopfield on the other side of a street might induce us to cross the street and greet him by name . for the billions of other possible people , there would be either be a different response or no response at all . this is why we defined the associations for a memory to include output as well as input . there must be sufficient information stored to enable the relevant computation to be done from the activation pattern corresponding to the stimulus . without being concerned with storage , it is perfectly possible to compare one activation pattern with a previous one , to identify whether or not it has changed . but to compare it with patterns of activation for all the people one remembers , and to take appropriate actions , one needs appropriate storage . it is simply not possible to evade the fundamental physical necessities given by information theory . [ [ coding - commonalities ] ] coding commonalities + + + + + + + + + + + + + + + + + + + + many remembered faces have features in common . can not this be used to reduce the number of synapses and neurons needed , by coding the common features in a special subsystem for the common features ? suppose we had a set of faces characterized by very short , dark , and curly hair . would there not be a gain by allocating a neuron to this combination of characteristics and using it instead of separate neurons for hair length , color and curliness ? if the different hair features were equiprobable and uncorrelated , the general argument would prevent any gain . as an example , suppose each of the three hair characteristics has 8 equiprobable values , for @xmath33 distinguishable combinations . if all the combinations were equiprobable , we would allocate 9 bits for the information content . probability here refers to a prior probability of occurrence , i.e. , before the creation of a memory trace . but if instead all the characteristics were perfectly correlated , then there would only be 8 combinations , which could be represented in 3 bits . when we construct memories , this gives a gain of a factor of 3 in storage if we only represent the combinations that actually occur in the input representation rather than all possible combinations . but this is exactly what is meant by using a correct measure of information to compute the minimum number of synapses . note that the task carried out by a memory system is not merely to identify the best fit to a current stimulus among stored memories ; for that very few bits are needed . for example , if a stimulus is represented by 100 bits , but only 8 memories are stored , then only @xmath34 suitably coded bits are needed to identify the stimulus , if it is assumed that the stimulus corresponds to one of the memories . but it is also necessary to identify the case that the stimulus fails to correspond to a stored memory , so that it is a candidate for a new memory . it is for this that the other 97 bits are needed . although our argument was used to show that gm coding can be optimally efficient in the use of synapses and neurons , the bounds on synapse and neuron number are independent of the coding method . it is therefore useful to verify that the bounds are obeyed by the polar opposite of gm systems , i.e. , by conventional distributed memory systems . calculations of the capacity of such systems have been made , e.g. , @xcite , and it can be checked that they do obey our bounds . but what appears to have been missed is that the capacity limit also removes the argument against using gm cells . hopfield s recent model @xcite provides an excellent example . in this model , each stored memory corresponds to the values of each of 50 categories , with 20 possible values per category . this gives a total of @xmath35 bits of information . retrieval of a memory results in a neural representation of this information in the firing of 50 out of 1000 binary neurons , for a biologically realistic sparsity of @xmath1 . storage is in binary synapses with all - to - all connectivity on the 1000 neurons , for a total of slightly under @xmath36 bits of storage capacity . this implies that the system can store at most @xmath37 separate memories . in fact , the simulations in @xcite show that , with the algorithms used in that paper , performance noticeably degrades when about 250 memories , i.e. , a factor of 20 below the physical limit . thus the information - theoretic bounds are obeyed . the problem is that the memories are stored on the synapses connecting the active neurons in a particular memory . synapses overlap between memory traces , which causes interference if too many memories are stored . consider in contrast the bmw model @xcite , which , as observed by its authors , is optimal in its number of synapses . suppose for input and output we use the same 1000 neurons as in the distributed model . then to store @xmath14 memories we add @xmath14 memory neurons and @xmath38 synapses ( for input and output ) . we also may use a relatively few extra interneurons and synapses to implement winner - take - all dynamics . the number of synapses is double our minimum estimate , because we treat input and output semantics separately : they could be different . this is much more efficient than the distributed model . there is also an extra neurons for each memory . but we can readily increase the capacity for stored memories by increasing the number of memory neurons . with @xmath39 , we would have the same number of synapses as in hopfield s model , @xmath40 more neurons , but double the capacity . with the distributed model , the capacity can be increased only either by simply duplicating the system , which is always a possibility , or by a change in architecture . an appropriate change in architecture would be to use the original input neurons to feed a separate layer which codes the same information more sparsely , after the style of a support vector machine . note that the bmw model reliably performs the same pattern completion task as hopfield s model . that is , a stimulus consisting of a few values is completed to the values of all categories for the corresponding memory . the bmw model essentially performs a comparison of the active bits in the sparse input pattern with the on - bits in the stored pattern . because of the @xmath1 sparseness of the input , the probability that an on - bit in the stimulus coincides with an on - bit in an unrelated stored pattern is also @xmath1 once more than a few bits are examined , the probability of a chance coincidence is extremely small , with a correspondingly small misidentification probability . we have seen that the efficiency argument against gm cells disappears , especially for semantically rich objects , like most people s grandmothers . but the efficiency analysis for the input representation shows that the gm cell population is necessarily accompanied by a population of cells carrying a distributed code . experimental characterizations of the different kinds of coding can be made by measuring the sparsity of neural responses to stimuli . by sparsity we mean , for each cell , the fraction of stimuli to which it responds . we assume that , as in @xcite , some threshold criterion is defined cell - by - cell as to whether a cell responds or not . thus the neuron is treated as binary . other definitions involving analog firing rates are possible , but we will not use them . note that with our definition and when all cells have the same sparsity , both the population and the lifetime sparsity are equal , unlike the case @xcite with other definitions of sparsity . for a system with fully distributed coding , we expect to measure sparsities characteristic of the input and output representations . for example , in hopfield s model , the sparsity is exactly @xmath1 . more realistically , there will be a range of sparsity . the input and output cells of a gm system will have naturally have similar sparsities to those of all the cells in a distributed - memory system . but the gm cells must respond much more rarely . so with a gm system we expect there to be two very different populations of cells distinguished by one population having a dramatically smaller sparsity than the other . whether or not the two populations are in the same area of the brain is not determined by general arguments . but we will find that , in fact , the putative gm cells of @xcite do have an accompanying distributed - code population . for our purposes it will be unimportant whether the detected distributed - code population is the one that provides the actual input and output for the detected gm - like cells . expectations for the sparsity of gm cells can be provided in terms of the repertoire size of a system , which has a connection to behavioral data . there are two somewhat different kinds of memory system we will consider . one is typified by face recognition , where a recognized input is categorized into one of @xmath22 categories , corresponding to distinct persons ; recognition is exclusive between categories . then for a random sample of faces in the repertoire of the system the sparsity of the gm cells is @xmath41 , if we assume that each person is allocated the same number of cells . the second case is for declarative memory ( episodes , facts , etc ) . recall is by a stimulus containing a few components of the original memory . since the same components could be part of other memories , the pattern of recognition firing need not be exclusive between memories . let us regard these patterns as priming the recall of the memories . full conscious recall of one particular memory requires some extra cues and modulation . with a gm system , non - exclusive priming recall would be at a relatively low level of firing above some threshold , with full recall involving exclusive firing at a much higher level . in a memory system , a typical stimulus can evoke multiple memories . if we let @xmath42 typify the number of memories evoked , then a given gm neuron is caused to fire by a fraction @xmath43 of stimuli . it is therefore convenient to define an effective repertoire size @xmath44 , so that the typical sparsity is @xmath45 . from standard psychological data , we envisage that @xmath46 is thousands to at least millions for interesting cases . in contrast , the distributed - code cells fire much more frequently ; this is known from data , and is necessary in order that this population can represent a sufficiently large number of stimuli . measurements of single - cell responses concern only a small fraction of cells and of all possible stimuli . so we will treat data as being from a sample over cells and stimuli , and deduce properties of the whole system : e.g. , the relative sizes of the cell populations and their sparsities , and hence the number of categories coded for by the gm population . in doing this , we will quantify , and hence compensate for , the strong biases against detection of gm cells . suppose we present a sample of @xmath47 stimuli that are randomly chosen from some broad class ( e.g. , pictures of famous people , images concerning movies that the subject has watched , pictures of buildings ) . any particular cell @xmath48 responds to some fraction of these , called the cell s ( lifetime ) sparsity @xmath49 . the number @xmath50 of stimuli that evoke a response by the cell is taken from a binomial distribution of mean @xmath51 : @xmath52 this simply corresponds to the probability that @xmath53 of the stimuli are in the response - causing class and @xmath54 are in the non - response - causing class , as regards cell number @xmath48 . these two classes of stimuli form fractions @xmath49 and @xmath55 of the whole set of stimuli . we now consider a sample of cells , thereby sampling the distribution of sparsity over cells , @xmath56 . this means that the fraction of cells with sparsity @xmath3 to @xmath57 is @xmath58 . then the probability of getting @xmath53 responses to the @xmath47 stimuli in some random chosen cell is obtained by integrating the single cell response with the sparsity distribution : @xmath59 this is a general result . the only necessary assumption is that the cells are randomly chosen out of some more global set of neurons ( e.g. , hippocampus ) and that the stimuli are randomly chosen out of some global class . the value of @xmath3 for a cell and the distribution @xmath56 depend both on the choice of stimulus class and on the choice of the threshold for a response . changing either will naturally affect the distribution . for the data we analyze , the response criterion is given in @xcite . if multiple sessions and multiple subjects are considered , eq.([eq : binomial.combo ] ) continues to apply , with @xmath56 being the distribution averaged over subjects . so this form is amenable for the analysis of aggregated data . observe also that the derivation of the formula does not require any assumption about the independence of the firing of different neurons : the formula is simply an average over all neurons in whatever area is being sampled . this allows the formula to be completely general , in contrast to the model of @xcite , which requires that neuron - neuron correlations be neglected . a common ansatz , as in @xcite , is to assume a fixed sparsity @xmath60 , i.e. , to set @xmath61 . such a model we term a single - population model . but for an analysis of a possible gm population , we must allow for at least two populations . from a mathematical point of view , eq . ( [ eq : binomial.combo ] ) expands @xmath62 in basis functions , with expansion coefficients @xmath56 . the significance to its use is four fold : ( 1 ) it relates the distributions for different pattern numbers @xmath47 via a common set of expansion coefficients @xmath56 . ( 2 ) the expansion coefficients are non - negative . ( 3 ) for a distributed - code population to represent all possible stimuli , efficiency is important , i.e , using the minimum of neurons . this will tend to maximize the sparsity subject to other constraints like keeping relatively low the metabolic costs @xcite of action - potential generation . thus we should expect the sparsity of the distributed - code population to vary over a fairly narrow range . ( 4 ) any gm population has an extremely small sparsity , so that it populates just the bins with @xmath63 and @xmath64 responses to the @xmath47 stimuli . therefore the two population property leads to the qualitative expectation for @xmath56 that is shown in fig . [ fig : d](a ) . the distribution of responses @xmath62 can be regarded as a smeared version of the sparsity distribution @xmath56 with @xmath65 . given this smearing , a useful approximation is to replace the distributed - code peak by a delta function at some fixed typical sparsity , fig.[fig : d](b ) . [ cols="^,^ " , ] a final piece of session - averaged data given in @xcite is the fraction of stimuli that produced a ( simultaneous ) response in at least two neurons . in in app . [ sec : evocative ] , we derive a formula for this quantity . as with the number of of evocative stimuli , a neglect of neuron - neuron correlations is needed . the results are shown in the last line of table [ table : comparison.session ] . as already observed in @xcite , the one - population model with their preferred value @xmath66 gives a fraction @xmath67 that is rather below the data ( @xmath68 ) . a comparably bad fit is obtained by our two - population model . in fact , the bad fit happens quite generally . the formulae for both @xmath69 and @xmath70 given them in terms of a single property of the model , the cell - averaged sparsity @xmath71 . we show in app . [ sec : evocative ] , that when @xmath72 is not two large , the two quantities obey an approximate relation @xmath73 this relation is obeyed to useful accuracy in the model calculations in the last two lines in table [ table : comparison.session ] , but it is violated by a factor of two by the data . the derivation is not affected by adding yet more populations of different sparsities , but only by including neuron - neuron correlations . there are in fact two simple ways to overcome this problem . one is simply that one fraction is proportional to the number of detected units in a session , and the other is proportional to its square . since this number varied quite widely @xcite between sessions ( 18 to 74 ) , the session average of @xmath74 can not be replaced by the square of the average of @xmath14 . this could easily account for the factor of two mismatch . in contrast , the fraction of evocative stimuli is amenable to a simple average over sessions . the second possibility is from neuron - neuron correlations , to which @xmath70 is much more sensitive than the other observables . if there were a small fraction of nearby neurons that always fired in pairs , these would disproportionately contribute to @xmath70 , but not nearly as much to @xmath69 . of course , both these suggestions can be tested by a closer examination of the data . we also examine how well the one - population model , with the parameters from @xcite , agrees with the data we used from the earlier paper @xcite . this is shown in table [ table : comparison.global ] . it can be seen that the observables we used are particularly sensitive to the differences between the models . a low _ average _ sparsity is necessary to keep the number of responsive neurons down to the experimental value . but with a one - population model this also implies that neurons with @xmath75 responses are many fewer than those with @xmath64 responses . moreover the number of cells with even more responses than 2 is minute , so that @xmath76 is close to its minimum value of 2 , whereas the data is much higher . this is a clear indication of the need for two populations of cells with very different sparsities . we conclude that working with the distribution of the number of responses by individual neurons ( or units ) , as we do , is preferable to the other distributions . the distribution @xmath77 is easy to work with , and data can be usefully aggregated over a whole experiment , given only that the number of stimuli is approximately the same in each session and that the stimuli are chosen at random in some large class . these conditions can be imposed by an experimenter . the model can always be systematically improved by changing the sparsity distribution , e.g. , by adding extra components . working with aggregate data keeps the sampling errors usefully low , and parameters for a model can be computed simply from properties of the aggregate data . other analyses of data , e.g. , @xcite , have reported that hippocampal facially responsive cells carry a distributed code as opposed to a gm - type code . see also the recent work of @xcite . these analyses might appear to contradict our calculation that distributed - code cells are a very small fraction , perhaps less than @xmath78 of the total . however , there is a strong bias against actually detecting gm - like cells . in this section , we use our fit to the more recent data to quantify this bias , at least roughly , to determine whether there is consistency between our results and the earlier data . the primary issue is that experiments typically only report those cells that are actually detected to respond to at least one of the stimuli used . for example , in a paper documenting place cell @xcite state `` the electrode assemblies were advanced until one or more hippocampal complex - spike cells were isolated extracellularly . '' then they observe that cells that do not produce any detectable spikes `` are excluded from analysis here due to our lack of ability to detect them '' . since gm - like cells respond to a very small fraction of stimuli , the ones that respond to no presented stimulus , i.e. , the vast majority , are typically omitted from an analysis . the resulting bias can be seen in the data that we analyzed earlier . a minority of the detected cells in @xcite are in the gm - like class ( 43 out of 132 ) , even though we have shown that the gm - like cells can be in the vast majority ( @xmath79 or more ) . with fewer stimuli , the bias becomes even stronger , as in the data used by @xcite . they used 20 face stimuli , and the total number of facially responsive neurons was 14 . a rather higher sparsity was reported than our result . but this is partly because a different definition of sparsity was used , applied to the spike numbers rather than to a binary response criterion . furthermore the cells have considerably larger background firing rates than those in the new data . despite the differences in cells and species , we blindly apply our model to give a rough test of consistency . in our two - population model , the fraction of gm cells in the _ detected _ cells is @xmath80 this is the probability that a cell is a gm - like cell _ conditional _ on the cell producing a detected response to one or more of @xmath47 stimuli . notice that the silent - cell ratio @xmath81 cancels in this formula ; we have a relation between numbers of different kinds of detected cell under different experimental conditions . we have estimates for the parameters of the model , so substituting @xmath82 predicts a detected gm - cell fraction of @xmath83 , i.e. , about 2 cells , in the set of cells investigated in @xcite . in fact , @xcite did report that two of their cells had a gm - like response . there is , of course , no significance to the fact that this number is exactly the value predicted : there are expected statistical fluctuations , and the measurements were done with different methods and in a different species than in @xcite . nevertheless it is very important that the previous report , viewed as evidence in favor of distributed representations , is completely compatible with gm cells being in the vast majority , with the parameters we have determined . the undetected gm cells simply appear to be silent within the experiment and are therefore classed as not facially responsive . statements about the neural representation being distributed apply only to ( most of ) those cells that the measurements actually detected , not to all cells in the relevant region of the brain . the results of @xcite clearly suggest the detection of grandmother cells in the classic sense . many other experiments have detected individual cells with strikingly specific responses ( e.g. , @xcite ) . therefore it is useful to hypothesize that some of these cells are indeed gm - like cells , even though the concept of gm - cell may need to be extended and modified . a purely experimental direct test of the idea needs too many stimuli to be practical , cf . so other arguments must be brought in , of which we have provided two . one uses an estimate of the actual storage requirements for a memory system . we showed that gm systems can be optimally efficient in the use of synapses and neurons . the usual efficiency argument applies only to the input representation , but now carries the implication that in a gm - like system there must be two populations of cells with widely different sparsities . our second argument is a method to analyze neural responses . a particular aim is to measure whether they are quantitatively consistent there being separate neurons coding for each recognized person , or , alternatively , for each individual declarative memory . our method enables one to determine whether or not individual cells necessarily code for multiple persons or memories . we derived a general formula eq . ( [ eq : binomial.combo ] ) for the neural responses in terms of an underlying distribution of sparsity . our expansion is a new result and is applicable independently of any detailed theory or model of neural function . in effect , the formula enables us to extrapolate from limited data to obtain the fraction to stimuli to which cells respond . it also allows us to compensate for the strong biases involved in detecting cells when sparsities differ by very large factors . thus we obtain valid estimates of the numbers of cells of different kinds . we thereby solve some of the issues raised by @xcite concerning the publication of data only about responsive cells . one primary remaining bias is that different neurons may have different electrical characteristics , with a consequent different maximum distance from the electrodes for detectability of spikes . but this is presumably a milder effect than that caused by orders of magnitude differences in sparsities . from the data we find indeed that the two - population property is obeyed . not only does the ultra - low - sparsity population comprise the vast majority of cells in the brain regions concerned ( hippocampus , etc ) , but its sparsity can be in a range compatible with the hypothesis of a gm - like system : roughly @xmath84 with a repertoire of @xmath0 . an important role is played by the many silent cells . it is obviously unreasonable to assume they have no function . but on the gm - cell hypothesis they naturally are to be interpreted as the majority of gm cells that are not relevant to the particular stimuli used in an experiment . the large number of these cells is what enables one to overcome the strong biases against detecting a response of any one gm cell to a limited set of stimuli . now the group responsible for the analyzed data argue @xcite that their data do not support the gm cell idea . in @xcite , they say `` if we assume that a typical adult recognizes between 10,000 and 30,000 discrete objects ( biederman , 1987 ) , @xmath85 implies that each neuron fires in response to 50 150 distinct representations . '' [ @xmath86 should be replaced by @xmath60 in the notation of the present paper . ] however their analysis assumed a single value of sparsity . while this is a suitable approximation for conventional mechanisms of distributed memory , it is very bad for gm - like systems . even though the explicit aim of @xcite was to test the gm - cell hypothesis , the use of a single sparsity in effect imposed an assumption that the hypothesis is wrong . we showed that the single - population hypothesis is a bad fit to the data . since our expansion ( [ eq : binomial.combo ] ) is very general , the fault is in the single - population hypothesis not in any assumption about neural properties . the rather low value of sparsity given by waydo et al . is merely a compromise between the widely different sparsities of the two populations . our results are consistent with an even higher number of recognized objects than in the estimates of @xcite . indeed , even _ without _ allowing for the silent cell correction , our fits allow a gm - cell population with a sparsity of @xmath87 corresponding to a number of objects not far from the lower edge of biederman s range . note that our basic estimate of the number of categories , @xmath0 , assumes that the cells are classic gm cells , each responding to a single individual person . but the number of categories could be substantially higher . if the cells are general memory cells , in the style of the model of @xcite , they could respond to images of several people . it could also be that more familiar stimuli , with richer associations , have more cells . in that case measurements with familiar stimuli , as is the case in the data , would be biased towards these memories with unusually large numbers of cells , with a corresponding reduction in our estimate of the number of categories compared with the true number . in gm systems , like the model of @xcite , the number of gm cells is very much larger than that of the input cells , as is consistent with the numbers we have deduced . an immediate implication is that each gm cell receives input from a modest number of input cells , but that each input cell sends output to a much larger number of memory cells . given also our finding that the gm cells in the relevant regions are in the vast majority , there are some striking anatomical implications . in fact striking disparities in synapse number are well known in the hippocampus @xcite : for example , each ca3 pyramidal cell gets about 50 input synapses from dentate granule cells , while other connections have tens of thousands of synapses . note that hippocampal neurogenesis results in dentate granule cells , highly appropriate if they are gm - like . however , general - purpose memories need a wider variety of ( processed ) input than does a face recognition system , and hippocampal - related regions are sufficiently complex that the real picture is undoubtedly much more complicated . even so , a careful analysis of the disparities in synapse number should provide critical information on neural function and the viability of gm - like systems . other less quantitative arguments have been advanced against the reality of gm systems , e.g. , @xcite . for example , distributed memory systems are said to be robust against partial destruction , since there is no single location for a single memories . but we do know that memories disappear . if there are multiple gm cells for a memory in different places , then we can overcome the robustness argument by simple redundancy . moreover memories form a network of knowledge , so that individual items of semantic memory can be readily reconstructed from other knowledge . episodic memory is really an ordered sequence of individual episodes , not necessarily remembered at all precisely . any one episode that disappears can be approximately filled in from neighboring episodes . distributed memory systems are also said to be good at filling in missing parts of input data , as in reconstructing a full remembered image from a stimulus containing only a part of the image . but this property can also be true for gm - like systems . for example , when the bmw architecture @xcite is used with a sparse input representation and suitable dynamics for its gm cells , it also performs pattern completion ; the completion property is actually associated with properties of sparse representations used for input data . it has been said that new memories are harder to construct in gm systems than in distributed - memory systems . but now that adult neurogenesis in the hippocampus is well established , it may well be that there is actually a pool of new neurons available for at least some uses that could include being gm - like cells for new memories . the new neuron rate may however be excessively small . in addition , it is possible that the gm nodes are on dendritic tree rather than being whole neurons . it is known that there can be substantial changes in dendritic topology , which could easily include the formation of new nodes . here the fundamental mode of operation is of a gm - like system while the neural code of memory neurons takes on some of the aspects of distributed memory . in any case , there are potential realistic mechanisms for the formation of new gm nodes , so that there is no insuperable obstacle here . jcc is supported in part by the u.s . he also thanks larry abbott and the volen center at brandeis university for hospitality during the initial stages that led to this work . dzj is supported by the alfred p. sloan fellowship and by the huck institute of life sciences at penn state . we thank larry abbott , jayanth banavar , gong chen , anne graybiel , john lisman , and michael wenger for useful conversations . our model for the statistics of neural firing has two cell populations : one that uses a conventional distributed code with a single sparsity @xmath86 and a second gm - cell population , as illustrated in fig . [ fig : gm - model ] . these from fractions @xmath88 and @xmath89 of the total number of cells . a remaining population of cells does not respond to any stimuli at all in the class used in the experiment . we let @xmath90 be the fraction of the images used which have stored representations in the gm population , we let @xmath22 be the repertoire of the gm cells , and we let @xmath42 be the typical number of categories ( or gm groups ) evoked by a stimulus in the system s repertoire . as before , we let @xmath44 . suppose first we record from some random cell known to be in the distributed population . we have seen that when we present @xmath47 images , the probability of getting @xmath53 responses is approximately the poisson distribution in eq . ( [ eq : single.cell ] ) with @xmath91 . if , instead , we pick a gm cell , then for each individual image it has a probability @xmath92 of responding . therefore over a set of @xmath93 unrelated images it has a probability @xmath94 of responding exactly once . there is a negligible probability of @xmath75 for such a cell . finally , if the cell is outside the above two populations , it is silent in the experiment and always gives @xmath63 . summing over the distributions of @xmath53 for cells of the different kinds , weighted by their fractional population size , gives @xmath95 \dfrac { f_{\rm gm } \ , k \ , p } { r_{\text{eff } } } + f_{\rm d } \,pa \ , e^ { -pa } & \text{if $ n=1 $ , } \\[2 mm ] f_{\rm d } ( pa)^n e^ { - pa } \dfrac { 1 } { n ! } & \text{if $ n\geq2$. } \end{cases}\ ] ] this is of the form of the general distribution eq.([eq : binomial.combo ] ) with @xmath96 and with the poisson approximation approximation for the distributed - code population and with the gm approximation that the gm - like cells fire so rarely that their responses for @xmath75 can be neglected . there are several meaningful parameters for the gm population , but only one combination affects the distribution @xmath62 . so we define @xmath97 , to find @xmath98 \dfrac { p } { \hat{r } } + f_{\rm d } \,pa \ , e^ { -pa } & \text{if $ n=1 $ , } \\[2 mm ] f_{\rm d } ( pa)^n e^ { - pa } \dfrac { 1 } { n ! } & \text{if $ n\geq2$. } \end{cases}\ ] ] the parameter @xmath99 has the meaning that if a cell is outside the distributed - code population then it responds to a stimulus in the chosen global class with probability @xmath100 \simeq 1/\hat{r}$ ] , where the last approximate equality applies in the realistic case that @xmath101 is small , according to our fit . we wish to extract the properties of the distributed - code cells without contamination from the gm cells . for that we need properties of the distribution for @xmath75 , for which we use the probability and the mean number of responses . the probability of @xmath75 is @xmath102 .\ ] ] the mean number of responses , in cells with @xmath75 , is @xmath103 these last two equations suffice to determine @xmath86 and @xmath88 from the data in @xcite see eqs . ( [ eq : fd ] ) and ( [ eq : a ] ) . several further observables are considered by @xcite . these observables refer to a session in which @xmath104 stimuli are presented to @xmath105 cells or @xmath14 units . one observable is the number of neurons @xmath106 that respond to at least one stimulus . its average is just the number of cells or units times the probability that one cell or unit responds , which in our notation is @xmath107 , in the notation of eq.([eq : binomial.combo ] ) . hence the number of responsive units in one - population model of @xcite is @xmath108 in our two - population model it is @xmath109 we are not quite sure how many cells correspond to each unit in the new data , so in table [ table : comparison.session ] we gave results for several choices of the ratio of cells to units , @xmath110 : 2 ( as we estimated for the earlier data ) , 2.5 , and 3 . a second observable is the number @xmath69 of stimuli that evoked a response in at least one neuron in the session . to derive this from the response distributions requires a further assumption that correlation between the firing of different detected neurons can be neglected . now the probability of one stimulus evoking no response in any of @xmath14 independent cells ( or units ) is @xmath111 , where @xmath112 is the probability of no response in one cell / unit on presentation of 1 stimulus . hence the average number of evocative stimuli in a session is @xmath113.\ ] ] from eq . ( [ eq : binomial.combo ] ) , we find that in general @xmath114 , where @xmath71 is the sparsity averaged over cells . in the one - population model of @xcite we therefore get @xmath115,\ ] ] while in our two - population model it is @xmath116.\ ] ] the appearance of @xmath81 in this last formula is misleading : the dependence on @xmath81 of @xmath99 and @xmath101 in our fit cancels the explicit factor of @xmath81 . we have used the number of cells @xmath105 in this formula rather than the number of units @xmath14 , since our fit is made with respect to cells . a final observable we consider is the number of stimuli in a session that evoked responses in 2 or more cells / units . this quantity , denoted @xmath70 can be obtained from the distribution of the number of neurone responding to a single stimulus : @xmath117^{n_{\text{r } } } } { n_{\text{r } } } e^{-n\,(1-\bar{\alpha})},\end{aligned}\ ] ] it follows that on average @xmath118.\ ] ] from eqs . ( [ eq : s.r ] ) and ( [ eq : s.r.nr.ge.2 ] ) , we get a relation between @xmath69 and @xmath70 valid when @xmath119 is less than about unity and @xmath14 is substantially larger than unity . we expand the powers of @xmath120 for small @xmath71 to obtain @xmath121 and then @xmath122 d.g . amaral , n. ishizuka , and b. claiborne , `` neurons , numbers and the hippocampal network '' , in progress in brain research vol . 83 , ed . j. storm - mathisen , j. zimmer and o.p ottersen , pp . 111 ( elsevier , 1970 ) . c.a . barnes , b.l . mcnaughton , s.j.y . mizumori , b.w . leonard , and l .- h . lin , `` comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing '' , progress in brain research vol . 83 , ed . j. storm - mathisen , j. zimmer and o.p ottersen , pp . 287300 ( elsevier , 1970 ) . henze , z. borhegyi , j. csicsvari , a. mamiya , k.d . harris , and g. buzski , `` intracellular features predicted by extracellular recordings in the hippocampus in vivo '' , j. neurosci . * 84 * , 390 ( 2000 ) . olshausen and d.j . field , `` how close are we to understanding v1 ? '' , neural computation * 17 * , 16651699 ( 2005 ) ; `` what is the other 85% of v1 doing ? '' , in `` 23 problems in systems neuroscience '' , j.l . van hemmen and t.j . sejnowski ( eds . ) ( oxford university press , 2006 ) .

consider two domains connected by a thin tube : it can be shown that the resolvent of the dirichlet laplacian is continuous with respect to the channel section parameter . this in particular implies the continuity of isolated simple eigenvalues and the corresponding eigenfunctions with respect to domain perturbation . under an explicit nondegeneracy condition , we improve this information providing a sharp control of the rate of convergence of the eigenvalues and eigenfunctions in the perturbed domain to the relative eigenvalue and eigenfunction in the limit domain . as an application , we prove that , again under an explicit nondegeneracy condition , the case of resonant domains features polinomial splitting of the two eigenvalues and a clear bifurcation of eigenfunctions .

1307.7930

the aim of this paper is to investigate the behavior of dirichlet eigenvalues in varying domains , when a shrinking cylindrical handle is attached to a smooth region , seeking not only for the rate of convergence but also for sharp asymptotics . since we consider a tubular handle with a cross - section of radius of order @xmath0 ( see figure [ fig:1 ] ) , it is quite natural to expect the rate of convergence of the eigenvalues to rely essentially on the capacity of the junction points and hence to be of order @xmath1 , being @xmath2 the space dimension . [ fig:1 ] ( -5.5,-3.5)(5.5,3.5 ) ( -0.75,1)(-0.6,3.5)(1.2,2)(1,0)(0.8,-2.4)(-0.6,-3.5)(-1.5,-1 ) ( -1.5,-1)(-1.2,0.5)(-1,-0.5)(-0.75,1 ) ( -0.25,-2.4)(1.05,0.5 ) ( 0,2.4)(1,0.5 ) ( -2.6,1)(-4.6,3.5)(-5.2,2)(-4.8,0)(-4.2,-3.5)(-2.8,-1 ) ( -2.6,1)(-2.4,0.5)(-2.65,-0.5)(-2.8,-1 ) ( -4.05,-2.4)(0.8,0.4 ) ( -4.25,2.4)(0.95,0.5 ) ( -4.05,-1.4)@xmath3 ( -0.05,-1.4)@xmath4 ( -5.5,-3.5)(5.5,3.5 ) ( -0.75,1)(-0.6,3.5)(1.2,2)(1,0)(0.8,-2.4)(-0.6,-3.5)(-1.5,-1 ) ( -1.5,-1)(-1.2,0.5)(-1,-0.5)(-0.75,1 ) ( 0,0.5)(0.2,0.4)(0.25,0.25)(0.15,-0.4)(0,-0.5 ) ( 0,0.5)(-0.1,0.4)(-0.15,0.25)(-0.25,-0.4)(0,-0.5 ) ( -0.25,-2.4)(1.05,0.5 ) ( 0,2.4)(1,0.5 ) ( -2.6,1)(-4.6,3.5)(-5.2,2)(-4.8,0)(-4.2,-3.5)(-2.8,-1 ) ( -2.6,1)(-2.4,0.5)(-2.65,-0.5)(-2.8,-1 ) ( -4,1)(-4.3,0.9)(-4.25,0.8)(-4.15,0)(-4,0 ) ( -4,1)(-3.8,0.9)(-3.7,0.8)(-3.65,0.1)(-4,0 ) ( -4,1)(-3.5,0.9)(-2,-0.5)(-0.2,0.5)(0,0.5 ) ( -4,0)(-3.6,-0.1)(-1.9,-1.2)(-1.5,-1)(-0.4,-0.5)(0,-0.5 ) ( -4.05,-2.4)(0.8,0.4 ) ( -4.25,2.4)(0.95,0.5 ) ( -4.05,-1.4)@xmath3 ( -0.05,-2)@xmath4 ( -2,-0.5)@xmath5 ( 0,0)1pt ( 0,0)@xmath6 ( -2,-0.85)(0.2,0.4 ) ( -2,-2)@xmath7 ( -2,-0.85)(-2.3,-1)(-2.5,-1.65 ) referring to figure [ fig:1 ] , let @xmath8 be the @xmath9-th eigenfunction on the limit ( disconnected ) domain @xmath10 completely supported only in the connected component @xmath4 . by the attachment of a handle @xmath5 with cross section of radius of order @xmath11 , its mass will be pushed into the channel in order to spread over the new entire domain @xmath12 . besides the tubular shape of the connecting tube , we require , as a basic assumption to start our analysis , that the handle is attached at a point @xmath13 of @xmath14 where @xmath8 has a zero of order one , i.e. its normal derivative is different from zero . if moreover @xmath8 is simple and suitably normalized , the corresponding eigenvalue @xmath15 can be continued into a family @xmath16 of eigenvalues corresponding to normalized eigenfunctions @xmath17 on the perturbed domain . we prove that , in such a setting , there exists the limit @xmath18 where @xmath19 is a positive constant depending only the geometry of the junction section @xmath7 ( see theorems [ teo_asintotico_autovalori ] and [ teo_asintotico_autofunzioni ] below ) . thinking to the eigenfunction @xmath8 as pushed into the channel , we can imagine that a force acts over the junction between the channel and the domain where @xmath8 is supported . the constant @xmath19 represents indeed the _ compliance _ of the channel s junction , under a constant force concentrated at the junction section ; the compliance , which can be expressed as the @xmath20-norm of the trace of a suitable harmonic function over the channel section , measures the faculty of an elastic membrane to adjust or to resist to a force applied on the section , see for a precise definition . our proof consists in a sharp differentiation with respect to the parameter , which requires first a careful analysis of the transition functions which have to be attached to @xmath8 in order to push it over the channel . in this way , we will prove that , once more , @xmath21 where @xmath17 and @xmath8 are trivially extended to the whole @xmath22 . as an application of the sharp asymptotics we are able to treat also the resonant case : if @xmath23 is a double eigenvalue on the limit disconnected domain which is a simple eigenvalue both on @xmath3 and on @xmath4 , an asymptotics for eigenvalues of type still holds if the limit problem is asymmetrical , e.g. under the assumption that the normal derivatives of the limit eigenfunctions at the junctions are different from each other ( see theorem [ t : reso ] ) . in this case , it turns out that the splitting of the two subsequent eigenvalues @xmath24 has the polynomial vanishing order @xmath1 ( see remark [ r : splitting_polinomiale ] ) ; such result complements those in @xcite , where it was proved that , in a symmetric dumbbell domain with a shrinking handle , the splitting of the first two eigenvalues vanishes with exponential rate . moreover , in contrast with @xcite , we can localize each approximating eigenfunction on its corresponding region , up to an exponentially vanishing tail , see theorem [ t : reso_autofun ] . for expository reasons , the present paper discusses the effect of attaching a thin handle on the spectral rate of convergence only for dumbbell domains . however , up to minor modifications , the results obtained here hold true in quite general contexts , since they rely essentially on the attachment of a shrinking handle at a point in which the limit eigenfunction has a zero of order @xmath25 ; therefore the presence / lack of a second domain beyond the channel and its shape seem to be irrelevant for the validity of the asymptotics we are going to derive . the choice of focusing on the dumbbell structure is motivated not only by the large attention devoted to this peculiar case of singularly perturbed domain in the literature , due to the many interesting related spectral phenomena ( see [ sec : motiv - refer - liter ] below ) , but also by the fact that some preliminary results required in our analysis have been obtained for dumbbell domains in @xcite , where the singular asymptotic behavior of eigenfunctions at the second junction of the tube is described . as a paradigmatic example , we consider a dumbbell domain where each `` chamber '' has a constant section , namely we straighten out the handle and assume its section @xmath7 to be constant along its whole length , whereas we spread out the two domains @xmath4 and @xmath3 assuming they are two entire half - spaces , see figure [ fig : dd ] . we observe that such a simplification of the domain s geometry does not imply a substantial loss of generality if a suitable weight is introduced in the eigenvalue problem under investigation : indeed , the effect of a diffeomorphism transforming a generic dumbbell in a dumbbell with two half - spaces as chambers is the transformation of the eigenvalue problem into a weighted one . let @xmath26 . we denote @xmath27 and , for all @xmath28 , @xmath29 where @xmath30 , @xmath31 , and @xmath32 denotes the ball of radius @xmath33 centered at @xmath34 . let @xmath35 be an open bounded set with @xmath36-boundary containing @xmath37 . for simplicity of notation , we assume that @xmath7 satisfies @xmath38 let @xmath39 be a weight satisfying @xmath40 assumption is required for technical reasons as in @xcite ; it is used in [ sec : point - wise - energy ] to prove some preliminary estimates of eigenfunctions on the perturbed domain . for every open set @xmath41 , we denote as @xmath42 the set of the diverging eigenvalues @xmath43 ( where each @xmath44 is repeated as many times as its multiplicity ) of the weighted eigenvalue problem @xmath45 it is easy to verify that @xmath46 . let @xmath47 be the domain formed by connecting the two half - spaces @xmath48 with a tube of length @xmath25 and cross - section @xmath49 , i.e. @xmath50 where @xmath51 and @xmath52 , see figure [ fig : dd ] . .,width=264 ] here and in the sequel , for every open set @xmath53 , @xmath54 denotes the functional space obtained as completion of @xmath55 with respect to the dirichlet norm @xmath56 . by standard minimization methods , it is easy to prove that the minimum @xmath57 is achieved , where @xmath58 and @xmath59 it is easy to verify that @xmath60 see corollary [ cor : min_phi1 ] . moreover , we notice that , denoting as @xmath61 the dual space of @xmath62 and letting @xmath63 defined as @xmath64 @xmath65 can be rewritten as @xmath66 where @xmath67 represents the compliance functional associated to the force @xmath68 concentrated on the section @xmath7 in the flavor of @xcite . in general , the compliance functional measures the rigidity of a membrane subject to a given ( vertical ) force : the maximal rigidity is obtained by minimizing the compliance functional @xmath19 in the class of admissible regions @xmath7 . with this notation and concepts in mind we state our main results . let us first assume that there exists @xmath69 such that @xmath70 we can then fix an eigenfunction @xmath71 associated to @xmath72 , i.e. solving @xmath73 such that @xmath74 from ( * ? ? ? * example 8.2 , corollary 4.7 , remark 4.3 ) ( see also ( * ? ? ? * lemma 1.1 ) ) , it follows that , letting @xmath75 where @xmath76 , so that @xmath77 , there holds @xmath78 we will denote @xmath79 furthermore , for every @xmath11 sufficiently small , @xmath80 is simple and there exists an eigenfunction @xmath17 associated to @xmath80 , i.e. satisfying @xmath81 such that @xmath82 where in the above formula we mean the functions @xmath83 to be trivially extended to the whole @xmath22 . we refer to @xcite for uniform convergence of eigenfunctions . [ teo_asintotico_autovalori ] under assumptions , , , , and , let @xmath84 be the @xmath85-th eigenvalue of problem on @xmath86 ( which is equal to the simple @xmath87-th eigenvalue on @xmath4 ) and @xmath88 be the @xmath85-th eigenvalue of problem on the domain @xmath89 defined in . then @xmath90 with @xmath8 as in and , and @xmath19 as in . [ teo_asintotico_autofunzioni ] under assumptions , , , , and , let @xmath17 and @xmath8 as in , , , , . then @xmath91 where @xmath17 and @xmath8 are trivially extended to the whole @xmath22 and @xmath19 is defined in . we observe that , once the measure of the section @xmath7 is fixed , the shape minimizing @xmath65 and hence maximizing both the limits @xmath92 and @xmath93 is the spherical one , as we will show in proposition [ p : steiner ] by steiner rearrangement . hence , the disk - shaped section of the tube is the one which makes as slow as possible the convergence of the eigenvalues on the perturbed domain to the eigenvalues on the limit domain , as the handle thickness shrinks to zero . in other words , this means that among all the admissible sections @xmath7 , the disk attains the minimum of the rigidity of the domain @xmath94 : from the opposite point of view , in the case of a round section , the eigenfunctions located in the right domain @xmath4 are the most sensitive to the attachment of the thin handle if compared to the case of more indented sections . this phenomenon can be read in theorems [ teo_asintotico_autovalori ] and [ teo_asintotico_autofunzioni ] , since the limits in and attain their maximal ( positive ) constant at a disk - shaped section : symmetrization of the section makes the difference @xmath95 and @xmath96 drift away from being @xmath97 . the proof of theorem [ teo_asintotico_autovalori ] , which is presented in section [ sec : proof - theor - reft ] , is based on the courant - fisher _ minimax characterization _ of eigenvalues : the estimates from above and below of the rayleigh quotient used to prove the theorem are based on the analysis of proper test functions introduced in section [ sec : prel - notat - techn ] . theorem [ teo_asintotico_autofunzioni ] is proved in section [ sec : rate - conv - eigenf ] , using some blow - up analysis developed in section [ sec : prel - notat - techn ] and the invertibility of an operator associated to the eigenvalue problem on @xmath4 ( see ) . in section [ sec : resonant - case ] we drop assumption and assume that @xmath98 is a simple eigenvalue on @xmath3 , a simple eigenvalue on @xmath4 , and a double eigenvalue on @xmath10 . we prove that by attaching the shrinking handle at two points where the normal derivatives of the limit eigenfunctions are different from each other , the double eigenvalue @xmath99 on the limit domain is approximated by two different branches of eigenvalues on the perturbed domain as @xmath100 . [ t : reso ] let us assume that , , hold and @xmath101 for all @xmath102 . let @xmath103 be a simple eigenvalue on @xmath3 with corresponding eigenfunctions having in @xmath104 a zero of order @xmath25 , a simple eigenvalue on @xmath4 with corresponding eigenfunctions having in @xmath105 a zero of order @xmath25 , and a double eigenvalue on @xmath10 . let @xmath106 and @xmath107 be the eigenfunctions associated to @xmath108 on @xmath4 and @xmath3 respectively satisfying @xmath109 if @xmath110 then @xmath111 and @xmath112 where @xmath89 is defined in and @xmath19 is defined in . in section [ sec : resonant - case ] , we also prove that , in the resonant case , under condition , each approximating eigenfunction is localized as @xmath100 on the corresponding component of the limit domain , i.e. an asymmetrical limit configuration prevents dumbbell eigenfunctions from spreading their mass over both components and forces them to concentrate in one of the two regions . [ t : reso_autofun ] under the same assumptions and with the same notations of theorem [ t : reso ] , there exist two continuously parametrized families @xmath113 of eigenfunctions on @xmath89 , i.e. soutions to @xmath114 for @xmath115 and @xmath116 respectively , such that @xmath117 moreover , for @xmath118 and @xmath119 there holds @xmath120 where @xmath121 and @xmath122 respectively , and @xmath123 denotes the first eigenvalue of the laplace operator on @xmath7 under null dirichlet boundary conditions . for the two families of eigenfunctions @xmath124 we provide a sharp asymptotics , extending the result of theorem [ teo_asintotico_autofunzioni ] in the resonant asymmetrical case . [ t : reso_autofun_conv ] under the same assumptions and with the same notations of theorem [ t : reso ] , let @xmath113 be as in theorem [ t : reso_autofun ] . then @xmath125 where @xmath126 , and @xmath127 are trivially extended to the whole @xmath22 and @xmath19 is defined in . the continuity of eigenvalues and eigenfunctions of the laplace operator under dirichlet boundary conditions in varying domains including the dumbbell case has been studied in @xcite . we also refer to @xcite for a first result about spectral continuity for less general domain s perturbations and to @xcite ( and references therein ) for a detailed survey . as far as estimates of the rate of convergence are concerned , we mention @xcite , where , among other results , the authors prove that , in the case of a helmholtz resonator with a cavity , the effect of adding a tubular region with a section of radius of order @xmath11 is to shift the eigenvalues by a small amount of order at most @xmath128 . this generalizes a previous result of @xcite where an @xmath128-rate of convergence for resonances of a helmholtz resonator was obtained in dimension @xmath129 . we stress that the case treated in present paper does not allow continuos spectrum for the dirichlet laplacian . as far as we know , no sharp estimates similar to ours can be found in the literature . similar to our settings , we mention @xcite which contains an @xmath130-bound from above for the a rate of convergence , but not the exact asymptotics . some other estimates on the rate of convergence of dirichlet eigenvalues for different domain s perturbations can be found in @xcite . we note that there exists an extensive literature dealing with neumann boundary conditions , but , in the case of dumbbell domains with thin handles , the eigenvalues of the laplacian may not be continuous , as observed in @xcite ( see also @xcite ) . spectral analysis in thin branching domains arises naturally in the study of models of propagation of waves in quasi one - dimensional systems : in this framework we meet the theory of quantum graphs which provide simplified models of quantum wires , photonic crystals , carbon nano - structures , thin waveguides and many other problems , see e.g. @xcite for details . similarly as in quantum graph theory , in this paper we address to systems which are composed by different chambers communicating by connecting regions and which are governed by certain differential equations . we mention that , besides their own theoretical interest in the framework of spectral theory for elliptic operators , such issues are also related to some engineering problems : elasticity problems in heterogeneous materials and limit problems at the junctions of several domains with different limit dimensions ( namely thin plates with beams or rods ) , see e.g. @xcite . the proof of theorem [ teo_asintotico_autovalori ] is based on the courant - fisher _ minimax characterization _ of eigenvalues and some estimates from above and below on the associated rayleigh quotient computed at suitable test functions . in this section we introduce the proper test functions on which the rayleigh quotient will be estimated to prove upper / lower bounds , and prove some properties ( i.e. point - wise estimates , blow - up analysis ) of such test functions and of eigenfunctions on the domain @xmath89 . we start by introducing some functions describing the domain s change of geometry at the junction , which will be used for the construction of super - solutions needed for deriving point - wise estimates on eigenfunctions and for estimating the rayleigh quotient associated to the eigenvalue problem . more precisely , we consider * the unique function @xmath131 which is harmonic in the domain @xmath94 , has finite energy in @xmath132 , and behaves as @xmath133 as @xmath134 in @xmath4 ( here @xmath135 denotes the positive part of @xmath136 for all @xmath137 ) ; * for every @xmath138 , the function @xmath139 defined as the harmonic extension of @xmath140 in the domain @xmath141 vanishing on @xmath142 ; * for every @xmath138 , the function @xmath143 defined as the harmonic extension of @xmath144 in the domain @xmath145 vanishing on @xmath146 . for all @xmath147 , we denote as @xmath148 the completion of @xmath149 with respect to the norm @xmath150\times{{\mathbb r}}^{n-1})\cup b_r^+}|\nabla v|^2dx\big)^{1/2}$ ] , i.e. @xmath148 is the space of functions with finite energy in @xmath151\times{{\mathbb r}}^{n-1})\cup \overline{b_r^+}$ ] vanishing on @xmath152 . in the sequel , we also denote as @xmath123 the first eigenvalue of the laplace operator on @xmath7 under null dirichlet boundary conditions , and as @xmath153 the corresponding positive @xmath154-normalized eigenfunction , so that @xmath155 being @xmath156 , @xmath157 . in ( * ? ? * lemma 2.4 ) , it is proved that there exists a unique function @xmath131 satisfying @xmath158 -\delta \phi=0\text { in a distributional sense in } \widetilde d , \quad \phi=0\text { on } \partial \widetilde d,\\[5pt ] \int_{d^+}|\nabla ( \phi-(x_1 - 1))(x)|^2\,dx<+\infty . \end{cases}\ ] ] furthermore @xmath159 in @xmath94 ( by the strong maximum principle ) and , by ( * ? ? ? * lemma 2.9 ) , there holds @xmath160 let @xmath161 being @xmath162 the unit @xmath163-dimensional sphere and @xmath164 where @xmath165 denotes the volume of the unit sphere @xmath166 , i.e. @xmath167 . here and in the sequel , the notation @xmath168 is used to denote the volume element on @xmath163-dimensional surfaces . we notice that , letting @xmath169 @xmath170 is the first positive @xmath171-normalized eigenfunction of @xmath172 on @xmath173 under null dirichlet boundary conditions satisfying @xmath174 [ lemma_phi_1 ] let @xmath131 , @xmath175 , and @xmath176 be as in , , and respectively , and , for every @xmath177 , let us define @xmath178 then @xmath179 part ( i ) is proved in ( * ? ? ? * lemma 2.2 ) . to prove ( ii ) we observe that @xmath180 so that the thesis immediately follows from differentiation of ( i ) and simple calculations . [ l : secondo_lemma_phi_1 ] let @xmath131 as in and @xmath181 as in . then @xmath182 { \rm ( ii)}\quad & j_\sigma ( \phi-(x_1 - 1)^+)=\min_{w\in \mathcal d^{1,2}(\widetilde d ) } j_\sigma(w);\\ { \rm ( iii)}\quad & \int_{\widetilde d}\nabla(\phi-(x_1 - 1)^+)\cdot\nabla v(x)\,dx=\int_\sigma v(1,x')\,dx',\quad\text{for every } v\in \mathcal d^{1,2}(\widetilde d).\end{aligned}\ ] ] let @xmath183 , @xmath184 . from we have that @xmath185 in @xmath4 , @xmath186 in @xmath4 , @xmath187 on @xmath188 , and @xmath189 . then , ( i ) follows from ( * ? ? ? * theorem 1.5 ) applied to the function @xmath190 . statements ( ii ) and ( iii ) are contained in ( * ? ? ? * lemma 2.4 ) and @xcite . as a consequence of the previous lemma , it is possible to characterize the minimum @xmath65 defined in in terms of the function @xmath131 . [ cor : min_phi1 ] let @xmath191 as in , @xmath65 as in , and @xmath131 as in . then @xmath192 from lemma [ l : secondo_lemma_phi_1](ii ) , we have that @xmath193 and the conclusion follows taking @xmath194 in lemma [ l : secondo_lemma_phi_1](iii ) . for every @xmath147 , we denote as @xmath139 the unique solution to the minimization problem @xmath195 which then solves @xmath196 [ lemma_z_r ] let @xmath139 , @xmath175 , and @xmath176 be as in , and respectively , and , for every @xmath138 and @xmath197 $ ] , let us define @xmath198 then @xmath199$}.\ ] ] we first observe that @xmath200.\ ] ] since @xmath139 is harmonic in @xmath141 , there exists @xmath201 such that @xmath202 in @xmath203 $ ] , so that @xmath204$.}\ ] ] from the regularity of @xmath139 in @xmath105 , we deduce @xmath205 and then @xmath206 . hence @xmath207,\ ] ] the thesis follows from and . for every @xmath147 , we denote as @xmath143 the unique solution to the minimization problem @xmath208 where @xmath209 is the completion of @xmath210 with respect to the norm @xmath211 . this function then solves @xmath212 [ lemma_v_r ] for every @xmath138 , let @xmath143 , @xmath175 , and @xmath176 be as in , , and respectively , and , for every @xmath213 $ ] , let us define @xmath214 then @xmath215 to prove ( i ) , for any @xmath216 , we estimate @xmath217 via the dirichlet principle since @xmath218 is harmonic in @xmath219 and @xmath220 , being @xmath221 a smooth cut - off function such that @xmath222 in view of and @xmath223 as @xmath224 , thus proving ( i ) . to prove ( ii ) , we observe that @xmath225.\ ] ] from , @xmath226 solves the equation @xmath227 in the interval @xmath228 $ ] , hence by integration we obtain that there exists @xmath201 such that @xmath229.\ ] ] replacing @xmath230 in the above identity and observing that the boundary condition in implies that @xmath231 we obtain that @xmath232 hence @xmath233,\ ] ] and then @xmath234.\ ] ] the conclusion follows by plugging @xmath230 in and . in order to prove theorem [ teo_asintotico_autovalori ] , quite precise decaying estimates of eigenfunctions on the varying domain are needed . in this subsection we pursue this analysis . [ controllo_supersol ] let @xmath235 and , for all @xmath236 , let @xmath237 solve @xmath238 1 . for every sequence @xmath239 there exist a subsequence @xmath240 and @xmath241 solving @xmath242 such that @xmath243 in @xmath244 . 2 . there exists @xmath245 and @xmath246 such that , for all @xmath247 and @xmath248 with @xmath249 , there holds @xmath250 from the spectral continuity analyzed in @xcite , @xmath251 ; hence the proof of ( i ) follows easily from classical compactness argument in view of the compactness of the map @xmath252 , @xmath253 . estimate follows by an iterative brezis - kato type argument ( see e.g. ( * ? ? ? * lemma 2.2 ) ) . from ( * ? ? ? * lemma 5.2 ) it follows that solutions to converging in @xmath244 actually converge in @xmath254 ; then @xcite and part ( i ) imply that for every sequence @xmath255 there exist a subsequence @xmath240 such that @xmath256 in @xmath257 and hence , for every @xmath147 , latexmath:[\[\lim_{k\to+\infty}\sup_{b_{r\eps_{n_k}}^+ \cup b_{r\eps_{n_k}}^-\cup \mathcal c_{\eps_{n_k } } } sequence nor on the subsequence , we deduce the limit as @xmath0 thus proving . estimate , together with classical elliptic estimates for @xmath259 over an annulus @xmath260 and @xmath261 over @xmath262 , yield . to prove estimate , let us consider the function @xmath263 where @xmath264 we note that @xmath265 as @xmath266 in view of estimate . for all @xmath267 , @xmath268 so that @xmath269 on @xmath270 . moreover @xmath271 whereas , via kato s inequality @xcite , @xmath272 so that there exists a constant @xmath273 independent of @xmath11 such that , for @xmath11 sufficiently small , @xmath274 weakly solves @xmath275 the boundary conditions imply that @xmath276 ; hence testing with @xmath277 and using hlder and sobolev inequalities , we obtain that @xmath278 where @xmath279 denotes the best constant in the sobolev inequality @xmath280 and @xmath281 is the lebesgue @xmath163-dimensional measure of @xmath7 . if @xmath282 , the above estimate would imply that @xmath283 for @xmath11 small , thus giving rise to a contradiction . then @xmath284 and @xmath285 in @xmath5 , which implies estimate . [ stime_integrali_vjeps ] let @xmath286 be as in lemma [ controllo_supersol ] . then for every @xmath287 and @xmath147 @xmath288 estimate is a a straightforward consequence of , whereas follows directly from . in order to prove , we observe that estimates and imply that @xmath289 as @xmath290 . for @xmath291 , we define @xmath292 we recall the following result stated in @xcite . [ lemma_zero ] let @xmath17 be as in under the assumptions , , , , , . then for every @xmath293 and @xmath294 , there exists @xmath295 such that for all @xmath296 and @xmath297 @xmath298 let @xmath17 be as in lemma [ lemma_zero ] . then , for every @xmath299 @xmath300 let @xmath301 be a smooth cut - off function such that @xmath302 in @xmath303 , @xmath304 in @xmath305 , @xmath306 , and @xmath307 . let @xmath308 . then , taking into account , @xmath309 from which the thesis follows invoking estimate of corollary [ stime_integrali_vjeps ] and lemma [ lemma_zero ] . for every @xmath247 and @xmath147 , let @xmath310 be the unique solution to the minimization problem @xmath311 where @xmath312 is the completion of @xmath313 with respect to the norm @xmath314 ; in particular @xmath310 solves @xmath315 in a similar way , for every @xmath247 and @xmath147 , we denote as @xmath316 the unique solution to the minimization problem @xmath317 in particular @xmath316 solves @xmath318 hence , for every @xmath247 and @xmath147 , we define @xmath319 as @xmath320 and @xmath321 as @xmath322 [ l : stimaeste ] let @xmath147 and @xmath323 be as in . then , for every @xmath324 , latexmath:[\[\label{eq:4 } as @xmath0 uniformly in @xmath326 . moreover @xmath327 from and the maximum principle , it follows that , for @xmath11 sufficiently small , @xmath328 we argue as in the proof of estimate in lemma [ controllo_supersol ] . let us consider the function @xmath329 from we obtain that , for all @xmath330 , @xmath331 so that @xmath332 on @xmath270 . moreover , @xmath333 whereas @xmath323 is nonnegative and harmonic in @xmath5 , so that @xmath334 on @xmath5 and , by the maximum principle , we deduce that @xmath335 from which estimate follows . let @xmath301 be a smooth cut - off function such that @xmath302 in @xmath303 , @xmath304 in @xmath305 , @xmath306 , and @xmath307 . let @xmath336 . then @xmath337 from which follows invoking . for all @xmath147 and @xmath247 , let us define @xmath338 [ blow_up_vari ] the following convergences hold as @xmath100 : @xmath339 with @xmath139 and @xmath143 being as in and respectively . the convergence follows from ( * ? ? ? * lemma 4.1 and corollary 4.4 ) and ( * ? ? ? * lemmas 2.1 and 2.4 ) . in order to prove , we notice that @xmath340 furthermore , for every @xmath28 , @xmath341 for some @xmath342 independent of @xmath11 and @xmath33 . then , by a diagonal process , one can easily prove that , up to subsequences , @xmath343 weakly converges in @xmath344 for all @xmath28 . by elliptic regularity theory we conclude that @xmath343 converges to its point - wise limit in @xmath345 ( since such a limit does not depend on the subsequence , the convergence actually holds as @xmath100 ) . in order to prove , we notice that @xmath346 solves @xmath347 and , by the dirichlet principle and , @xmath348 where @xmath301 is a smooth cut - off function such that @xmath304 in @xmath349 , @xmath302 in @xmath350 . then @xmath351 as @xmath100 in @xmath352 and convergence is proved . in order to prove , we first notice that , in view of , @xmath353 some @xmath354 independent of @xmath11 . then , up to subsequences , @xmath355 weakly in @xmath148 and strongly in @xmath356 as @xmath357 for some @xmath358 which is is harmonic in @xmath359 . since , by , @xmath360 in @xmath356 , we conclude that @xmath361 ; in particular , since the weak @xmath148-limit of @xmath362 does not depend on the subsequence , the convergence actually holds as @xmath100 . moreover , by standard interior elliptic estimates , it is easy to prove that the convergence is strong in @xmath363 for every @xmath364 . in addition , we can prove that @xmath365 indeed , since @xmath366 is harmonic on @xmath367 , then its energy is less or equal to the energy of any other @xmath368-function with the same boundary conditions on @xmath369 . in particular , letting @xmath301 be a smooth cut - off function such that @xmath304 in @xmath349 and @xmath302 in @xmath350 , and @xmath370 be a smooth cut - off function such that @xmath371 in @xmath349 and @xmath372 in @xmath373 , in view of and @xmath363-convergence for @xmath364 , we obtain that @xmath374 hence @xmath375 in @xmath376 , which , together with @xmath363-convergence for @xmath364 , implies . [ rem : hdiv ] convergences and together with the normal trace embedding theorem for @xmath377 ( see e.g. ( * ? ? ? * chapter 20 ) ) , imply that , for all @xmath138 , @xmath378 where @xmath379 is the normal external unit vector to @xmath380 . as a straightforward corollary of the blow - up analysis performed in lemma [ blow_up_vari ] , we obtain the following result , which will play a crucial role in the proof of theorem [ teo_asintotico_autofunzioni ] . [ stima_autofunzioni_via_blow_up ] under assumptions ( [ eq : p][eq:54 ] ) , let @xmath17 and @xmath8 as in ( [ problema][convergenza_u_0 ] ) and ( [ eq : u0][eq:13 ] ) . then @xmath381 for all @xmath138 . the thesis follows from and through a change of variable . let us recall and fix some notation we are going to use throughout this section . we recall that @xmath88 denotes the @xmath85-th eigenvalue of problem on the domain @xmath89 and @xmath382 denotes the @xmath85-th eigenvalue on @xmath86 which is equal to the simple @xmath87-th eigenvalue on @xmath4 . let @xmath17 be the eigenfunction on @xmath89 associated to @xmath80 satisfying and . for every @xmath383 , we fix an eigenfunction @xmath384 associated to @xmath385 on @xmath89 such that @xmath386 and an eigenfunction @xmath387 associated to the eigenvalue @xmath388 on @xmath10 such that @xmath389 . in particular , we can choose such eigenfunctions in such a way that @xmath390 in the sequel we will denote @xmath388 as @xmath391 and @xmath385 as @xmath392 ( we recall that the eigenvalues are repeated as many times as their own multiplicity ) . the proof of theorem [ teo_asintotico_autovalori ] is based on the following preliminary result . [ t : asintotico_autovalori1 ] under assumptions , , , , and , let @xmath88 be the @xmath85-th eigenvalue of problem on the domain @xmath89 defined in and @xmath84 be the @xmath85-th eigenvalue of problem on @xmath86 ( which is equal to the simple @xmath87-th eigenvalue on @xmath4 ) . then @xmath393 where @xmath131 is defined in . we observe that a straightforward consequence of the minimax principle for eigenvalues is that @xmath394 . we are going to prove first two estimates for the quantity @xmath395 , one from below and one from above , in order to reach , for every @xmath138 , an estimate of the type @xmath396 for some constants @xmath397 depending @xmath11 and @xmath398 ; secondly , we will prove that @xmath399 thus implying the stated asymptotics . * step 1 : * estimate from below . from the courant - fisher _ minimax characterization _ of the dirichlet eigenvalues , we have that @xmath400 let @xmath138 . if we choose the space @xmath401 ( where the functions @xmath402 are trivially extended to the whole @xmath89 ) , we have that @xmath403 and then @xmath404 in view of the estimate @xmath405 which holds by lemma [ l : stimaeste ] and sobolev inequality . then @xmath406 we observe that , since @xmath407 is simple , for all @xmath408 @xmath409 from convergences and established in lemma [ blow_up_vari ] and remark [ rem : hdiv ] , it follows that @xmath410 where @xmath411 for every @xmath408 let us denote @xmath412 in view of the orthogonality in @xmath413 between @xmath402 and @xmath8 @xmath414 taking into account lemma [ l : stimaeste ] and the fact that @xmath415 by dirichlet principle and . we claim that @xmath416 to prove , let @xmath417 , @xmath418 , @xmath419 , be such that @xmath420 and @xmath421 we first prove that @xmath422 indeed from @xmath423 , and , it follows that @xmath424 which implies that @xmath425 . assuming by contradiction that does not hold , there should exists a sequence @xmath426 such that @xmath427 . then , up to subsequences , there would exist @xmath428 such that @xmath429 . therefore and would imply @xmath430 i.e. @xmath431 as @xmath432 , thus contradicting @xmath428 . estimate is thereby proved . from we deduce that @xmath433 as @xmath100 , then from , , , and it follows that @xmath434 thus proving claim . from and , we deduce that @xmath435 and hence , for every @xmath138 , @xmath436 where , for every @xmath138 , @xmath437 * step 2 : * estimate from above . by the courant - fisher _ minimax characterization _ of the eigenvalue @xmath438 , we have that @xmath439 let @xmath138 and @xmath440 be a smooth cut - off function such that @xmath441 in @xmath442 , @xmath443 in @xmath444 , @xmath445 and @xmath446 in @xmath10 . we choose the @xmath85-dimesional space @xmath447 in . then @xmath448 we notice that @xmath449 while , from lemma [ lemma_zero ] , corollary [ motivazione_o_grande ] , assumption , and corollary [ stime_integrali_vjeps ] it follows that @xmath450 then @xmath451 where we have set @xmath452 let us study each coefficient of the quadratic form above . from , , and corollary [ motivazione_o_grande ] , we have that @xmath453 and hence , in view of lemma [ blow_up_vari ] and remark [ rem : hdiv ] , @xmath454 where @xmath455 for every @xmath456 , in view of estimates and of lemma [ controllo_supersol ] , we have that @xmath457 from , estimate of lemma [ controllo_supersol ] and estimates and of corollary [ stime_integrali_vjeps ] we deduce that @xmath458 for every @xmath459 such that @xmath460 , from , estimates and of lemma [ controllo_supersol ] and the orthogonality of @xmath461 and @xmath462 in @xmath463 , it follows that @xmath464 from , , and ( * ? ? ? * lemma 2.10 and 2.11 ) , we have that @xmath465 from , , , , and ( * ? ? ? * lemma 2.10 ) ( which in particular yields @xmath466 as @xmath0 ) , it follows that , for every @xmath456 , @xmath467 we claim that @xmath468 as @xmath0 . to prove , let @xmath418 , such that @xmath469 and @xmath470 from @xmath471 it follows that @xmath472 and hence , in view of and , @xmath473 from , , and , it follows that @xmath474 since @xmath475 , we obtain that , for every @xmath456 , @xmath476 from , , and , it follows that @xmath477 in the case @xmath478 , and imply that @xmath479 as @xmath100 , hence for all @xmath480 @xmath481 . therefore , in view of and , we obtain that @xmath482 as @xmath100 , thus improving estimate for @xmath478 . then , for any dimension @xmath26 we obtain @xmath483 arguing a third time in the same way , from and the improved estimate on the mixed terms , we can improve and obtaining @xmath484 from , , , and , we thereby obtain @xmath485 thus implying @xmath486 from , , , and , it follows that @xmath487 thus proving claim . from and , we deduce that @xmath488 and hence , for every @xmath138 , @xmath489 where , for every @xmath138 , @xmath490 * step 3 : * asymptotic behavior . up to now we have proved the following estimate @xmath491 for any @xmath492 , where @xmath493 with @xmath494 and @xmath495 defined in and respectively . we now claim that @xmath496 as far as @xmath497 is concerned , we first observe that @xmath498 therefore , from lemma [ lemma_v_r](ii ) , we have that @xmath499 therefore , from lemma [ lemma_v_r](i ) , it follows that @xmath500 let us now study the limit of @xmath495 as @xmath224 . we split as @xmath501 we first prove that the second term in the sum vanishes as @xmath502 . indeed , @xmath503 testing equation @xmath504 in @xmath505 with @xmath506 , we have that @xmath507 as @xmath224 thanks to . on the other hand , testing the equation @xmath508 in @xmath141 with the function @xmath509 ( being @xmath221 a cut - off function as in ) , we have that @xmath510 thanks to the dirichlet principle and estimate . therefore , from , , , , lemmas [ lemma_z_r ] and [ lemma_phi_1 ] , and the fact that , in view of , , and , @xmath511 for all @xmath138 , it follows that @xmath512 combining and we prove claim . the conclusion follows from and , observing that @xmath513 due to the fact that , by the strong maximum principle , @xmath159 in @xmath4 . we are now ready to prove theorem [ teo_asintotico_autovalori ] . theorem [ teo_asintotico_autovalori ] from lemma [ lemma_phi_1 ] it follows that @xmath514 on the other hand , testing first equation @xmath515 in @xmath516 with @xmath517 and then equation @xmath518 in @xmath516 with @xmath519 , we obtain that @xmath520 combining and , we deduce that @xmath521 the conclusion follows from theorem [ t : asintotico_autovalori1 ] , , corollary [ cor : min_phi1 ] , and . steiner rearrangement allows proving that the shape of the section @xmath7 minimizing @xmath65 and hence maximizing @xmath522 is the spherical one . [ p : steiner ] for every @xmath35 being an open bounded domain containing @xmath37 , let @xmath65 be defined in . then , for every @xmath523 , @xmath524 where @xmath525 denotes the lebesgue measure of @xmath526 , @xmath527 denotes the @xmath163-dimensional ball of radius @xmath528 centered at @xmath104 , and @xmath529 denotes the volume of the unit @xmath530-dimensional sphere . for every @xmath35 being an open bounded domain containing @xmath37 , let us consider the domain @xmath531\times \sigma\big)\cup d^+$ ] and its steiner symmetral @xmath532 in codimension @xmath533 defined as @xmath534\times b'({\mathbf 0},r_\sigma ) \big)\cup d^+=d_{b'({\mathbf 0},r_\sigma)},\ ] ] where @xmath535 . for every @xmath536 , @xmath537 a.e . , its steiner rearrangement in codimension @xmath533 is the function @xmath538 defined as @xmath539 for every open bounded domain @xmath540 containing @xmath37 and @xmath536 such that @xmath541 a.e . , the plya - szeg inequality for the steiner rearrangement ( see e.g @xcite and @xcite ) implies that @xmath542 whereas the cavalieri principle yields @xmath543 therefore , letting @xmath544 , @xmath545 , we have that @xmath546 since the minimum of @xmath191 over @xmath547 is attained by a nonnegative function , we then conclude that @xmath548 for every open bounded domain @xmath540 containing @xmath37 , thus completing the proof . in this section we prove a sharp estimate for the rate of convergence of eigenfunctions . in view of corollary [ stima_autofunzioni_via_blow_up ] , it will be sufficient to obtain an estimate of @xmath549 . to this aim , we consider the following operator @xmath550 where the symbol @xmath551 stands for the @xmath552-norm , i.e. @xmath553 @xmath554 is the dual space of @xmath555 , and , for all @xmath556 , @xmath557 acts as @xmath558 we recall from that @xmath559 and hence @xmath560 . therefore @xmath561 . [ f_frechet ] under assumptions , , , , and , let @xmath84 be the @xmath85-th eigenvalue of problem on @xmath86 ( which is equal to the simple @xmath87-th eigenvalue on @xmath4 ) and @xmath8 be as in and . then , the operator @xmath562 defined in is frecht - differentiable at @xmath563 and its frecht - differential @xmath564 is invertible . for all @xmath565 , there holds @xmath566 as @xmath567 in @xmath568 . therefore @xmath562 is frecht - differentiable at @xmath563 and @xmath569 it remains to prove that @xmath570 is invertible . to this aim , by exploiting the compactness of the map @xmath571 , @xmath253 , it is easy to prove that , if @xmath572 is the riesz isomorphism and @xmath573 denotes the identity on @xmath574 , then the operator @xmath575 is a compact perturbation of the identity . therefore , from the fredholm alternative , @xmath576 is invertible if and only if it is injective . let @xmath565 be such that @xmath577 vanishes , i.e. @xmath578 i.e. @xmath579 therefore , @xmath580 and then , choosing @xmath581 in , we obtain that @xmath582 . it follows that @xmath583 is a weak @xmath552-solution to @xmath584 in @xmath4 . since , by assumption , the eigenvalue @xmath407 is simple on @xmath4 , we conclude that @xmath585 for some @xmath586 . from it follows that @xmath587 which implies @xmath588 and then @xmath589 in @xmath4 . we conclude that @xmath576 is injective and then invertible . [ stima_teo_inversione ] let @xmath590 be as in definition . then , for every @xmath591 and @xmath138 there exist @xmath592 such that @xmath593 \lim_{\eps\to 0^+}k(\eps , r)=k(r),\quad \text{for every } r>2,\quad \text{and}\quad \lim_{r\to+\infty}k(r)=0.\end{gathered}\ ] ] let us fix @xmath138 and notice that @xmath594 in @xmath552 as @xmath100 . indeed , from , , , , and , we deduce that @xmath595 as @xmath100 . therefore @xmath596 in view of lemma [ f_frechet ] , the operator @xmath576 is invertible ( and its inverse is continuous by the open mapping theorem ) , then implies that @xmath597 as @xmath100 . in order to prove the theorem , we are going to estimate the norm of @xmath598 as far as @xmath599 is concerned , from theorem [ teo_asintotico_autovalori ] and it follows that @xmath600 where @xmath601 is as in the proof of theorem [ t : asintotico_autovalori1 ] . in particular @xmath602 . as far as @xmath603 is concerned , we observe that , for every @xmath604 , @xmath605 thus , letting @xmath606 and @xmath607 as in and respectively , we have that @xmath608 where @xmath609 is defined as @xmath610 . since @xmath611 is an isometry of @xmath552 , we deduce that @xmath612 from the convergences and established in lemma [ blow_up_vari ] it follows that @xmath613 we observe that , for every @xmath614 such that @xmath615 , lemma [ l : secondo_lemma_phi_1](iii ) implies that @xmath616 since @xmath617 is harmonic in @xmath141 , @xmath618 , and vanishes on @xmath142 , if @xmath221 is a smooth cut - off function satisfying , from the dirichlet principle , , and , we can estimate @xmath619 as @xmath224 . from , , and we deduce that @xmath620 the conclusion follows combining with , , , , and . theorem [ teo_asintotico_autofunzioni ] let @xmath621 . from theorem [ stima_teo_inversione ] and , there exists @xmath622 such that @xmath623 from theorem [ stima_teo_inversione ] , it follows that @xmath624 and hence , from corollary [ stima_autofunzioni_via_blow_up ] and theorem [ stima_teo_inversione ] , we deduce that there exists @xmath625 such that , for all @xmath626 , @xmath627 thus proving that @xmath628 on the other hand , lemma [ l : secondo_lemma_phi_1](iii ) , corollary [ cor : min_phi1 ] , and imply that @xmath629 the proof is thereby complete . in this section we drop assumption and treat the case in which @xmath407 is a double eigenvalue on @xmath630 and a simple eigenvalue on each of the components @xmath4 and @xmath3 . to this aim , we exploit twice the sharp asymptotics provided by theorem [ teo_asintotico_autovalori ] in two domains obtained by attaching small handles to each chamber @xmath48 . besides and , we assume that @xmath39 satisfies @xmath631 and that there exist @xmath632 such that @xmath633 since @xmath634 , , , and imply that @xmath407 is a double eigenvalue on @xmath630 and hence there exists @xmath635 such that @xmath636 let @xmath637 and @xmath638 be the eigenfunctions associated to @xmath407 on @xmath4 and @xmath3 respectively , i.e. solving @xmath639 such that @xmath640 let us introduce the following domains @xmath641 we observe that the asymptotics of eigenvalues stated in theorem [ teo_asintotico_autovalori ] can be proved , up to minor modifications , replacing the dumbbell perturbed domain @xmath89 defined in with either the domain @xmath642 or @xmath642 , since the proof just relies on the attachment of a shrinking handle at a point in which the limit eigenfunction has a zero of order @xmath25 ; therefore , arguing as in the proof of theorem [ teo_asintotico_autovalori ] , we can prove that , under assumptions , , , , , and , there holds @xmath643 as @xmath100 , where @xmath19 is defined in . under the non - symmetry condition that the normal derivatives of the limit eigenfunctions at the two junctions are different , we observe that the double eigenvalue @xmath407 is approximated by two different branches of eigenvalues in @xmath644 , see figure [ fig:2 ] . [ prop_risonanza ] under assumptions , , , , , and , let @xmath645 and @xmath127 be as in and , and let @xmath646 , @xmath647 be as in , respectively . if @xmath648 then , for @xmath11 sufficiently small , @xmath649 where @xmath644 is defined in and @xmath85 is as in . we note that @xmath652 . expansions and together with assumptions , yield @xmath653 and hence implies that @xmath654 for @xmath11 sufficiently small , which gives the conclusion in view of the convergence of eigenvalues on @xmath644 proved by daners in @xcite . for the sake of brevity , we prove the lemma just for @xmath658 ; the proof for @xmath659 is similar . by the courant - fisher _ minimax characterization _ of eigenvalues we have that @xmath660 let @xmath661 be a smooth cut - off function such that @xmath662 for every @xmath663 and @xmath11 small , we fix an eigenfunction @xmath384 associated to @xmath385 on @xmath89 such that @xmath386 and @xmath664 if @xmath460 . choosing the @xmath85-dimesional space @xmath665 in we obtain that @xmath666 from corollary [ stime_integrali_vjeps ] it follows that , for every @xmath667 , @xmath668 whereas , exploiting the orthogonality of eigenfunctions , if @xmath460 we have that @xmath669 from , , and it follows that @xmath670 from corollary [ stime_integrali_vjeps ] it follows that @xmath671 @xmath672 for all @xmath673 , and , if @xmath460 , @xmath674 where in the third equality we have tested @xmath675 with @xmath676 for @xmath460 and integrated by parts over @xmath677 , using assumption . [ r : splitting_polinomiale ] we observe that , under assumption , expansions , , proposition [ prop_risonanza ] and lemma [ l : tildeomegaeps ] imply that the splitting of the two subsequent eigenvalues @xmath679 approximating the same double eigenvalue @xmath680 has a polynomial vanishing order , i.e. @xmath681 we emphasize that non - symmetry assumption is crucial for having a polynomial splitting : indeed it was proved in @xcite that in the case of a symmetric dumbbell domain the splitting of the first two eigenvalues vanishes with exponential rate . [ l : spectral_estimate ] under the same assumptions as in theorem [ t : reso ] , let @xmath687 be defined as @xmath688 where @xmath689 is the dual space of @xmath690 and @xmath691 is the dual space of @xmath692 . then , for @xmath11 sufficiently small , @xmath693 and @xmath694 are invertible and @xmath695 from , , and theorem [ t : reso ] , it follows that , for @xmath11 sufficiently small , @xmath696 and @xmath697 in particular , by assumption , we have that @xmath698 and @xmath699 , and therefore @xmath693 and @xmath694 are invertible . moreover , by classical spectral estimates , , and , it turns out that @xmath700 the proof is thereby complete . theorem [ t : reso_autofun ] due to simplicity of the eigenvalue @xmath701 on each component @xmath3 and @xmath4 , it is is enough to prove estimates and for any family of eigenfunctions @xmath702 on @xmath89 associated to @xmath703 and any family of eigenfunctions @xmath704 on @xmath89 associated to @xmath705 such that @xmath706 we prove only , being the proof of analogous . let @xmath661 be a smooth cut - off function such that @xmath707 a direct computation shows that , letting @xmath694 as in lemma [ l : spectral_estimate ] , @xmath708 where @xmath709 acts as @xmath710 from corollary [ stime_integrali_vjeps ] it follows that @xmath711 and then lemma [ l : spectral_estimate ] implies that @xmath712 estimate is thereby proved . theorem [ t : reso_autofun_conv ] we prove only , being the proof of analogous . to this aim , we first observe that , in view of the simplicity of the eigenvalue @xmath713 and by theorem [ teo_asintotico_autofunzioni ] ( adapted to the easier case of the perturbed domain @xmath646 ) , there exists a family of eigenfunctions @xmath714 on @xmath646 associated to @xmath651 such that @xmath715 and @xmath716 in view of , theorem [ t : reso_autofun ] , and corollary [ stime_integrali_vjeps ] , to prove it is enough to show that @xmath717 for some @xmath661 being a smooth cut - off function such that @xmath718 to prove , we argue as in section [ sec : rate - conv - eigenf ] and consider the operator @xmath719 where , for all @xmath720 , @xmath721 acts as @xmath722 we observe that @xmath723 and @xmath724 is frecht - differentiable at @xmath725 ; moreover , arguing as in the proof of lemma [ f_frechet ] , we can prove that , since @xmath651 is simple per small @xmath11 , the frecht - differential @xmath726 is invertible . due to the fact that @xmath651 converges to a simple eigenvalue on @xmath4 , it is also easy to verify that @xmath727 since @xmath728 from we deduce that , for @xmath11 small , @xmath729 we have that @xmath730 by direct calculations and using proposition [ prop_risonanza ] , lemma [ l : tildeomegaeps ] , corollary [ stime_integrali_vjeps ] , and theorem [ t : reso_autofun ] , we have that @xmath731 in order to estimate @xmath732 we observe that @xmath733 acts as @xmath734 from corollary [ stime_integrali_vjeps ] it follows that @xmath735 combining , , , and , we obtain , thus completing the proof . a. a. arsenev , _ on the singularities of the analytic continuation and the resonator properties of a solution of the dispersion problem for the helmholtz equation _ , zh . vychisl . fiz . 12 ( 1972 ) , 112138 . v. felli , a. ferrero , s. terracini , _ asymptotic behavior of solutions to schrdinger equations near an isolated singularity of the electromagnetic potential _ , j. eur . ( jems ) 13 ( 2011 ) , no . 1 , 119174 . t. kato , _ schrdinger operators with singular potentials _ , israel j. math . 13 ( 1972 ) , 135148 . proceedings of the international symposium on partial differential equations and the geometry of normed linear spaces ( jerusalem , 1972 ) .

the supranova model of gamma - ray bursts ( grbs ) , in which the grb event is preceded by a supernova ( sn ) explosion by a few months to years , has recently gained support from fe line detections in the x - ray afterglows of some grbs . a crucial ingredient of this model yet to be studied is the fast - rotating pulsar that should be active during the time interval between the sn and the grb , driving a powerful wind and a luminous plerionic nebula . we discuss some observational consequences of this precursor plerion , which should provide important tests for the supranova model : 1 ) the fragmentation of the outlying sn ejecta material by the plerion and its implications for fe line emission ; and 2 ) the effect of inverse compton cooling and emission in the grb external shock due to the plerion radiation field . the plerion - induced inverse compton emission can dominate in the gev - tev energy range during the afterglow , being detectable by glast from redshifts and by agile from redshifts , and distinguishable from self - compton emission by its spectrum and light curve . the prospects for direct detection and identification of the precursor plerion emission are also briefly considered . naoj - th - ap 2001 , no.66

astro-ph0111591

in the currently standard interpretation of gamma - ray bursts ( grbs ) , the central engine gives rise to a highly relativistic outflow , the ` fireball ' . although the presence of relativistic outflows in grbs has been amply demonstrated by multiwavelength observations of afterglows ( piran 1999 , van paradijs , kouveliotou & wijers 2000 , mszros 2001 ) , the nature of the central engine itself is still a great mystery . one model of the central engine that has recently gained attention is the ` supranova ' model of vietri & stella ( 1998 , hereafter vs98 ) , in which a massive star ends up in a supernova ( sn ) , but the subsequent black hole formation and grb event is delayed by some time @xmath2 , typically expected to be a few months to years . the sn is assumed to leave behind a rotationally - supported , ` supramassive ' neutron star ( smns ; hereafter also referred to simply as ` pulsar ' ) which then slowly shrinks by shedding angular momentum via a magnetospheric wind ( and/or gravitational radiation ; see [ sec : sum ] ) . after a spin - down time @xmath2 when roughly half its initial angular momentum has been lost , the configuration becomes unstable and collapses to a black hole , possibly with a surrounding disk , leading to the grb proper . the model s major advantage lies in the potential realization of a very baryon - clean pre - grb environment ( mandatory for generating sufficiently relativistic fireballs ) : first the sn ejects the majority of the outlying mass of the progenitor star , and second the smns driven - wind effectively sweeps up remaining baryonic matter in the vicinity of the central object . support for this model comes from recent detections of strong fe emission features in the x - ray afterglow spectra of some grbs ( piro et al . 1999 , antonelli et al . 2001 , yoshida et al . 2001 ) , particularly that of grb991216 ( piro et al . 2000 ) , as well as a transient fe absorption feature in the prompt emission of grb990705 ( amati et al . 2000 ) , indicating surprisingly large amounts of fe - rich material existing nearby , yet relatively removed from the grb site . observational constraints on its location , quantity , density and velocity are compatible with the pre - ejected shell sn remnant ( snr ) in the supranova model ( lazzati et al . 1999 , vietri et al . 2001 , lazzati et al . 2001 ) , but may be difficult to accomodate in other models ( see however rees & mszros 2000 , mszros & rees 2001 , bttcher & fryer 2001 , mclaughlin et al . 2002 ) . the precursor pulsar wind should be extremely powerful . before collapsing , the smns must inevitably expel a substantial fraction of its rotational energy @xmath3 as a magnetically - driven wind , @xmath4 where @xmath5 and @xmath6 are the smns s mass and angular velocity , @xmath7 is its angular momentum in units of @xmath8 , and the numbers are typical smns model values ( salgado et al . 1994 , cook , shapiro & teukolsky 1994 , vs98 ) . the spin - down time ( i.e. the sn - grb delay time ) and corresponding wind luminosity can be estimated from the magnetic dipole formula ( pacini 1967 , vs98 ) @xmath9 and @xmath10 where @xmath11 is a typical smns equatorial radius . the surface magnetic field @xmath12 is unconstrained from model calculations and can be considered a free parameter . equivalently , we may take @xmath2 as the free parameter , and vary @xmath13 accordingly with @xmath3 fixed as in eq.[eqn : ep ] ; the fiducial value we choose below is @xmath14 ( implying @xmath15 ) so as to be consistent with observations of grb991216 ( see [ sec : aftenv ] ) . during @xmath2 , @xmath16 is expected to be relatively constant , and the wind should energize a plerionic nebula in the pre - grb surroundings , a more compact yet much more luminous version of the crab nebula . the consequences of such a precursor plerion in the supranova scenario has not been considered previously , and this paper addresses some important dynamical and radiative effects it may induce , each providing important observational diagnostics for the supranova model . we discuss the acceleration and fragmentation of the sn ejecta material by the plerion - snr interaction and its implications for fe line emission in [ sec : snr ] , and inverse compton scattering of the ambient plerion radiation field in the grb external shock and the resulting high - energy afterglow emission in [ sec : ec ] . a brief consideration of the direct detection and identification of the precursor plerion emission is given in [ sec : dir ] . we will assume a flat lambda cosmology with @xmath17 , @xmath18 and @xmath19 . after this paper was submitted , we became aware of the work of knigl and granot ( 2002 , hereafter kg02 ) , who also investigated the properties of grb afterglows occurring inside plerionic nebulae in the context of the supranova model . they stressed the advantages of such a picture in realizing the relatively high magnetic fields and electron injection efficiencies required in the grb blastwave to explain observed afterglows , and demonstrated this by constructing magnetohydrodynamical ( mhd ) models for the plerion under various assumptions . here we are interested in some characteristic observable effects caused by the precursor plerionic activity which are peculiar to the supranova model and through which the model may be tested by future observations . the emphasis of this paper will be on the relevant radiative processes and their observational implications ; we choose to keep the discussion of the plerion dynamics relatively simple and leave more detailed modeling of this aspect to future studies . a study following our approach but with a more realistic treatment of the plerion has recently been carried out by guetta & granot ( 2002 ) . an exemplary case of the dynamical interaction between a plerion and a snr can be seen in the well - studied crab nebula , which is known to be accelerating and fragmenting the surrounding snr , resulting in the prominent optical filaments instead of a clear shell ( e.g. davidson & fesen 1985 , hester et al . 1996 ) . in this respect , the powerful supranova plerion can be even more effective than the crab . we model the plerion in a simple way following pacini & salvati ( 1973 , hereafter ps73 ; see also bandiera , pacini & salvati 1984 , chevalier & reynolds 1984 , chevalier 2000 ) , considering a homogeneous , spherical bubble into which energy is injected at a constant rate @xmath20 , a fraction @xmath21 going into magnetic field and the rest @xmath22 into relativistic electrons . the electrons here are mostly radiative ( see [ sec : ec ] ) , and we neglect their pressure for simplicity . as the plerionic bubble initially plows through the expanding core and envelope of the progenitor star , it should accelerate the swept - up ejecta material ( chevalier 1977 , chevalier & fransson 1992 , kg02 ) . if a fair fraction of the total plerion energy @xmath23 can be conveyed to sn ejecta of several @xmath24 , their final attained velocities could reach @xmath25 , consistent with those inferred from the observed width of the fe line in grb991216 ( piro et al . 2000 ) . as in the crab , the sn ejecta should not entirely remain as a spherical shell during this acceleration phase due to rayleigh - taylor ( rt ) instabilities operating at the plerion - snr interface . the growth timescale of the rt instability on a spatial scale @xmath26 is @xmath27 , and since @xmath28 , @xmath29 being the plerion magnetic pressure , its ratio to the expansion timescale @xmath30 is @xmath31 ( bandiera , pacini & salvati 1983 ) , where @xmath32 , @xmath33 and @xmath34 are the snr ejecta mass , velocity and kinetic energy , and @xmath35 is time after the sn in units of 10 days . in the case of the crab , @xmath36 and the rt instability is currently setting in , consistent with other lines of evidence ( hester et al . 1996 , sankrit & hester 1997 ) . however , it can potentially develop much more rapidly for the supranova plerion . the final outcome of the rt instability is not easy to predict , as it will depend on the instability s non - linear behavior as well as other processes such as radiative cooling and thermal conduction . we refer to the work of jun ( 1998 ) as a guideline , who carried out 2-dimensional numerical simulations of the plerion - snr interaction and associated rt instabilities , choosing crab - like parameters for @xmath16 and @xmath37 and neglecting cooling . his results demonstrate that the rt instability leads to dramatic effects at late times , strongly disrupting the snr shell and transforming it into pronounced rt `` fingers '' which carry the majority of the sn ejecta mass and kinetic energy . to be identified with the crab s optical filaments , these fingers protrude inward from the outer ejecta shell and end in dense `` heads '' with slower expansion velocities than the plerion , which can eventually become decoupled from the shell . in jun s simulations , the density of the heads was on average @xmath38 times that of the rest of the sn ejecta , and the radial extent of the fingers was @xmath39 times that of the shell radius . however , as mentioned in his discussion , the inclusion of radiative cooling should increase the density contrast of the heads , and both this and the radial extent of the fingers should grow with time . these quantities could then be much larger for the supranova plerion , since the relative timescale of the instability here might be much shorter ( eq.[eqn : trt ] ) . although a firm conclusion warrants high resolution simulations of the relevant conditions , it is not implausible that the plerion can effectively shred a substantinal portion of the sn ejecta into condensed fragments and engulf them during its lifetime @xmath2 . note that arons ( 2002 ) , in a somewhat different context , has recently discussed the issue of snr fragmentation by a powerful pulsar wind in greater depth . if this is indeed the case ( see 5 for comments when this is not ) , there are important implications for the fe line emission in afterglows . the simplest explanation for the lines is fe k@xmath40 multiple recombination radiation from fe - rich matter photoionized by the x - ray afterglow continuum on timescales of days ( lazzati et al . 1999 , weth et al . 2000 , paerels et al . 2000 ) . consistency with observations require that the material to be illuminated is very dense and possesses a large covering factor , and yet that the grb line of sight is devoid of any such matter so that the grb blastwave will not be decelerated too quickly ( lazzati et al . the filamentary fragmentation of the snr may naturally account for such a geometry . the condensation of the snr matter may also enhance the line emissivity . the emission rate of line photons through recombination when a sufficient photoionizing flux irradiates material with electron density @xmath41 and temperature @xmath42 containing an fe mass @xmath43 is @xmath44 where @xmath45 is the total number of fe atoms and @xmath46 is the recombination time for h - like fe ( paerels et al . 2000 , vietri et al . 2001 ) . by the time of the grb at @xmath47 ( see [ sec : ec ] ) , a snr of @xmath48 expanding at @xmath49 will have reached a radius @xmath50 , and if distributed in a spherical shell of width @xmath51 , its electron density would be @xmath52 typically observed line fluxes can be obtained for @xmath53 ( eq.[eqn : nl ] ) , but only if the entire spherical shell is irradiated and emits line photons efficiently . the grb blast wave must then be isotropic and no beaming is allowed ( piro et al . 2000 ; see however kg02 ) , which is incongruous with recent evidence to the contrary ( e.g. frail et al . 2001 , panaitescu & kumar 2002 ) . however , if the snr ejecta can be condensed into filaments of much higher density , @xmath54 is shortened , and irradiating a smaller fraction of material may suffice to produce the observed lines . for comparison , the density of the the crab s optical filaments averaged over the volume of the whole nebula is @xmath55 , whereas the actual density of the individual filaments is @xmath56 ( davidson & fesen 1985 ) . although detailed numerical modeling is necessary for quantitative predictions , we may speculate that the density contrast could be even higher for the supranova plerion due to the instability being more developed ( see above ) , and beaming fractions of order @xmath57 may be entirely compatible with the observed fe lines . note that in explaining the transient fe absorption feature in grb990705 , lazzati et al . ( 2001 ) have inferred the presence of clumpy snr ejecta material with density contrasts of order 100 - 1000 surrounding the burst site ( see also bttcher , dermer & fryer 2002 ) . the observed line energy in grb991216 can also be consistent with beaming if identified with blueshifted emission from he - like fe ( for which the rest frame energy is 6.7 kev , and @xmath54 is similar to h - like fe at somewhat lower @xmath58 ) in material approaching toward the observer with @xmath59 ( ballantyne & ramirez - ruiz 2001 ) . we mention that even when circumstances conducive to fe line emission can not be realized , the action of the plerion for @xmath60 may expel much of the baryons surrounding the grb site out to radii @xmath61 and clear the path for the ensuing fireball , which is a great advantage of the supranova model . another important implication of the precursor plerion in the supranova model is that the grb should be exploding into a ` radiation - rich ' environment , i.e. into the luminous radiation field of the plerion , which can constitute a large fraction of the pulsar wind luminosity . such ` external ' photons impinging into the relativistically expanding grb shock from outside would be seen highly doppler - boosted in the shock comoving frame , and act as efficient seeds for inverse compton scattering ( referred to as external comptonization , or ec ) , with important observational consequences for the afterglow emission . since our focus here is on the possible manisfestations of the ec emission process in the supranova model , we do not attempt a detailed physical modeling of the the plerion dynamics , employing instead a simplified description for illustrative purposes . as discussed by kg02 , such modeling would entail various uncertainties and requires a dedicated study by itself , which is out of our current scope . once a suitable dynamical model of the plerion can be constructed , the formulation of this paper may be used to make more reliable predictions of the afterglow emission . alternatively , if an afterglow ec emission component is actually observed in the future , we may interpret the data with our formulation so that the unknown properties of the supranova plerion can be probed . we continue the discussion of the plerion as a roughly homogeneous , spherical bubble with constant energy injection . following the arguments of the preceding section , it is postulated that the plerion can first penetrate effectively through the expanding sn ejecta , shredding and entraining the ejecta material within a relatively short time . thereafter , the plerion can be considered to expand into the ambient medium , here taken to be uniform for brevity . the consequences of the dense , entrained baryonic clumps for the global dynamics of the plerion is unclear , but lacking a realistic model , we do not delve into this complication and make simplified assumptions regarding its evolution . if the snr filaments are expanding at velocities of order @xmath59 ( as implied by the observed fe line width in grb991216 ) , the expansion velocity of the plerionic bubble @xmath62 may be several times larger than this , judging from the observed expansion velocities in the crab as well as the numerical simulations of jun ( 1998 ) . as long as the ejecta material is not dynamically well - coupled to the plerion as a whole , we may also estimate @xmath62 from the relativistic blastwave solution of blandford & mckee ( 1976 ) in the case of steady , adiabatic energy injection into a uniform medium , which gives @xmath63 , where @xmath64 is the expansion bulk lorentz factor , @xmath65 is the ambient density and @xmath66 is a numerical factor of order unity . , @xmath67 ) ; in the nonrelativistic limit ( @xmath68 , @xmath69 ) , the deceleration goes as @xmath70 and @xmath71 . ] a mildly relativistic expansion is then inferred for ages @xmath72 and the values of @xmath16 and @xmath73 considered here . we assume , at least for the conditions of our interest , that the plerion has been expanding at a nearly constant velocity of @xmath67 , so that its radius @xmath74 . the luminosity and spectrum emitted by the plerion may be evaluated following ps73 . the magnetic field inside the plerion ( taken to be isotropically tangled ) is @xmath75 , with the value of @xmath3 fixed as in eq.[eqn : ep ] . relativistic electrons / positrons ( hereafter simply electrons ) are injected into the plerion with a power of @xmath76 , having an isotropic , power - law energy distribution @xmath77 in the lorentz factor range @xmath78 . we fiducially take @xmath79 , @xmath80 and @xmath81 , parameters which lead to consistent fits when similar models are applied to the multiwavelength spectra of the crab and other known plerionic nebulae ( ps73 , bandiera et al . 1984 , amato et al . 2000 , chevalier 2000 ) . the maximum lorentz factor @xmath82 is set by equating the synchrotron cooling time @xmath83 with the acceleration time , here assumed to be @xmath84 ( consistent with fermi acceleration at relativistic shocks ) , so that @xmath85 and the corresponding maximum synchrotron emission frequency @xmath86 . , synchrotron - self - compton cooling is at most comparable to synchrotron cooling , so this is neglected for simplicity . ] the distribution should develop a break at the lorentz factor @xmath87 where @xmath88 equals the adiabatic expansion time @xmath89 , leading to @xmath90 . our fiducial parameters give @xmath91 , implying that injected electrons of all but the lowest energies are radiative and cool within @xmath92 ; the cooled distribution has index @xmath93 and emission spectrum of energy index @xmath94 . the specific flux at the surface of the plerion is @xmath95 synchrotron self - absorption will cause an effective minimum cutoff in the spectrum at a frequency @xmath96 where @xmath97 is a factor of order unity weakly depending on @xmath98 ( e.g. rybicki & lightman 1979 ) , and @xmath99 is the normalization factor of the cooled electron distribution ( ps73 ) . the expressions in the last lines of eqs.[eqn : nupa ] and [ eqn : kc ] are for our fiducial choice of @xmath80 . as the flux falls steeply below @xmath100 , the overall plerion spectrum is that of eq.[eqn : fnup ] between @xmath100 and @xmath101 , with constant luminosity per logarithmic frequency interval , and the total emitted luminosity is of order @xmath76 . which may arise from self - compton emission is irrelevant for electrons in the grb shock due to klein - nishina effects ; see below . ] note that some quantities regarding the plerion contain subscripts or superscripts @xmath102 in order to avoid confusion with those for the grb blastwave . we mention that the plerion should also drive a forward shock into the ambient medium , whereby the associated synchrotron radiation may give rise to an additional emission component . since the magnetic field and the electron injection efficiency in the shocked external medium are expected to be lower than that inside the plerion , such emission should be of subdominant luminosity appearing at frequencies below @xmath100 . ec scattering of this component by grb blastwave electrons is unimportant for the observed afterglow , as it is likely to be masked by the synchrotron and ssc radiation occurring in the pertinent frequency range ( see below ) ; however the emission at radio frequencies may play a role in the direct detectability of the plerion before the grb ( 4 ) . a proper account of this plerion forward shock emission mandates a more detailed description of the plerion expansion , so in this study we choose to neglect this component and restrict ourselves to emission from the plerion interior . at time @xmath2 after the sn , the pulsar collapses and the grb goes off , sending a fireball and relativistic blastwave into the plerion . we may distinguish two different situations for how the deceleration of the grb blastwave , and hence the afterglow , initiates . as in conventional discussions of afterglows , the deceleration may begin when it has swept up enough outlying baryonic material , which may occur if the ( probably collimated ) fireball in our line of sight happens to strike the denser parts of the entrained sn ejecta fragments inside the plerion . alternatively , the deceleration may result from the blastwave accumulating enough inertia in relativistic particles and magnetic field of the pulsar wind material , as discussed by kg02 . to describe the afterglow in this case , kg02 showed that standard formulae ( e.g. sari , piran & narayan 1998 ) can still be used if one replaces parameters such as the ambient medium density and electron and magnetic energy density fractions by equivalent quantities related to the plerion properties ( see below ) . in considering either case , we will assume for simplicity that the decelerating medium is roughly uniform , at least within the timescales we consider . ( note that this is not a bad approximation for most of the mhd plerion models discussed by kg02 . ) the standard expressions for the radius @xmath103 and bulk lorentz factor @xmath104 of the shocked material in an adiabatic , spherical blastwave decelerating self - similarly in a uniform medium are @xmath105^{1/4 } \sim 3.6 \times 10^{17 } { \rm cm } ( e_{52}/n)^{1/4 } [ t_d/(1+z)]^{1/4}$ ] and @xmath106^{1/8 } \sim 5.9 ( e_{52}/n)^{1/8 } [ t_d/(1+z)]^{-3/8}$ ] , respectively ( e.g. mszros & rees 1997 , piran 1999 ) . here @xmath107 is the blastwave energy , @xmath108 the external medium density , @xmath109 the observer time elapsed after the grb , @xmath110 is the grb redshift , and we have adopted the kinematic relation @xmath111 . the deceleration starts at radius @xmath112 at observer time @xmath113 , where @xmath114 is the initial bulk lorentz factor . to describe the time - dependent , multiwavelength afterglow spectrum from the blastwave , we follow standard discussions of the synchrotron emission ( e.g. mszros & rees 1997 , sari , piran & narayan 1998 , wijers & galama 1999 ) , and extend it to include cooling and emission by the ec process in addition to the synchrotron - self - compton ( ssc ) process ( panaitescu & kumar 2000 ; sari & esin 2001 , hereafter se01 ; zhang & mszros 2001 , hereafter zm01 ) . it is assumed that constant fractions @xmath115 and @xmath116 of the postshock energy are imparted to magnetic field and relativistic electrons , respectively . the comoving magnetic field is @xmath117^{-3/8}$ ] . electrons are accelerated in the shock to a power - law distribution @xmath118 in the lorentz factor range @xmath119 . we assume @xmath120 , our standard choice being @xmath121 . the minimum lorentz factor @xmath122 is given by @xmath123 \nonumber\\ & & \times ( e_{52}/n)^{1/8 } [ t_d/(1+z)]^{-3/8 } \label{eqn : gammn } \end{aligned}\ ] ] ( note that @xmath124=1 $ ] when @xmath121 ) . the electrons radiatively cool by the combination of the synchrotron , ssc and ec processes , the timescales of which are @xmath125 , @xmath126 and @xmath127 respectively , the total cooling time being @xmath128^{-1 } = t'_{sy}(1+y+x)^{-1}$ ] . the maximum lorentz factor @xmath129 is determined by balancing @xmath130 with the acceleration time @xmath131 , @xmath132^{3/16 } \label{eqn : gammx } \end{aligned}\ ] ] ( zm01 ) . the cooling lorentz factor @xmath133 is where @xmath130 equals the comoving adiabatic expansion time @xmath134 , @xmath135^{1/8 } . \label{eqn : gamc } \end{aligned}\ ] ] the observed characteristic synchrotron emission frequency for an electron of lorentz factor @xmath136 is @xmath137 ( wijers & galama 1999 ) , so those for each of @xmath122 , @xmath129 and @xmath133 are respectively @xmath138 ^ 2 \nonumber\\ \times e_{52}^{1/2 } t_d^{-3/2 } ( 1+z)^{1/2 } , \label{eqn : numn } \end{aligned}\ ] ] @xmath139 and @xmath140 at early times ( @xmath141 ) , @xmath142 , and all electrons cool within @xmath143 ( fast cooling regime ) . the synchrotron flux @xmath144 then peaks at @xmath145 and mainly consists of three power - law segments : @xmath146 for @xmath147 , @xmath148 for @xmath149 and @xmath150 for @xmath151 . at late times ( @xmath152 ) , @xmath153 , and only electrons of @xmath154 cool within @xmath143 ( slow cooling regime ) . then the spectrum peaks at @xmath155 , and again has three parts : @xmath156 for @xmath157 , @xmath158 for @xmath159 and @xmath160 for @xmath161 . the peak synchrotron flux @xmath162 for either fast or slow cooling is @xmath163 where @xmath164 is a numerical constant and @xmath165 is the luminosity distance to the grb . the transition from fast to slow cooling ( @xmath166 ) occurs at the time @xmath167 ^ 2 \nonumber\\ \times e_{52 } n ( 1+z ) . \label{eqn : tt } \end{aligned}\ ] ] to account for the effects of ec and ssc cooling , we have introduced @xmath168 and @xmath169 , which are in general time - dependent . the ec parameter @xmath168 equals the ratio of comoving frame energy density in plerion radiation field @xmath170 to that in magnetic field @xmath171 . if the blastwave decelerates inside the plerion ( @xmath172 ) , which is the case mainly considered below , we may presume that the radiation field is nearly uniform and isotropic , obtaining @xmath173 where @xmath174 is the frequency above which klein - nishina ( kn ) effects suppress compton scattering , depending on @xmath136 . observe here that the factors of @xmath175 in @xmath170 and @xmath171 cancel out so that aside from the weak @xmath37-dependence through @xmath176 , @xmath168 is constant in time . generally @xmath177 , but we do not make large errors in taking @xmath178 in the logarithmic factor for simplicity . for a blastwave decelerating outside the plerion ( @xmath179 ) , the radiation field is anisotropic and falls with @xmath103 , and @xmath168 differs from eq.[eqn : xpar ] by a factor of @xmath180 where @xmath181 ( which applies for @xmath182 ; see e.g. sikora et al . 1996 , inoue & takahara 1997 ) , decreasing with time . the ssc parameter @xmath169 can be evaluated as in se01 , @xmath183 \sim \eta \epsilon_e/[\epsilon_b ( 1+y+x)]$ ] , where @xmath184 and @xmath185 are the comoving frame energy densities in relativistic electrons and synchrotron radiation , and @xmath186 is the fractional energy radiated by electrons . in the fast cooling phase , @xmath187 and we get the expression @xmath188 in the slow cooling phase , @xmath169 decreases with time but only very slowly for typical values of @xmath102 ( se01 ) , so we simplify by assuming eq.[eqn : yf ] at all times . the ssc spectrum has a similar shape to the synchrotron spectrum , and we approximate it as comprising broken power - laws ( se01 , zm01 ) , with characteristic frequencies at @xmath189 ^ 4 \nonumber\\ & & \times e_{52}^{3/4 } n^{-1/4 } t_d^{-9/4 } ( 1+z)^{5/4 } \label{eqn : numinsc } \end{aligned}\ ] ] and @xmath190 the maximum ssc frequency should occur at @xmath191 where kn effects completely cut off the spectrum . ( generally , @xmath192 . ) for fast cooling , the ssc flux @xmath193 peaks at @xmath194 , the spectrum being @xmath195 for @xmath196 , @xmath197 for @xmath198 and @xmath199 for @xmath200 ; for slow cooling , the peak is at @xmath201 , and the spectrum is @xmath195 for @xmath202 , @xmath203 for @xmath204 and @xmath199 for @xmath205 . the peak ssc flux @xmath206 is equal to the peak synchrotron @xmath162 multiplied by the optical depth @xmath207 of the shocked material , @xmath208 the ec spectrum reflects the shapes of both the electron distribution and the plerion spectrum . considering deceleration inside the plerion , and again adopting the broken power - law approximation , spectral breaks arise at the frequencies where electrons of @xmath122 and @xmath133 upscatter plerion photons of @xmath100 , @xmath209 ^ 2 \nonumber\\ & & \times e_{52}^{1/2 } n^{-1/2 } t_d^{-3/2 } ( 1+z)^{1/2 } \label{eqn : numinec } \end{aligned}\ ] ] and @xmath210 where the fiducial numerical value for @xmath100 in eq.[eqn : nupa ] has been used . note the factor @xmath175 which results from two lorentz transformations , one into the comoving frame from the observer frame and another vice - versa . the fast cooling ec spectrum has a peak flux at @xmath211 , is @xmath212 for @xmath213 from the behavior of the compton scattering cross section , @xmath197 for @xmath214 and @xmath215 for @xmath216 , mirroring the flat plerion spectrum . the kn limit discussed above also applies here for the maximum ec frequency , @xmath101 being irrelevant . likewise , the slow cooling ec spectrum peaks at @xmath217 , and is @xmath212 for @xmath218 , @xmath203 for @xmath219 and @xmath215 for @xmath220 . the peak ec flux @xmath221 may be obtained analogously to the ssc case , but using the relation @xmath222 in the comoving frame , where @xmath223 is the the peak plerion flux emitted at @xmath224 . accounting for lorentz transformations to and from the comoving frame , @xmath225 where the last line assumes fiducial plerion parameters . we caution that being isotropic in the observer frame , the plerion radiation field should actually be highly anisotropic in the shock frame moving at @xmath182 . instead of an accurate but cumbersome calculation of the compton upscattered photon distribution including the full angle - dependence , we have approximated by using quantities averaged over angles in the comoving frame ; see e.g. dermer 1995 , inoue & takahara 1996 , sikora 1997 . the broken power - law representations of the ec and ssc spectra should be adequate to within an order of magnitude , but more accurate calculations properly integrating over the broad seed photon frequency distribution will result in smoother spectral breaks and somewhat larger fluxes ( above @xmath211 and @xmath194 for fast cooling , or above @xmath217 and @xmath201 for slow cooling ) due to logarithmic terms contributed by a range of electron lorentz factors ( se01 ) . regarding kn effects , a gradual steepening from @xmath226 instead of an abrupt cutoff should actually occur , as parts of the seed photon spectra will be in the kn regime even for electrons with @xmath227 . we have not included the effects of internal pair attenuation , which could be important for the highest emission energies at relatively early times in the afterglow . however , significant differences are expected only for tev energies and above ( zm01 ) , whereby pair degradation by the infrared background during intergalactic propagation should be serious anyway for grbs with @xmath228 ( e.g. stecker & de jager 1998 , totani 2000 ) . having laid out the tools to calculate the broadband afterglow emission , we now discuss some constraints to be imposed on the local environment of the blastwave in the context of our model . for an afterglow to produce a strong fe line as observed in grb991216 , it is necessary that the grb blastwave decelerates inside the plerion near relatively dense snr fragments , so that they are irradiated efficiently by afterglow continuum x - rays . in case the decelerating medium comprises mostly entrained baryons ( referred to as the `` baryon case '' ) , we opt to directly use the above equations with @xmath229 and @xmath230 , since the physics governing these quantities here should be similar to that for conventional afterglows decelerating in external media . although an inhomogeneous spatial distribution is actually expected for the baryons , we surmise that the formulae are applicable for a certain timescale during which the medium can be approximated as locally uniform . the local baryon density in the plerion interior may range anywhere from values as high as @xmath231 for the densest snr clumps , to much lower values @xmath232 for regions efficiently swept out by the magnetized plerionic plasma . if the local density in our line of sight @xmath233 , the blastwave turns subrelativistic in less than @xmath234 and is unable to generate the observed afterglows ( lazzati et al . 1999 ; see however masetti et al . 2001 , in t zand et al . 2001 ) , whereas if the density @xmath235 , the blastwave does not decelerate significantly until it reaches the medium outside the plerion at @xmath236 . we fiducially choose @xmath237 , implying that if the medium was globally uniform at this value , the blastwave travels from @xmath238 to @xmath239 during @xmath240 s to @xmath241 days inside the plerion , illuminating nearby fe - rich condensations along the way , and turns subrelativistic ( @xmath242 ) at @xmath243 days . in reality , a sufficiently clumpy baryonic distribution may cause a sporadic type of variability in the afterglow which should be detectable by future instruments , and could perhaps be relevant to the x - ray flaring behavior observed in the afterglows of grb970508 ( piro et al . 1999 ) and grb970828 ( yoshida et al . 2001 ) . for the case of deceleration occurring mainly within the magnetized plerionic material ( referred to as the `` plerion case '' ) , kg02 showed that the usual afterglow formulae can be utilized with some modifications , such as substituting @xmath244 with an equivalent baryon density @xmath245 where @xmath246 is the electron pressure and @xmath247 and @xmath248 are the magnetic and electric fields in the frame comoving with the shocked pulsar wind . for our simplified plerion model , we estimate @xmath249 by taking @xmath250 , @xmath251 and @xmath252 , which gives @xmath253 . thus , as with the baryon case , we fiducially choose @xmath254 for the plerion case . spatial uniformity should be a good approximation , even for the more sophisticated mhd models considered by kg02 . one must also make corresponding changes for @xmath255 and @xmath256 , as well as insert correction factors for some quantities , but these depend on the detailed properties of the pulsar wind shock ( see kg02 ) for which our model lacks information . for want of a better alternative , we choose to employ @xmath257 and @xmath258 , and ignore the correction factors . with these parameters , the blastwave should be partially radiative during the fast cooling stage ; however , an accurate treatment of such cases is rather cumbersome ( e.g. bttcher & dermer 2000 ) , so we proceed with the adiabatic assumption for the sake of simplicity . a further requirement from the observed fe lines is that @xmath2 should be neither much shorter nor much longer than @xmath259 : @xmath260 would not allow time for enough sn - synthesized radioactive co to decay to fe ( vietri et al . 2001 ) , whereas for @xmath261 the snr fragments would be of too low density to generate the observed fe line flux , unless they can be made extremely condensed ( eqs.[eqn : nl],[eqn : nes ] ) . this justifies our fiducial choice of @xmath262 . between the continuum source and the irradiated , line - emitting matter , which can be smaller than the radius @xmath103 of the matter from the grb site . the blastwave ( i.e. the continuum source ) itself should have moved out to @xmath263 by @xmath264 , but fe - rich material could be present in its immediate vicinity at @xmath265 . ] guided by the observed characteristics of afterglows such as grb991216 , our fiducial parameters were chosen to represent conditions favorable for generating strong fe lines . however , we will not attempt model fits to the available data on individual objects , since i ) realistically , additional complicating effects of jet geometry , etc . can be important and must be taken into account , and ii ) full parameter searches for fitting particular observations are out of our scope ( c.f . panaitescu & kumar 2002 ) the results given here for the fiducial parameter sets are intended to exemplify some notable features of afterglows wherein ec cooling and emission induced by the supranova plerion are important . changing these parameters can lead to either stronger or weaker ec effects ( see below ) . we first discuss the baryon case . figs.[fig : specb ] and [ fig : lcb ] show our fiducial results in terms of the time - dependent broadband afterglow spectra at different observer times @xmath37 , and the afterglow light curves at selected fixed frequencies , respectively , assuming the grb to be at @xmath266 in the flat lambda cosmology ( i.e. @xmath267 ) . the values of @xmath168 and @xmath169 here are fiducially @xmath268 and @xmath269 , and since they were approximated as being constant , the relative luminosities of the synchrotron , ssc and ec components stay roughly the same . however , each component is prominent in a different energy band which changes with @xmath37 : at the beginning of the afterglow ( @xmath270 ) , the ssc component dominates at the highest energies ( @xmath271 tev ) whereas the ec component does so at somewhat lower energies ( @xmath271 gev ) ; in the latter stages ( @xmath272 ) , this is reversed . the lower energies are always dominated by the synchrotron component , although the energy of the luminosity peak progressively decreases from @xmath271 mev down to the radio band . there are clear differences between the spectral shapes of the ssc and ec components : the ssc peaks in @xmath273 at @xmath274 and has a steep high energy slope above this break identical to that of the synchrotron emission ( energy index @xmath275 ) , but the ec component is flat in @xmath273 above @xmath276 , reflecting the harder plerion spectrum . this spectral signature is one way to observationally distinguish between ssc and ec emission , even though this distinction can be blurred if values of @xmath102 happen to be close to @xmath98 . here the fast to slow cooling transition occurs at @xmath277 , and the afterglow is still fast cooling at @xmath264 . this is longer than in standard discussions , and is partly due to the extra cooling process of ec , but primarily as a consequence of the relatively high external density adopted , @xmath237 . the light curves at fixed frenquencies can be described by breaks at characteristic times @xmath278 and @xmath279 corresponding to the passage of the minimum and cooling frequencies for each spectral component , as well as at @xmath280 . the synchrotron light curves are as in previous studies ( sari et al . 1998 ) : for high frequencies in which @xmath281 , @xmath282 for @xmath283 , @xmath284 for @xmath285 and @xmath286 [ @xmath287 for @xmath288 ; for low frequencies in which @xmath289 , @xmath282 for @xmath290 , @xmath291 for @xmath292 , @xmath293 [ @xmath294 for @xmath295 and @xmath286 [ @xmath287 for @xmath296 . in these expressions and below , the @xmath37-dependence inside the brackets are for our standard choice of @xmath80 and @xmath121 . defining @xmath297 , @xmath298 as times when @xmath299 , @xmath201 respectively , the ssc light curves are : for high frequencies in which @xmath300 , @xmath301 & t > t_m^{sc } ; \end{array } \right . \label{eqn : fhsc } \end{aligned}\ ] ] for low frequencies in which @xmath302 , @xmath303 & t_m^{sc } < t < t_c^{sc}\\ t^{(10 - 9p)/8 } \ \ [ t^{-1.56 } ] & t > t_c^{sc } . \end{array } \right . \label{eqn : flsc } \end{aligned}\ ] ] the ec light curves depend on the plerion spectral index ( @xmath304 ) . with @xmath305 , @xmath306 as times when @xmath307 , @xmath217 respectively : for high frequencies in which @xmath308 , @xmath309 & t_m^{ec } < t < t_t\\ t^{(2 - 2p - s)/4 } \ \ [ t^{-1.2 } ] & t > t_t ; \end{array } \right . \label{eqn : fhec } \end{aligned}\ ] ] for low frequencies in which @xmath310 , @xmath311 & t_m^{ec } < t < t_c^{ec}\\ t^{(2 - 2p - s)/4 } \ \ [ t^{-1.2 } ] & t > t_c^{ec } . \end{array } \right . \label{eqn : flec } \end{aligned}\ ] ] differences can be seen in the high frequency decay indices at late times between the ssc and ec components : @xmath312 [ @xmath313 both before and after @xmath280 for ssc , compared to @xmath314 [ @xmath315 before @xmath280 breaking to @xmath316 [ @xmath317 after @xmath280 for ec . these features should be important observational diagnostics in the x - ray and gamma - ray range , although here again , the discrimination is less clear if @xmath102 approaches @xmath98 . ( note also that there should actually be a slight break in the ssc light curve at @xmath280 from the time - dependence of @xmath169 ; se01 . ) the generally harder spectrum of the ec emission relative to ssc should make it increasingly dominant in the higher energy bands at later times . the early x - ray and optical emission is predominantly synchrotron , but the entry of the minimum frequency of the ssc emission ( and to a lesser extent , the ec emission ) gives rise to ` bumps ' in the light curves , which appear at @xmath318 days in x - rays and @xmath319 days in the optical for our fiducial calculation . in particular , we see that x - ray emission after @xmath320 day is ssc plus ec emission . so far , there have been no clear indications for such x - ray bumps , except for a flattening in the x - ray light curve of grb000926 at @xmath321 days . together with multiwavelength spectra , this has been interpreted as ssc emission from a blastwave in a moderately dense medium of @xmath322 ( harrison et al . 2001 ) , but a different explanation with an even denser medium ( @xmath323 ) is also possible ( piro et al . 2001 ) . in the optical band , it is interesting to recall that a small number of afterglows have been seen to exhibit bumps at few tens of days , attributable to an underlying supernova light curve superimposed on the afterglow ( e.g. bloom et al . 1999 , 2002 , galama et al . 2000 , lazzati et al . the observed bumps seem to manifest simultaneous spectral reddening , as opposed to the spectral hardening expected for the ssc induced bumps . however , with the exception of grb980326 and grb011121 , the combined evidence for light curve bump plus reddening is not very strong , so an ssc explanation may not be ruled out , at least for some of the observations . in any case , realistic modeling of light curve breaks should also include other effects such as jet geometry , so firm conclusions in comparison with observations are not possible at the moment . the detection of the ec component , which is distinctive to the supranova model , should be best achieved at gev energies and above . we can assess the detectability in gev gamma - rays by the egret and glast instruments , following the discussion of zm01 . the fluence threshold of each intrument are described by a constant value for photon - limited , short integration times and one proportional to @xmath324 for background - limited , long integration times . this can be compared with the model fluence light curves , intergrated over the energy range of 400 mev to 200 gev . our fiducial baryon case model would have been observable by egret hours after the grb from @xmath1 , providing a viable explanation for the extended gev emission seen in grb940217 ( hurley et al . similar observations should also be possible with agile ( e.g. mereghetti et al . 2001 ) , having a slightly better sensitivity than egret . glast should be able to detect this emission from redshifts as high as @xmath325 up to @xmath271 1 day after the burst . our results for the fiducial plerion case are shown in figs.[fig : specp ] and [ fig : lcp ] . since the equivalent baryon density @xmath326 , the only difference here with the fiducial baryon case stems from the parameter values of @xmath255 and @xmath256 . many of the trends described above are also valid here qualitatively , but important quantitative differences are to be seen . the much larger magnetic field ( by a factor of 50 in energy density ) induces greatly increased synchrotron losses , leading to smaller values of @xmath133 and the corresponding cooling frequencies , while the larger electron injection efficiency ( by a factor of 5 ) causes higher values of @xmath122 and the associated miminum frequencies . this implies very large @xmath280 , and the electrons remain fast cooling throughout the duration of the afterglow . the synchrotron and ssc luminosities are larger and occupy higher energy bands ; the ec emission is relatively much less prominent , appearing at gev over the ssc emission only during the early time interval from @xmath38 s to @xmath327 s , and again at later times @xmath328 days . detection of this component should be feasible with future gamma - ray instruments as with the baryon case , but here it may be more difficult to entangle from the ssc emission . the light curve breaks due to the passage of the minimum frequecies occur much later , and the ssc bumps are not observable . we now discuss the effect of changing our main free parameters , particularly those regarding the plerionic enviroment , @xmath244 and @xmath2 . keeping all others constant for the baryon case , decreasing @xmath244 from @xmath237 leads to a greater dominance of the ec component and less synchrotron emission , as this implies a smaller @xmath329 and larger @xmath168 for the same @xmath256 . higher values of @xmath244 may also occur inside the plerion ; note the recent observations of grb010222 indicating @xmath330 for this burst ( masetti et al . 2001 , in t zand et al . 2001 ) , although alternative interpretations with lower values of @xmath244 are possible ( e.g. panaitescu & kumar 2002 ) . more ` conventional ' parameters ( lower @xmath244 , small @xmath168 ) may be accomodated in the context of our model by taking a larger @xmath2 to make @xmath331 smaller , say @xmath332yr , and assuming @xmath333 . for the plerion case , @xmath244 is not an independent free parameter as @xmath334 ; here @xmath332yr directly implies @xmath333 . however , for this larger plerion the volume - averaged sn ejecta density should also drop ( eq.[eqn : nes ] ) , so that strong fe lines are less likely , requiring extreme clumping of ejecta material . note that clear upper limits on fe line emission obtained for some bright afterglows point to a variety of grb environments ( yonetoku et al . 2000 , 2001 ) , which could correspond to a range of @xmath2 and/or @xmath244 in our model . we also mention that the well observed afterglow of grb970508 has an fe line feature detection ( piro et al . 1999 ) , as well as arguments for consistency with more or less ` standard ' parameters ( wijers & galama 1999 ) , but including the effects of ssc and ec cooling in the model fitting may allow different sets of parameters ( e.g. se01 ) . finally , we briefly remark on the possibility of the direct detection and identification of the precursor plerion emission , which would be a straightforward test for the supranova model ( vs98 ) . a more detailed assessment can be found in guetta & granot ( 2002 ) . as discussed above , the plerion should be a luminous , steady source during a period @xmath2 prior to the burst , typically with bolometric luminosity @xmath335 and power - law spectrum of energy index @xmath336 between frequencies @xmath337 and @xmath338 . this is comparable to high - luminosity quasars and readily observable ; e.g. it should have already been detected out to @xmath339 as bright kev x - ray sources during the rosat all - sky survey ( rass ; voges et al . 1999 ) the event rate of grbs detectable by batse from @xmath340 may be roughly @xmath341 all - sky ( e.g. porciani & madau 2001 ) , so there should always be at least @xmath342 active plerions all - sky above the rass sensitivity . here @xmath343 is inverse of the fractional solid angle subtended by the grb beam , which could be as large as @xmath344 ( frail et al . 2001 ) , and we have assumed that the plerions are not strongly beamed . however , the problem lies in ascertaining these sources as precursors to grbs , which must be shown to be positionally coincident as well as correlated in time within @xmath2 before the grb . even though the operation periods of batse and rass were close enough in time , the large batse error boxes ( @xmath345 ) preclude discrimination from the large number of agns with similar x - ray spectra . more precise grb localizations by present and next generation instruments ( e.g. hete-2 with @xmath346 or swift with @xmath347 ) , combined with concurrently conducted wide field x - ray surveys ( e.g. xmm - newton serendipitous survey ) should improve the prospects on searching for grb - precursor plerion associations . the plerion should also be conspicuous in the optical , e.g. its r - band magnitude would be @xmath348 at @xmath349 . ( the precursor supernova may be observable as well , but could be masked by the luminous plerion . ) again , identifying these with the ensuing grb out of the numerous other objects at similar magnitudes requires accurate grb positions and nearly contemporaneous wide field optical surveys . this may be possible through serendipitous studies of observations made by e.g. the subaru suprime - cam , the sloan digital sky survey , and the future supernova acceleration probe mission . detection at radio frequencies may also be possible if there is additional emission from the plerion forward shock ( [ sec : ple ] ) . we summarize the salient points of our work . in the supranova model of grbs , the fast - rotating smns active during the time between the sn and the grb should drive a luminous plerionic nebula into the preburst environment , with a number of important consequences for the ensuing grb afterglow . rayleigh - taylor instabilities acting at the plerion - sn ejecta interface may induce significant filamentary clumping of the ejecta material , allowing a favorable geometry and enhanced emissivity for the afterglow fe line emission . the plerion radiation field can act as seeds for ec cooling and emission in the grb external shock , leading to prominent gev - tev gamma - rays during the afterglow , which is detectable by glast out to typical grb redshifts , and distinguishable from ssc emission by its characteristic spectrum and light curve . a direct search for the plerion emission prior to the grb may be conducted through accurate grb positions and concurrent wide field surveys , e.g. in the optical and x - rays . all of these should provide critical tests for the supranova model in the near future . the ability of the plerion to effectively fragment the snr and penetrate between the ejecta material is an uncertain aspect of our scenario . we may therefore consider the consequences of the plerion being completely confined inside the snr shell instead . in the case of a spherical geometry , such a circumstance would be at odds with constraints derived from the observed fe lines , since : 1 ) the snr would likely be optically thick to thomson scattering at the time of the grb , so that no prompt or afterglow emission can be seen ( eq.[eqn : nes ] ) ; 2 ) the snr shell , as well as the plerionic bubble material , would be dense enough to decelerate the grb blastwave in a time shorter than the observed afterglow ; and 3 ) the radiation density of the plerion would be so large as to cause excessively fast cooling of electrons in the afterglow blastwave . a possible alternative is a snr - confined plerion with a globally anisotropic geometry , being elongated in the direction of the grb beam , as discussed by kg02 and guetta & granot ( 2002 ) . the true efficacy of rt instabilities at the snr - plerion interface should be clarified through future high - resolution numerical simulations including all the relevant microphysics . arons ( 2002 ) has recently discussed in more detail the plausibility of a powerful pulsar wind shredding the surrounding snr in the context of a model for cosmic ray acceleration by magnetars . although we have concentrated on the case of the pulsar spinning down exclusively via a magnetospheric wind , there is a possibility that significant spindown can also occur through gravitational wave emission , such as those driven by r - mode instabilities ( e.g. andersson et al . 2000 , stergioulas & font 2001 , lindblom , tohline & vallisneri 2001 and references therein ) . however , many aspects of the present theoretical calculations regarding gravitational waves are uncertain ( fryer & woosley 2001 ) . our work can be viewed as exploring the maximum possible effects of electromagnetic spindown , which can be modeled with some more confidence . in reality , both effects could be important to varying degrees , and we leave an investigation of such cases to the future . we are thankful to m. salvati for an initial suggestion leading to the present work , and e. amato , j. granot , d. lazzati , n. omodei and m. vietri for helpful discussions . s.i . also expresses his heartfelt thanks to members of the arcetri astrophysical observatory for a very pleasant and hospitable environment where part of this work was carried out . this work was partly supported by a grant from the italian ministry of university and research ( cofin-99 - 02 - 02 ) . amati , l. et al . 2000 , sci . , 290 , 953 amato , e. , salvati , m. , bandiera , r. , pacini , f. & woltjer , l. 2000 , , 359 , 1107 andersson , n. , jones , d. i. , kokkotas , k. & stergioulas , n. 2000 , , 534 , l75 antonelli , l. a. et al . 2001 , , 545 , l39 arons , j. 2002 , submitted to apj ( astro - ph/0208444 ) ballantyne , d. r. & ramirez - ruiz , e. 2001 , , 559 , l83 bandiera , r. , pacini , f. & salvati , m. 1983 , , 126 , 7 bandiera , r. , pacini , f. & salvati , m. 1984 , , 285 , 134 blandford , r. d. & mckee , c. f. 1976 , phys . fluids , 19 , 1130 bloom , j. s. et al . 1999 , , 401 , 453 bloom , j. s. et al . 2002 , , 572 , l45 bttcher , m. & dermer , c. d. 1999 , astropart . phys . , 11 , 113 bttcher , m. & dermer , c. d. 2000 , , 532 , 281 bttcher , m. & fryer , c. l. 2001 , , 547 , 338 bttcher , m. , fryer , c. l. & dermer , c. d. 2002 , , 567 , 441 chevalier , r. a. 1977 , in supernovae , ed . d. n. schramm , ( dordrecht : reidel ) 53 chevalier , r. a. 2000 , , 539 , l45 chevalier , r. a. & fransson , c. 1992 , , 395 , 540 cook , g. b. , shapiro , s. l. & teukolsky , s. a. 1994 , , 424 , 823 davidson , k. & fesen , r. a. 1985 , , 23 , 119 dermer , c. d. 1995 , , 446 , l63 frail , d. a. et al . 2000 , , 538 , l129 frail , d. a. et al . 2001 , , 562 , l55 fryer , c. & woosley , s. 2001 , , 411 , 31 galama , t. j. et al . 2000 , , 536 , 185 granot , j. , piran , t. & sari , r. 2000 , , 534 , l163 guetta , d. & granot , j. 2002 , submitted to ( astro - ph/0208156 ) harrison , f. a. et al . 2001 , , 559 , 123 hester , j. j. et al . 1996 , , 456 , 225 hurley , k. et al . 1994 , , 372 , 652 inoue , s. & takahara , f. 1996 , , 463 , 555 inoue , s. & takahara , f. 1997 , prog . phys . , 98 , 807 in t zand , j. j. m. et al . 2001 , , 559 , 710 jun , b .- 1998 , , 499 , 282 kluniak , w. & ruderman , m. 1998 , , 505 , l113 knigl , a. & granot , j. 2002 , , 574 , 134 ( kg02 ) lazzati , d. et al . 2001a , , 556 , 471 lazzati , d. et al . 2001b , , 996 , 1002 lazzati , d. , campana , s. & ghisellini , g. 1999 , , 304 , l11 lindblom , l. , tohline , j. e. & vallisneri , m. 2001 , , 86 , 1152 masetti , n. et al . 2001 , , 374 , 382 mclaughlin , g. c. , wijers , r. a. m. j. , brown , g. e. & bethe , h. 2002 , , 567 , 454 mereghetti , s. et al . 2001 , in exploring the gamma - ray universe , 4th integral workshop , ed . b. battrick , esa sp-459 , p603 mszros , p. 2001 , sci , 291 , 79 mszros , p. & rees , m. j. 1997 , , 476 , 232 mszros , p. & rees , m. j. 2001 , , 556 , l37 pacini , f. 1967 , nature , 216 , 567 pacini , f. & salvati , m. 1973 , , 186 , 249 ( ps73 ) paerels , f. , kuulkers , e. , heise , j. & liedahl , d. a. 2000 , , 535 , l25 panaitescu , a. & kumar , p. 2000 , , 543 , 66 panaitescu , a. & kumar , p. 2002 , , 571 , 779 piran , t. 1999 , physrep , 314 , 575 piro , l. et al . 1999 , , 514 , l73 piro , l. et al . 2000 , sci , 290 , 955 piro , l. et al . 2001 , , 558 , 442 porciani , c. & madau , p. 2001 , , 548 , 522 rees , m. j. & mszros , p. 2000 , , 545 , l73 reynolds , s. p. & chevalier , r. a. 1984 , , 278 , 630 rybicki , g. b. & lightman , a. p. 1979 , radiative processes in astrophysics , wiley salgado , m. , bonazzola , s. , gourgoulhon , e. & haensel , p. 1994 , , 291 , 155 sankrit , r. & hester , j. j. 1997 , , 491 , 796 sari , r. & esin , a. a. 2001 , , 548 , 787 ( se01 ) sari , r. , piran , t. & narayan , r. 1998 , , 497 , l17 shibata , m. , baumgarte , t. w. & shapiro , s. l. 2000 , , 61 , 044012 sikora , m. in 4th compton symposium , eds . c. d. dermer , m. s. strickman & j. d. kurfess , aip conf . 410 , p494 sikora , m. , sol , h. , begelman , m. c. & madejski , g. m. 1996 , , 280 , 781 stecker , f. w. & de jager , o. c. 1998 , , 334 , l85 stergioulas , n. & font , j. a. 2001 , , 86 , 1148 totani , t. 2000 , , 536 , l23 van paradijs j. , kouveliotou c. , wijers r.a.m.j . , 2000 , , 38 , 379 vietri , m. et al 2001 , , 550 , l43 vietri , m. & stella , l. 1998 , , 507 , l45 ( vs98 ) voges , w. et al . 1999 , , 349 , 389 weth , c. , mszros , p. , kallman , t. & rees , m. j. 2000 , , 534 , 581 wijers , r. a. m. j. & galama , t. j. 1999 , , 523 , 177 yonetoku , d. et al . 2000 , , 52 , 509 yonetoku , d. et al . 2001 , , 557 , l23 yoshida , a. et al . 2001 , , 557 , l27 zhang , b. & mszros p. 2001 , , 559 , 110 ( zm01 )

we present spectra of the optical transient of grb 021004 obtained with the hobby - eberly telescope starting 15.48 , 20.31 hours , and 4.84 days after the-ray burst and a spectrum obtained with the h. j. smith 2.7 m telescope starting 14.31 hours after the-ray burst . grb 021004 is the first burst afterglow for which the spectrum is dominated by absorption lines from high ionization species with multiple velocity components separated by up to 3000 km s . we argue that these absorption lines are likely to come from shells around a massive progenitor star . the high velocities and high ionizations arise from a combination of acceleration and flash - ionization by the burst photons and the wind velocity and steady ionization by the progenitor . we also analyze the broad - band spectrum and the light curve so as to distinguish the structure of gas within 0.3 pc of the burster . we delineate six components in the medium surrounding the-ray burst along the line of sight : ( 1 ) the absorption lines arise from the innermost region closest to the burst , where the ionization will be highest and the 3000 km s velocity comes from the intrinsic velocity of a massive star wind boosted by acceleration from the burst flux . for a mass loss rate of , this component also provides the external medium with which the jet collides over radial distances 0.0040.3 pc to create the afterglow light . ( 2 ) a second cloud or shell produces absorption lines with a relative velocity of 560 km s . this component could be associated with the shell created by the fast massive star wind blowing a bubble in the preceding slow wind at a radial distance of order 10 pc or by a clump at 0.5 pc accelerated by the burst . ( 3 ) more distant clouds within the host galaxy that lie between 30 - 2500 pc and have been ionized by the burst will create the absorption lines . ( 4 - 6 ) if the three bumps in the afterglow light curve at 0.14 , 1.1 , and 4.0 days are caused by clumps or shells in the massive star wind along the line of sight , then the radii and over - densities of these are 0.022 , 0.063 , and 0.12 parsecs and 50% , 10% , and 10% respectively . the immediate progenitor of the-ray burstcould either be a wc - type wolf - rayet star with a high velocity wind or a highly evolved massive star the original mass of which was too small for it to become a wn - type wolf - rayet star . in summary , the highly ionized lines with high relative velocities most likely come from shells or clumps of material close to the progenitor and these shells were plausibly produced by a massive star soon before its collapse .

astro-ph0211189

the large absorption line red shift measurements settled the issue that most @xmath0-ray bursts are at cosmological distances . the physical mechanism of the origin of the @xmath0-ray bursts , questions of whether there are more than one mechanism , and the issue of whether @xmath0-ray bursts interact only with the interstellar medium or with a circumstellar environment are still very open . correlations with star formation and observations of putative iron lines with the chandra observatory @xcite have put new focus on the connection to a supernova - like core collapse event . currently , most observations of @xmath0-ray burst afterglows are photometric , with the effort being to find the power law behavior of the afterglow brightness as a function of wavelength and time . most of the 25 known red shifts have been measured by deep spectroscopy on the host galaxy long after the afterglow has faded . prior to grb 021004 , only eight @xmath0-ray burst afterglows have had early - time spectra that reveal absorption lines from the host galaxy ( e.g. , vreeswijk et al . 2001 ) . of these , virtually all have been low excitation lines in one velocity system that are likely to arise in gas far from the burster . two exceptions to this are grb 011211 which shows and in addition to lower excitation lines @xcite , and grb 010222 which displayed two low - ionization systems separated by @xmath6 km s@xmath1 that could arise from clouds at different positions within the host galaxy @xcite . grb 020813 has a spectrum which displays absorption lines in both a z=1.223 and z=1.255 systems @xcite . with so little known , and many afterglows revealing new properties , more prompt spectroscopy is clearly needed . afterglow spectra could show absorption lines arising from the host galaxy ( especially the fe and mg ism lines ) that will give the red shift . the red shift is the key parameter for determining the burst energetics and hence further testing the conclusions of @xcite and @xcite that @xmath0-ray bursts have a nearly constant total energy . one can also improve the calibration ( currently based on 9 events , see also schaefer , deng , & band 2000 ) that the spectral lags and light curve variability are cepheid - like luminosity indicators that could turn @xmath0-ray burstsinto a premier tool for cosmology out to z@xmath410 @xcite . intrinsic afterglow spectral features of any kind would be a boon to interpretation . if there are underlying supernova - like explosions , corresponding spectral features could be superposed on the continuum and revealed by careful subtraction of the well - defined power - law continuum . detection of , for instance , features of red shifted ca h & k , which tend to be among the strongest lines in all types of supernovae , would yield both a red shift and clues to the nature of the explosion itself . other possibilities would be to discover fe emission lines , which would yield temperatures , red shifts , and abundances . perhaps the most important possibility is that of serendipitous discovery of some unexpected feature . grb 021004 provided an unprecedented opportunity to obtain prompt spectroscopy of the early afterglow of a @xmath0-ray burst . grb 021004 was discovered by instruments on the hete ii satellite at 12:06 ut on 2002 october 4 @xcite . prompt discovery of the optical transient by @xcite allowed rapid spectroscopic observations . @xcite obtained the first spectra and identified two intervening systems at z = 1.38 and 1.60 from and absorption . @xcite confirmed those features , noted several absorptions at these redshifts , and pointed out four absorption lines at @xmath4 4633 , 4664 , 5109 , and 5152 ( see also sahu et al . 2002 ; castander et al . these were identified by @xcite as and features at red shift of @xmath7 . chornock & filippenko also identified ly@xmath8 emission at z = 2.323 and absorption components and perhaps ly@xmath9 at similar red shifts . @xcite identified four absorption components of at z = 2.295 , 2.298 , 2.230 and 2.237 , noting that the total spread is about 3000 km s@xmath1 . they remarked that this velocity dispersion would be difficult to interpret as due to a single cluster of galaxies and noted that if associated with a supernova shell the supernova must have substantially preceeded the @xmath0-ray burst to be optically thin . @xcite suggested the identification of one absorption feature as low excition at z = 2.328 and identified that , and the ly@xmath8 emission , as the red shift of the host galaxy . they raised the issue of the velocity dispersion being due to a wind in a wolf - rayet like progenitor . @xcite noted that the ly@xmath8 flux may imply a strong star formation rate in the host galaxy . @xcite reported a very high resolution spectrum ( r = 50,000 ) that revealed the complex fine structure of many of the high ionization lines . they determined that the individual features show complex flow spanning about 1000 km s@xmath1 and conclude that the @xmath10 system is probably close to the @xmath0-ray burst . they note that of the four high red shift systems , the low ionization lines , and possibly , only occur in the highest red shift , @xmath10 , component . @xcite and mller et al . ( 2002 ) have provided an excellent compendium of line identifications along with detailed line profiles . @xcite give a spectral index in the optical of @xmath11 , and report a reddening of the continuum in the blue region of the spectrum over the first three days . here we report spectra obtained within the first day after outburst and five days later . we concentrate on constraining the circumburst medium in the context of the high ionization lines and the afterglow light curve . spectra of the optical afterglow were obtained with both the 2.7 m harlan j. smith telescope using the igi imaging spectrograph @xcite and the 9.2 m hobby - eberly telescope ( het ) with the marcario low - resolution spectrograph @xcite at mcdonald observatory . three consecutive exposures of 900 seconds each were taken with the 2.7 m telescope beginning 14.31 hours after onset of the @xmath0-ray burst , followed by sets of three 900 second exposures taken with the het beginning at 15.48 hours , 20.31 hours , and 4.84 days after the burst . a 2.5 arcsecond slit was used for the 2.7 m data , while the het observations utilized a 2.0 arcsecond slit , giving all data sets a 16 resolution . the spectra were reduced using iraf , and wavelength calibrated using argon , neon , cadmium , and mercury lamps . the combined spectrum from the first night is displayed in figure 1 . our first night spectrum shows a collection of absorption lines longward of 5800 that arise from iron and magnesium of relatively low ionization at red shifts of 1.38 and 1.60 . these are due to intervening galaxies with no relevance to the @xmath0-ray burst analysis . the lines at red shifts of @xmath12 are presented with their measured properties in table 1 . the lines are separated into two primary velocity components , with @xmath13 and @xmath14 . we identify the 4587.2 line as being from ( c.f . matheson et al . 2002 ) . there are few resonance lines from the ground state of highly - ionized elements of high cosmic abundance in the observed wavelength range , and the only one that has not produced a detected absorption line is with rest wavelengths 1238.8 and 1242.8 for which we can only present an upper limit to its equivalent width . the separation of the two high redshift primary velocity components corresponds to a relative velocity of 3000 km s@xmath1 in the frame of the host galaxy . the unblended lines for the @xmath10 velocity component are significantly resolved in our spectrum , with a fwhm of @xmath15 km s@xmath1 . from our own data , this implies either multiple components or one component with a velocity range that large . other spectra @xcite have resolved the high red shift component into at least two sub - components with a velocity separation of 560 km s@xmath1 . thus , the @xmath16 system consists of at least three separate velocity components , with velocities of 0 km s@xmath1 , 560 km s@xmath1 , and 3000 km s@xmath1towards earth relative to the highest red shift component . @xcite presented detailed calculations of the complex of absorption lines produced from 1110 - 1705 by vibrationally excited molecular hydrogen in a molecular cloud near a @xmath0-ray burst . individual lines have a typical depth of 20% for normal low - resolution sprecta and might be difficult to detect without very high signal . fortunately , the entire pattern of lines is characteristic of the process and so all of the lines ( with their unique positions and depths ) can be simultaneously sought in any spectrum . the idea is that a spectrum in which any one line would be highly insignificant might yet have the presence of many lines detectable at a high confidence level . even a noisy spectrum might have its apparent noise following the characteristic molecular hydrogen features hence resulting in a significant detection . many of the individual lines are extremely saturated with optical depths of @xmath17 in the line center and with equivalent widths of 0.1 or greater , so that the strength of the absorption is quite insensitive to the column density of vibrationally excited molecular hydrogen , @xmath18 . b. draine has kindly provided us with the transmission spectrum for excited molecular hydrogen for our spectral resolution . we have used this as a template to cross correlate with our spectrum . for a red shift of 2.3 , we cover the range from 1240 - 1705 . we find no significant correlation peak for any red shift of the template . within the red shift range of 2.292.33 , the largest peak in the cross correlation is never more than the rms of the cross correlation over a much wider red shift range . the implication of this negative result will be discussed in section 3.5 . we found no significant differences among our spectra within the first night . specifically , the equivalent widths of the absorption lines were all identical to within the errors . we have constructed broad - band photometry from our reduced spectra using the photometric pass - bands of landolt ( 1992 ) and normalizing the colors to match the observations concurrent to the 2.7 m spectrum . the resulting synthetic photometry for the b , v , and r magnitudes yields colors that are constant to within 0.04 mag for the first night . for a comparison between our observations on the first and fifth nights , we are handicapped by the poor signal on the fifth night due to the faintness of the afterglow . the only lines for which we can quote line widths on the fifth night are the two lines which have equivalent widths in the rest frame of the host ( the observed width divided by 1+z ) of @xmath19 and @xmath20 at the same wavelengths as for the first night . this is consistent with the first night . we confirm the report by @xcite that the blue end of the continuum became redder from the first night to subsequent nights . in particular , our synthetic photometry shows the b - v color to have reddened by @xmath21 mag from the first to the fifth night , while the v - r color was unchanged , @xmath22 mag , over the same interval . the lyman alpha _ emission _ line is likely to represent the average velocity of the host galaxy , although its exact wavelength will shift redward somewhat due to the absorption component . the red shift of the ly@xmath8 line is variously given as 2.3351 , 2.332 , 2.328 , and 2.323 @xcite . the average velocity of the host galaxy is also likely represented by the low - ionization feature of at 2.328 . the highest red shift components of the and lines correspond to this host galaxy velocity . the velocity of the burster itself might vary by up to several 100 km s@xmath1from the average velocity for the host . the other primary components of these high - ionization lines are at velocities of 560 km s@xmath1 and 3000 km s@xmath1blueward of the host galaxy velocity . the presence of the @xmath12 lines of and shows that the absorbing gas is highly ionized . such lines are commonly seen in the spectra of distant quasars through intervening galaxies where the ionized material is in a collisionally ionized halo . could the absorbing gas for grb 021004 just be in halos of chance galaxies along the line of sight ? the presence of three velocity components within the @xmath12 system with separation of up to 3000 km s@xmath1 argues strongly that this can not be . the probability of two galaxies lying in so small a red shift range just blueward of the host is rather small even in a cluster of galaxies , and the velocities are too large to allow for a bound cluster . sargent , steidel , & boksenberg ( 1988 ) present statistics on lines with rest frame equivalent widths of more than 0.3 ( i.e. , a threshold much smaller than the observed lines for grb 021004 ) , and find that absorption systems with @xmath12 occur with a rate of @xmath23 per red shift unit . the probability of getting two absorbers within a range of 0.033 in @xmath24 in front of the host galaxy is thus 0.0024 . such an occurrence is not impossible ( c.f . the @xmath25 system for q1511 + 091 , sargent , steidel , & boksenberg 1988 ) , but it is too small to be plausible for grb 021004 . it is also improbable that random intervening galaxies will have equivalent widths as thick as 1.4 ( as for the observed lines from the systems blueward of the host galaxy ) . the distribution of equivalent widths ( for both members of the doublet ) falls off exponentially such that the probability for any one galaxy being above this threshold is @xmath26 , or 4% for both systems . in all , the probability of getting two chance galaxies with a red shift within 0.033 of the host and with rest frame equivalent widths of @xmath27 is 0.00010 . we conclude that the high - ionization absorption lines are not from intervening galaxies . the only other situation for which the and lines are seen in absorption is when gas is subjected to a high flux of ionizing radiation , as in an agn , a wolf - rayet star , or a @xmath0-ray burst . an agn at the position of the @xmath0-ray burst would already have been detected , and so this possibility can be rejected . with only @xmath28 wolf - rayet stars in our own galaxy , the odds of the line of sight passing close to 13 such stars is small ( ignoring clustering ) provided that the star is not the progenitor of the @xmath0-ray burst . that the progenitor is itself a wolf - rayet star is plausible , because the progenitor is likely to be a massive star just before a core - collapse alternatively , the ionizing radiation could come from the @xmath0-ray burst itself . thus , we think that the most likely means to ionize the gas to high levels is either radiation from the burst progenitor or the burst itself , with both cases requiring the ionized gas to be physically associated with the burster . wolf - rayet stars have temperatures from 30,00070,000 k and supergiant luminosities . for the hotter stars , this is sufficient to ionize the surrounding nebulosity to levels where a significant fraction of the carbon and silicon atoms can produce and absorption and emission . thus , a wolf - rayet - like progenitor might already ionize the surrounding gas such that later light from the afterglow will have silhouettes of these high - excitation lines . alternatively , the @xmath0-ray burst itself is an obvious source of a huge ionizing flux . the most frequent mode of ionization of carbon and silicon will be for the @xmath0-ray to knock out an inner electron , whereupon autoionization will cause ejection of the valence electrons . in addition the ejected electrons will have sufficient energy to ionize other atoms . in the remainder of this section , we will estimate a crude distance range over which the burst flux can ionize enough gas so as to produce the observed absorption lines . the cross section for photoelectric interactions varies strongly with photon energy , so the probability that any given atom is ionized by burst flux requires an integral over energy . the cross sections for carbon and silicon are tabulated in @xcite along with a prescription for interpolation . for carbon , the cross sections are 0.0213 , 40.3 , and 44500 barns ( one barn equals @xmath29 ) at photon energies of 100 , 10 , and 1 kev respectively . with this , we see that most of the ionization will be from burst photons in the x - ray band . the burst spectrum has only been measured down to a photon energy of 7 kev ( 23 kev in the frame of the host galaxy ) by hete-2 . above 7 kev , @xcite state that the spectrum is well fit by a single power law such that @xmath30 and has a fluence from 7400 kev of @xmath31 . below 7 kev , the spectrum will likely turn over to asymptote at a slope close to the synchrotron limit of @xmath32 . here , we will model this behavior as a broken power law with indices of -1.64 and -0.66 above and below a break energy , @xmath33 . we also impose a lower energy cutoff in our calculations at @xmath34 , below which we assume that no flux reaches the gas . this is to model the possible photoelectric absorption of photons before they reach the gas . we normalize this spectrum to the 7400 kev fluence reported by hete-2 and correct the photon energies of this observed spectrum by a factor of @xmath35 to get the spectrum as seen by gas in the host galaxy frame . the expectation value for the number of times that a given atom will be ionized by burst flux is @xmath36 , where @xmath37 is the burst photon spectrum ( in units of photons @xmath38 ) as a function of distance from the burster , @xmath39 is the cross section for ionization as a function of photon energy @xmath40 , and @xmath41 is the time . for these calculations , we adopt @xmath42 kev . we find that the result is highly sensitive to the value of @xmath34 ( due to the large cross section for ionization at low energies ) , which itself is highly uncertain since we do not know the degree of absorption of the low energy photons before the radiation hits the gas . for @xmath43 kev in our rest frame ( 1 kev in the host galaxy ) , we expect @xmath44 ionizing interactions per carbon atom at a distance of 1 pc . for @xmath45 kev in our rest frame ( 0.1 kev in the host galaxy ) , we expect @xmath46 ionizing interactions per carbon atom at a distance of 1 pc . the expected number of ionizing interactions will scale as the inverse square of the distance . for this range of cases , the expected number of ionization interactions is unity at distances ranging from 10 - 25 pc . this provides an order - of - magnitude estimate of the minimal range out to which the burst flux will completely ionize carbon . the radius at which significant high ionization absorption can be caused may be substantially farther than this calculated distance for two reasons . first , only a small fraction of the carbon need be ionized to result in significant absorption . since the cross section for the resonance line is large , the intervening gas need only have a column density of @xmath47 in the state to account for the observed absorption line . this corresponds to a fraction of @xmath48 of the total carbon expected in a smooth wind ( see 3.4 ) . if only 0.01% of the carbon must be ionized to to create the necessary column density , then the burst s effective radius for ionization is larger than the previous distance estimate by a factor of 100 . this factor depends on the column density of the cloudy ism . second , the electrons ejected from the neutral atoms at high velocity will also have a high cross section for ionizing further atoms . this effect depends on the density of the gas and the spectrum of the electrons @xcite . this effect will roughly increase the ionizing radius by a factor of the square root of the ratio of the electron energy to the ionization energy , which might be up to a factor of order ten . in all , there are many uncertainties and dependencies on unknown conditions , but the effective ionization radius of the burster might be of order 10100 times larger than calculated in the previous paragraph . this would imply that any gas cloud within perhaps 1002500 pc of the burster will suffer enough ionization to create a sufficient column density of so as to form a detectable absorption line . the complementary question is how close to the burster some gas will be incompletely ionized at the epoch of observation . this is important since we later consider the possibility of the absorbing gas being quite close ( @xmath26 pc ) to the burst . @xcite have presented a detailed and extensive calculation that shows that even a very thick wind will have all of its hydrogen completely ionized by a grb flash on time scales of milliseconds or faster . the implication is that most ( if not all ) of the atoms of carbon in a wind will also be completely ionized , and thus there may be no atoms in the state to create absorption lines near the burst . whereas lazzati & perna considered a smooth wind , the column density of @xmath49 @xmath50 could be accounted for by clumps in the gas . the existence of such clumps is plausible since images of wolf - rayet winds show structure on all size scales . also , the quasi - stationary flocculi ( clumps of gas created in the wolf - rayet progenitor wind ) in cas a and the kepler supernova remnants have densities of @xmath51 @xcite after correcting by a factor of 4 for the shock jump conditions , with a typical projected size of @xmath52 cm in width and @xmath53 cm in length . for grb 021004 in particular , the existence of clumps is indicated by the three bumps in the afterglow light curve ( see 5 ) where the clumps might be large in size with moderate contrast or might be small in size yet with large central densities . the existence of clumps in the winds of massive stars is thus expected , and these clumps can have densities @xmath54 @xmath55 . clumps can serve in two ways to allow for the existence of close to the @xmath0-ray burst . first , the clump will shield portions of the wind material ( both within the clump and within the shadow of the clump ) from the high burst flux . for example , a clump with size @xmath56 cm with a density of @xmath57 at a distance of a fraction of a parsec from the burster should absorb the ionizing flux . second , the higher density within clumps can allow for faster recombination that can generate a day after the burst even if the burst had completely ionized the carbon . the degree of recombination that will have occured when we see the absorption a day or so after the burst has ionized the gas depends strongly on the density . for ionized carbon , the recombination time scale is @xmath58 days for densities of @xmath59 that would be typical of a wolf - rayet wind at a distance of @xmath60 pc . the recombination time scale would be @xmath44 days for densities of @xmath57 that are typical of the densities observed in supernova remnant knots that are presumed to have formed in the progenitor s wind . within such clumps , the fraction of carbon in the state should be @xmath61 a day after the burst . a clump with density @xmath57 and size @xmath56 cm will produce a column of @xmath62 in . this is @xmath63 times larger column than is required to produce the observed absorption lines . detailed calculations of the ionization structure are difficult without knowledge of the density structure in the wind . nevertheless , it appears that shadows and recombination in the expected clumps can yield a column density of @xmath47 even quite close to the burster . the calculation of the ionization structure is not straight forward even if the density structure is given . one trouble is that the burst spectrum below 7 kev is unknown , and this is the critical energy region for ionizing the gas due to the large cross sections for photo - ionization . a related problem is accounting for the ionizing flux from the afterglow , which is comparable to that from the burst itself . an additional problem is that stimulated recombination will ensure that some fraction of the carbon will not be completely ionized even in a radiation field of the highest intensity . yet another issue is that the photo - ionized electrons likely have roughly a power - law energy distribution , whereas the cross sections for ordinary and stimulated recombination are calculated for thermal electron distributions . the correct recombination rates are likely to be significantly different than the usual " calculations . in principle , these problems can all be surmounted by substantial theoretical effort . in the mean time , we conclude that it is difficult to make any definite statements regarding the ionization state of gas close to the burster . a cloud far from the @xmath0-ray burst will have a low relative velocity corresponding to ordinary galactic rotation . such a cloud would be an obvious candidate for creating the high - excitation low - velocity absorption component . another obvious candidate to contribute to the high - excitation low - velocity lines is the shell piled up by the o main sequence star wind as it blows a bubble in the local ism at a distance of perhaps 30 pc . thus , there are two adequate sources of material along the line of sight with distances from 302500 pc that can be ionized by the burst flux . gas near the burster will receive a tremendous blast of photons , and interactions will transfer some of the photons momentum to the gas . the total radiative acceleration of the gas that determines its final velocity will occur during the few seconds when the burst flux passes through the gas a few hours before the time at which the absorption lines are produced by the gas . how much momentum gets transferred depends on the cross section of the gas , the spectrum of the burst , and the flux of the burst photons . the final velocity will fall off as the inverse - square of the distance from the burst . at some distance , the blast will accelerate the gas to 3000 km s@xmath1 , and this process could account for the observed high relative velocity of one of the absorption components . in this subsection , we present a detailed calculation of the radiation - induced velocity as a function of distance from the burster . the radiative acceleration @xmath64 ( as a function of distance r and time t ) by a central energy source is given by @xmath65 where @xmath66 is the cross section for interaction with a burst photon of energy @xmath40 , @xmath67 is the mass per nucleon , and @xmath68 is the eddington flux which is the first moment of the intensity ( with units of erg @xmath69 . the velocity due to radiative acceleration is @xmath70 where @xmath71 is the initial velocity of the gas , and @xmath72 is the time interval covering the burst duration as the photons pass through the gas being accelerated . this integral over time will change the burst flux into a fluence , which is to say that the acceleration only depends on the number of incident photons and not on their time distribution . this change in velocity structure in the gas around the burst will not lead to any significant changes in the density structure on time scales for which we are concerned . that is , a parcel of gas accelerated to , say , 3000 km s@xmath1 will move only 1.7 au in one day , so as to cause a @xmath73 density change within a gas clump of size , for example , @xmath56 cm if the backside suffers zero acceleration . the @xmath0-ray cross sections for the photon - matter interaction have been included as detailed in hflich ( 1991 ) . because the photon energies far exceed the binding energies of the electrons in atoms , the interaction of electrons and photons is given by the klein - nishina cross section per nucleon . for lower energies , we determine the angle dependent klein - nishina cross section by a standard rejection technique and obtain the cross section by formal integration over angle . bound - free cross sections have been included according to veigel ( 1973 ) . the x - ray opacities are dominated by bound - free transitions from inner shell electrons ( which are always populated since when an inner shell electron is ejected it is rapidly filled by one of the outer electrons ) , and are independent of the conditions in the gas cloud . thus , the bound - free opacity is rather model independent . the final velocities are a function of composition and the burst spectrum . for composition , we adopted either a normal solar abundance @xcite , a wolf - rayet composition @xcite , or a pure gas of ionized hydrogen . for the spectrum , we adopted broken power laws with a low - energy cutoff , as described in the previous subsection . the results for @xmath74 are displayed in figure 2 . we find that the composition of the gas has little effect . we find that the flux around one kev ( in the frame of the host galaxy ) is very important , due to the large cross sections associated with bound - free interactions . unfortunately , the flux for energies below 7 kev are not known and we can only assume plausible spectral shapes , as discussed in the previous section . our favorite model of the final velocity as a function of the radial distance from the burster is represented by the thick line in figure 2 . to accelerate gas from rest to 3000 km s@xmath1 implies that the gas is at a distance of 0.2 parsec . if the gas were already expanding at a velocity of 2500 km s@xmath1 , then the radial distance of the gas would be @xmath4 0.5 parsec . gas initially at rest at @xmath4 0.5 pc , would be accelerated to @xmath4 500 km s@xmath1 , corresponding to one of the observed absorption components . if the cloud has a substantial extent along the line of sight , then the inner edge will be accelerated to a higher velocity than the outer edge . this would result in a broad line , instead of the narrow lines observed . the radiative acceleration mechanism thus requires that the intervening cloud be physically narrow . this could arise from a small cloud ( as a clumpy part of a wind ) or from a thin shell ( presumably created by the progenitor ) . for quantitative limits , the width of the carbon lines appears to be less than 400 km s@xmath1 , for which the dominant absorbing region in the shell must be thinner than 7% of its radius provided the gas is accelerated from rest to around 3000 km s@xmath1 by the burst radiation . if the gas already had an initial velocity of 2000 km s@xmath1 , then the dominant absorbing region in the shell must be thinner than 21% of its radius . if an absorbing gas has a wind - like density distribution ( that falls off as the inverse - square of the radial distance ) , then 50% of the absorption will come from between the inner edge of the absorbing region to twice that radius . thus , for absorption from a smooth wind , the radiative acceleration must be small compared to the initial wind velocity to avoid excessively broad absorption lines . however , if the absorbing gas has a significant clump superposed on a wind - like structure ( see 5 ) , then such a clump may be sufficiently thin to form a narrow absorption line despite a large radiative acceleration even in a wind . this acceleration of gas close to the burster must occur , the only question is the size of the effect . this will depend primarily on unknown details of the low energy photons in the burst spectrum after they have passed through material at small radial distances . for reasonable assumptions ( @xmath75 kev ) , the radiative acceleration will induce velocity changes of 1000 km s@xmath1 or greater for gas within a fraction of a parsec from the burster . this accelerated gas must include that which is creating the afterglow , as well as the dense gas interior to the afterglow shock that represents the inner region of any wind that was driven off the progenitor . wolf - rayet stars are very hot ( 30,000 - 70,000 k ) stars shedding a thick wind for which the characteristic spectra involve bright and broad emission lines with p cygni profiles . they are highly evolved stars , the main sequence masses of which were larger than 35 @xmath76 or so . for the most massive stars ( @xmath77 ) the evolutionary path is from an o main sequence star to a luminous blue variable to a wolf - rayet star of the wn class to a wolf - rayet star of the wc class to a supernova collapse , while a less massive star ( @xmath78 ) progresses from an o main sequence star to a red supergiant to a wolf - rayet star of the wn class to a supernova collapse . significant evidence and theoretical models support the hypothesis that long - duration @xmath0-ray bursts are caused by the collapse of massive stars @xcite , consistent with the notion that the immediate progenitors of bursts are wolf - rayet stars . each stage in the evolution of massive stars has a distinct wind @xcite . the o main sequence star wind is characterized by a velocity of @xmath79 km s@xmath1 and a mass loss rate of order @xmath80 to @xmath81 which lasts for a few million years . the luminous blue variable phase has a massive ejection ( at a rate of up to @xmath82 ) over a short time ( roughly 10,000 years ) with moderate velocities ( around 300 km s@xmath1 ) . the red supergiant wind is characterized by a velocity of 10 - 100 km s@xmath1 and a mass loss rate of up to @xmath83 which lasts for around 200,000 years . the wolf - rayet wind is characterized by velocities of 1000 - 3000 km s@xmath1 and a mass loss from @xmath84 to @xmath85 over a lifetime of fractions of a million years . the interactions between the winds in each successive stage of evolution will create shells with various properties @xcite . the o main sequence star will blow a bubble into the surrounding ism that has a thin shell and a slow expansion velocity at a distance of order 30 pc . the massive ejection of a luminous blue variable will produce a thick shell that expands with a velocity of around 300 km s@xmath1 . the red supergiant wind will bunch up at its outer edge and expand with a velocity of 10 - 100 km s@xmath1 . the wolf - rayet wind blows a bubble in the slower and denser wind around it , creating a thin shell that expands with a velocity of order 500 km s@xmath1 . this bubble will overtake the slower shell in @xmath86 years and coalesce to form a combined turbulent shell . at the time of the core collapse of the star , the inner several parsecs will consist of the wolf - rayet wind expanding at 1000 - 3000 km s@xmath1 , surrounded by a dense and turbulent shell expanding at perhaps 500 km s@xmath1at a distance of around 10 pc , surrounded by a relatively low density region evacuated by the o main sequence star wind , all inside a geometrically thin bubble of material piled up from the ism . the winds of wolf - rayet stars are likely to be anisotropic , and the absorption lines in grb 021004 were created by only a small geometrical cross section , presumably along the rotation axis of the progenitor star . can this wind structure around a very massive star account for the observed velocity components in the absorption lines of grb 021004 ? the blue shifted 3000 km s@xmath1 component would arise from the dense inner wind driven during the wolf - rayet phase . the o main sequence star wind also has a velocity of 3000 km s@xmath1 , but its density has been greatly reduced by its expansion into space so that it should produce only negligible absorption . the requisite density in this radius range could also be provided in principle by the slower wind of a lower mass star with lower mass loss rate ; this case would subsequently require more radiative acceleration . if the density in the wind falls off as the inverse - square of the radial distance from the star , the absorption from the wind will be dominated by its inner region , say 0.20.4 pc . this means that the absorption lines would primarily sample a wolf - rayet wind emitted over a brief time interval somewhat before the core collapse . over its lifetime , a wolf - rayet wind is roughly constant in velocity , while the wind velocity across the 0.20.4 pc region should be even more precisely constant . this could produce a narrow absorption line just as is observed . the blue shifted 560 km s@xmath1 absorption component in grb 021004 could arise from the dense shell created at the outer edge of a wolf - rayet wind . this shell is expected to be composed of geometrically thin structures that form a turbulent front width of which is roughly a tenth of its radius which expands at roughly 500 km s@xmath1@xcite . the velocity structure within the shell is chaotic and will lead to some broadening of the resultant absorption lines . thus , expected wolf - rayet wind structures can produce the observed 3000 km s@xmath1 and 560 km s@xmath1 components . what are the expected column densities in the wolf - rayet winds ? for a mass loss rate of @xmath87 , a wind velocity of 3000 km s@xmath1 , and an average atomic weight equal to that of helium , a wolf - rayet wind will produce a column density of roughly @xmath88 . the total column density of the shell at the outer edge of a wolf - rayet wind is uncertain , but it is likely to be roughly the mass ejected @xcite in the red supergiant phase ( @xmath4 20 @xmath76 ) or the luminous blue variable phase ( @xmath4 10 @xmath89 ) spread over a shell of radius 10 pc . for an average atomic mass like that of helium , the total column density will be of order @xmath90 atoms @xmath50 or greater . the two shell components thus have comparable column densities , just as the two absorption lines have comparable equivalent widths . are these column densities enough to produce significant absorption lines ? for the oscillator strengths of and an assumed velocity dispersion of order 100 km s@xmath1 , a column density of @xmath91 for atoms in their ground state will produce an absorption line with an optical depth of unity . with the total column density for a component of a wolf - rayet wind at around @xmath92 atoms @xmath50 , the carbon line would produce a significant feature provided that in the ground state is more than @xmath48 of the nucleons in the wind . a large fraction of the ejecta from wolf - rayet stars will be carbon . the various ionization states will all spend almost all the time at their ground level because the decay times are of order @xmath93 seconds . the question of calculating the fraction of carbon that is ionized to any particular degree ( by the radiation from the burst and from the progenitor ) is a difficult problem both for the physics and for the lack of knowing the conditions of the gas ( 3.2 ) . nevertheless , it is plausible that some significant fraction of the carbon will be three - times ionized . for one - tenth of the carbon being three - times ionized , the total column of atoms capable of producing the line in the wolf - rayet wind will be @xmath94 atoms @xmath50 . for such a case , the line will be highly saturated . this estimate has many uncertainties , but it is clear that wolf - rayet winds can have sufficient carbon to create the observed absorption lines . wolf - rayet stars often show p - cygni line profiles , with the emission component arising from portions of the shell off the line of sight . if the wind structure described above exists around grb 021004 , we do not expect to see p - cygni profiles . there are two reasons for this . the first is that the @xmath0-ray burst radiation is strongly beamed by the jet , so most of the shell off the line of sight is not illuminated . the second reason is that a flash of illumination on the shell will produce emission lines that are delayed by an amount that varies with position in the shell . for typical shell sizes , the emission component would be spread out over many years and will become too faint to be observable . thus the lack of p - cygni profiles in the high - excitation lines of grb 021004 is not an argument against their origin in a wolf - rayet wind . what is the lower limit on the mass of the original star such that it can still produce the required dense wind ? o - type stars with masses of @xmath95 can produce dense winds with surrounding shell structures @xcite , and hence might account for the observed lines in grb 021004 . these stars can have wind velocities up to 3000 km s@xmath1 but with mass loss rates that might be smaller than allowed by our analysis of the spectral energy distribution ( see 4 ) . the evolution of such stars would be from an o main sequence star to a red supergiant to a blue supergiant with a core collapse at some point . while such a progenitor can not be called a wolf - rayet star , it is still a massive star surrounded by shells and hence could be the progenitor of grb 021004 . since the clumpy density profile is so uncertain , the only rigorous lower limit we can place on this process is that for core collapse itself , that is @xmath96 . for instance , a star that blew up as a red giant might have a carbon - rich wind and hence have the necessary density to give the column depth of . such a structure would have to rely on radiatively accelerated clumps of matter at @xmath4 0.5 pc to produce the lower velocity features in the absence of a shell generated by the blue / red wind boundary . we now have two scenarios that can explain the velocity and ionization structure of the high - excitation absorption lines seen in grb 021004 . the first scenario postulates two gas shells or clumps initially nearly at rest at distances of @xmath26 pc and @xmath97 pc that are radiatively accelerated to velocities of 3000 km s@xmath1and 560 km s@xmath1 , respectively , as well as ionized by the flash from the burst itself . the second scenario proposes that the progenitor is a wolf - rayet - like star with a wind velocity of 3000 km s@xmath1 that is surrounded by a shell with velocity 560 km s@xmath1 , the gas of which is ionized by the flux from the progenitor . these two scenarios are actually part of a continuum of similar scenarios where the ionizations and velocities vary from being caused entirely by the progenitor to being caused entirely by the burst . for example , in the second scenario , the original wind might have started at a lower ionization state at a velocity of 1500 km s@xmath1 and been accelerated and ionized to the final values by the burst . in all these related scenarios , the absorption lines with near zero velocity are caused by chance clouds along the line of sight as far as 2500 pc from the burster that have been ionized in part by the burst radiation . in the first scenario , we have to ask why a @xmath0-ray burst progenitor would have two geometrically - thin gas clouds at distances of @xmath26 pc and @xmath98 pc ? this is improbable unless there is some causal connection between the clouds and the burster . the presence of two geometrically - thin clouds along the line of sight could be caused by two shells around the progenitor or by some number of clumps of gas around the progenitor . in the second scenario , we also have shells of gas centered on the progenitor . we conclude that the progenitor had shells or clumps of gas centered on it and associated with it . models where the @xmath0-ray burst arises from the collision of two compact objects in close orbit are substantially less likely because these have no ready supply of gas close to the system . the presence of shells is an unusual situation for stars , as few have multiple shells , although clumps are more likely . the only stellar objects with multiple shells that have any realistic hope of being associated with @xmath0-ray bursts are supernovae and massive stars . any shell from a supernova would have to be pre - existing ( as in the supranova model of vietri & stella , 1998 ) since the afterglow jet would be far outside any normal supernova ejecta if the supernova and burst exploded simultaneously . @xcite point out there are difficulties with producing narrow lines with the observed equivalent width from an old supernova remnant . the narrowness of the absorption lines would require that the ejecta has already cooled and condensed into filaments . the question is then the likelihood of getting multiple filaments of greatly different velocity on our line of sight and the column density of any such filaments . ideally , the absorption lines from the @xmath2 system can tell us about the abundances of carbon , silicon , hydrogen , and nitrogen in the innermost shell ejected by the progenitor ; and this can then tell us about the progenitor itself . for example , if the composition is that of a wolf - rayet shell of a particular type ( perhaps with some swept up ism material ) , then we would even get a good idea of the mass of the progenitor . unfortunately for this program , we have few lines from the @xmath2 system ( see the many lines for grb 010222 , @xcite ) . also , the likelihood that our unresolved lines are saturated implies that the measured equivalent widths will be difficult to convert to abundances . nevertheless , a full analysis based on our data or the data of , for example , @xcite , might place useful limits on the abundances of the shell and its possible origin . an example of such an analysis can be made with respect to the carbon - to - nitrogen abundance . both atoms have similar ionization potentials and should have somewhat similar ionization fractions into either or , the oscillator strengths of the two resonance lines are similar , and the cosmic abundance of nitrogen is down from that of carbon by a factor of three . so simplistically , we would expect the line to have an equivalent width that is roughly three times deeper than that of , whereas ( see table 1 ) we do not see any nitrogen lines and the limit on the equivalent widths is a factor 4.6 . with the likelihood that the line has a saturated core , the implication is that nitrogen is significantly under - abundant compared to carbon . this would appear to rule out wn type wolf - rayet stars . the lower mass wolf - rayet stars undergo core collapse while the wind is still nitrogen rich @xcite , whereas the higher mass wolf - rayet stars have core collapse after the wolf - rayet has turned from a wn to a wc type @xcite . thus , the lack of a line in the spectrum of grb 021004 appears to imply that the burst progenitor must be amongst the more massive of wolf - rayet stars . alternatively , the progenitor might be a star for which the original mass was too low ( @xmath99 @xmath100 ) to allow it to become a wn - type star , yet it would still have fast and dense winds that are not nitrogen - rich . the lack of molecular hydrogen absorption can be used to place further limits on the density of the gas near the burster . in the model of @xcite , the column density of excited @xmath101 , @xmath18 , arises from an ionization / dissociation front the dynamics of which imply that there are natural values of the column depth . for example , an increase in the burst fluence would produce a similar front ( only deeper into the molecular cloud ) with a similar @xmath18 , while an increase in the molecular density would produce a front with a narrower width yet with a similar @xmath18 . the only way to produce a smaller column density is to have uv fluxes that are not sufficient to establish a proper ionization / dissociation front in the molecular hydrogen . this could arise if the molecular cloud were at a great distance ( say , 1 kpc ) from the burster or if the burster were near the edge of the molecular cloud . draine calculates that a typical burst will destroy molecular hydrogen in gas with a hydrogen column density of @xmath102 ( corresponding to @xmath103 or so ) . for densities from 10010000 @xmath104 , the burst will destroy the molecular hydrogen that could serve as an absorber if it is within roughly 2200 pc of the edge of the cloud . for the general case of the burst erupting well inside a molecular cloud , the @xmath18 will be the value used to calculate the template ( see section 2 ) , and this is rejected by our spectrum . we conclude that the lack of molecular hydrogen absorption lines in our spectra indicate that the @xmath0-ray burst is not inside a molecular hydrogen cloud unless it is within @xmath42 - 200 pc from its edge . we have constructed broad - band fluxes from the radio to the x - ray for several epochs ( see figure 3 ) with the data collected from the gcn notices . the major features of the radio emission are : @xmath105 at 0.53.5 days the 15 ghz light - curve rises as @xmath106 , @xmath107 at 1.5 days the 1586 ghz spectrum is @xmath108 , appearing flat from 86 to 230 ghz , and @xmath109 at 5.7 days the 1.48.5 ghz spectrum is @xmath110 . the last two features indicate that the synchrotron self - absorption frequency @xmath111 is not significantly above 15 ghz at 1.5 days and not much below 8 ghz at 5.7 days , which would suggest a wind - like medium ( for which @xmath112 ) instead of a homogeneous environment ( for which @xmath111 is constant ) . however , with only a few measurements and the possibility of significant interstellar scintillation affecting the lower frequency ( under 15 ghz ) observations , it is premature to favor a wind - like medium based on the apparent decrease of @xmath111 over a factor of 4 in time . the rise of the 15 ghz emission is consistent with both the expectations from a homogeneous medium ( where @xmath113 irrespective of the location of @xmath111 ) and a behavior intermediate to the asymptotic @xmath114 and @xmath115 expected for a wind - like medium . within the framework of relativistic fireballs , consistency between the decay rate of the optical emission ( @xmath116 with @xmath117 ) and spectral slope ( @xmath118 with @xmath119 ) requires that the cooling frequency @xmath120 is below or within the optical range , as in this case the expected relationship ( @xmath121 , irrespective of the type of external medium ) is satisfied . furthermore , that @xmath120 is below optical is also consistent with the equality of the optical and x - ray decay indices ( @xmath122 ; sako & harrison 2002 ) and spectral slopes ( @xmath123 ) . we note , however that the difference between the observed optical and x - ray decay indices , @xmath124 , is also consistent ( even if only marginally ) with the theoretically expected value @xmath125 in the case where @xmath120 is above the optical domain but below x - rays , and that the intrinsic optical spectral slope @xmath126 can be smaller than observed if there is a significant dust reddening within the host galaxy . therefore , the cooling frequency could be above the optical domain , in which case one expects that @xmath127 for a homogeneous medium , and @xmath128 for a wind - like medium . furthermore it is also expected that @xmath129 , which favors a homogeneous medium when compared with the above @xmath126 inferred from the optical decay . concluding , the currently available radio and optical data for grb 021004 do not constrain sufficiently the type of external medium , homogeneous or wind - like . however , the average density of the medium or the mass - loss rate to wind speed ratio can be determined if the cooling frequency lies within the optical range at about 2 days , as suggested by matheson et al 2002 . together with the optical measurements , the 15 , 86 and 230 ghz fluxes reported by pooley ( 2002 ) and bremer & castro - tirado ( 2002 ) indicate that , at 1.5 days , the peak of the spectrum of grb 021004 is @xmath130 mjy , at a frequency between @xmath131 and @xmath132 hz . adding that @xmath133 ghz at a few days , these four constraints on the synchrotron spectrum are sufficient to determine the external medium density , fireball isotropic - equivalent energy , and fractional energies in electrons and magnetic fields . we obtain an external density in the range @xmath134 for a homogeneous medium , consistent with the values found by panaitescu & kumar ( 2002 ) for other grb afterglows . for a wind - like medium , we find that density of the wind corresponds to that resulting from a mass - loss rate to wind speed ratio in the range 1 to several times @xmath135 km s@xmath1 ) , within the range expected for wr stars . we have constructed light curves and color curves for grb 021004 from data published in the gcn notices ( see figure 4 ) . overall , the light curve behaves approximately as a @xmath136 power law which is normal for many afterglows . however , there are many significant bumps and wiggles about any simple single power law . to help visualize these bumps , the middle panel of figure 4 shows the r - band light curve after a @xmath136 power law has been subtracted , and we see three bumps with peaks at 0.14 , 1.1 , and 4.0 days after the burst . the v - r color is plotted in the lower panel of figure 4 . the color appears to remain essentially constant from 0.2 to 10 days . afterglow models suggest that light curves should consist of broken power laws , so there is an urge to try to interpret figure 4 in terms of broken power laws . perhaps the basic behavior is @xmath136 with bumps and some separate mechanism for the dimness at the earliest times . alternatively , perhaps the basic behavior is a @xmath137 power law up until a break at 3 days to a @xmath138 power law with a particularly strong bump from 0.10.7 days . for a third alternative , perhaps the basic behavior is a @xmath138 decline over the entire time range except that some mechanism inserts more energy three times during the decline ( at 0.14 , 1.1 , and 4.0 days ) . in general , @xmath0-ray burst afterglows have simple power law declines in brightness , often with a break ascribed to collimation of the outflow . nevertheless , many @xmath0-ray bursts have bumps in their afterglows . for example , the optical brightness of the afterglow of grb 970508 increased by 1.5 magnitudes 12 days after the @xmath0-ray burst @xcite . grb 970508 was seen by the bepposax satellite to have a significant brightening in its x - ray light curve starting at 16 hours after the burst @xcite . grb 000301c displayed an achromatic bump that peaked around 4 days after the burst that has been ascribed to gravitational microlensing by @xcite . in addition , three @xmath0-ray bursts with bumps around 30 days after the burst have been ascribed to an underlying supernova ( reichart 1999 , bloom et al . 1999 ; 2002 ; but see dermer 2002 ) or dust echos ( reichart 2001 ) . the presence of bumps in otherwise - smooth afterglow light curves has previously been attributed to microlensing , dust echos and supernovae . now we have a burst with three bumps , and they can not all be due to microlensing , dust echos , or supernovae , so we have a proof that afterglows can makes bumps on various time scales for other reasons . thus , the mere presence of a single bump can only provide poor support for any particular proposed cause unless other possibilities can be excluded by some means . for example , the red color of a bump ( compared to the underlying afterglow ) can not be taken as substantial evidence of a supernova origin until the cause and color - distribution of other classes of bumps are known . there are two generic scenarios ( in addition to microlensing , dust echos , and supernovae ) for producing variability in the afterglow emission : an inhomogeneous circumburst medium and non - uniform properties of the @xmath0-ray burstejecta ( see also nakar , piran & granot 2003 ) . inhomogeneities in the external gas could be either in the form of clumps of denser material @xcite , which enhance the dissipation of the kinetic energy of the @xmath0-ray burst remnant and , implicitly , the afterglow brightness , or in the form of a stratified medium , consisting of shells of various densities @xcite , perhaps produced by fluctuations in the wind blown by the @xmath0-ray burst progenitor not long before the release of the @xmath0-ray burst ejecta . non - uniformities in the @xmath0-ray burst ejecta could be manifest either as a range of initial lorentz factors within an impulsive ejection or as a non - isotropic distribution of the ejecta energy per solid angle . in the former case , the slower part of the outflow catches up with the faster part as the latter is decelerated , enhancing the afterglow kinetic energy and , hence , its brightness . in the latter case , the regions of higher angular energy density are bright spots on the ejecta surface , moving slightly off the direction toward the observer , so that fluctuations in the afterglow emission are produced as the spots decelerate and the widening cone of their relativistically beamed emission starts to include the direction toward the observer . the variable - lorentz - factor and anisotropic outflow scenarios have been used to explain the brightening of the afterglow of grb 970508 by @xcite and @xcite , repectively . an energy input that is extended in time prior to the brightening at 0.14 days could explain the shallower decline prior to 0.14 days , while an inhomogeneous shell with a density decreasing with radius @xcite could account for the steeper light - curve decay after 0.1 days in the stratified medium scenario . the faster decline after a brightening is consistent with the expectations from scenarios involving dense clumps or bright spots , where the contribution of the clump or spot to the total afterglow emission decreases in time as the fireball decelerates , and more of the uniform ejecta surface becomes visible for the observer . in the latter two scenarios where the afterglow emission from clumps or spots declines with time , the afterglow light curve should eventually recover its initial decay rate . in this context , the observed light curve is consistent with the proposal that the @xmath137 fall - off seen in the first 15 minutes reappeared at 0.7 days , just about the time the spectra reported here were obtained , and lasted until 13 days after the burst , when the light curve steepened to @xmath139 . if this interpretation is correct , grb 021004 would display the shallowest light - curve break ever observed . the change in the break index would be @xmath140 , suggesting that the break is caused by the passage of the cooling break through the optical domain @xcite . the existence of a break in the optical spectrum at 13 days is supported by the spectroscopic observations by @xcite and this paper ( see 2 ) , that the blue part of the optical spectrum reddens between @xmath141 and @xmath142 days . this interpretation is consistent with a break frequency that decreases in time , implying that the external medium is homogeneous rather than wind - like . in the case of a wind , the cooling frequency should increase at least as fast as @xmath143 @xcite . we note , however , that the break in the optical spectrum seems too sharp in wavelength to be due to the cooling break ( across which the slope of the power - law spectrum changes by 0.5 ) , and that the spectral changes may not sufficiently affect the r - band light - curve at a few days to explain the r - band light curve behavior . the radial distance of the jet from the center of the progenitor as a function of the observer s time should be correlated with the location of clumps and the epoch of the light curve bumps . this can be modelled given both the broad - band spectrum and the light curve by the methods presented in @xcite . unfortunately , only preliminary input numbers are available ( for example , no jet break has yet been definitely timed ) and details of the external medium are not known ( for example , whether or not the average density falls off like a wind ) . nevertheless , it is possible to give estimates that are useful for producing a reasonable picture . for this , we adopt the isotropic equivalent energy of the fireball equal to @xmath144 erg @xcite . if the jet is expanding into an external medium with a wind - like structure , then we found in the previous section that the mass - loss - rate to wind - speed ratio is roughly @xmath145 per 1000 km s@xmath1 . in this case , the radius of the jet will be @xmath146 pc , with @xmath41 measured in days . if the jet is expanding into a homogenous medium of constant density , then we found in the previous section that the density is roughly 30 @xmath147 . in this case , the radius of the jet will be @xmath148 pc . if the light curve bumps are due to density enhancements in the surrounding medium , then we can estimate the approximate radial distances from the burst and over - densities for each clump . for peaks in the light curve at 0.14 , 1.1 , and 4.0 days , we find approximate radii for the density clumps of 0.022 , 0.063 , and 0.12 pc , respectively , for a wind - like medium . we get similar results for expansion into a homogenous medium with parameters as derived above . the excess light in the light curve bumps will scale roughly as the square root of the over - density compared to the background " density , for the case where the clump is relatively large compared to the visible region ( as for a shell ) . from figure 4 , the extra light in each of the bumps is roughly 0.6 , 0.3 , and perhaps 0.3 mag ; this corresponds to over - densities of roughly 50% , 10% , and 10% . if the bumps are due to clumps , the broad - band spectral slopes should remain the same and thus the colors would not change across the bump , as observed . at some time the afterglow jet will collide with the gas that is causing the @xmath2 absorption . when this happens , we would expect that the afterglow will brighten significantly if the gas is in a relatively narrow shell . we also expect that the associated absorption lines will disappear . we note that our late het spectrum still showed the @xmath149 line with moderately low significance at a time of 4.84 days after the burst when the jet would have reached @xmath150 pc . in principle , we can predict the date at which this bump in the afterglow decline should occur . in practice , such a prediction will have substantial uncertainties , primarily in the exact radius of the shell and the density structure of the medium inside the shell . in the case where a shell of material has been radiatively accelerated to 3000 km s@xmath1 , the shell will have a radius not much larger than 0.1 pc and it should be over - run sometime between roughly 240 days after the burst . in the case where the absorption is due to a wolf - rayet wind that extends out to several parsecs , the 3000 km s@xmath1 component will be completely eliminated only after many years . the absorption will slowly diminish during the first few months after the burst as the jet catches up with the densest inner part of the wind . in summary , the absorbing gas might be eliminated during the early days after the burst ( an event that could be directly observable ) or it might take many years to be eliminated ( and be totally unobservable ) . the grb 021004 spectrum displays high - ionization and high - velocity absorption lines . we argue that these lines are likely to have arisen from shells of gas around the progenitor star , and further , that these shells plausibly come from a progenitor that was descended from a massive star . the gas in these shells was partially ionized and accelerated by the photons from the burst . we have used the optical spectrum , the broad - band spectrum , and the light curve to identify six components in the medium external to the @xmath0-ray burst . the @xmath2 absorption lines arise from a shell at a radial distance of less than a parsec that is moving at a velocity of 3000 km s@xmath1after being ionized and accelerated by the burst radiation . we identify this component as being the wind from the progenitor star , perhaps a wolf - rayet star with a mass loss rate of @xmath151 or a lower mass star wind accelerated by the bust . this same component provides the external medium with which the jet is colliding to produce the afterglow light . this inner component is probably significantly clumped in order to prevent complete ionization and to facilitate recombination , thus allowing a significant column depth of . the second component is associated with the absorption lines moving 560 km s@xmath1 with respect to the host galaxy velocity . we identify this component as arising from a shell of material plowed up by the wolf - rayet - like wind at a radius of order 10 parsecs . alternatively , this feature could be a shell or clump at @xmath4 0.5 pc accelerated by the burst . the third component gives rise to the low - velocity high - excitation absorption lines which must come from a distance between tens and hundreds of parsecs from the burster . this component might be from the shell created by the o main sequence star wind or it might be from chance ism clouds along the line of sight , perhaps part of the same star forming region that gave birth to the @xmath0-ray burstprogenitor . the fourth , fifth , and sixth components are three density enhancements that occur in the wolf - rayet - like wind at radial distances of 0.022 , 0.063 , and 0.12 pc so as to produce the three bumps in the light curve . these clumps could be related to the clumping we deduce for component 1 , that is , all could be associated with knots in the progenitor wind . the results summarized in the previous paragraph offer a plausible description of the density structure near this burst . what we see is a variety of density enhancements along the line of sight , with the evidence coming from bumps in the light curve decline and absorption lines in the spectrum . this picture is apparently not universal for @xmath0-ray bursts , since only two out of nine bursts with well - observed afterglow spectra show both high - excitation and high - velocity lines . the other bursts must have some mechanism for suppressing similar absorption lines . a full analysis is needed for the @xmath12 lines of grb 021004 , with the possibility of determining the abundance of the progenitor s shell and hence to obtain information about the progenitor itself . as a start to this program , we have shown that the nitrogen in the shell is under - abundant with respect to the carbon , and this demonstrates that the progenitor was not a wn - type wolf - rayet star and further that the progenitor was either one of the more massive wolf - rayet stars or a massive star that is below that required to make a wn - type star . a further program for our community is to obtain high dispersion spectroscopy so as to be able to measure fine velocity structures and faint lines in the afterglow spectra of many bursts . the most important result from our data and analysis is that a plausible way to create the high - excitation and high - velocity absorption lines is to have shells of gas surrounding the progenitor , and that a realistic way to achieve this is for the progenitor to be a star descended from a massive o - type main sequence star . this strongly suggested connection to a massive star is much more direct than the moderate statistical association of @xmath0-ray bursts with parts of galaxies undergoing star formation and is much more convincing than attributing poorly observed bumps in afterglow light curves to some associated supernova event . as such , we believe that the absorption lines in the grb 021004 spectra provide the first strong and direct argument for associating a normal @xmath0-ray burst with a massive progenitor star surrounded by shells . we thank bruce draine , bev wills , greg shields , and d. j. hillier for their discussion on a wide range of topics . this work was supported in part by nasa grant nag59302 and nsf grant ast-0098644 . the marcario low resolution spectrograph is a joint project of the hobby - eberly telescope partnership and the instituto de astronomia de la universidad nacional autonoma de mexico . the hobby - eberly telescope is operated by mcdonald observatory on behalf of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen . ccccc @xmath152 & si iv & 1393.8 & @xmath153 & 2.291 + @xmath154 & si iv & 1393.8 & @xmath155 & 2.321 + @xmath154 & si iv & 1402.8 & @xmath155 & 2.299 + @xmath156 & si iv & 1402.8 & @xmath157 & 2.326 + @xmath158 & c iv & 1548.2 & 1550.8 & @xmath159 & 2.296 + @xmath160 & c iv & 1548.2 & 1550.8 & @xmath161 & 2.325 + @xmath162 & al ii & 1670.8 & @xmath163 & 2.328 + @xmath164 & n v & 1238.8 & 1242.8 & @xmath165 & @xmath166

we have used the locally self - consistent green s function ( lsgf ) method in supercell calculations to establish the distribution of the net charges assigned to the atomic spheres of the alloy components in metallic alloys with different compositions and degrees of order . this allows us to determine the madelung potential energy of a random alloy in the single - site mean field approximation which makes the conventional single - site density - functional- theory coherent potential approximation ( ss - dft - cpa ) method practically identical to the supercell lsgf method with a single - site local interaction zone that yields an exact solution of the dft problem . we demonstrate that the basic mechanism which governs the charge distribution is the screening of the net charges of the alloy components that makes the direct coulomb interactions short - ranged . in the atomic sphere approximation , this screening appears to be almost independent of the alloy composition , lattice spacing , and crystal structure . a formalism which allows a consistent treatment of the screened coulomb interactions within the single - site mean - filed approximation is outlined . we also derive the contribution of the screened coulomb interactions to the formalism and the generalized perturbation method .

cond-mat0202370

the coherent potential approximation ( cpa ) as implemented on the basis of multiple scattering theory @xcite and combined with density functional theory ( dft ) @xcite constitutes the basis for _ ab inito _ calculations of the electronic structure and physical properties of random metallic alloys . this combination of the cpa with dft , or , in most cases , with the local density approximation ( lda ) , seems to be quite transparent @xcite leading to expressions for the one - electron potential and total energy which are very similar to those for ordered systems . however , there is , by now , a well - recognized problem @xcite with this description caused by the fact that the atomic or `` muffin - tin '' spheres , which _ artificially _ divide the crystal into regions associated with particular alloy components , may poses non - zero net charges . the problem stems from the fact that the conventional single - site ( ss ) dft - cpa method is based on the effective medium model of a random alloy which considers only conditionally averaged quantities and leads to the use of the single - site approximation not only in the electronic structure part of the problem during the solution of the cpa equations , but also in the dft self - consistent loop in the calculations of the electrostatic contributions to the one - electron potential and energy . the single - site approximation provides no information as to the charge distribution beyond the atomic sphere of each alloy component and , since the surrounding effective medium is electroneutral , poisson s equation can not be solved properly if the atomic spheres have non - zero net charges . hence , to find the correct solution to poisson s equation one must somehow describe the effect of the missing charge . since the electron density inside each atomic sphere is well - defined , any such description may be associated with a modification of the effective medium specifically for _ each _ alloy component . this may be regarded as an inconsistency since , in that case , the cpa and the electrostatic part of the dft are based on different effective media . one obvious solution to the problem is to use electroneutral spheres ( see , for instance , ref . however , in the methods based on the atomic sphere approximation ( asa ) this frequently leads to large sphere overlaps and a quite poor description of the electronic structure , especially in the case of inhomogeneous system , such as partially ordered alloys or surfaces with an inhomogeneous concentration profile . a more general solution can be found , however , in which the electrostatic potential is modified without making effective media for each alloy component in contradiction to the assumptions of the cpa . the way to do this is to introduce an additional shift of the one - electron potential due to the electrostatic interaction of the electrons inside each atomic sphere with the missing charge distributed outside of the sphere and postulate that the interaction comes from the boundary between the atomic sphere and the effective medium . such a shift may be associated with an _ intrasite _ interaction , which has no connection , at all , to the effective medium . this is exactly what is done in the locally self - consistent green s function method @xcite where one goes beyond the single - site approximation for poisson s equation by means of a supercell which models the spatial distribution of the atoms in a random alloy while a cpa effective medium is used in the electronic structure calculations beyond a local interaction zone ( liz ) . if the liz consists of only one atom , the lsgf method becomes equivalent to the cpa method with a properly defined electrostatic potential and energy @xcite . in this case , however , each atom in the supercell has its own electrostatic shift given by the madelung potential from all the other atoms in the supercell while the effective medium is the same for all atoms . it is clear that such an additional shift for each alloy component does not interfere with the cpa because the cpa effective medium is determined on the basis of the one - electron potentials including these shifts and because the cpa itself does not impose any restriction on the one - electron potentials of the alloy components . following the above arguments two groups have proposed an _ ad hoc _ expression for the electrostatic shift of the one - electron potential due to non - zero net charges in the atomic spheres of the alloy @xcite . although the basic models are seemingly different and based on different observations , either i ) , the net charge of an impurity in a metal is screened beyond the first coordination shell @xcite or ii ) , the net charge of an alloy component is proportional to the number of the nearest neighbors of the opposite type @xcite , they lead to exactly the same expression for the one - electron potential , i.e. , @xmath1 where @xmath2 is the additional electrostatic shift of the one - electron potential of the @xmath3-th alloy component of net charge @xmath4 and @xmath5 the radius of the first coordination shell . in fact , the models described above are practically identical to the model proposed more than three decades ago @xcite to account for charge transfer effects in the self - consistent hartree scheme based on the tight - binding cpa . in this scheme the variation of the @xmath3-th atom energy level , @xmath6 , is proportional to the corresponding charge transfer @xmath4 , i.e. , @xmath7 , where @xmath8 is some average _ intra - atomic _ coulomb interaction . the non - self - consistent limit corresponds to @xmath9 , while @xmath10 provides local neutrality @xcite . in the present context one may identify @xmath6 with @xmath2 and it therefore follows that @xmath11 may be considered an intrasite coulomb interaction . although there is at least some consensus concerning the definition of the additional electrostatic shift ( [ eq : v_sim ] ) , which gives charge transfers quite close to the values obtained in supercell calculations @xcite , different workers do not agree on the corresponding electrostatic contribution to the total energy of the random alloy . some completely deny even the possibility of having such a term in a `` consistent '' ss - dft - cpa theory @xcite while others argue about the details of how this term should be defined . it would seem that the presently suggested models of charge transfer effects in the single - site approximation to the electrostatic problem , except the trivial elimination of the net charges by adjusting the radii of the atomic spheres of the alloy components , may be considered neither exact nor even `` a consistent theory '' . it is the main purpose of the present paper to shown that a consistent ss - dft - cpa theory including a correct description of the charge transfer effects does indeed exist . here , we define the electrostatic shift of the one - electron potential and the corresponding contribution to the total energy in a form which is very similar to that proposed by korzhavyi _ _ @xcite as well as johnson and pinski @xcite , and which provides a practically _ exact _ solution to the electrostatic problem in the single - site approximation . the actual expression for the electrostatic shift in the single - site model for poisson s equation includes one adjustable parameter the value of which is to be obtained in supercell calculations by the locally self - consistent green s function method in which the madelung problem is solved exactly . it turns out that the value of the adjustable parameter is practically independent of lattice structure , volume , and alloy composition due to the fact that the screening of the electrostatic part of the problem , in the asa at least , is almost universal . the paper is organized as follows . in sec . ii we outline the main concepts behind the lsgf used in this work and the details of the calculations . in sec . iii we present a pragmatic solution to the problem of finding the madelung shift in ss - dft calculations on the basis of the average values of the net charges and coulomb shifts , @xmath12 and @xmath13 , from supercell calculations . we also demonstrate that the linear relation between the net charges , @xmath4 , and the corresponding coulomb shifts , @xmath2 , of the alloy components discovered for metallic alloys by faulkner _ et al . _ @xcite is practically universal in the effective medium approach for the green s function . this means that the response of the electron system to the coulomb field is linear and universal in such systems , and that the screening must be universal too . that this is indeed the case is demonstrated in the next section where we calculate the distribution of the screening charge in several systems and show that it is almost independent of the crystal structure , the alloy constituents , and the composition . in sec . v we present a formalism for the screened coulomb interactions in the single - site mean - field approach for the electrostatic potential and energy and demonstrate that the conventional assumption of a vanishing madelung potential and energy is not valid in general . instead , one must include an additional term due to _ intrasite _ interactions which are , in fact , exactly the screened coulomb interactions . we also discuss the ordering contribution to the madelung energy and show why the screening contribution may be obtained in supercell calculations for ordered structures . the contribution from the screened coulomb interaction to the generalized perturbation method and the @xmath0 formalism is also determined . finally , in sec . vi , we demonstrate that the total energy of a random alloy may be reproduced exactly in single - site cpa - dft calculations with corrections due to the screening intrasite interaction . in this paper we will consider only such alloy systems that on an underlying crystal lattice with perfect translational symmetry satisfy two conditions : i ) spatial homogeneity and ii ) no correlations between the one - electron potentials at sufficiently large distances . for the coulomb interactions in a random alloy both conditions may be formulated explicitly in terms of the average monopole electrostatic potential @xmath2 in the atomic sphere around site @xmath3 due to the charge distribution in the whole of the remaining system . in an ordered alloy this potential is the madelung potential . specifically , the first condition means that any average values of products of potentials must be translationally invariant , i.e. , @xmath14 where @xmath15 is the translation operator @xmath16 , and the second condition is @xmath17 for @xmath18 . according to lifshitz _ @xcite the space formed by the complete set of distinct realizations of the potential @xmath2 on the lattice , the operator @xmath15 , and the property of spatial homogeneity plays the same role in the theory of disorder as the phase space , the operator of dynamical evolution , and liouville s theorem do in statistical mechanics . moreover , according to birkhoff s ergodic theorem , for any functional @xmath19 $ ] , where @xmath20 is some random realization of the potential on the lattice we have @xmath21 = < f[{v_i } ] > , \ ] ] i.e. , the spatial and the phase - space average are equivalent @xcite . this equation constitutes the principle of spatial ergodicity , according to which all possible finite atomic arrangements may be realized in a single infinite sample if the conditions ( [ cond1 ] ) and ( [ cond2 ] ) are satisfied . what makes the above principle work in practice is the fact that for self - averaging or `` measurable '' quantities , which have a well - defined limit when the volume of the systems approaches infinity , all the correlations of the atomic distribution become unimportant at some distances and , hence , the sample may be chosen finite . this may be formulated explicitly by means of the cluster expansion theorem @xcite which defines the corresponding measurable quantity in terms of the site occupation correlation functions @xmath22 where @xmath23 are the coefficients or interaction parameters , @xmath24 the correlation function of the figure or cluster @xmath25 which corresponds to a specific position of the sites @xmath3 , @xmath26 , and @xmath27 in the lattice , and @xmath28 = @xmath29 is the fluctuation of the site occupation numbers @xmath30 taking on values 0 and 1 depending on whether site @xmath3 is occupied by one or the other component . according to ( [ clexp ] ) there are two practical ways of calculating the properties of a random system for which we have @xmath31 and , thus , @xmath32 : i ) the cluster or supercell approach , where @xmath31 is satisfied on average only for those clusters @xmath25 for which @xmath33 , or ii ) the effective medium approach , which directly gives @xmath34 from some knowledge of the alloy components . the first approach is realized , for instance , in the so - called special quasirandom structure ( sqs ) method @xcite while the second approach is realized by the coherent potential approximation where the real atoms are substituted by a specifically chosen effective medium on the lattice . the supercell and the effective medium approaches are combined into a single computational scheme in the locally self - consistent green s function ( lsgf ) method @xcite . in the lsgf method the supercell approach is used to provide the correct solution to the madelung problem for a given alloy modeled by an appropriate supercell . it is also used in part in the electronic structure calculations , which are performed separately for every atom in the supercell by means of the local interaction zone centered at each atom . inside the liz the multipole scattering equations are solved exactly , while the region outside the liz is represented by the effective medium , which is usually taken to be the cpa effective medium built on all the one - electron potentials in the supercell . this means that every atom of the supercell `` sees '' only the cpa effective medium outside the liz , which according to the cpa definition represents a random alloy . in other words , the one - electron green s function of the supercell ( sc ) obtained in lsgf calculations may be presented as @xcite @xmath35 , \ ] ] where @xmath36 is the green function of the @xmath3th atom in the supercell embedded in the cpa effective medium , @xmath37 the contribution to @xmath38 due to the presence of the specific atomic arrangements on the figure @xmath25 in the liz as specified by the correlation function @xmath39 , which is equal to that of the supercell @xmath40 if the figure @xmath25 is circumscribed by the liz , i.e. , it can be put inside the liz in such a way that one of its vertices coincides with the central atom of the liz , otherwise @xmath39=0 . equation ( [ g_lsgf ] ) clearly shows how the lsgf works , and in particular , how the effective medium approach , represented by @xmath36 , is combined with the cluster or supercell approach , represented by by the second term . it follows from ( [ g_lsgf ] ) that the liz allows one effectively to cut off the contributions from the clusters which are not circumscribed by the liz ( an equivalent formulation in terms of effective interactions is given in ref . if the liz is single - site , i.e. , it consists of only one atom ( liz=1 ) , the contributions from the second term in ( [ g_lsgf ] ) vanish and we are left with the usual cpa or pure effective medium approach to the electronic structure problem . it is this single - site approach , referred to as ss - lsgf , which will be used in most of the present work . it has the advantage over the usual ss - dft - cpa method that poisson s equation is solved exactly within a given approximation for the form of the electron density . note , however , that the lsgf method by no means is restricted to the single - site approximation . in fact , it allows us to include local environment effects in the electronic structure calculations for the figures circumscribed by the liz if on average @xmath41 which is the case in a random alloy . in this respect the lsgf method may be considered a _ self - consistent _ embedded cluster method ( ecm ) of the kind proposed by gonis _ @xcite more than two decades ago . with a proper choice of the supercell used to model a given random alloy @xcite the lsgf solves two major problems of the ecm : i ) it provides a set of clusters to represent an alloy with a given short range order and ii ) it allows one to close the dft loop with the correct treatment of the electrostatics . there is one important point concerning the electronic structure obtained in the lsgf with the cpa effective medium which should be mentioned : despite the fact that it can be quantitatively accurate , it is _ qualitatively _ different from the electronic structure which would result from direct supercell calculations with periodic boundary conditions . that is , the electronic spectrum in the lsgf - cpa method is always complex , unless all the atoms are equivalent in the supercell ( pure metal ) or the size of the liz is infinite . thus , in the lsgf - cpa the electronic structure of an _ ordered _ alloy is never correct , although it may be calculated with arbitrarily high accuracy . on the other hand , since blochs theorem is not applied to the supercell during the electronic structure calculations , the lsgf method is a perfect tool for calculating sqs as opposed to the ordinary band structure methods , which in this case lead to a real , i.e. , qualitatively incorrect , electronic structure of the random alloys due to the fact they present purely a supercell approach . it is possible to obtain the madelung potential and energy by a combined supercell ( cluster ) and effective medium approach similar to that used in the green s function approach to the electronic structure problem within the lsgf method . however , this requires some knowledge of the charge - charge correlations or the screening in the alloy . hence , the only way to solve the problem is to use a supercell model with the madelung potential and energy determined exactly from the _ bare _ electrostatic interactions , as it is usually done . here , another problem arises : the supercell should be constructed such as to provide zero correlation functions up to the distance where the net charges of the alloy components become uncorrelated or completely screened . in the calculations presented below we assume at the outset the existence of a short - range screening which occurs over the distance of the first several coordination shells . this assumption is based on results obtained by the charge - correlated model @xcite , on single - impurity calculations @xcite as well as on the most recent lsgf calculations by ujfalussy _ the latter authors demonstrated that a 16-atom supercell for an fcc equiatomic random alloy , in which the sro parameter at the 8th coordination shell must be equal to 1 due to the translation symmetry , i.e. , all the atoms in the 8th coordination shell are the same as that at the central site , yields practically the same average charge transfer and total energy as a 250-atom supercell , in which the sro parameter at the 8th coordination shell corresponding to a random number generator distribution of the alloy components on the lattice should be @xmath42 1 ( see also @xcite ) . thus , in all the random alloys considered below the distribution of the atoms in the supercell was chosen such that the sro parameters ( or pair correlation functions , @xmath43 ) were exactly zero at least in the first 6 coordination shells and small ( not greater 0.01 in absolute value ) up to the 8th coordination shell . although the multisite correlation functions have not been optimized , they should not play a significant role in the asa where only monopole intersite coulomb interactions are taken into account . the electronic structure calculations were in all cases performed in the scalar relativistic approximation by the kkr - asa technique @xcite with an @xmath44- , @xmath45- , and @xmath46basis in the framework of either the usual ss - dft - cpa or the lsgf methods with a cpa effective medium . the asa ( no multipole corrections to the madelung potential and energy ) has been used in the electrostatic part of the problem . the integration of the green s function over energy was performed in the complex plane over 16 - 20 energy points on a semicircular contour . the local density approximation was used in the dft part with the perdew and zunger @xcite parameterization of the results by ceperly and alder @xcite . here we discuss a pragmatic solution to the following problem : can one devise a madelung potential for the alloy components to be used in ss - dft - cpa calculations such that the charge transfer effects , i.e. , the net charges of the alloy components , are consistent with those obtained in ss - lsgf calculations where charge transfer effects are treated properly ? the fact , that such a potential can be found , may seem surprising in view of the principle difference between the lsgf and the ss - dft - cpa methods . in the lsgf approach all the atoms in the supercell are different due to their different local environment while in the usual ss - dft - cpa approach one deals only with average quantities , i.e. , in terminology of the ref . @xcite , the lsgf supercell approach is equivalent to the _ polymorphous _ model of the alloy while the effective - medium approach is equivalent to the _ isomorphous _ model . however , it is obvious , that this can be done _ on average_. it was discovered by faulkner _ et al . _ @xcite from supercell calculations that the net charges on different sites @xmath3 , @xmath4 , and the corresponding madelung potentials , @xmath2 , obey a linear relationship . in fig . [ qv_alli ] we show such a @xmath47 relation for a 512-atom supercell which models a random al@xmath48li@xmath48 alloy on an underlying fcc lattice . for comparison we also show the charge and madelung shift for alli in the ordered l1@xmath49 structure . all results are obtained by lsgf calculations with the cpa effective medium in i ) the single - site approximation for the electronic part of the problem , ss - lsgf , i.e. , liz=1 , ( upper panel ) and ii ) with the perturbation in the electronic structure caused by the local environment up to the second coordination shell , i.e. , embedded cluster(ec)-lsgf , liz=3,(lower panel ) , included in the green s function . for random alloys the inclusion of more distant coordination shells do not affect the results significantly and thus the liz=3 results may be considered to be converged in the liz size . the most striking feature of the @xmath47 relation obtained in the ss - lsgf calculations is the perfect alignment of the @xmath47 points along two almost straight lines , one for each alloy component . this is , in fact , very similar to what has been observed by pinski @xcite in model calculations using the thomas - fermi approximation . furthermore , a change of the ratio of the atomic sphere radii of the alloy components , @xmath50 , leads to a rescaling of the @xmath47 points . hence , for a specific ratio , @xmath51 in the present case , the @xmath47 relation collapses into the single point : @xmath52 . the existence of this point in the ss - lsgf is a consequence of the fact that all al atoms as well as all li atoms become indistinguishable if the net charges of the alloy components are zero : the difference between the atoms caused by local environment effects is solely due to the madelung shift , which is zero in this case . thus , for this particular choice of @xmath53 the polymorphous model is identical to the usual isomorphous model , or the ss - lsgf method is identical to the ss - dft - cpa method . on the other hand , it is clear that the two models are not equivalent when local environment effects are included in the electronic structure part of the lsgf method , i.e , for liz@xmath541 . this is demonstrated in the lower panel of fig . [ qv_alli ] where the local environment effects are clearly seen to destroy the strict alignment of the @xmath47 points and , as consequence , the possibility of choosing electroneutral atomic spheres by a single @xmath53 value . however , even if this were possible , all the atoms , or the corresponding one - electron potentials , would still be different . the discussion of local environment effects is beyond the scope of the present paper , and the results are included only to demonstrate the _ qualitative _ difference between the correct results and those obtained by the lsgf method in the single - site approximation : inclusion of intersite correlations in the electronic structure calculations leads to a real polymorphous description of random alloys which can not be mimicked by a single - site liz . as a consequence , as we will show later the ss - dft - cpa method can reproduce the results of the ss - lsgf exactly , but will , in general , reproduce only approximately the correct solution to the supercell or polymorphous model of a random alloy . the two @xmath47 points for the ordered l1@xmath49 structure are seen to fall on the @xmath47 lines for the random alloy as already noted in ref . @xcite ) , and , in fact , all the points on the @xmath47 relation obtained by the ss - lsgf method may be reproduced by a series of ordinary ss - dft - cpa calculations , by using the shift of the one - electron potential defined in a way similar to ( [ eq : v_sim ] ) , i.e. , @xmath55 where @xmath4 is the net charge of the alloy components , @xmath56 the wigner - seitz radius , and @xmath57 a parameter which may be varied arbitrarily in the ss - dft - cpa calculations without specifying its physical meaning . however , it is important to note , that @xmath58 corresponds to the electroneutral case ( @xmath59 ) and @xmath60 to the limit where there is no response of the system to charge transfer effects . as we will see later , the values of the net charges @xmath61 obtained in the ss - dft - cpa calculations with @xmath60 are important scaling parameters . it is also useful to note , that for the l1@xmath49 structure @xmath62=0.8811575 @xcite , and in the screened impurity model ( [ eq : v_sim ] ) @xmath63 0.552669 and 0.568542 for the fcc and bcc crystal structures , respectively . [ qv_alli ] shows the @xmath47 relation , indicated by the black line , obtained in the ss - dft - cpa calculations including ( [ eq : v ] ) with @xmath57 varying from @xmath641.5 to 5 together with the ss - lsgf results , gray circles . it is clearly seen that the _ `` isomorphous '' _ and _ polymorphous _ @xmath47 relations coincide , and this allows one to make an isomorphous model consistent with the polymorphous results . the point is that all the net charges and corresponding coulomb shifts in the polymorphous model have significance only in terms of the average values they produce . this is so , because every supercell has its own set of net charges and madelung shifts and , in the case of an infinite system , there is an infinite number of different @xmath47 points . their average values , @xmath12 and @xmath13 , however , have a well - defined physical meaning as conditional averages of self - averaging quantities , and thus it is the average @xmath65 point which must be reproduced by the isomorphous model . hence , for a random alloy @xmath57 is given by @xmath66 note , that in a binary ab alloy , it clearly does not matter , for which alloy component , @xmath67 , @xmath68 is determined , since @xmath69 @xcite . the same is true for multicomponent alloys , but in this case , rather than being a trivial consequence of the charge neutrality condition , it follows from the physical origin of @xmath68 , which will be discussed in the next two sections . for an fcc al@xmath48li@xmath48 alloy at @xmath56=2.954 a.u . we find from the lsgf calculations @xmath700.60716 . of course , the coincidence of ( @xmath71 ) is a necessary , but not sufficient condition for the equivalence of the isomorphous and polymorphous models . the two models may be called equivalent only if the electronic structure of the random alloy and its conditional averages agree . in fig . [ dos ] we show that this is indeed the case : the local densities of states ( dos ) for the al and li atoms in al@xmath48li@xmath48 calculated by the ss - dft - cpa method coincide with the corresponding conditional average state densities obtained in the ss - lsgf calculations for the 512-atom supercell . for comparison we also show the dos obtained with @xmath720 corresponding to the `` conventional '' cpa . although the latter differs from the correct state density , it is obvious , that the neglect of the electrostatic shift ( [ eq : v ] ) has only a minor effect on the dos . the reason why the average state densities coincide is the following : in the ss - lsgf method the difference between the atoms of the same type comes only through the corresponding madelung shift . a shift in potential leads to a change in the charge transfer through a _ skewing _ of the local dos as seen in fig . [ dos2 ] . therefore , when the conditionally averaged dos is obtained , the skewing contributions from the individual atomic sites caused by @xmath2 will cancel and leave only the dos given by the average @xmath13 . of course , this is true only in the ss - lsgf method ( liz=1 ) . in fact , the local environment effects in concentrated random alloys may influence quite strongly the electronic structure of the central site of the liz . to investigate how the @xmath47 relation depends on the system we show in fig . [ qv_univ ] @xmath47 relations for five different systems including a cu impurity in pt ( @xmath73 3 a.u . ) and four random alloys : fcc cu@xmath48pt@xmath48 ( @xmath73 3 a.u . ) , fcc al@xmath48li@xmath48 ( @xmath73 2.954 a.u . ) , bcc cu@xmath48zn@xmath48 ( @xmath73 2.7 a.u . ) , and ternary fcc cu@xmath48ni@xmath74zn@xmath48 ( @xmath73 2.65 a.u . ) . in the plot all charges have been normalized by @xmath61 obtained in the no response limit , i.e. , @xmath60 or @xmath75 , and all madelung shifts have been normalized by @xmath76 . to partly simplify the plot we have used @xmath77 in the normalizations , thereby separating the @xmath47 relations into two lines rather than one . the results presented in fig . [ qv_univ ] show the existence of a universal @xmath47 relation . or in other words , the linear - response function @xmath78 which gives the change in the net charge relative to @xmath61 caused by @xmath2 , i.e. , @xmath79 , is a universal constant in metallic alloys in the asa . from the results presented in the figure we find that @xmath800.63 . this unavoidably leads to the existence of a single , unique @xmath68 as witnessed by the coincidence of all the @xmath65 points in fig . [ qv_univ ] . strictly speaking , the slopes of the @xmath47 lines are not exactly identical and , in fact , @xmath68 varies from 0.6 in cu@xmath48zn@xmath48 and cu@xmath48ni@xmath74zn@xmath48 to 0.615 in limg alloys , not included in the figure . however , for most practical purposes the choice @xmath81 provides a sufficiently accurate description of the electronic structure of random alloys in the ss - dft - cpa method in the asa for the electrostatic part . in the previous section we have , in effect , defined a procedure whereby ss - dft - cpa calculations may provide the exact solution to the electrostatic problem in random alloys . the only requirements are that the madelung shift ( [ eq : v ] ) is included and that the constant @xmath68 is obtained from ( [ eq : alpha ] ) with the average madelung potential and net charges of the alloy components determined in supercell calculations by the ss - lsgf method . in the derivation of the procedure we have used some general arguments which do not clarify the physical origin of the universal value of @xmath68 . however , it is clear , that ( [ eq : v ] ) accounts for the missing charge @xmath82 in the single - site poisson equation for the @xmath3th atomic sphere . thus , @xmath68 must be connected to the screening . the linear character of the @xmath47 relation indicates that the screening in the impurity case as well as in the case of a random alloy may be very well described by linear response theory . owing to enhanced electron scattering at opposite regions of the fermi surface linear response predicts in the case of a free - electron gas the existence of long - range friedel oscillations , which however decrease relatively fast ( @xmath83 ) with the distance . in a random alloy , on the other hand , the screening is much more efficient due to the finite life time of the bloch states for the underlying crystal lattice and the spatial distribution of the screening density decays exponentially . in this respect charge correlated model @xcite adopted by johnson and pinski @xcite in the cc- and scr - cpa method or the equivalent screened impurity model @xcite may be viewed as the first approximations for the screening . based on the fact , that a single impurity in a metallic host is a particular case of a dilute random alloy , one would expect , and the results for a single cu impurity in pt presented in the previous section unambiguously indicate this , that the screening effects in the two cases are similar . it is therefore surprising that faulkner _ _ @xcite and ujfalussy _ et al . _ @xcite claim that the screening in a random alloy is _ qualitatively _ different from that found in a single impurity system . in fact , these authors found extremely long - range correlations between the madelung potential at some particular site and the net charges at the other sites . since the madelung shift on a site is proportional to the net charge on the site , this may happen only if there are extremely long - ranged correlations between net charges or , in other words , there is no screening . however , this result has been obtained on the basis of summations of the direct or _ bare _ coulomb interactions which is , at best , an ill - defined procedure , even mathematically . to clarify the issue of screening we will perform the following computer experiment which will allow us to establish the range of the net - charge correlations or the screening in random alloys for one particular site . we set up a 512-atom supercell which represents an fcc cu@xmath48pt@xmath48 random alloy ( all sro parameters are equal to zero up to the sixth coordination shell and @xmath84 0 for at least the next 10 coordination shells ) and perform self - consistent ss - lsgf ( liz=1 ) and ec - lsgf ( liz=3 ) calculations . we then substitute one pt atom with one cu atom in some site which , in general , may be chosen arbitrarily . however , to keep the atomic distribution as close as possible to the random distribution we chose a site the local environment of which corresponds to the random alloy ( having equal number of cu and pt atoms ) for the first three coordination shells . we then repeat the self - consistent lsgf calculations for the supercell with the substituted atom and find new values for the net charges in the supercell . it is clear that , when liz=1 , the difference between the net charges in the two calculations , @xmath85 , gives the charges induced by the change of the net charge at the substitution site ( in the case of liz=3 the local environment effects also effect the charge transfer ) . this charge is simply the screening charge . in the upper panel of fig . [ q_scr ] we have plotted the normalized , induced charges @xmath86 at the first eight coordination shells around the substitution site @xmath87 for cu - pt substitution in a cu@xmath48pt@xmath48 random alloy and for cu - pt substitution in pure pt ( @xmath73 3 a.u . ) obtained in the single - site approximation for the electronic structure ( liz=1 ) as well as with local environment effects included ( liz=3 ) . one may see that , while the local environment in a pure metal hardly affects the distribution of the net charges , it does introduce a dispersion in the distribution of the net charges in the random alloy , which is quite substantial at the first coordination shell but which practically disappears beyond the 5th coordination shell where , in fact , all the induced charges almost vanish . to demonstrate that the net charge of the cu `` impurity '' indeed becomes screened we show in the lower panel of fig . [ q_scr ] the total normalized induced charge in the @xmath3th shell @xmath88 where @xmath89 is the coordination number of the @xmath26th shell . i is seen that @xmath90 vanishes beyond the 7th coordination shell in all cases , and we conclude that the screening in a random alloy in the single - site approximation is practically the same as the screening in the case of a single impurity in a pure metal . there is neither qualitative nor quantitative differences between the impurity and the alloy cases . in the upper panel of fig . [ q_scr_all ] we show the distribution of the screening charge ( not to be confused with the screening _ density _ : the screening charge is , in fact , the screening density integrated in the corresponding atomic sphere ) for a cu impurity in fcc , bcc , and bct pt plotted as a function of the distance from the impurity site in units of the wigner - seits radius , @xmath56 . it is clear that the screening charge follows a single , common curve which does not depend on the structure . in fact , by changing the @xmath91 ratio in the bct structure one may completely fill the remaining gaps in the calculated curve . in the lower panel of the figure we have collected the results for the distribution of the screening charges in seven different systems including such hosts as pt , al , cu , v , na , and k. it appears that the screening in metallic alloys depends neither on the crystal structure nor on the nature of the alloy components , at least , when described within the asa . the universal picture of the screening in alloy systems found above is partly destroyed when the electrostatics is treated more correctly , for instance by including multipole moment contributions to the one electron madelung potential and energy . however , the asa still gives a qualitatively correct picture and catches the main physics behind the phenomenon . hence , it is worth to discuss the origin of such a universality in both the screening and the response function . first of all , it was understood long time ago that the net charges in the atomic spheres of the alloy components has very little in common with the `` charger transfer '' in terms of the redistribution of the electron charge between the alloy components ( see , for instance ref . @xcite ) . even in the case of the so - called ionic solids the self - consistent charge distribution is very close to that obtained from a linear superposition of the free - atom electron - densities @xcite and this is the reason for the success of the charge - correlated model @xcite , in which the net charge is proportional to the number of nearest neighbors of the opposite type ( see also ref . @xcite ) . what we are seeing is basically a size effect : the net charges originate from the redistribution of the electron density in the _ interstitial _ region between the atomic spheres . the electron density in metals and their alloys in this region is very smooth and may be well described by a free - electron model , even for transition metal alloys . the interstitial density is much easier to perturb than the density closer to the atomic nuclei and it participates in the screening . on this basis one may , in fact , developed a model based on linear response theory which leads to a semi - analytical description of the universal screening . however , this is beyond the scope of the present paper . the calculated distribution of the screening net charge may be used to obtain the screening contribution to the one - electron potential in the single - site model for the poisson equation given by ( [ eq : v ] ) with @xmath57 equal to @xmath92 where @xmath93 is the radius of the @xmath3th shell with coordination number @xmath94 . using the results for the screening charge , @xmath90 , in the case of cu - pt substitution for one particular site ( not on average ! ) in the cu@xmath48pt@xmath48 random alloy we find from ( [ eq : alpha_scr ] ) after summation up to the 8th coordination shell that @xmath95 = 0.60572 . at the same time , the average values of the net charges and the madelung potentials , @xmath96 and @xmath97 , in conjunction with ( [ eq : alpha ] ) gives @xmath68 = 0.60530 . that is , @xmath98 . since , there is only one effective medium in the supercell lsgf calculations , it is obvious that the screening is the same for all the sites and , thus , should not depend on the alloy component , that is , the ratio @xmath99 does not depend on the alloy components . it is important to understand that @xmath68 appears in the formalism due to the _ intrasite _ interactions between the electron density inside an atomic sphere and its screening ( or missing ) charge . therefore the inclusion of ( [ eq : v ] ) in ss - dft - cpa calculations does not _ in any way _ contradict the `` conventional '' mean - field picture according to which the contribution from _ intersite _ interactions to the madelung potential is zero in the ss - dft - cpa . in fact , neglecting the term will lead to incorrect results . on the other hand , the parameter @xmath62 which allows one to perform ss - dft - cpa calculations _ identical _ to those of the ss - dft - cpa for the l1@xmath49 structure , has an origin different from that of @xmath95 determined by ( [ eq : alpha_scr ] ) . the former is a constant , which apart from the @xmath100 symmetry does not depend on anything , and in particular , not on the screening , while the latter is a system dependent parameter which is entirely determined by the type of intersite interactions or the screening in a given system . this means that there is no connection between the madelung constants for ordered structures and @xmath95 . the reason why it is possible to obtain @xmath95 from the supercell calculations for the completely ordered structures , as we have just done , is the fact that at the large distances , where the atomic - distribution correlation - functions are not zero anymore , the `` real '' _ intersite _ interactions do not contribute to the electrostatics due to the short - ranged screening . we will return to this point in the discussion of the coulomb energy of a random alloy , but here we would like to comment on the use of the single - site approximation in the green s function calculations . our lsgf calculations of impurities in different metals indicate that the problems observed in the single - site green s function impurity calculations by stefanou @xcite and drittler _ et al . _ @xcite do not originate from the single - site approximation for the dyson equation , but from poisson s equation , which these authors also solve in the single - site approximation . the use of the screening electrostatic shift for the one - electron potential allows one to solve the impurity problem in the single - site green s function formalism in the asa or in the mt approximation almost exactly . this is so because the impurity case corresponds to the dilute limit of a random alloy where the concentration of one of the alloy components approaches zero . in this case , the contribution to the electronic structure due to the local environment effects becomes negligible , and the electronic structure of the impurity obtained by the single - site or the cluster dyson equations become almost identical . such an effect may be seen , for instance , in fig . [ q_scr ] : the dispersion of the screening net charge found in the concentrated alloy case , i.e. , cu@xmath48pt@xmath48 ( liz=3 ) , vanishes in the case of a cu impurity in pt . the effect is , in fact , the origin of the increasing accuracy of the cpa with decreasing concentration of one of the alloy components . the formalism describing the electrostatics of random alloys in the single - site mean - field approximation , where all the a and all the b atoms are represented only by the appropriate conditional averages , is indeed a trivial one , although it has remained quite a confusing issue for more than 30 years , starting from the definition of the electrostatic energy of a random alloy given by harrison in 1966 in connection with pseudopotential theory @xcite . according to harrison the madelung energy , which may be associated with the electrostatic interactions of the net charges of the ions of a random a@xmath101b@xmath102 alloy , is given by @xcite @xmath103 where @xmath104 are the ion charges of the alloy components , @xmath105 is the madelung constant of the underlying lattice , @xmath56 the radius of the wigner - seitz sphere , @xmath106 the average charge equal to @xmath107 , and @xmath108 some constant . equation ( [ eq : e_harr ] ) was , in fact , not derived specifically for coulomb interactions . rather , it was suggested on the basis of a more general consideration of the band - structure contribution to the total energy of a random alloy within second order perturbation theory . later , it was shown by krasko @xcite that in a system with randomly distributed a and b ions , having charges @xmath109 and @xmath110 , respectively , embedded in a medium of compensating charges , the electrostatic energy included only the second term in ( [ eq : e_harr ] ) while the first term vanished . it is obvious that ( [ eq : e_harr ] ) is valid also in the cpa - dft if one substitutes the ion charges @xmath104 by the net charges @xmath4 of the atomic spheres . in the asa the average charge @xmath111 is zero , and in this particular approximation the contribution to the madelung energy of a random alloy given by the second term in ( [ eq : e_harr ] ) will also vanish . note , however , that the term vanishes neither in inhomogeneous systems @xcite nor in the mt - approximation @xcite . this result was criticized by magri _ @xcite who a decade ago deduced from calculations for ordered compounds that the net charge of the atomic sphere of an alloy component is proportional to the number of nearest neighbors of the opposite type , and on this basis developed the charge correlated model in which the madelung energy of a binary random alloy is given by @xmath112 where @xmath113 is a constant equal to 0.5947049 and 0.60817846 for the fcc and bcc structures , respectively . exactly the same result has been obtained later by korzhavyi _ _ @xcite and by johnson and pinski @xcite in their `` screened '' models for the single - site cpa - dft . the difference between these models lies only in the way the parameter @xmath113 is determined ( from 0.4397212 for bcc in ref . @xcite to 0.54282038 for fcc in ref . @xcite ) . a discussion of the issues involved may be found in refs . @xcite . it is certainly surprising , that ( [ eq : e_cc ] ) is exactly the first term in ( [ eq : e_harr ] ) which was shown to vanish in the case of two charges , here @xmath114 and @xmath115 , randomly distributed on an underlying lattice in a compensating homogeneous effective medium . this means that either the models are inconsistent with general theory or that there should be some reason , not accounted for in the derivation by krasko , for the presence of the first term in ( [ eq : e_harr ] ) . assuming the existence of only on - site and pair - wise interactions the hamiltonian of a binary a@xmath101b@xmath102 alloy may be written @xmath116 + \frac{1}{2}\sum_{r \neq r ' } [ v^{aa}_{rr'}c_rc_{r ' } + v^{ab}_{rr'}(1-c_r)c_{r ' } \\ \nonumber & + & v^{ab}_{rr'}c_r(1-c_{r ' } ) + v^{bb}_{rr'}(1-c_r)(1-c_{r ' } ) ] , \end{aligned}\ ] ] where @xmath117 are on - site or intrasite interactions , which we will assume depend only on the type of atom on site @xmath118 , @xmath119 are pair potentials acting between x and y atoms at site @xmath118 and @xmath120 , respectively , and @xmath121 is the site - occupation operator taking on the value 1 if there is an a atom on site @xmath118 and 0 otherwise . using @xmath122 , defined by @xmath123 , we may rewrite the hamiltonian in the equivalent form @xmath124 , \ ] ] where the first term includes the intrasite interaction ( @xmath125 ) @xmath126\ ] ] as well as the intersite interactions ( @xmath127 ) @xmath128 upon fourier transformation of the first term we find @xmath129,\end{aligned}\ ] ] where the second term is the average contribution to the energy due to pair interactions which in the case of direct coulomb ion - ion interactions is @xmath130 combined with the corresponding contribution from the interaction between the ions and the homogeneous compensating charge . this is exactly the second term in ( [ eq : e_harr ] ) . the first term in ( [ eq : h_3 ] ) is usually associated with the configurational contribution to the energy of the system , but this is correct only if the contribution from intrasite interactions is zero . it is easily evaluated in a completely random alloy , where all the occupation numbers are uncorrelated and therefore @xmath131 ( which provides the normalization of the ordering energy per atom ) . one finds @xmath132 which according to ( [ eq : h_1 ] ) is equal to @xmath133 . it now remains to define the on - site interaction term @xmath134 , which results from the interaction of the net charge @xmath4 in the alloy with the corresponding screening charge , in such a way that @xmath134 and the corresponding on - site coulomb potential @xmath2 given by ( [ eq : v ] ) are consistent within dft , i.e. , @xmath135 : @xmath136 using this definition the first term in the hamiltonian ( [ eq : h_3 ] ) may be written @xmath137 = -\frac{e^2}{2 } c(1-c ) \frac{\alpha_{scr}}{s } ( q_a - q_b)^2 \\ \nonumber & \equiv & c(1-c)v_{scr}(r=0),\end{aligned}\ ] ] which is exactly the first term in ( [ eq : e_harr ] ) and which , according to the above derivation , arises from on - site or intrasite interactions such as the screening interactions in metallic alloys . it is now clear why the first term in ( [ eq : e_harr ] ) is absent in the work by krasko : his derivation is based on inter - site coulomb interactions only . thus , krasko s result is valid in the absence of the screening intrasite interactions , which is the case , for instance , in the pseudopotential formalism within second order perturbation theory @xcite . however , at the same time harrison s result is in fact correct under the more general assumptions needed in the ss - dft - cpa calculations . the previous section may seem trivial : first we define our inter- and intrasite interactions in real space , then we fourier transform , and from the fourier transform we return to the initially defined intrasite term . however , in some formalisms , such as pseudopotential theory or the @xmath0 formalism @xcite , @xmath138 is already defined and this may lead to problems with the correct definition of the configurational part of the total energy . the point is that the intrasite interactions _ do not _ contribute to the _ configurational _ part of the total energy which in real space may be written as ( here , we do not consider the contribution from multi - site interactions ) @xcite @xmath139 therefore , if the configurational hamiltonian is written in terms of @xmath138 , e.g. , in the concentration wave formalism , it must be corrected by the subtraction of the corresponding intrasite interaction , i.e. , @xmath140 c_q c^*_q , \end{aligned}\ ] ] where we have used the sum rule for the concentration wave density @xmath141 : @xmath142 . the subtraction of the intrasite term in ( [ eq : h_conf_q ] ) is crucial for obtaining the correct ordering energy in pseudopotential theory and the @xmath0 formalism @xcite as well as for making the whole theory consistent . let us , for instance , consider the madelung energy of a binary completely ordered alloy with two non - equivalent sublattices . it is easy to show that its madelung energy can be presented exactly in the form ( [ eq : e_harr ] ) . for instance , the madelung energy of the l1@xmath49 ordered phase is @xmath143 where the last term is zero in the asa since @xmath144=0 as in the random alloy case , but now @xmath62 is a constant which appears due to the intersite coulomb interactions . in the appendix of ref . @xcite it is shown that , in fact , @xmath145 is the fourier transform of the effective direct electrostatic interaction at the corresponding superstructure vector @xmath146 , i.e. , @xmath147 . on the other hand , @xmath148 may also be found as the sum of the electrostatic energy of the completely random alloy , @xmath149 , given by ( [ eq : e_scr_rand ] ) and the ordering energy , @xmath150 : @xmath151 . since the madelung energy of the ordered l1@xmath49 alloy , @xmath148 is uniquely defined in terms of the corresponding madelung constant , which has nothing to do with the screening in the alloy , it is obvious that such a screening term must be present in the ordering energy , @xmath150 , to compensate the screening contribution . indeed , as shown in the appendix of ref . @xcite the ordering energy in the l1@xmath49 structure can be written in the form as @xmath152 , \ ] ] from which it is easy to see that in the in the completely ordered state , where the long range order parameter @xmath153=1 , the last term in ( [ u_ord ] ) is exactly the madelung energy of random alloy at the stoichiometric composition ( @xmath154 = 1/4 ) , and thus @xmath155 . this illustrates an important point : the ordering energy represented in reciprocal space in the concentration wave formalism must be corrected by the exclusion of the intrasite term , otherwise the theory will not be consistent . equation ( [ eq : h_conf_q ] ) gives the correct definition of the ordering energy considered more thoroughly in the appendix of ref . the intrasite interaction must also be subtracted when one considers the energy of sro effects , and thus the correct krivoglaz - clapp - moss expression must have @xmath156 instead of @xmath138 , which is exactly the case in krivoglaz s derivation @xcite . note , however , that this problem does not exist if the krivoglaz - clapp - moss expression is used together with the so - called onsager correction @xcite under the condition that it is properly defined . the reason , why it was possible to calculate @xmath95 on the basis of the _ ordered _ structures , is the fact that , in an ordered binary alloy with only _ two _ non - equivalent sublattices , one has an exact cancellation of the screening contribution to the madelung energy and potential . this does not happen , however , in the general case of a supercell with @xmath157 non - equivalent sublattices . here , the madelung energy may written as sum of the contribution from the intrasite screening interactions @xmath158 \\ \nonumber [ & \neq & \frac{e^2}{2 } \frac{\alpha_{scr}}{s } ( c < q_a>^2 + ( 1-c ) < q_b>^2 ) ] , \end{aligned}\ ] ] where @xmath159 and @xmath160 are the number of a and b atoms , respectively , and the ordering energy due to the intersite interactions @xmath161 here , @xmath162 is a normalizing coefficient , @xmath163 a constant due to the _ bare _ electrostatic interactions between the net charges for the superstructure vector @xmath164 which may be calculated from the madelung constants @xmath165 of the corresponding supercell similar to the @xmath62 considered above , and @xmath166 the difference between the charges in the crests and in the troughs of the concentration wave in the supercell . in the case of a binary alloy with two non - equivalent sublattices , there is only one @xmath164 and @xmath167 . if @xmath163 depends only on the structure and describe the _ bare _ electrostatic interaction between net charges , then @xmath166 `` dresses '' these interactions according to the real charge distribution in the alloy ( an equivalent description in real space in the charge correlated model is given by wolverton and zunger @xcite , who also show that the madelung energy of the random alloy has intrasite character ) . if the net charges in the supercell are screened ( or uncorrelated ) at distances less than half the period of the concentration wave with wave - vector @xmath164 then @xmath166=0 and the corresponding contribution to the ordering energy vanishes . if the supercell includes only long - range concentration waves , the corresponding ordering contribution to the madelung potential and energy becomes zero . let us finally mention the fact that the madelung energy of a random alloy obtained in supercell calculations ( [ eq : e_intra_n ] ) is not equal to the madelung energy in the single - site calculations , and thus it can not be used to obtain @xmath95 . the reason is simply that the madelung energy is not a self - averaging quantity . however , the madelung potential is , and it is clear that @xmath168 which allows one to use ( [ eq : alpha ] ) to obtained @xmath95 in the supercell calculations and shows why @xmath68 is exactly equal to @xmath95 . although the screened coulomb interactions have an intrasite character , they may contribute to the effective _ pair intersite _ interactions of the kind obtained in the generalized perturbation method ( gpm ) @xcite because the screening charge is located on several of the coordination shells around each atom . this was , in fact , already recognized by ducastelle @xcite who derived the contribution to the gpm potentials from the screened coulomb interactions in the framework of the hartree - fock tight - binding cpa theory . the existence of an additional electrostatic term due to the screening is also consistent with andersen s the force theorem @xcite , which states that the change in the total energy of a system due to some perturbation to first order is given by the change in the sum of the one - electron energies obtained from frozen one - electron potentials plus the change of the electrostatic energy due to the perturbation . in fact , this latter contribution from the screened coulomb interaction has been completely neglected in a number of first - principles calculations of gpm interactions @xcite . here , we will therefore show how the screening contribution to the gpm potentials may be defined and obtained on the basis of the calculated spatial distribution of the screening charge . gpm - like pair interactions , usually defined by ( [ v ] ) for a specific lattice vector @xmath169 , may be determined as the site - projected part of the change in the total energy when two atoms of different types in a completely random alloy are exchanged between sites infinitely far apart in such a way that their neighbors at the relative position @xmath169 are of the opposite type after the exchange . this is schematically illustrated in fig . [ fig : exch ] . that part of the total energy which should be accounted for is half the site - decomposed total energy written in terms of the intersite interactions or interatomic potentials , i.e. , @xmath170 .\ ] ] here , @xmath171 is the total energy due to pairwise interactions of the unperturbed system projected onto site @xmath172 and @xmath173 is the same quantity after the exchange . a similar expression is also valid in the case of multisite interactions , but this will involve a more complex exchange of atoms and will not be considered here because the screened coulomb interactions do not contribute to the effective multisite interactions in the asa . within multiple scattering theory as well as in the tight - binding approximation a green s function formulation allows both site- and `` path''-decomposition of the electron density and thereby makes it possible to write down an analytical expression for the one - electron contribution to the @xmath174-site interactions , @xmath175 , in the cpa . concerning the screened coulomb interactions one must , however , proceed differently . there are several ways to do so , but here we will present a straightforward approach . in the sense of the cpa and single - site mean - field theory we will use an effective medium approach , assuming that at all sites , i.e. , within the atomic spheres assigned to each site , there is an electroneutral effective medium except at the two sites @xmath176 and @xmath177 under consideration . in those two sites we must use the actual values of the net charges of the alloy components , which in the effective medium approach , are the average net charges , @xmath114 and @xmath115 , of the alloy components . in the first - principles methods , however , these net charges depend on the specific choice of the size of the atomic spheres and thus they must , in principle , go together with the corresponding screening cloud . since we calculate the change in the electrostatic energy of the two systems shown in fig . [ fig : exch ] projected onto site @xmath176 due to the exchange of a and b atoms in positions @xmath169 , we must include only the interaction of the net charge at site @xmath176 with the net charge at site @xmath169 and _ its screening _ charge . that is , the interaction of the net charge with its own screening charge must be excluded as it is included in the definition of the screened on - site interactions , see ( [ eq : alpha_scr ] ) . thus , the first term in ( [ v2 ] ) for the system before the exchange of atoms has been made , @xmath178 , is @xmath179 here , @xmath180 is either the net charge of the @xmath3th component , @xmath4 , if @xmath181 or the corresponding screening charge if @xmath182 . a similar expression may be written for @xmath183 , after the exchange of the a and b atoms in the @xmath169-sites , i.e. , @xmath184 the resulting expression for the screened coulomb interactions which should be added to the usual one - electron term is therefore @xmath185 \\ \nonumber & = & \frac{e^2}{2}(q_a - q_b)^2 \sum_{{\bf r } ' \neq 0 } \frac{q(|{\bf r}'-{\bf r}|)}{r ' } , \\ \nonumber \end{aligned}\ ] ] where @xmath186 is the normalized screening charge defined in ( [ q_i ] ) , and where we have used the condition that the screening does not depend on the type of the atom . finally , performing the summation in ( [ v_scr_gpm_1 ] ) one may define the screened coulomb interactions as @xmath187 it is easy to see from ( [ v_scr_gpm_1 ] ) and ( [ eq : alpha_scr ] ) that @xmath188 and therefore , @xmath189 is exactly the on - site screened interactions that defines the madelung energy of the binary alloy which has exactly the same form ( [ eq : e_scr_rand ] ) . this on - site interaction must be included in the definition of the @xmath0 interactions @xcite , as has been demonstrated in the previous section ( see also ref . @xcite ) . when @xmath190 , @xmath191 defines the intersite screened coulomb interaction contribution to the gpm - like effective interactions . since the screening in the asa , is practically universal these interactions have the universal form presented in fig . [ fig : vscr ] . the fact that the madelung energy of a random alloy described either by the effective medium model defined by the ss - dft - cpa method or by the supercell model in conjunction with the ss - lsgf method differ from each other , has neither consequences for the final result for the total energy of the random alloy nor even for the partial and local contributions to the total energy . this follows simply from the fact that the density of states and its average local contributions are the same in the two methods , as shown above . in table [ e_iso_pol ] we compare the total energy and its components in a cu@xmath48pt@xmath48 random alloy calculated by the ss - dft - cpa method with @xmath1920.60572 and by the ss - lsgf method on the basis of a 512-atom supercell , in which the atomic positions of cu and pt have been chosen such that the sro parameters are equal zero at the first 7 coordinations shells ( lsgf-1 ) . the agreement between the two calculations is seen to be excellent if one combines the electron - nucleus , the electron - electron , and the madelung contributions to form a total coulomb energy , @xmath193 . the accuracy of the ss - dft - cpa method with the appropriate screening contribution to the madelung potential and energy may be appreciated if one compares the results of a 512-atom supercell calculation performed by the ss - lsgf method ( lsgf-2 ) where the distribution of the cu and pt atoms have not been optimized after the application of the random number generator leading to quite small , but not zero , sro parameters . the values of the sro parameters for the first 7 coordination shells are @xmath640.005208 ( 1 ) , 0.026041 ( 2 ) , 0.007161(3 ) , -0.014323(4 ) , -0.021484(5 ) , 0.0390625(6 ) , @xmath640.0136718 ( 7 ) , respectively , which are approximately the same , as in the lsms calculations in ref . the agreement between ss - dft - cpa results and ss - lsgf calculations with a properly chosen supercell ( lsgf-1 ) is obviously better than between two ss - lsgf calculations . the screened coulomb interactions which are due to the interaction between the net charge of an alloy component and its screening charge must be included in a consistent single - site mean - field theory of the electrostatics in random alloys . in this paper we have shown how this may be done and we have calculated the spatial distribution of the screening charge which in the asa is found to be practically universal for homogeneous systems . a formalism that describes the contribution from for the screened coulomb interaction to madelung potential and energy as well as to the effective interactions of the gpm - type is presented . valuable discussions with dr . p. a. korzhavyi , i. a. abrikosov , a. yu . lozovoi , prof . a. gonis and prof . s. faulkner are greatly acknowledged . center for atomic - scale materials physics is sponsored by the danish national research foundation . a. gonis and j. w. garland , phys . rev . b * 16 * , 2424 ( 1977 ) ; c. w. myles and j. d. dow , phys . lett . * 42 * , 254 ( 1979 ) ; a. gonis , w. h. butler , and g. m. stocks , phys . lett . * 50 * , 1482 ( 1982 ) ; a. gonis , g. m. stocks , w. h. butler , and h. winter , phys . rev . b * 29 * , 555 ( 1984 ) . . the total energy ( in ry ) of cu@xmath48pt@xmath48 random alloy and corresponding contributions obtained in three different calculations : by the single - site cpa - dft method ( ss - cpa - dft ) , in the 512-atom supercell lsgf calculations with optimized atomic distribution , providing zero sro parameters up to the 7th coordination shell ( lsgf-1 ) , and with atomic configuration immediately after random number generator ( lsgf-2 ) . ( @xmath194 ) [ cols=">,>,>,>,>",options="header " , ]

we compare eighteen binary systems with precisely determined radii and masses from 23 to 1.1 , and stellar evolution models produced with our newly revised code tycho . `` overshooting '' and rotational mixing were suppressed in order to establish a baseline for isolating these and other hydrodynamic effects . acceptable coeval fits are found for sixteen pairs without optimizing for heavy element or helium abundance . the precision of these tests is limited by the accuracies of the observed effective temperatures . high dispersion spectra and detailed atmospheric modeling should give more accurate effective temperatures and heavy element abundances . pv cas , a peculiar early a system , ek cep b , a known post - t tauri star , and rs cha , a member of a young ob association , are matched by pre - main sequence models . predicted mass loss agrees with upper limits from iue for cw cep a and b. relatively poor fits are obtained for binaries having at least one component in the mass range , whose evolution is sensitive to mixing . these discrepancies are robust and consistent with additional mixing in real stars . the predicted apsidal motion implies that massive star models are systematically less centrally condensed than the real stars . if these effects are due to overshooting , then the overshooting parameter increases with stellar mass . the apsidal motion constants are controlled by radiative opacity under conditions close to those directly measured in laser experiments , making this test more stringent than possible before .

astro-ph0103390

prior to any rigorous investigation of thermonuclear yields , mixing , rotation , mass loss , and other complex phenomena in the evolution of stars , it is necessary to insure that the methodology used can reproduce observations at the current state of the art . detached , double - lined eclipsing binaries provide the most accurate source of information on stellar masses and radii , and as such provide a crucial test for models of stellar evolution @xcite . the predictive power of such simulations depends upon the extent to which the validity of their oversimplifications can be tested . phenomenology is particularly pernicious in that good tests , which could give rise to falsification of the models and thus to progress , can be nullified by parameter adjustment ( `` calibration '' ) . similarly , the inclusion of new processes in the simulations is often contentious in that there may be several candidates put forward as the cause of the puzzling data . the treatment of the boundaries of convection zones ( `` overshooting '' ) is ripe for re - examination in terms of the underlying hydrodynamics @xcite . to aid this development , we wish to understand just how well a standard convective treatment can do ( i.e. , we turn overshooting off ) . a more physically sound procedure will be presented subsequently . parameterized overshooting has been widely discussed ( see @xcite for an early discussion , and @xcite and @xcite for a recent ones with extensive references ) . rotation also may cause mixing @xcite . within the context of a plasma , rotation and convection may generate magnetic fields , which by their buoyancy and angular momentum transport may provide additional causes for mixing . this paper represents a first step toward testing our extensively revised stellar evolution and hydrodynamics code , tycho . our goals are ( 1 ) the understanding the predictive capability of stellar evolution theory by the critical re - evaluation of its assumptions , and of its underlying basis in observations and in laboratory data , ( 2 ) the examination of the tricky problem of mixing in stars to help design experiments @xcite and numerical tests @xcite , and ( 3 ) the development of an open source , publicly available stellar evolution code with modern capabilities for community use . the most comprehensive list of binary systems with accurately measured masses and radii is given in the review by @xcite . a subset of the original binaries was chosen for this exploratory effort . the systems with the smallest uncertainties were picked such that the range of masses from 23 to 1.1 @xmath0 was well sampled . the upper mass limit is the largest mass present in the data , while the lower limit is safely above the point at which the equation of state used in the modeling becomes inaccurate . this is primarily due to approximate treatment of coulomb contributions to the pressure ; we use only the weak screening limit for the plasma . the coulomb interaction leads to a negative pressure correction of @xmath3 8% in the outer part of the convection region and @xmath3 1% in the core for a star of 1 @xmath0 @xcite , and is less important for more massive stars . we obtain a comparable correction to the central pressure for a solar model ( -1.7% ) . aside from the exclusion of stars of still lower mass , no bias was applied in the selection process . for stars of @xmath4 or less , the issue of possible overshooting in the convective core is moot because their cores are radiative . also , because they rotate rapidly only for a brief part of their lives , rotational mixing is expected to be less than for our selected stars . the binaries used in the study along with their fundamental parameters are presented in table [ tbl-1 ] . @xcite have revised the @xcite temperature estimates , and revised estimates for the masses of em car @xcite and cw cep @xcite have been presented . @xcite have revised the parameters for dm vir . all these changes are incorporated into table [ tbl-1 ] . if the purpose of a stellar evolution code is to make testable predictions of the behavior of stars , then the adjustment of the mixing length by fitting stellar data is repugnant . alternatives are to constrain it by experiment or by simulation . at present we know of no definitive experimental results which determine the mixing length appropriate to stars , although a variety of experiments do test other aspects of stellar hydrodynamics and the codes used to simulate them @xcite . however , it is becoming possible to simulate turbulent , compressible convection with sufficient realism to constrain the range allowed for the mixing length @xcite . @xcite find a mixing length @xmath5 in units of pressure scale height ; this is based upon simulations having mesh resolutions as high as @xmath6 and corresponding rayleigh numbers as high as @xmath7 . @xcite based their work on resolutions up to @xmath8 , but with a more realistic treatment of radiative transfer and ionization . they were able to synthesize the line profile of feii @xmath95414 which compared well with that observed . their results agreed with standard 1-d models , although they suggest that this might be `` the right result for the wrong reason . '' standard models use @xmath10 to @xmath11 , which is significantly smaller than the value of @xcite . for a red giant , @xcite inferred @xmath12 , based upon 2-d simulations but with fairly realistic microphysics . @xcite have proposed a serious model to replace mixing length theory ; this has had the salutary effect of shifting the debate to the physics of convection and away from the best choice of mixing length . finally , @xcite have shown from 2-d simulations that the underlying physical picture for stellar convection is incomplete , even in the deep interior where the complication of radiative transfer is minor . we simply choose @xmath12 , and look forward to the convergence of these efforts to provide a convection algorithm which is independent of stellar evolutionary calibrations . the evolutionary sequences were produced with the tycho stellar evolution code . the code was originally developed for one dimensional ( 1d ) hydrodynamics of the late stages of stellar evolution and core collapse @xcite . it is being completely rewritten as a general purpose , open source code for stellar evolution and hydrodynamics . the present version is written in structured fortran77 and is targeted for linux machines . it has been successfully ported to sunos and sgi irix operating systems . it has extensive online graphics using pgplot , an open source package written by t. j. pearson ( ) . a library of analysis programs is being built ( modules for apsidal motion , pulsational instability , reaction network links , and history of mass loss are now available ) . the code is being put under source code control to allow versioning ( this will allow particular versions of the code for example the one used in this paper , to be resurrected accurately at later times ) , and to improve the reproducibility of results . knowledge of the radiative opacity of plasma at stellar conditions has changed qualitatively in the past decade . historically , solar and stellar atmospheres provided much of the empirical data on hot plasmas . for example , @xcite tabulates a range of effective temperatures from 2,000 to 200,000 k. these temperatures and the corresponding ( low ) densities characterized what was directly observable . higher temperatures could be found at lower densities ( in non - lte ) or indirectly inferred by use of theoretical models . terrestrial tests involved explosions which were difficult to quantify with adequate precision to determine opacity . measurements of opacity in a well characterized , hot , dense , laser - produced plasma have become possible @xcite . the first experiments to simultaneously quantify temperature and density with good precision @xcite involved temperatures @xmath13 and densities @xmath14 , which are directly relevant to stellar evolution and to apsidal motion . the range of conditions which are experimentally accessible is expanding with the development of new instruments and techniques . not only can direct measurements be made , but complex and sophisticated theories of the physical state of the plasma can be tested , giving more reliable extrapolations into conditions not yet experimentally accessible @xcite . the conditions just quoted are encountered in stars of about @xmath15 , and are important for stellar evolution ( see ch . 7 , @xcite ) . the opacities used here are from @xcite and @xcite , for a solar abundance pattern @xcite . the @xcite opacities were computed with 21 elemental species ; @xcite have shown that the remaining elements are so rare as to have only a marginal effect on the rosseland mean opacities ( for solar relative abundances of the heavier elements ) . while the opal opacities were constructed for astrophysical use , the underlying experiments and theoretical models are determined by the inertial confinement fusion ( icf ) community , reducing the danger of unconscious bias from astronomical puzzles leaking back into the construction of opacities . extension of the opacity library to lower temperatures and lower entropies is planned . tycho is designed to use an adaptable set of reaction networks ; for these calculations , two networks were used . at higher temperatures ( @xmath16 ) , an 80 element reaction network was solved . the reaction rates were from f. k. thielemann ( private communication ) ; see also @xcite . for lower temperatures this was replaced by a 15 element network which was designed for deuterium , lithium , beryllium , and boron depletion . the reaction rates were from @xcite . the match at the temperature boundary was sufficiently good as to require no smoothing . more recent compilations of nuclear rates are available @xcite , but were not used here to simplify the comparison with previous work . the outer boundary condition was determined by use of the eddington approximation to a grey , plane parallel atmosphere , integrated in hydrostatic equilibrium inward to a fitting point for the interior . for the most extended model considered here , the ratio of mean free path to radius was @xmath17 , so that spherical effects are negligible . the most vigorous mass loss considered was so mild that the ram pressure @xmath18 at the photosphere was @xmath19 of the total thermal pressure , which is consistent with the hydrostatic assumption . such integrations were used to define the pressure and temperature at the fitting points , @xmath20 and @xmath21 , for the stellar luminosity @xmath22 and radius @xmath23 . their derivatives with respect to stellar luminosity and radius were approximated by finite differences constructed between three such integrations ( at @xmath24 , @xmath25 , and @xmath26 ) . typically , of order 200 to 400 steps were used in the envelope integration . we used schwarzchild convection as formulated by @xcite , and our treatment of convective overshooting was turned off . mass loss was included and based on the theory of @xcite for @xmath27 and the empirical approach of @xcite for lower effective temperatures . r. kudritzki kindly provided appropriate subroutines for the hotter regime . even for em car , the most massive system in the list , the effects of mass loss were modest ( @xmath28 ) . the equation of state was that discussed in @xcite , augmented by the solution of the ionization equilibrium equations for h , he , and a set of heavier elements scaled from the solar abundance pattern . both the equation of state and the thermonuclear reaction rates are affected by coulomb properties of the plasma . only weak screening was necessary here . extension to include both weak and strong screening consistently in the equation of state and thermonuclear reaction rates is planned ; previous versions of the code included strong screening as well . models were run for each mass , starting with a fully convective initial model on the hayashi track and ending well beyond hydrogen depletion in the core . a more realistic approach would have been to form the stars by accretion ( a. g. w. cameron , private communication ; @xcite ) . we justify our choice by its simplicity , and by noting that only the last stages of the pre - main sequence are relevant here , for which the two cases give similar results . zoning in the interior typically ranged from 200 to 500 zones . all runs had solar heavy element abundance @xcite and used a ratio of mixing length to scale height of @xmath29 for convection . this choice gave a reasonably good solar model when compared to @xcite and @xcite ; inclusion of element settling by diffusion and adjustment of the helium abundance would give improved consistency for the present - day sun , but diffusion would have less time to operate in the more massive stars considered here . rotational mixing was turned off . the stars are all at or near the main sequence , so that the possible list of citations is enormous ; the efforts of the padua group @xcite , the geneva group @xcite , and the franec group @xcite have comparable input physics and form a useful context . we focus discussion on @xcite and @xcite , who consider many of the same binaries , and @xcite who examined the apsidal motion test . the largest differences in microphysics between @xcite and this paper are our use of @xcite rather than @xcite opacities , a more realistic nuclear network , and a different approach to the equation of state , but none of this seems to be particularly significant for the issues here . we do include mass loss , but these effects are not large . @xcite do not calculate the pre - main sequence evolution ( which is relevant to several of our binaries ) . they define an overshooting parameter which is fixed by previous work on @xmath30 aurigae binaries @xcite . they construct a grid of models in mass ( 0.5 to @xmath31 ) and heavy element abundance ( @xmath32 , 0.02 and 0.03 ) , assuming @xmath33 and @xmath34 for the abundances of hydrogen and helium . they minimize a @xmath35 error estimator in four parameters : the masses @xmath36 and @xmath37 , the age @xmath38 of the binary , and @xmath39 the heavy element abundance . as a test of consistency for later evolution , we have reproduced the @xmath40 and @xmath41 sequences of @xcite , which did not use overshooting . the notoriously sensitive blue loops were reproduced to graphical accuracy ( their figure 4 ) for the same input physics . the codes seem highly consistent . @xcite used the models of @xcite and @xcite , which use @xcite opacities and a 14 isotope network and include overshooting and mass loss . they too interpolated in a grid , minimizing a @xmath35 error estimator . this procedure was more complex than that used by @xcite , and need not be described here . both heavy element abundance and helium abundance were freely varied . our strategy differs from both @xcite and @xcite , which may provide a useful contrast . here we are interested in isolating the possible inadequacy of the standard formulation of stellar evolution , so we avoid optimization of parameters as much as possible . by using ( 1 ) solar abundances and ( 2 ) the measured masses , we reduce the degrees of freedom , and hopefully make the possible flaws in our stellar evolution prescriptions easier to see . by the same token , our models should fit the data less well because we have not optimized abundances or masses . mathematically , optimization will almost always improve the fit , but not necessarily for the correct reasons . however , the actual abundances may be different from our assumptions , and the masses do have error bars . @xcite used @xcite opacities , solar abundances , a mixing length ratio @xmath42 , and overshooting of @xmath43 pressure scale heights ( that is , essentially the same physics as the models used by @xcite ) , and computed structure constants for apsidal motion for seven of the binaries we consider ( em car , cw cep , qx car , u oph , @xmath30 phe , iq per , and pv cas ) . detailed comparisons will appear in the discussion below . the first step in comparing the binary data with the computations is the choice of the best models . this was done by examining a @xmath35 quantity for each binary pair , defined by @xmath44 where @xmath45 and @xmath46 denote the primary and the secondary star , respectively . here @xmath47 and @xmath48 are the observationally determined luminosity and radius of the primary , with @xmath49 and @xmath50 being the observational errors in @xmath51 and in @xmath52 . we convert the observational data for the radii to logarithmic form for consistency . correspondingly , @xmath53 and @xmath54 are the luminosity and radius of the model . this @xmath35 was evaluated by computing two evolutionary sequences , one for a star of mass @xmath55 and one for @xmath56 , and storing selected results from each time step . then these files were marched through , calculating @xmath35 at consistent times ( @xmath57 to a fraction of a time step , which was a relative error of a few percent at worst ) . the smallest @xmath35 value determined which pair of models was optimum for that binary . note that if the trajectories of both @xmath45 and @xmath46 graze the error boxes at the same time , @xmath58 . these error parameters along with the corresponding uncertainties from the observations are presented in table [ tbl-2 ] . figure 1 displays the resulting @xmath35 for each binary pair , in order of descending mean mass . the binaries fall into three separate groups : ten have excellent fits ( @xmath59 ; em car , v478 cyg , cw cep , cv vel , u oph , pv cas , rs cha , pv pup , dm vir and ux men ) , six are marginal ( @xmath60 ; qx car,@xmath61 phe , iq per , my cyg , ek cep , and v1143 cyg ) , and two are poor fits ( @xmath62 , denoted _ offscale _ in figure 1 ; vv pyx and ai hya ) . the boundaries between these groups are indicated by vertical lines . the weakness of a @xmath35 measure is that it is most meaningful if the errors have a gaussian distribution around the mean ( @xcite , chapters 14 and 15 ) , which does not seem to be the case here . in particular , systematic shifts in the empirical data , due to new analyses , can give significant shifts in the error estimation . @xcite have re - estimated the effective temperatures of 13 of the 18 binaries we have examined . five ( qx car , u oph , pv cas , ai hya , and rs cha ) were changed by more than twice the error estimates of either @xcite or @xcite . further , @xcite and @xcite have analyzed additional data ( from iue ) and find masses of cw cep and em car which lie beyond twice the error estimates . this is to be expected if the errors are dominated by systematic effects , and warns us to distrust all but our most robust inferences . because the fractional errors in mass and in radius are much more restrictive , it strongly supports the need for renewed efforts to pin down the effective temperatures of these stars . the choice of @xmath22 and @xmath23 rather than @xmath22 and @xmath63 in our definition of @xmath35 is significant : the smaller errors for @xmath23 make the @xmath35 more discriminating . @xcite use @xmath23 and @xmath63 which has the slight advantage here of involving less propagation of observational errors , but because @xmath23 is much more precise than @xmath63 , the effect is small for the present data . we have chosen to update the original data of @xcite , incorporating the changes made by the @xcite effective temperatures and the @xcite and @xcite masses . we have used the new data for dm vir @xcite . our general conclusions are unaffected by which of these sets of data we use . the comparison of observed and computed stars may be presented as an goodness of fit vector , which has the advantage of being directly representable in the hr diagram for the stars . the observed points with error bars are plotted along with an arrow indicating the distance and direction to the best model point . the way in which the models differ from the observations can then be taken in at a glance . figure 2 shows the goodness of fit vectors , from the observed points ( shown with error bars ) to the best model star ( chosen as described above ) . the largest discrepancy is the secondary of vv pyx . of the eight binaries ( qx car , @xmath30 phe , iq per , vv pyx , ai hya , ek cep , my cyg , v1143 ) which have mediocre or poor fits , seven ( qx car is the exception ) have at least one component lying in the range @xmath64 . @xcite noticed similar behavior . figure 3 shows the luminosity differences between the models and the stars . the vertical axis is mass in solar units ; binary components are connected by a line . the two binaries with @xmath62 ( vv pyx and ai hya ) are denoted by crosses ; they are poor fits and should be given little weight . considering the best fits , @xmath59 ( solid squares ) , there is a dramatic trend : the highest mass models ( for example , em car ) are underluminous relative to the actual binaries , while the lower mass models are not . given the indications that the errors may be dominated by systematic effects , we approach a statistical discussion with caution . the two binaries which have @xmath62 are eliminated from this statistical discussion on the basis that these fits are too poor to be meaningful . in principle , the mean errors could show a systematic shift in the models relative to the data , but because we choose an optimum pair of models , the choice masks any absolute shift . the error should reappear as a larger rms difference . for luminosity , the first moment of the difference between model and stellar logarithmic luminosity is just the mean of this difference , which is @xmath65 in the base ten logarithm ( the models are too dim by this amount ) . the shift is smaller than the rms error of the observational data , which is 0.056 . if there were a bad global mismatch , the rms difference in `` model minus star '' would be much larger than the average error in the observations . however , the rms difference between the models and stars is 0.054 , which is almost the same as the observational error . the luminosity is basically a measure of the leakage time for radiation , which is dominated by the value of the opacity in the radiative regions . it samples the whole star , including the deep interior . in a global sense , our mean leakage rate seems correct to the level of the statistical and observational error . the shifts in radius between the models and the stars are shown in figure 4 . the vertical axis and the symbols are the same as in the previous figure . the mean shift is 0.0053 in the logarithm ( the models are too large by this small amount ) ; the corresponding standard error of the andersen data for which the fits are acceptable is 0.016 , to be compared with an rms differnce between models and stars of 0.014 . except for a few outlying cases , the distribution is fairly uniformly distributed around zero . if only the best fits ( squares ) are considered , a subtle trend might be inferred : 9 of 12 of the models above @xmath66 have radii which are too large . the corresponding mean shift in @xmath67 is @xmath68 ( the models are too cool by this amount ) . again , this is small in comparison to the standard error of the observations ( @xmath69 ) . the corresponding rms difference between models and stars is @xmath70 . the effective temperature is a surface quantity , and is more sensitive to the outer layers which contain little of the stellar mass . these numbers suggest that standard stellar evolutionary sequences of these stages should be able to produce luminosities , radii , and effective temperatures within 11 , 3 and 4 percent , respectively , of good observational data . otherwise , new physics is indicated . because the standard stellar evolutionary models do this well , small `` improvements '' may contain no information . we will emphasize systematic trends , and those implications which emerge from several independent tests . .,scaledwidth=85.0% ] figure 5 shows the evolution of the model stars in log luminosity and log radius , for em car , v478 cyg , cw cep , qx car , cv vel , and u oph , corresponding to a mass range from 23 to 4.6 @xmath71 . except for qx car ( @xmath72 ) , models of these binaries have @xmath59 , and so represent good fits . the error bars are centered on the observed stars ; the arrows point from them to the optimum models . notice the the fits can be multivalued because the trajectories may pass through the error boxes multiple times . this is shown occurring first as the model descends from the pre - main sequence ( pre - ms ) , and again during main sequence hydrogen burning . if the stellar masses are significantly different , this ambiguity is removed by the condition that both components have the same age . all the model stars are too dim ( all the arrows point downward ) , a signal that the standard stellar evolution prescription is systematically wrong . em car , being the most massive system , also has the most significant mass loss . the model evolutionary sequences are set by the choice of initial mass , but the observational constraint on mass is applied after the best fitting model is determined . this is an implicit function of the choice of initial mass , and iteration is required . initial masses of 22.91 and 20.91 produce masses at fit of 22.25 and 20.12 @xmath71 , respectively . this mass ratio of 0.904 is consistent with the observational value of @xmath73 @xcite , and the masses agree with observation to within the estimated errors ( @xmath74 ) . however , even this loss is still small . a loss of @xmath75 is about twice the uncertainty in mass determination , @xmath76 . such a change in mass , since @xmath77 , corresponds to a shift in luminosity of @xmath78 , to be compared to the observational error in luminosity of @xmath79 , which is still larger . this is due to the fact that effective temperature is less well determined than the radius . a concentrated effort to refine the effective temperature determinations for em car , v478 cyg , cw cep , qx car , cv vel , and u oph would translate directly into much sharper constraints on massive star evolution . @xcite have revised the effective temperature for qx car ( upward ) by twice the quoted error , so that the inferred luminosity increases . prior to this revision , the fit to qx car was good ( @xmath59 ) . this larger discrepancy for qx car is in the same sense as noted for the other massive systems ; the models are dimmer than the stars . taking the larger masses from @xcite , with or without mass loss , still gives good fits , and the models are still dimmer than the stars . the result seems robust . @xcite have used high resolution spectral images obtained with the international ultraviolet explorer ( iue ) to study the winds from cw cephei ( hd 218066 ) . they place upper limits on the mass - loss rates of @xmath80 for the primary and @xmath81 for the secondary . the model masses start at 12.9 and 11.9 and decrease only to 12.8 and 11.88 respectively , to be compared to @xmath82 and @xmath83 @xcite . the mass loss predicted by the @xcite theory for our models at the point of minimum error is @xmath84 for the primary , and @xmath85 for the secondary . if the upper limits were detections , this could be considered good agreement , considering the complexity of the problem of interpreting the system @xcite . the net loss of mass up to this point is no larger than the error in mass determination , @xmath86 . because the mass loss rate is restricted by these observations to be at or below the value we use , the mass loss process should have no larger effect than we compute . hence , the remaining discrepancy must come from some other effect . table [ tbl-3 ] gives the instantaneous mass loss rates from the models , at the point of optimum fit , for the most massive binary systems . at lower masses , the mass loss rates are smaller still . additional observational data on mass loss for these systems could prove crucial in clarifying the role of mass loss in stellar evolution . @xcite estimate ages for em car and cw cep . their procedure not only gives ages , but also error estimates for those ages . our ages agree with theirs to within these errors , even though we use no overshooting and they do . it may be that the convective region in high mass stars is sufficiently large that the gross evolutionary properties of stars on the main sequence are not greatly affected by the overshooting correction . the understanding of the physics of overshooting is still too preliminary to do more than speculate on this issue . @xcite find acceptable fits for em car , v478 cyg , cw cep , qx car , and u oph ( it is probable that qx car would not have been a good fit with the revised effective temperatures ) , but with increasingly lower heavy element abundances with increasing mass . em car and v478 cyg have fits at the limit of the heavy element abundance range . @xcite find a similar effect : the heavy element abundances of their massive binaries are marginally smaller than those of the less massive ones . their effect is not quite as obvious as in @xcite , perhaps because @xcite do not force the helium abundance to correlate with heavy element abundance , and it fluctuates for these systems . the added degree of freedom may allow the fitting procedure to obscure the trend . this behavior could be interpreted as a galactic evolutionary effect , which would be extremely interesting , but there is another possibility . the problem with the massive models ( see figure 2 and figure 3 ) is that they are too dim . lower heavy element abundance gives higher luminosity because of reduced opacity . the fitting algorithms , having little freedom for mass variation ( thanks to the high quality of the data ) , must find lower heavy element abundance preferable , whether or not the heavy element abundances are actually smaller . more effective mixing , giving larger cores , also results in higher luminosities even if the abundances are unchanged . _ it is crucial to obtain spectroscopic information to decide the issue . _ @xcite have recently examined v380 cyg , which is a binary of disparate masses ( @xmath87 and @xmath88 ) and evolutionary state . they conclude that more mixing is needed ( @xmath89 ) . however , this system is complicated . guinan et al . estimate that the system is approximately ten thousand years from roche lobe overflow by the primary . thus , conclusions based on this system should be made with caution . the overshoot parameter @xmath2 may be a function of mass ( at least ) . similar behavior can also be seen in @xcite , which finds different values of the best fit overshoot parameter for five different masses . alternatively , rotational mixing might be increasingly effective for larger masses . phenomenological prescriptions are valuable if they capture the essential physics of the phenomena ; if fitted parameters turn out to be variable , a new formulation is needed . we have at least three causes for one effect ; sorting this out is an interesting theoretical and observational challenge . phe , iq per , pv cas , ai hya , vv pyx and rs cha . the masses range from 3.93 to 1.74 @xmath71.,scaledwidth=85.0% ] figure 6 shows @xmath30 phe , iq per , pv cas , ai hya , vv pyx , and rs cha , the group which has some of the most challenging binaries . the masses range from 3.9 to 1.1 @xmath71 . both @xmath30 phe and iq per have a mass ratio significantly different from one : 0.65 and 0.49 respectively . because the more massive components will evolve more rapidly , common age is a stringent constraint . in both cases , the error is dominated by the less massive component . for @xmath30 phe , the @xmath90 secondary is brighter than the model ; @xcite have the same problem . @xcite avoid it by using a lower heavy element abundance ( 0.013 ) and a higher helium abundance ( 0.29 ) . the heavy element abundance might be tested by high resolution spectroscopy . for iq per , the @xmath91 secondary is too blue ; its evolutionary track never gets so hot . @xcite attribute the difficulty in fitting @xmath30 phe and iq per to problems in determining @xmath63 of the secondary @xcite , which is much dimmer because of the large mass ratios of the components . for @xmath30 phe and iq per , it is clear that much of our `` difficulty '' is due to the relatively small error bars ; see figure 2 . consequently , small changes may improve the @xmath35 significantly , even if they do not correspond to the physics of the system . in the case of iq per , use of the @xcite value of effective temperature improves the fit , compared to @xcite , as does adjustment of the abundances . ai hydrae is particularly interesting because the primary is fitted by a model which is swiftly evolving , so that catching it in such a stage is unlikely . overshoot from the convective core would broaden the main sequence band and increase the age of the fast evolving primary , allowing the possibility of a fit with a more probable , slower evolutionary stage . this is consistent with the conclusions of @xcite , who find that ai hya is the only binary for which the overshooting models give a greatly improved fit . some of the ai hya behavior can be attributed to heavy element abundance effects . both members are classified as peculiar metal line stars . the heavy element abundance of this system ( from multi - color photometry ) is 0.07 @xcite , which is 3.5 times the value used in the models . the true interior composition can not be this metal rich . we have examined a sequence which had a heavy element abundance of 0.03 rather than 0.02 . this modest change gave a dramatic shift toward lower luminosity ( @xmath92 , which is three times the observational error ) and cooler effective temperatures ( @xmath93 , which is also three times the observational error ) . however , if the heavy element abundance were high only near the surface of the star , the opacity effects would produce a shift to the red in the evolutionary tracks , which would bring the models much more in line with the observations . vv pyx has almost identical components , so that their coeval origin has almost no effect on the fit . they track the same path at essentially the same time . the fit is simply the point that the observational error box is most closely approached , and should be viewed with caution , especially as the models give a poor fit . only pv cas and rs cha have good fits ( @xmath94 ) , and they are pre - main sequence ( pre - ms ) . rs cha has previously been suggested to be in a pre - main sequence stage @xcite . the pre - ms identification would have important theoretical implications . if true , it implies that the error in the models occurs after the core convection has been established in these stars . in any case , convection is an interesting possible cause for the problem ; these binaries have at least one component with convective core burning . pv cas has sufficiently different masses to require us to examine the pre - ms fit seriously . questions have been raised about the evolutionary status of pv cas since @xcite . previous attempts to fit the system to main sequence models @xcite have been unsatisfactory , mainly due to a large and irreconcilable age discrepancy between the members . fitting both components to pre - ms models , however , produces excellent agreement . to test the case for pv cas being pre - ms , we looked for other observation clues . the double - lined eclipsing binary system rs cha was recently found by @xcite to be pre - ms . not only were pre - ms tracks for rs cha a better fit than post - ms tracks , but two other observations strengthened the argument : ( 1 ) rs cha had several nearby _ rosat _ all - sky survey x - ray sources nearby which were found to be very young , low - mass , weak - lined t tauri stars , and ( 2 ) rs cha s proper motion matched that of the t tauri stars , suggesting a genetic tie . pv cas is at a distance of 660 pc @xcite , and a young stellar aggregate or membership within an ob association could have been previously overlooked . searching the hipparcos and tycho-2 catalogs , as well as examining pv cas on the digitized sky survey , we found no evidence for pv cas being a member of a known ob association . more massive members of a putative association would be included in the hipparcos catalog with proper motions similar to that given in the tycho-2 entry for pv cas , but none were found . we searched for known groups of young stars with vizier at cds : the compilations of ob associations by @xcite , @xcite , and @xcite , and open clusters by @xcite and @xcite . the only possible known associations that pv cas could belong to are cep ob3 ( d=840 pc , @xmath95 ) and cep ob2 ( d=615 pc , @xmath96 ) , however their projected separations from pv cas are large ( 600 pc and 100 pc , respectively ) , and their average radial velocities are far from popper s value for pv cas ( @xmath97 ) . hence , pv cas does not appear to be connected to any known ob associations or clusters which help us to infer its nature . the _ rosat _ all - sky survey ( rass ) bright source catalog ( bsc ; @xcite ) and faint source catalog ( fsc ; @xcite ) were searched to see whether there was any evidence for a clustering of x - ray - emitting t tauri stars in the vicinity of pv cas . no concentration of sources near pv cas was detected , although the sensitivity of rass at 660 pc is about l@xmath98 @xmath99 10@xmath100 erg s@xmath101 , corresponding to the very high end of the x - ray luminosity function for t tauri stars @xcite . only one rass - bsc source was within 30 ( @xmath36 pc projected ) of pv cas , but its f@xmath102/f@xmath103 ratio was 2 magnitudes too high to be a plausible t tauri star candidate . the only rass - fsc source within 30 of pv cas appeared to be related to a galaxy cluster on the digitized sky survey . we conclude that we currently have no evidence for a pre - ms aggregate around pv cas which could strengthen the argument for its pre - ms status . however , the taurus clouds also are forming low mass pre - ms stars without high mass cluster counterparts . .,scaledwidth=85.0% ] figure 7 shows ek cep , my cyg , pv pup , dm vir , v1143 cyg , and ux men , whose masses range from 2.03 to 1.12 @xmath71 . the @xcite effective temperatures improve the fit for ux men . ek cep has a large mass ratio . the optimum fit occurs as ek cep b is still on the pre - ms track , in agreement with @xcite and @xcite . we find that the surface abundance of @xmath104 is depleted to about @xmath105 of its initial value , while @xmath106 is depleted from @xmath107 to @xmath108 . this corresponds to a depletion of elemental lithium of about 0.57 dex ( base 10 logarithm ) . this is somewhat larger than found by @xcite ( 0.1 dex ) , but may be due to differences in the nuclear reaction rates used . in this range , the depletions are almost linear in the net cross section for @xmath106 destruction . a careful analysis with a variety of rates is warranted : @xcite suggest that the observations are in conflict with pre - ms models giving a li depletion greater than 0.3 dex . although ek cep has a large @xmath35 ( @xmath109 ) if the radii are used in determining the fitting function , the situation is different for @xmath110-@xmath67 , the conventional hr plane . the observational errors are now larger , and the corresponding @xmath35 approaches 4 . this confirms the importance of using the radii directly as a discriminant@xcite . my cyg and ux men are found to be well into main sequence hydrogen burning . my cyg is underluminous relative to the models . a higher heavy element abundance would remove the discrepancy ; observational tests of this are needed . @xcite found @xmath111 and @xcite found @xmath112 , which are consistent with this suggestion . pv pup and v1143 cyg are on the pre - ms / ms boundary . the fitting procedure chooses the cusp at which the star settles down to main sequence burning . this cusp shifts with small changes in abundance , so these fits would benefit from independent measurement of the abundances in these binaries . dm vir has been updated for @xcite . although the changes were small , the new fit is in the middle of main sequence hydrogen burning instead of pre - ms contraction . the track lies well within the error bars ; the previous data also had @xmath59 , although a much younger age estimate . because the masses are almost equal , the coeval birth requirement has little effect , and the ages have a corresponding uncertainty . observational selection favors binaries with a small separation . in order to determine the true usefulness of these systems as tests of models of single star evolution it is necessary to know to what extent these systems are detached ( noninteracting ) , and how far into the past and the future this condition is satisfied . in order to answer this question to first order for the systems in our sample , the average roche lobe radius for each star was calculated using @xmath113 where @xmath114 is the binary separation and @xmath115 is the mass ratio with the star in question in the numerator @xcite . this average radius was then compared to the model radii to estimate when each star overflows its roche lobe . dynamical evolution of the orbits was not taken into account . none of the models indicated significant mass transfer prior to the ages of the models closest to the observed points since early in the pre - main sequence evolution . four of the binaries in the sample have at least one member which , according to the model radii , will overflow their roche lobes when they are between 1.3 and 2 times older than their current age . the results of the roche lobe comparisons are given in table [ tbl-4 ] . all stars labeled `` post '' do not exceed their roche lobe radius until well into their post - main sequence evolution . two stars are in contact early in the pre - main sequence evolution ( ek cep b and my cyg a ) . the times given for these stars correspond to when they contract below the critical radius and mass transfer ends . these numbers should be taken as a rough guideline at best , since the dynamical evolution of protostars is undoubtedly more complex than the simple algorithm used here . the models corresponding to the primaries em car and v478 cyg exceed their roche lobe radii in less than @xmath116 years . these values are approximate in that dynamical evolution is not taken into account , the model radii do not match exactly the observed radii , and an approximate roche lobe geometry was used to facilitate comparison to the spherically symmetric models . apsidal motion in binaries allows us to infer constraints on the internal mass distributions @xcite . apsidal motion , that is , rotation of the orientation of the orbital ellipse relative to an inertial frame , does not occur for binary orbits of point particles interacting by newtonian gravity . @xcite showed that the general relativistic expression for the periastron shift of a double star is the same as for the perihelion shift of mercury . following @xcite ( see pages 194 - 7 ) , the shift is @xmath117 where @xmath118 is the speed of light and @xmath119 the gravitational constant . using units of solar masses and radii , and with the period @xmath120 in days , this dimensionless number becomes @xmath121 apsidal orbits per orbit . tests of general relativity have reached high precision @xcite ; the perihelion shift has now been tested to about @xmath122 . there has been some controversy as to a possible breakdown of general relativity because of a discrepancy between observations and predictions of the apsidal motion of some systems . this has been clarified by claret ( see @xcite for a recent discussion ) , who pointed out errors in theoretical models and difficulties in observations , especially for systems whose apsidal periods are too long for much to be measured with modern equipment . we adopt the point of view that general relativity is better tested than subtleties in the evolution of binary stars , and ascribe errors to other causes ( tidal effects not included , rotational effects , and systematic errors in observational interpretation , for example ) . tides induced by each companion give an additional interaction which is not purely inverse square in the separation and cause apsidal motion . @xcite have discussed the validity of the classical formula , which we use , @xmath123f(e),\ ] ] where @xmath120 is the period of the orbit , @xmath124 the period of apsidal motion , @xmath125 the mass and @xmath126 the radius of the star @xmath127 , and @xmath128 where @xmath129 is the eccentricity of the orbit . the separation of the pair in solar radii is @xmath130 if the period @xmath120 is measured in days and the masses in solar units . the classical apsidal motion formula gives accurate results when the periods of the low - order quadrupole @xmath131 , @xmath132 and @xmath133modes are smaller than the periastron passage time by a factor of about 10 or more @xcite . for em car , the lowest order pulsational mode of the primary has a period of @xmath134 days compared with the orbital period of 3.414 days and an eccentricity of @xmath135 , so that this condition is just satisfied . if we assume that the observed apsidal motion is due only to these two effects , classical simple tides and general relativity , we have @xmath136 we use the products @xmath137 directly for greater precision , but quote the apsidal constants @xmath138 for comparison . @xcite has pointed out that accuracy problems may exist because the relevant parameter is @xmath139 , where @xmath138 is the apsidal constant and @xmath23 is the stellar radius , not just @xmath138 alone . figure 8 shows the integrand of the apsidal constant , which approaches an asymptotic value as the integration exceeds about 0.7 of the radius . inner regions contribute little because of their small radii ; outer regions have little mass . the change from the interior ( henyey ) integration to envelope integration occurs around @xmath140 , and is visible in the change in the density of points . at the join , the temperature is about @xmath141 , and the density @xmath142 this temperature is about ten times the value attained in the early opacity experiments @xcite on the nova laser , and is about half the goal for the national ignition facility ( nif ) . for such main sequence ( and pre - main sequence ) stars , the apsidal constants are most sensitive to the range of density and temperature which is directly accessed by high energy density laser experiments ( see @xcite and discussion above ) . in this range , the new opacities show significant deviation from those previously used in astrophysics @xcite . the @xcite catalog contains orbital elements for 128 binaries , including most ( 11 of 18 ) of the binaries in our list . table [ tbl-5 ] gives apsidal constants @xmath138 , as well as the observed and predicted ratios of @xmath143 . given the significant improvement in the opacities , a critical re - examination of these data seems warranted . figure 9 shows the dimensionless rate of apsidal motion , @xmath144 , which would be due to classical apsidal motion , plotted versus log of half the total binary mass . @xmath120 is the orbital period and @xmath124 the apsidal period . the observational data ( corrected for general relativity ) are shown as diamonds , with vertical error bars . the model predictions are shown as solid squares ( @xmath59 ) for the best fits , open triangles for @xmath145 , and crosses for @xmath62 . the massive binaries with good fits ( em car , v478 cyg , and cw cep ; @xmath146 ) have predicted apsidal motion in excess of that observed , and qx car also follows that trend . these models are not as centrally condensed as the stars . this may be related to the underluminosity of these models found above . additional mixing would give more massive , convective cores , which would result in both greater luminosity and more centrally condensed structure . of the lower mass binaries with measured apsidal motion , only pv cas has a good fit model . its predicted apsidal motion is also larger than that observed ( the stars are more centrally condensed ) . the other binaries need better fitting models before the tests can be convincing . note that at the lowest tick mark in figure 9 , the apsidal period is measured in centuries . @xcite have shown that inclusion of ( a ) overshooting , ( b ) variation in heavy element abundance , and ( c ) rotation can produce models consistent with the apsidal data ; see also @xcite . this represents good progress toward establishing the apsidal motion data as a useful test of stellar evolution . our results , while not based upon identical assumptions , are consistent . the challenge is that of correctly determining the relative importance of the several different small effects which can give consistency with the observations . standard stellar evolution , without embellishments such as overshooting and rotation , does fairly well on these quantitative tests . while clear discrepancies exist , they are relatively subtle . this makes it difficult to uniquely identify exactly which additional physics is needed . we find a detailed consistency with similar calculations by @xcite and @xcite . it is important to test observationally the abundance variations implied by @xmath35 optimization , as such procedures may hide missing physics in parameter variation . laser experiments now explore the regions of temperature and density which are relevant not only to conventional stellar evolution , but also to apsidal motion tests . massive stars require more mixing than given by standard stellar evolution , and probably more than the prescriptions for overshooting used by @xcite and @xcite . rotational mixing , overshooting which is mass dependent , or something else is needed . our mass loss prescription is near the observational upper limit , so that additional mass loss is an unlikely solution . lower mass stars with convective cores are not well fit by standard stellar evolution . again , additional mixing is a promising answer . several of these binaries seem to be pre - main sequence ; this will allow some interesting tests of depletion of light nuclei and the mixing processes . we find two serious challenges : ( 1 ) disentangling conflicting solutions of the relatively subtle discrepancies , and ( 2 ) controlling shifts in the observational `` target areas '' due to systematic errors , which seem to be larger than the statistical errors . improved determinations of effective temperature , and of heavy element abundances ( e.g. , @xmath147 $ ] ) , would greatly improve these tests . stimulating discussions with brigitta nordstrom and johannes andersen are gratefully acknowledged . special thanks are due to marcel arnould , shimon asida , john bahcall , audra baleisis , al cameron , joergen christensen - dalgaard , paul drake , rolf kudritzki , robert kurucz , t. j. pearson , ted perry , thomas rauscher , marc rayet , bruce remington , friedel thielemann , frank timmes , and cliff will , whose work and/or advice contributed to aspects of this paper . this work was supported in part by doe , grant number de - fg03 - 98dp00214/a001 , and a subcontract from asci flash center at u. of chicago . press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. , 1992 , numerical recipes in fortran , second edition , university press : cambridge , quataert , e. j. , kumar , p. , & on , chi , 1996 , , 463 , 284 crrllllll em car & @xmath148 & a & o8v & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + hd97484 & & b & o8v & @xmath154 & @xmath155 & @xmath156 & @xmath152 & @xmath157 + v478 cyg & @xmath158 & a & o9.5v & @xmath159 & @xmath160 & @xmath161 & @xmath162 & @xmath163 + hd193611 & & b & o9.5v & @xmath164 & @xmath160 & @xmath165 & @xmath166 & @xmath163 + cw cep & @xmath167 & a & b0.5v & @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 + hd218066 & & b & b0.5v & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 + + qx car & @xmath178 & a & b2v & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 + hd86118 & & b & b2v & @xmath184 & @xmath185 & @xmath186 & @xmath187 & @xmath188 + cv vel & @xmath189 & a & b2.5v & @xmath190 & @xmath191 & @xmath192 & @xmath193 & @xmath194 + hd77464 & & b & b2.5v & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 + u oph & @xmath200 & a & b5v & @xmath201 & @xmath202 & @xmath203 & @xmath204 & @xmath205 + hd156247 & & b & b6v & @xmath206 & @xmath207 & @xmath208 & @xmath209 & @xmath210 + + @xmath30 phe & @xmath211 & a & b6v & @xmath212 & @xmath213 & @xmath214 & @xmath215 & @xmath216 + hd6882 & & b & b8v & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 + iq per & 1.74 & a & b8v & @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 + hd24909 & & b & a6v & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 + pv cas & 1.75 & a & b9.5v & @xmath232 & @xmath233 & @xmath234 & @xmath235 & @xmath236 + hd240208 & & b & b9.5v & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + + ai hya & 8.29 & a & f2 m & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 + + 02259 & & b & f0v & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 + vv pyx & 4.60 & a & a1v & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 + hd71581 & & b & a1v & @xmath257 & @xmath253 & @xmath258 & @xmath255 & @xmath256 + rs cha & 1.67 & a & a8v & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 + hd75747 & & b & a8v & @xmath264 & @xmath265 & @xmath266 & @xmath267 & @xmath268 + + ek cep & 4.43 & a & a1.5v & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 + hd206821 & & b & g5vp & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 + my cyg & 4.01 & a & f0 m & @xmath279 & @xmath280 & @xmath281 & @xmath249 & @xmath282 + hd193637 & & b & f0 m & @xmath283 & @xmath280 & @xmath284 & @xmath285 & @xmath286 + pv pup & 1.66 & a & a8v & @xmath287 & @xmath288 & @xmath289 & @xmath290 & @xmath291 + hd62863 & & b & a8v & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 + + dm vir & 4.67 & a & f7v & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath301 + hd123423 & & b & f7v & @xmath302 & @xmath298 & @xmath303 & @xmath304 & @xmath301 + v1143 cyg & 7.64 & a & f5v & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 + hd185912 & & b & f5v & @xmath310 & @xmath311 & @xmath312 & @xmath313 & @xmath314 + ux men & 4.18 & a & f8v & @xmath315 & @xmath310 & @xmath316 & @xmath317 & @xmath318 + hd37513 & & b & f8v & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath323 + + crrrrrrr em car & a & @xmath324 & 0.972 & 4.509 & 4.933 & 6.666 & 1.67 + hd97484 & b & @xmath325 & 0.937 & 4.504 & 4.843 & 6.668 & + v478 cyg & a & @xmath326 & 0.879 & 4.464 & 4.566 & 6.807 & 2.67 + hd193611 & b & @xmath327 & 0.866 & 4.462 & 4.530 & 6.808 & + cw cep & a & @xmath328 & 0.766 & 4.440 & 4.245 & 6.788 & 2.39 + hd218066 & b & @xmath329 & 0.720 & 4.421 & 4.075 & 6.798 & + + qx car & a & 9.267 & 0.649 & 4.362 & 3.698 & 6.986 & 11.3 + hd86118 & b & 8.480 & 0.611 & 4.343 & 3.544 & 7.000 & + cv vel & a & 6.100 & 0.614 & 4.231 & 3.103 & 7.604 & 1.30 + hd77464 & b & 5.996 & 0.603 & 4.228 & 3.070 & 7.607 & + u oph & a & 5.198 & 0.538 & 4.198 & 2.820 & 7.687 & 0.43 + hd156247 & b & 4.683 & 0.480 & 4.177 & 2.623 & 7.699 & + + @xmath30 phe & a & 3.930 & 0.457 & 4.136 & 2.409 & 7.831 & 11.4 + hd6882 & b & 2.551 & 0.283 & 4.028 & 1.633 & 7.836 & + iq per & a & 3.521 & 0.380 & 4.119 & 2.189 & 7.656 & 6.92 + hd24909 & b & 1.737 & 0.195 & 3.891 & 0.906 & 7.688 & + pv cas & a & 2.827 & 0.362 & 4.015 & 1.736 & 6.576 & 1.12 + hd240208 & b & 2.768 & 0.351 & 4.011 & 1.698 & 6.577 & + + ai hya & a & 2.145 & 0.539 & 3.866 & 1.494 & 9.023 & 21.4 + @xmath330 & b & 1.978 & 0.395 & 3.865 & 1.204 & 9.025 & + vv pyx & a & 2.101 & 0.349 & 3.920 & 1.331 & 6.850 & 32.2 + hd71581 & b & 2.099 & 0.350 & 3.920 & 1.330 & 6.850 & + rs cha & a & 1.858 & 0.324 & 3.893 & 1.174 & 6.925 & 1.57 + hd75747 & b & 1.821 & 0.363 & 3.876 & 1.183 & 6.928 & + + ek cep & a & 2.029 & 0.228 & 3.952 & 1.217 & 7.429 & 10.3 + hd206821 & b & 1.124 & 0.086 & 3.761 & 0.166 & 7.432 & + my cyg & a & 1.811 & 0.347 & 3.851 & 1.052 & 9.117 & 6.69 + hd193637 & b & 1.786 & 0.329 & 3.853 & 1.025 & 9.121 & + pv pup & a & 1.565 & 0.183 & 3.858 & 0.750 & 7.980 & 1.66 + hd62863 & b & 1.554 & 0.182 & 3.855 & 0.738 & 8.101 & + + dm vir & a & 1.460 & 0.241 & 3.802 & 0.639 & 7.177 & 0.15 + hd123423 & b & 1.454 & 0.240 & 3.800 & 0.633 & 7.180 & + v1143 cyg & a & 1.391 & 0.169 & 3.789 & 0.447 & 7.323 & 9.13 + hd185912 & b & 1.347 & 0.151 & 3.783 & 0.388 & 7.327 & + ux men & a & 1.238 & 0.140 & 3.781 & 0.356 & 9.266 & 2.31 + hd37513 & b & 1.198 & 0.119 & 3.773 & 0.283 & 9.303 & + rcrr em car & a & @xmath331 & @xmath332 + & b & @xmath333 & @xmath334 + + v478 cyg & a & @xmath335 & @xmath336 + & b & @xmath337 & @xmath338 + + cw cep & a & @xmath339 & @xmath340 + & b & @xmath341 & @xmath342 + + qx car & a & @xmath343 & @xmath344 + & b & @xmath345 & @xmath346 + + crrrrrrrr em car & a & @xmath324 & 33.75 & @xmath347 & 9.34 & 12.9 & 6.666 & 6.795 + & b & @xmath325 & & & 8.33 & 12.6 & 6.668 & 6.832 + v478 cyg & a & @xmath326 & 27.31 & @xmath348 & 7.42 & 10.5 & 6.807 & 6.949 + & b & @xmath327 & & & 7.42 & 10.2 & 6.808 & 6.965 + cw cep & a & @xmath328 & 24.22 & @xmath349 & 5.68 & 9.33 & 6.788 & 7.085 + & b & @xmath329 & & & 5.18 & 8.91 & 6.798 & 7.168 + qx car & a & @xmath350 & 29.82 & @xmath351 & 4.29 & 11.5 & 6.986 & post + & b & 8.480 & & & 4.05 & 11.0 & 7.000 & post + cv vel & a & 6.100 & 34.97 & @xmath352 & 4.09 & 17.0 & 7.604&post + & b & 5.996 & & & 3.95 & 13.2 & 7.607 & post + u oph & a & 5.198 & 12.76 & @xmath353 & 3.44 & 4.90 & 7.687 & 7.910 + & b & 4.683 & & & 3.01 & 4.68 & 7.689 & 8.030 + @xmath30 phe & a & 3.930 & 11.04 & @xmath354 & 2.85 & 4.57 & 7.831 & 8.276 + & b & 2.551 & & & 1.85 & 3.80 & 7.836 & 8.797 + iq per & a & 3.521 & 10.58 & @xmath355 & 2.45 & 4.68 & 7.656 & 8.405 + & b & 1.737 & & & 1.50 & 3.39 & 7.688 & post + pv cas & a & 2.827 & 10.85 & @xmath356 & 2.30 & 4.07 & 6.576 & 8.669 + & b & 2.768 & & & 2.26 & 4.07 & 6.577 & 8.697 + ai hya & a & 2.145 & 27.63 & @xmath357 & 3.91 & 10.7 & 9.023 & post + & b & 1.978 & & & 2.77 & 10.2 & 9.025 & post + vv pyx & a & 2.101 & 18.77 & @xmath358 & 2.17 & 7.08 & 6.850 & 9.066 + & b & 2.099 & & & 2.17 & 7.08 & 6.850 & post + rs cha & a & 1.858 & 9.14 & @xmath359 & 2.14 & 3.47 & 6.925 & post + & b & 1.821 & & & 2.34 & 3.47 & 6.928 & post + ek cep & a & 2.029 & 16.64 & @xmath360 & 1.58 & 7.08 & 7.429 & post + & b & 1.124 & & & 1.32 & 5.50 & 7.432 & 5.669 + my cyg & a & 1.811 & 16.27 & & 2.19 & 6.17 & 9.117 & 5.646 + & b & 1.786 & & & 2.19 & 6.17 & 9.121 & post + pv pup & a & 1.565 & 8.62 & @xmath361 & 1.54 & 6.17 & 7.980 & post + & b & 1.554 & & & 1.50 & 6.17 & 8.101 & post + dm vir & a & 1.460 & 16.79 & @xmath362 & 1.73 & 6.31 & 7.177 & post + & b & 1.454 & & & 1.73 & 6.31 & 7.180 & post + v1143 cyg & a & 1.391 & 22.83 & @xmath363 & 1.35 & 8.71 & 7.323 & post + & b & 1.347 & & & 1.32 & 8.51 & 7.327 & post + ux men & a & 1.238 & 14.69 & @xmath364 & 1.35 & 5.62 & 9.266 & post + & b & 1.198 & & & 1.27 & 5.50 & 9.303 & post + rcrrrrrrr em car & a & @xmath331 & 2.240 & @xmath365 & @xmath366 & @xmath367 & 2.74 & @xmath368 + & b & @xmath333 & 2.180 & @xmath369 + + v478 cyg & a & @xmath335 & 2.185 & @xmath370 & @xmath371 & @xmath372 & 3.43 & @xmath373 + & b & @xmath337 & 2.175 & @xmath374 + + cw cep & a & @xmath339 & @xmath375 & @xmath376 & @xmath377 & @xmath378 & 1.79 & @xmath379 + & b & @xmath341 & @xmath380 & @xmath381 + + qx car & a & @xmath343 & @xmath382 & @xmath383 & @xmath384 & @xmath385 & 0.341 & @xmath386 + & b & @xmath345 & @xmath387 & @xmath388 + + u oph & a & @xmath389 & @xmath390 & @xmath391 & @xmath392 & @xmath393 & 1.93 & @xmath368 + & b & @xmath394 & @xmath395 & @xmath396 + + @xmath30 phe & a & @xmath397 & @xmath398 & @xmath399 & @xmath400 & @xmath401 & 0.827 & @xmath402 + & b & @xmath403 & @xmath404 & @xmath405 + + iq per & a & @xmath406 & @xmath407 & @xmath408 & @xmath409 & @xmath410 & 0.418 & @xmath411 + & b & @xmath412 & @xmath413 & @xmath414 + + pv cas & a & @xmath415 & @xmath416 & @xmath417 & @xmath418 & @xmath419 & 0.597 & @xmath420 + & b & @xmath421 & @xmath422 & @xmath423 + + vv pyx & a & @xmath424 & @xmath425 & @xmath426 & @xmath427 & @xmath428 & 0.0876 & @xmath429 + & b & @xmath430 & @xmath431 & @xmath432 + + ek cep & a & @xmath433 & @xmath434 & @xmath435 & @xmath436 & @xmath437 & 0.0728 & @xmath438 + & b & @xmath439 & @xmath440 & @xmath441 + + v1143 cyg & a & @xmath442 & @xmath443 & @xmath444 & @xmath445 & @xmath446 & 0.0929 & @xmath447 + & b & @xmath448 & @xmath449 & @xmath450 + +

in the literature , the study of electron transport in quantum devices is mainly devoted to dc properties . the fluctuations of the electrical current around these dc values , the so - called quantum noise , are much less analyzed . the computation of quantum noise is intrinsically linked ( by temporal correlations ) to our ability to understand / compute the time - evolution of a quantum system that is measured several times . there are several quantum theories that provide different ( but empirically equivalent ) ways of understanding / computing the perturbation of the wave function when it is measured . in this work , quantum noise associated to an electron impinging upon a semitransparent barrier is explained using bohmian mechanics ( which deals with wave functions and point - like particles ) . from this result , the fundamental understanding and practical computation of quantum noise with bohmian trajectories are discussed . numerical simulations of low and high frequency features of quantum shot noise in a resonant tunneling diode are presented ( through the bitlles simulator ) , showing the usefulness of the bohmian approach .

1410.0530

historically , the definition of noise was related to the sound : a noise is an unwanted , unpleasant and confusing type of sound . however , such definition is ambiguous . what does it mean unwanted , unpleasant or confusing ? an attempt to provide a more academic definition comes from music : noise is a non - harmonious or discordant group of sounds . again , however , the definition is not free from ambiguities because one man s noise is another man s music @xcite . a more scientific definition closer to our interest in electrical devices comes from communications : a noise is an electric disturbance that interferes with or prevents reception from a signal or information . for example , the buzz in a telephone call . thus , we realize that once we have a precise definition of what is a signal , the meaning of what is noise becomes perfectly clear : it is the difference between the measured value and the signal . as discussed above , the answer to what is noise in electrical devices depends on our definitions of the electrical signal . for most dc applications , the signal is just a time average value of the current . for frequency applications , the signal is equivalently defined as a time average value , but using a shorter time interval ( related to the inverse of the operating frequency ) . in other applications , mainly digital applications , the signal is related to a time average value of the voltage in a capacitor . hereafter , we will assume that the electrical signal is the dc value of the current , referenced by the symbol @xmath0 . all fundamental and practical issues discussed here for the dc signal ( and its noise ) can be easily and straightforwardly extended to those other types of electrical signals . what is measured in a laboratory for the dc signal is the time average value of the instantaneous current @xmath1 in a unique device during a large period of time @xmath2 : @xmath3 once we have defined the signal @xmath0 as the dc value , in principle , the noise can be quantified by time averaging the difference between the measured value of the current @xmath1 and the signal in a unique device : @xmath4 the square of the difference avoids positive and negative cancellations . + at this point , it is very important to realize that @xmath1 presents very rapid fluctuations that can not be captured by the standard laboratory apparatuses . any experimental setup that measures the current fluctuations behaves as a low - pass filter ( i.e. the current fluctuations at frequencies higher than the apparatus cut - off frequency are not measured ) . therefore , the experimentally accessible information about the current fluctuations is not given by eq . ( [ noise2 ] ) , but by the power spectral density of the fluctuations @xmath5 ( and its related magnitudes ) . from the wiener - khinchine relation , the power spectral density can be defined as the fourier transform of the time average definition of this autocorrelation function @xmath6 : @xmath7 where @xmath8 . a straightforward calculation shows that eq . ( [ noise3 ] ) can be rewritten as @xmath9 with : @xmath10 then , the fourier transform of eq . ( [ noise3 ] ) gives the noise power spectral density @xmath5 : @xmath11 it is quite trivial to realize that the definition of the spectral density @xmath5 in eq . ( [ noise5 ] ) and eq . ( [ noise3 ] ) is consistent with the definition of the total noise defined in eq . ( [ noise5 ] ) is non - negative and symmetric with respect to @xmath12 . then , since only positive frequencies @xmath12 are measured in a laboratory , the measured density includes our @xmath5 and @xmath13 , and the integral of the noise spectrum measured in a laboratory runs from @xmath14 till @xmath15 . ] in eq . ( [ noise2 ] ) : @xmath16 where we have used the definition of the delta function @xmath17 . + it is very relevant for the rest of the paper to realize that the measurement of @xmath5 through the function @xmath18 defined in eq . ( [ noise4 ] ) requires the knowledge of the measured value of the current during all @xmath19 . thus , we have to make predictions about the evolution of the electronic device while being ( continuously ) measured . in a classical scenario , such discussion about measurement is generally ignored . on the contrary , for quantum systems , it has very relevant implications because the evolution of a system with or without measurement can be dramatically different . if the electronic device satisfies the ergodic theorem @xcite , a continuous measurement of the system can be avoided in practical computations . let us see in what sense ergodicity can simplify our noise computations . in general , the _ mathematical _ concept of a random process is used to deal with noise . a random process requires a sample space . in our case , we can define an ensemble of _ identical _ electrical devices for the instantaneous current measured during the first time interval , @xmath20 for the second interval , and so on . ] , each one labeled by the sample space variable @xmath21 . then , the ( instantaneous ) current is labeled by the random process @xmath22 . for a fixed time , @xmath23 , the quantity @xmath24 is a random variable . for a fixed device @xmath25 , the function @xmath26 is a well - defined non - random function of time . finally , @xmath27 is just a real number . often the sample space variable @xmath21 is omitted in the notation . the dc value of the current in eq . ( [ noise1 ] ) can be alternatively defined for an ergodic system as : @xmath28 where @xmath29 is the probability of getting @xmath30 at time @xmath23 . these probabilities are defined as the ratio of the number of devices providing @xmath30 divided by the total number of devices . it is important to realize that the experimental evaluation of eq . ( [ noise7 ] ) requires only one measurement of the current at @xmath23 in a large number of _ identical _ @xmath21-devices . then , the theoretical predictions of eq . ( [ noise7 ] ) do only need to determine the free ( without measuring apparatus ) evolution of the electronic device from the initial time @xmath31 till @xmath23 . see a detailed discussion in appendix [ app - noise ] about how ergodicity avoids the complications of the measurement in a quantum system . obviously , we can compute the total noise represented in eq . ( [ noise2 ] ) from a unique measurement in ergodic systems : @xmath32 however , the noise measured in a laboratory is not given by @xmath33 , but by @xmath5 in eq . ( [ noise5 ] ) . we repeat the reason explained in eq . ( [ noise2 ] ) . the amount of noise generated by an instantaneous current evolving for example from @xmath34 ma to @xmath35 ma during a time interval of @xmath36 fs , is not captured from the state - of - the - art laboratory apparatuses ( which already have difficulties to capture noise at frequencies higher than a few of terahertzs ) . from an experimental point of view , in fact , it is easy to get @xmath37 , but impossible to get @xmath38 . we can compute the noise power spectral density @xmath5 from the ensemble average version of the autocorrelation defined in eq . ( [ noise4 ] ) as : @xmath39 in general , we can assume that the instantaneous current in an electronic device behaves as a wide - sense stationary random process . then , @xmath0 in eq . ( [ noise7 ] ) is constant and time - independent . identically , then , the autocorrelation function in eq . ( [ noise8 ] ) depends only on the time difference @xmath40 with @xmath41 . finally , we use eq . ( [ noise5 ] ) with @xmath9 computed from eq . ( [ noise8 ] ) , to get the noise power spectral density @xmath5 . it is important to emphasize ( for a posterior discussion ) that the probability @xmath42 implies a two - measurement process for each electronic device . the system evolves freely ( without interaction with the measurement apparatus ) from @xmath31 till @xmath23 when the current is measured , giving the value @xmath30 . then , the system evolves freely again until time @xmath43 , when the system is measured again giving @xmath44 . in summary , even if the ergodicity argument is invoked , the noise computation through the autocorrelation function requires , at least , two measurements at different times in a single device ( and the average over all @xmath21-devices ) . we anticipate that our computations with bohmian mechanics will not assume ergodicity ( which is not an obvious property for open systems out of equilibrium @xcite ) , but the prior expressions requiring a continuous measurement of the current . let us emphasize that the previous discussion is valid for either classical or quantum devices . the adjective _ quantum _ emphasizes that the signal and the noise are computed or measured in an electrical device governed by quantum laws @xcite . if the electronic device is not ergodic , expression ( [ noise4 ] ) requires a continuous measurement of the current @xmath1 . on the contrary , for an ergodic electron device , expression ( [ noise7 ] ) requires one unique measurement , while expression ( [ noise8 ] ) requires a two - times measurement when dealing with the power spectral density @xmath5 . up to here , we realize that the definition of quantum noise seems very trivial . then , why does the concept of quantum noise have a halo of mystery around it ? our previous definition about what is quantum noise does not answer the question of how we compute it . if we want to predict the values @xmath1 used in eq . ( [ noise2 ] ) and eq . ( [ noise4 ] ) or the probabilities @xmath45 and @xmath42 for eq . ( [ noise7 ] ) and eq . ( [ noise8 ] ) , we require a quantum theory . there are several quantum theories available in the literature that , by construction , are empirically equivalent when explaining all quantum phenomena . among others , the so - called copenhagen or orthodox interpretation @xcite , bohmian mechanics @xcite or the many - worlds theory @xcite . any theory has usually two different planes . first , the formalism , which is a set of mathematical rules ( using elements such as wave functions , operators , trajectories ) that allow us to make practical computations that reproduce experimental results . the formalism of a theory provides an answer to the question : how do we compute quantum noise ? the second plane of a theory is its interpretation . it tries to provide a deep connection on how the mathematical rules and its elements explain how nature works . the interpretation of the theory provides answers to the question : which is the physical origin of quantum noise ? each quantum theory will provide its own answers to both questions . many people argue that the only important part of a quantum theory ( once we know it is empirically valid ) is its formalism because it is the only part we need to make computations . certainly , one can make noise computations using any of the available formalisms without worrying about its interpretation . at the end of the day , by construction , each theory should give the same predictions . other people argue that even when one is only interested in computations , a correct understanding of the interpretational issues of each theory is fruitful because it provides an enlarged vision about how correctly apply the theory in unsolved problems ( abandoning the _ shut up and calculate _ we will return to this very point later , at the conclusions in sec . [ sec5 ] . now , we want to clarify why quantum noise is specially sensible to fundamental quantum mechanical issues . any electrical device ( or any experiment ) is connected to a measuring apparatus . in our case , an ammeter to get the electrical current . quantum noise is sensible to the ( ammeter ) measuring process . as stated in eq . ( [ noise8 ] ) , in order to obtain the noise , the quantum system has to be measured , at least , twice . this two - time measurement faces directly with one of the most complex issues in quantum mechanics , namely , which is the perturbation of the quantum wave function when a measurement is performed . historically , this perturbation is known with the somehow scary name of _ the collapse of the wave function_. can we ignore it ? definitively not if temporal correlations need to be correctly predicted . see for example , ref . @xcite:_the fluctuations ... are a consequence of a probabilistic reflection and transmission probability ( a wave phenomena ) and are a consequence of the fact that detectors register either a transmitted or a reflected particle ( a particle phenomena)_. the measurement process is hidden in the word _ we also mention that the fundamental understanding / computing of the measurement process can be largely relaxed when dealing with dc predictions . they can be computed from an ensemble of devices with only one measurement in each device , so we can ignore the evolution of the quantum system after the measurement . see appendix [ app - noise ] to enlarge this point . in this paper , we will provide an explanation to the role of collapse in quantum noise from a bohmian perspective . we emphasize that we are not saying that the bohmian answer is the best one . answers from other theories are equally satisfactory , and provide the same predictions . we are just defending that it is a consistent answer that in the authors opinion provides a quite intuitive and understandable explanation of quantum noise and also a numerically accessible formalism . in sec . [ sec2 ] we explain how the copenhagen interpretation explains the multi - time measurement process in a experiment with a flux of electrons impinging upon a tunneling barrier , by introducing the notion of operators . in sec . [ sec3 ] we provide an explanation of the same experiment using bohmian mechanics , without using operators . then , in sec . [ sec4 ] we illustrate how the formalism of bohmian mechanics exposed in sec . [ sec3 ] can be applied in practical problems to calculate the quantum noise in electrical devices , including coulomb and exchange interactions . finally , we conclude in sec . [ sec5 ] explaining how the different theories explain the origin of quantum noise . a typical scenario when discussing quantum noise in electrical devices is a flux of electrons impinging upon a partially transparent barrier ( located in the middle of the active region ) . electron transport through the barrier takes place by tunneling . electron is either transmitted or reflected , but not both ! @xcite we get a transmitted electron with a probability @xmath2 , while a reflected one with probability @xmath46 . to simplify the discussion , we consider a constant injection of electrons ( at zero temperature ) , one by one . each electron , after measurement at time @xmath23 , will appear randomly at the left or at the right of the barrier . the time averaged number of transmitted electrons will be proportional to @xmath2 , but the number of transmitted electrons fluctuates instantaneously because of the randomness of the transmission . these fluctuations of the number of transmitted electrons ( when compared with the dc signal ) are named partition noise @xcite . there are many other sources of noise in electrical devices , for example , the @xmath47 noise which become very relevant at low frequencies @xcite . in this paper , we will only deal with partition noise due to a tunneling barrier . in sec . [ sec4 ] , we will discuss partition noise considering also the injection of electrons at a finite temperature ( the so - called thermal noise ) . the fluctuations due to both processes simultaneously are known in the literature as shot noise @xcite . in this section we discuss how the partition noise is understood within the orthodox interpretation of quantum mechanics , also known as copenhagen interpretation @xcite . let us specify that most available formulations of shot noise are developed within this orthodox interpretation @xcite . we consider a very simple example , but with a detailed discussion of the role played by the measuring apparatus ( the ammeter ) . the copenhagen interpretation associates a wave function @xmath48 to a system of @xmath49 particles . in principle , such wave function _ lives _ in a @xmath50 dimensional configuration space . within the first non - relativistic quantization language , the evolution of this wave function is defined by two laws @xcite . the first law , known as schrdinger equation , states that ( when the system is not measured ) the wave function evolves unitarily and deterministically according to the following equation @xmath51 where @xmath52 $ ] . with @xmath53 we denote a generic interaction potential in the position representation , with @xmath54 the mass of the @xmath55-th particle and with @xmath56 the multidimensional vector in the configuration space . to provide a simple discussion of the partition noise in a tunneling barrier , let us assume that each electron in our experiment can be described by a single - particle wave function ( we neglect the exchange and the coulomb interaction among electrons ) . in fig . [ figure1 ] we plot the ( unitary ) evolution of such wave function solution of eq . ( [ tdse ] ) . is the ( unitary ) schrdinger equation alone depicted in fig . [ figure1 ] enough to understand quantum noise ? the orthodox theory has a second law , known as the _ collapse _ of the wave function , that takes into account the effects of the interaction of a measuring apparatus with the quantum ( sub)system . this second law can be found in many textbooks @xcite . it requires a new non - unitary operator @xmath57 . this operator is different from the unitary schrdinger evolution , which is generated by the hamiltonian seen below eq . ( [ tdse ] ) , and it must be able to encapsulate all the interactions of the quantum systems with the rest of the particles ( including the ammeter , the cables , the environment , etc ) . this new operator @xmath57 is the only tool provided by the theory to determine the possible results of a measurement . in principle we do not know anything about this operator except that it is a ( hermitian @xmath58 ) function whose ( real ) eigenvalues @xmath59 of its spectral decomposition are the possible results of the measurement . once the system in fig . [ figure2 ] is measured ( and not before ) , the wave function is projected to one of the eigenstates of the mentioned operator in a non - unitary evolution . after the collapse , the _ new _ wave packet evolves again according to the time - dependent schrdinger equation until a new measurement is done . when it occurs the first measurement and ( d ) time @xmath43 corresponding to the second measurement . at time @xmath23 and @xmath43 , because of the unitary evolution , the electron can be detected at both sides of the barrier ( color figure online ) ] for simplicity , in our present conceptual discussion let us assume a reasonable ( but ad - hoc ) operator ( why this operator is reasonable will be clarified in sec . [ sec3 ] ) . such operator provides the following perturbation of the wave function . if the electron is _ randomly _ measured as a reflected electron at @xmath23 , the transmitted part of the wave function is eliminated . this measuring process corresponds to figs . [ figure2 ] ( c ) and ( d ) where only the reflected wave function survives after @xmath23 . equivalently , the measurement process associated to _ randomly _ getting a transmitted electron corresponds to eliminating the reflected part , as seen in figs . [ figure2 ] ( g ) and ( h ) . now , by comparing the evolutions of the wave functions in fig . [ figure1 ] and fig . [ figure2 ] , it is obvious that the former is wrong . by looking at fig . [ figure1 ] , it could be the case that an electron found at time @xmath23 at the right ( transmitted ) can be found in a later time @xmath43 at the left as a reflected electron ( see the evolution of the probability density in fig . [ figure1 ] ) . this sequence of possibilities is wrong . experimental results confirm that once , say time @xmath23 , the electron is detected at one side , in a later time @xmath43 it is always found at the same side . then , we get a very valuable lesson from the copenhagen explanation : the ( unitary ) schrdinger equation alone is not able to explain completely quantum noise . it is necessary to include the collapse of the wave function to understand properly what is quantum noise ( temporal correlations ) . the popular arguments that _ shot noise is a consequence of quantization of charge _ @xcite or _ `` this is the noise that arises from the graininess of the current '' _ @xcite emphasize exactly this very point . at the left side . ( e ) , ( f ) , ( g ) and ( h ) non - unitary evolution of the wave function for a transmitted electron detected at time @xmath23 at the right side . symbols are the same as in fig . [ figure1 ] ( color figure online ) ] all mentioned orthodox formalisms dealing with quantum noise reproduce experimental results successfully because they include the measurement process inside @xcite . most of them do not discuss explicitly which is the operator associated with the ammeter . over the years , physicists have identified the operators , by developing instincts on which are the effects of measurements in the wave function . there are scenarios ( as the one depicted in fig . [ figure2 ] ) where it is quite obvious which operator is the _ right _ one . on the contrary , for example , when measuring the total ( conduction plus displacement ) current it is not at all obvious which are the relevant operators . is this measurement process _ continuous _ or _ instantaneous _ ? does it provide a _ strong _ or a _ weak _ perturbation of the wave function ? the answers to these questions are certainly not simple . the copenhagen theory itself does not answer these _ technical _ questions about how to find the _ right _ operator . can other quantum theories provide additional help ? in the previous section , we discussed how the copenhagen formalisms can be successfully used to understand quantum noise in electrical devices . one technical difficulty with this formalism is the proper definition of the _ right _ operator that determines the collapse of the wave function when , for example , the total ( conduction plus displacement ) current is measured . there are alternative theories which account for the perturbation of the wave function during a measurement process in a different way , without operators . the one that we will develop here is bohmian mechanics . let us emphasize again that both ( copenhagen and bohmian ) theories are empirically equivalent so that the preferences of one in front of the other are related to computational abilities , intuitive results , etc . @xcite . in the bohmian theory , the complete description of a quantum system of @xmath49 particles is given by the ( same ) wave function , @xmath48 mentioned in sec . [ sec2 ] , and by the actual positions of the point - like particles , @xmath60 the actual position of the particle , while the lower case letter @xmath61 is used to indicate generic positions . with the barred letter we refer to a multidimensional vector in the configuration space , while a letter without bar denotes the 3-dimensional vector in physical space . ] ( see appendix [ app - bohm ] and refs . @xcite for a more detailed discussion on this theory ) . we emphasize that the evolution law for the wave function @xmath48 is the same as in standard quantum mechanics : the schrdinger equation ( eq . ( [ tdse ] ) ) . the wave function _ guides _ the movement of the actual positions of the particles in time , according to the so - called _ guidance equation _ , which defines the velocity of the @xmath55-th particle as @xmath62 where @xmath63 is the usual probability current density , defined as @xmath64 , associated to the @xmath55-particle , @xmath65 is the usual probability distribution and i m denote the imaginary part . we note that eq . ( [ eq - guidance ] ) describes the _ velocity field _ for the @xmath55-particle and depends on the actual position of all particles of the system @xmath66 . each particle follows a definite trajectory which can be obtained integrating in time the velocity field @xmath67 where @xmath68 is the initial position of particle @xmath55 . + a proper ensemble of these trajectories ( proper means that the initial position of each trajectory of the ensemble is selected according to the initial squared modulus of the wave function , see eq . ( [ qeh ] ) in appendix [ app - bohm-3 ] ) reproduces the time - evolution of the many - particle wave function at any later time . + in sec . [ sec2 ] , we saw that in order to reproduce the experimental results , we have used the notion of operators to describe how the wave function of a measured system is modified under a measurement process . in the bohmian theory , we simply consider the apparatus as another ( big and complex ) quantum system interacting with our measured system . the interaction among them is then included in the hamiltonian of eq . ( [ tdse ] ) as any other interaction . then from the unitary evolution of the many - particle wave function ( system plus apparatus ) we can look at the behavior of the wave function of the measured system . the latter is called _ conditional wave function _ ( an exclusive concept belonging to bohmian mechanics ) and it is defined from the many - particle wave function in the configuration space , fixing all the actual particles positions excluding that of our subsystem ( see appendix [ app - bohm-2 ] for more details ) . let us provide a quite realistic ( in particular , non - instantaneous , but in some ways schematic ) example in which we can numerically track the behavior of the conditional wave function during the measurement process of the partition noise discussed in sec . the quantum system is an electron labelled as @xmath69 impinging on an external tunneling barrier . behind the barrier there is a measuring device , that we call transmitted charge detector " modeled as a single degree of freedom @xmath70 ( thought as the center of mass of a complex system ) , which can detect the successful transmission of an electron . first , we have an interaction of the electron with the potential barrier and , subsequently , an interaction with the transmitted charge detector . it is important to stress that both interactions are regarded at the very same level within bohmian mechanics . the measurement interaction introduces a channelling of the wave function in the configuration space such that the desired property of the `` quantum system '' ( here , whether the electron is reflected or transmitted ) can be read off from the final position @xmath70 of a particle , thought of as the pointer of the apparatus . the interaction between the electron and the pointer can be modeled as : @xmath71 where @xmath72 is the momentum operator of the detector and @xmath73 is the interaction constant . @xmath74 is a function that is equal to zero when the electron is outside the detector , ( @xmath75 in fig . [ conditional wave function - figure1 ] ) , and is equal to one when the particle is inside the detector ( @xmath76 ) . , from zero to one , is done softly in order to minimize the perturbation of the `` quantum system '' as explained in @xcite . ] in fig . [ conditional wave function - figure1 ] the region in the configuration space in which this function is different from zero is represented by a rectangle . the many - particle schrdinger equation reads @xmath77 where @xmath78 is the effective mass of the electron , @xmath79 is the mass of the apparatus pointer and @xmath80 is the external potential energy barrier . at four different times . the configuration space region where the _ transmitted charge detector _ is present is indicated by a rectangle and the barrier by a solid line . the @xmath81 line indicates the modulus of the conditional wave function @xmath82 , while the @xmath83 line corresponds to @xmath84 . four trajectories @xmath85 with different initial positions are presented with @xmath86 , @xmath87 , @xmath88 and @xmath81 . the actual detector position associated with the reflected trajectory ( @xmath81 ) with @xmath89 does not move because there is no interaction between this trajectory and the detector ( color figure online ) ] the main feature of a transmitted charge detector is that the center of mass of the wave function in the @xmath90 direction has to move if the electron is transmitted and it has to be at rest if the electron is reflected . we solve eq . ( [ eq - schr2d ] ) numerically considering as initial wave function the products of two gaussian wave packets , i.e. @xmath91 . all details of this simulation can be found in @xcite . in particular we are considering @xmath92 . in fig . [ conditional wave function - figure1 ] the numerical solution of the squared modulus of @xmath93 is plotted at four different times . at the initial time @xmath94 , fig . [ conditional wave function - figure1 ] ( a ) , the entire wave function is at the left of the barrier . at a later time @xmath31 the wave function has split up into reflected and transmitted parts due to the barrier , see fig . [ conditional wave function - figure1 ] ( b ) . then , because the electron has not yet arrived at the transmitted charge detector , the wave function has the following form : @xmath95\phi(x_2,t_0).\ ] ] after that , figs . [ conditional wave function - figure1 ] ( c ) and ( d ) , the interaction of the detector with the transmitted part of the wave function appears . for time @xmath96 the transmitted part of the wave function is shifted up in the @xmath90 direction while the reflected part does not move . the interaction with the apparatus thus produces two channels in the configuration space , one corresponding to the electron being transmitted and the other corresponding to the electron being reflected , getting an entangled superposition among the electron and the apparatus . line in ( a ) , ( b ) , ( c ) and ( d ) is the time evolution of the squared modulus of the conditional wave function associated to the trajectory @xmath97 in fig . [ conditional wave function - figure1 ] , i.e. @xmath98 . the @xmath83 line in ( e ) , ( f ) , ( g ) and ( h ) is the squared modulus of the conditional wave function associated to the trajectory @xmath99 in fig . [ conditional wave function - figure1 ] . i.e. @xmath100 . the actual detector position @xmath101 is plotted at each time in order to compare these results with those in fig . [ conditional wave function - figure1 ] ( color figure online ) ] in fig . [ conditional wave function - figure1 ] we also plot the actual positions of the system and detector @xmath102 for four different possible initial positions @xmath103 , corresponding ( say ) to four distinct runs of the experiment ( labelled by @xmath104 ) . of the four possible evolutions shown , three show the electron being transmitting ( @xmath105 ) and one being reflecting ( @xmath89 ) . while the pointer position @xmath101 does not move for the reflected particle , its evolution for the transmitted ones clearly shows a movement . in conclusion , looking at the _ detector _ position we can perfectly certify if the particle has been reflected ( @xmath106 and @xmath107 ) or transmitted ( @xmath108 and @xmath109 ) . we hope the reader will realize how trivially we have been able to explain the measurement , using only a _ channelized _ ( unitary ) time - evolution of 2d wave function plus two bohmian trajectories , one for the system and another for the measuring apparatus . once we have solved the complete problem of the measurement in ( 2d ) configuration space , we can describe the same measurement in ( 1d ) physical space with the help of the conditional wave function . the key point illustrated here is that the collapse of the one - particle wave function for the electron , which collapse is of course postulated through the second law seen in sec . [ sec2 ] in ordinary quantum theory , instead arises naturally and automatically in bohmian mechanics . it is simply a consequence of slicing the unitary - evolving ( 2d ) wave function @xmath110 along the ( moving ) line @xmath111 , resulting @xmath112 . in fig . [ conditional wave function - figure1 ] we have plotted two solid horizontal lines corresponding to a slice of the wave function at two different values of @xmath101 . in fig . [ conditional wave function - figure2 ] we report the evolution of these ( time - dependent ) slices of the many - particle wave function , the _ conditional wave function _ for the electron , for the trajectories @xmath97 and @xmath99 from fig . [ conditional wave function - figure1 ] . we clearly see that if the particle is reflected , as it is the case for @xmath97 , the position of the pointer does not change with time and , after the interaction with the detector has been performed , the electron s _ conditional wave function _ includes only a reflected part . [ conditional wave function - figure2 ] ( c ) and ( d ) . on the other hand , when the particle is transmitted ( e.g. , @xmath113 ) , it is the reflected part of the _ conditional wave function _ which collapses away , leaving only the transmitted packet . [ conditional wave function - figure2 ] ( g ) and ( h ) . note in particular that the 1d evolution of @xmath114 ( the electron s _ conditional wave function _ ) is not unitary , even though the 2d evolution of @xmath110 is . while a wave function formulation of quantum mechanics provides only statistical information about the experimental results , with the help of the bohmian trajectories , we have been able to recover the individual result of each experiment . in fact , for each experiment the pointer of the detector is either moving ( corresponding to a transmitted electron ) or not ( reflected electron ) , while an ensemble of repeated experiments ( where the initial positions of the particles , both the electron @xmath115 and the detector @xmath116 , are selected according to the squared modulus of the wave function at the initial time @xmath117 ) reproduce the same statistical results . thus with the previous numerical example we have reproduced the collapse - behaviour of the wave function of a transmitted ( or reflected ) electron . this allows us to conclude that the same results of standard formalism ( explained in sec . [ sec2 ] ) are obtained within bohmian mechanics ( see @xcite for a formal derivation of the empirical equivalence of the two theories ) . apart from irrelevant technicalities ( related to how we define the measuring apparatus ) the results in fig . [ figure2 ] and fig . [ conditional wave function - figure2 ] are conceptually identical . we emphasize that , the collapse in bohmian theory is naturally derived . such a natural derivation of the collapse behavior demystifies the measurement process ( and the quantum noise ) . we underline that we achieve the non - unitary evolution of the wave function of a measured system simply slicing the enlarged wave function ( which includes the apparatus ) in the configuration space . let us return to the questions posed at the end of sec . [ sec2 ] about the measurement of the total ( conduction or displacement ) current . is this measurement process _ continuous _ or _ instantaneous _ ? does it provide a _ strong _ or a _ weak _ perturbation of the wave function ? the bohmian theory does not provide simple answers to these questions , but it clearly indicates the path . we need to include ( somehow ) the measuring apparatus in the hamiltonian . here , the electrostatic interaction between the electrons in the system and those in the contacts , cables , etc . see a preliminary work in this direction in ref . @xcite . a powerful simulator which uses bohmian mechanics to compute dc and quantum noise is the bitlles simulator @xcite . this simulator allows us to work with a lot of flexibility , being able of introducing any sort of potential , including coulomb correlations and exchange interaction . the next section is dedicated to expose the fundamental ideas of the simulator and an example of the calculation of noise with it . we have previously exposed the main features of bohmian mechanics explaining in a quite trivial way the partition noise in a tunneling barrier . however , as it occurs for all theories , there is a huge step between its general formulation and its practical application . in fact , it happens many times that the practical problem we want to solve is unsolvable both analytically and numerically , and some kind of approximations are required . the paradigmatic example of the need of approximations in quantum theories is the well - known _ many - body problem _ that reminds us that the celebrated schrdinger equation in eq . ( [ tdse ] ) ( or any alternative formulation ) can only be solved exactly for very few ( one , two , three , .. ) degrees of freedom @xcite . in principle we have seen in sec . [ sec3 ] that to reproduce the collapse of the wave function in bohmian mechanics we have to include a suitable interaction with an external apparatus . then we can write down the schrdinger equation for our complete system including all the electrons in the active region of the device plus all the particles composing the detector . but solving numerically this problem is obviously impossible . again , the _ many - body problem _ appears . then we should look for suitable approximations able to reduce the complexity of our problem . let us emphasize that the ( technical ) approximations that we will show , do not alter the general framework we have previously presented . the first kind of approximation regards the inclusion of the apparatus in our simulations . it seems that its inclusion is unavoidable in order to provide the collapse of the wave function . and this is true , but in the particular case of quantum noise in electrical devices , the fact of _ playing _ with ( bohmian ) trajectories will greatly simplify the problem . in appendix [ app - bohm-3 ] , we have reported how any experimental value is calculated in bohmian mechanics . the important thing is that any expectation value of a given operator is simply calculated as a function of the actual particles positions over an ensemble of repeated experiments ( see eq . ( [ mean - value ] ) in appendix [ app - bohm-3 ] ) . thus what really matters in the computation of a property of the quantum system are only the trajectories of the bohmian particles ( not the wave functions ) . therefore , if the trajectories without measuring apparatus are enough accurate ( this means if the error on these trajectories due to neglecting the apparatus is reasonably small compared to the exact solution ) we can get accurate results with a minimal computational effort . in the case of the transmitted charge detector of sec . [ sec3 ] , it has been demonstrated @xcite that the error due to the exclusion of the apparatus from the simulations is almost negligible for the computation of the trajectories . in this way we can decrease enormously the computational burden , removing all the degrees of freedom related to the apparatus from our computations . we can provide a more didactic discussion on why the previous technical approximation for the measuring apparatus works quite well when using bohmian trajectories . in sec . [ sec2 ] , we conclude that the reason why the wave function evolution in fig . [ figure1 ] was wrong is due to the wrong possibility that an electron that is transmitted at time @xmath23 is later reflected at time @xmath43 . this unphysical result simply disappears when using bohmian trajectories : the dynamic of a transmitted electron at time @xmath23 will be determined _ locally _ by the guidance law eq . ( [ eq - guidance ] ) that only takes into account the transmitted part of the wave function . we can , for all practical purposes , completely ignore the reflected part of the wave function . therefore , at time @xmath43 , this electron will remain as a transmitted electron with full certainty . + finally , let us emphasize that , in principle , the measuring apparatus has also a role in the classical simulation of electronic devices . such interaction with the apparatus is included at a classical level , at best , by a proper boundary conditions for the scalar potential of the hamiltonian ( i.e. the poisson equation ) ensuring overall charge neutrality . obviously , this kind of approximation can also be included here . once we have _ practically _ eliminated the apparatus from our computations , a second kind of approximation regards the interactions among the electrons of our device . the active region of the electronic device can contain hundreds of electrons . also in this case , as we mentioned , the many - particle schrdinger equation can be solved only for very few degrees of freedom . a standard way to proceed consists then on reducing the complexity of the problem by _ tracing out _ certain degrees of freedom . this process ends up with what is called the _ reduced density matrix_. when the reduced density matrix is used , its equation of motion is no longer described by the schrdinger equation but in general by a non - unitary operator . the reduced density matrix is no longer a pure state , but a mixture of states and their evolution is in general irreversible @xcite . now we discuss how bohmian mechanics allows us to proceed in a very different way . as it will be seen below , the concept of _ conditional wave function _ @xcite provides an original tool to deal with many - body open quantum systems @xcite . + as said , once again the key instrument is the conditional wave function . in order to use the conditional wave function to reduce the degrees of freedom of a system we must know how it evolves in time . it can be demonstrated @xcite that the single - particle conditional wave function of particle @xmath118 , @xmath119 , for a system of @xmath49 interacting particles , obeys the following wave equation : @xmath120 the explicit expression of the potentials @xmath121 and @xmath122 that appears in eq . ( [ conditional_eq ] ) can be found in reference @xcite . however , their numerical values are in principle unknown and need some educated guesses . on the other hand , the total electrostatic potential energy among the @xmath49 electrons that appears in eq . ( [ tdse ] ) , has been divided into two parts : @xmath123 the term @xmath124 can be any type of many - particle potential defined in the position - representation , in particular it can include short - range and long - range coulomb interactions . the other term @xmath125 in eq . ( [ potential_energy ] ) without dependence on @xmath69 , is contained in the coupling potential @xmath126 in eq . ( [ conditional_eq ] ) . the same procedure can be done for the rest of the @xmath127 particles , for example for particle @xmath128 we fix the positions of particle @xmath129 obtaining the analogous of eq . ( [ conditional_eq ] ) for @xmath130 . from a practical point of view , all quantum trajectories @xmath66 have to be computed simultaneously . in order to gather all the above concepts , let us discuss a practical computation with conditional wave functions by detailing a sequential procedure : 1 . at the initial time @xmath131 , we fix the initial position of all @xmath55-particles , @xmath68 , according to the initial probability distribution ( @xmath132 ) , and their associated single - particle wave function @xmath133 . 2 . from all particle positions , we compute the exact value of the potential @xmath134 for each particle . an approximation for the terms @xmath135 and @xmath136 is required at this point . we use the simplest one @xcite . 3 . we then solve each single - particle schrdinger - type equation , eq . ( [ conditional_eq ] ) , from @xmath131 till @xmath137 . 4 . from the knowledge of the single - particle wave function @xmath138 , we can compute the new velocities @xmath139 for each @xmath140-particle ( see eq . ( [ eq - guid - cwf ] ) in appendix [ app - bohm-2 ] ) . 5 . with the previous velocity , we compute the new position of each @xmath55-particle as @xmath141 . 6 . finally , with the set of new positions and wave functions , we repeat the whole procedure ( steps 1 till 5 ) for another infinitesimal time @xmath142 till the total simulation time is finished . the advantage of the above algorithm using eq . ( [ conditional_eq ] ) instead of the many - particle schrdinger equation ( eq . ( [ tdse ] ) ) is that , in order to find approximate trajectories , @xmath143 , we do not need to evaluate the wave function and potential energies in the whole configuration space , but only over a smaller number of configuration points , @xmath144 , associated with those trajectories defining the highest probabilities according to @xmath145 . + the exchange interaction is naturally included in eq . ( [ conditional_eq ] ) through the terms @xmath135 and @xmath146 . due to the pauli exclusion principle , the modulus of the wave function tends to zero , @xmath147 , in any neighborhood of @xmath148 such that @xmath149 . thus , both terms , @xmath150 and @xmath151 , have asymptotes at @xmath152 that _ repel _ the @xmath153 particle from other electrons . however , in order to exactly compute the terms @xmath154 and @xmath146 we must know the total wave function , which is in principle unknown . there are however a few ways to introduce the symmetry of the wave function without dealing directly with these two coupling terms @xcite . clearly , the complexity of the algorithm increases as we go beyond the single - particle quantum transport scenario mentioned in sec . [ sec2 ] and sec . [ sec3 ] . an electron device is an open system , where many parameters can only be estimated from the knowledge of their statistical ( typical ) distribution . apart from the uncertainty in the initial position in the quantum trajectories ( the @xmath155 distribution explained in appendix [ app - bohm ] ) , we also take into account the uncertainty on the properties of the injected electrons ( initial energies , momentums , etc ) which we refer to the parameter @xmath156 . the random process @xmath22 mentioned in sec . [ sec1.1 ] is now written as @xmath157 . at finite temperature , the thermal noise introduces fluctuations on the energies of the electrons entering inside the device . as discussed in the introduction of sec . [ sec2 ] , the study of the noise in electrical devices due to the partition noise of the barrier plus the thermal noise of the injection are traditionally known as quantum shot noise @xcite . this is the noise studied in this sec . [ sec4 ] . in many systems , one obtains the well known schottky s result @xcite or poissonian shot noise , @xmath158 , for the noise power spectral density defined in eq . ( [ noise5 ] ) at zero frequency , i.e. @xmath159 . + we select a particular ( large ) set of wave packets with values @xmath155 and @xmath156 for selecting their initial conditions . we refer to such selection as @xmath160 and @xmath161 . we evolve the wave packets and trajectories as explained in previous paragraphs . within the approximation mentioned in sec . [ sec4.1 ] and sec . [ sec4.2 ] , the total current value can be calculated as the sum of the particle or conduction current plus the displacement current : @xmath162 where @xmath163 is the surface where we want to calculate the current , @xmath164 are the points of the chosen surface , @xmath165 is the dielectric constant in the same surface and @xmath166 is the electric field in the surface @xmath163 which depends on the actual position of all the electrons . once we know @xmath167 for a large interval of time , the algorithm to compute the current fluctuations is quite simple following eq . ( [ noise3 ] ) and eq . ( [ noise5 ] ) . this discussion can be familiar for those people who works in semi - classical approaches . in fact , the bohmian procedure explained here for quantum transport is very similar to that of , for instance , the monte - carlo simulations of the boltzmann equation . but instead of being the electric - field the one who _ guides _ the electrons , it is the wave function , through the guiding velocity field in eq . ( [ eq - guidance ] ) . computed analytically and numerically ( color figure online ) ] as a practical example of the computation of the fluctuations , we show here the current response to a step input voltage in the negative differential conductance region of a rtd . the input signal is the step voltage @xmath168 $ ] where @xmath169 is the heaviside ( step ) function . the voltages @xmath170 and @xmath171 are constant . then the current response can be expressed as @xmath172 $ ] where @xmath173 and @xmath174 are the stationary currents corresponding to @xmath170 and @xmath171 respectively , and @xmath175 is the _ intrinsic _ transient current . the results are reported in fig . [ f6_alarcon : figure6_9a ] where @xmath176 manifests a delay with respect to the step input voltage , due to the dynamical adjustment of the electric field in the conductors . after this delay , the current response becomes a rlc - like response ( dot - dashed line rlc response @xmath128 ) i.e. purely exponential . performing the fourier transform of @xmath176 in fig . [ f6_alarcon : figure6_9b ] and comparing it with the single pole spectra ( fourier transform of the rlc - like responses , dashed and dashed dotted lines ) , we are able to estimate the cut - off frequency and the frequency offset due to the delay @xcite . of fig . [ f6_alarcon : figure6_9a ] . a logarithmic scale is used to resolve the cut - off frequency offset ( color figure online ) ] in order to understand how the many - body coulomb interaction affects the noise in rtds , we also investigate the correlation between an electron trapped in the resonant state during a dwell time @xmath177 and those remaining in the left reservoir . this correlation occurs essentially because the trapped electron perturbs the potential energy felt by the electrons in the reservoir . in the limit of non - interacting electrons , the fano factor will be essentially proportional to the partition noise , however , if the dynamical coulomb correlations are included in the simulations ( see fig . [ f6_alarcon : figure6_10 ] ) this result is no longer reached , it shows super - poissonian values . finally , we are also interested in the high frequency spectrum @xmath5 given by eq . ( [ noise5 ] ) revealing information about the internal energy scales of the rtd that is not available from dc transport ( see fig . [ f6_alarcon : figure6_12 ] ) . defined as @xmath178 , evaluated using the current fluctuations directly available from the bitlles simulator ( color figure online ) ] in the present paper we discuss the understanding and the computation of quantum noise in electrical devices from a bohmian perspective . computations of quantum noise are quite complex because noise is generally quantified in terms of temporal correlations . such correlations must include the time - evolution of a quantum system during and after a measurement . usually , many other quantum computations do only require a final measurement , so that their time - evolution from the initial until the final time is uniquely determined by the unitary ( schrdinger like ) evolution . as discussed in fig . [ figure2 ] , this unitary evolution is not enough to compute time correlations which require mixing unitary and non - unitary ( the so - called collapse of the wave function ) time evolutions . there are several ( empirically equivalent ) quantum theories . each quantum theory has its own formalism that is able to connect the experimental values with some abstract elements such as wave functions , operators , trajectories , etc . that are able to satisfactorily reproduce ( or predict ) experimental results . we discuss how the copenhagen ( also known as standard or orthodox ) interpretation and bohmian mechanics give explanation to the partition noise . for a flux of electrons impinging upon a tunneling barrier , we analyze how a measurement process affects partition noise in a quantum device . for simplicity , to focus on the importance of the measurement process , we consider spinless electrons without coulomb and exchange interaction . in sec . [ sec2 ] we explain how standard quantum theory provides an answer for the measurement problem by means of the introduction of the notion of operators . we see that this notion is not always satisfactory even for practical purposes , because the definition of which is the _ right _ operator is not obvious . then in sec . [ sec3 ] we discuss an alternative way to deal with the collapse without introducing the idea of operators . in fact within bohmian mechanics , a theory of wave _ and _ particles , the collapse is derived trivially by means of the introduction of the _ conditional wave function _ ( the wave function of a subsystem ) , a tool exclusively belonging to bohm s theory . obviously , each theory gives a different formalism to compute quantum noise and different interpretation of its origin . in any case , at the end of the day , the same empirical predictions are achieved by using the orthodox quantum theory or bohmian mechanics . in sec . [ sec4 ] , because the bohmian formulation uses trajectories to compute experimental results , we see that a very reasonable approximation to include collapse can be achieved with a very small computational effort . finally , details of the simulator named bitlles based on bohmian mechanics and numerical results for low and high frequency noise of the current in a resonant tunneling diode are presented . we emphasize that the presented formalism and the procedure for computing the properties of a system ( in our case current , noise , etc . ) have many similarities with the one used in semi - classical simulations ( for example monte - carlo of the boltzmann equation @xcite ) . in any case , bohmian formalism is not at all a semi - classical approach but a complete quantum theory that can be applied to study any non - relativistic quantum phenomena , quantum noise and collapse among them . finally , we wish to discuss which is the ultimate origin of the quantum noise according to orthodox and bohmian interpretations . before entering into details , let us recall that the definition of noise given in eq . ( [ noise2 ] ) in sec . [ sec1 ] is just the difference between the experimental value @xmath1 and what we define as the signal . therefore , even a sinusoidal current @xmath1 provides a value of @xmath33 different from zero . what we want to discuss hereafter are not all the possible sources of fluctuations in @xmath1 , but only if there is any new type of randomness in quantum devices that is not present in classical ones and what is its origin . given this last specification , we can provide an answer to the question : _ what is the ultimate origin of quantum noise according to the orthodox interpretation ? _ as we see in sec . [ sec2 ] , the transmission or reflection of a single electron impinging upon a tunneling barrier becomes unpredictable . this is an example of a new source of randomness present only in quantum devices . according to the orthodox theory , this randomness appears because of the collapse of the wave function due to measurements . without the collapse ( that is put in by hand as an additional postulate in the orthodox theory ) , the wave function follows a deterministic law dictated by the schrdinger equation . the partition noise in the tunneling barrier discussed in sec . [ sec2 ] is due to the action of the operator which implements the random collapse of the wave function ( selecting the final wave function stochastically among the set of available eigenstates ) . alternatively , we can also answer the question : _ what is the ultimate origin of quantum noise according to bohmian mechanics ? _ the randomness in the values of the current in @xmath1 provided by the bitlles simulator comes from the @xmath156 and @xmath155 distributions mentioned in sec . [ practical ] . the @xmath156 distribution is due to the uncertainty of the initial energies , momentums , entering times , etc . of the electrons . this source of `` extrinsic '' randomness can be minimized imagining technological improvements of the setup ( for example , well - controlled single electron sources ) . on the contrary , the @xmath155 distribution of the conditional wave function ( explained in appendix [ app - bohm ] ) is an unavoidable source of randomness . this randomness of the @xmath155 distribution can not be minimized by any technological improvement @xcite . therefore , whether the particle is transmitted or reflected becomes unpredictable in bohmian theory too . thus , though bohmian mechanics is deterministic , an _ appearance _ of randomness emerges in the subsystems @xcite . it is important to notice that the measurement of the system does not introduce any additional randomness . the ultimate origin of the unpredictability is the fact that the uncertainty principle does not allow us to known the ( well - defined in the bohmian theory ) initial position of the particles in each experiment . in summary , according to the orthodox interpretation , the partition noise has its origin in the stochasticity of the orthodox measurement process . on the contrary , bohmian mechanics says that the origin of noise is the uncertainty of the initial position of the trajectory in each realization of the experiment . although both theories give the same predictions , in the authors opinion , the latter has a more natural , common and understandable explanation of the origin of quantum noise . while following deterministic laws , the transmission or reflection of a bohmian electron is unpredictable in a given experiment . a classical dice is a simpler example of a system following deterministic laws that becomes unpredictable . collapse in bohmian mechanics is so naturally derived that the quantum measurement problem , in general , and quantum noise , in particular , are somehow demystified . we underline that bohmian mechanics achieves the non - unitary evolution of the wave function of a measured system simply slicing the enlarged wave function in the configuration space ( without introducing any measurement - associated randomness ) . we accept that preferences between the explanation of the origin of quantum noise in terms of the orthodox or bohmian interpretations are subjective . therefore , in this paper we have also developed objective arguments about the computational advantages of the bohmian formalism . the facts that the measuring apparatus , what we call the ammeter , is directly included into the hamiltonian of the schrdinger equation and that the current values are computed from trajectories ( not from the wave functions ) allow us to study system plus apparatus scenarios ( or look for reasonable approximations ) . this ability is very relevant , for example , in the computation of high frequency currents where it is difficult to find the _ right _ operator . for all these reasons , we conclude that quantum noise is easily understood and computed from a bohmian perspective in many practical scenarios . we want to acknowledge t. norsen , n. zangh , g. albareda and f.l . traversa for insightful discussions.this work has been partially supported by the ministerio de ciencia e innovacin through the spanish project tec2012 - 31330 and by the grant agreement no : 604391 of the flagship initiative graphene - based revolutions in ict and beyond . z. z acknowledges financial support from the china scholarship council ( csc ) . is supported in part by infn and acknowledges the support of cost action ( mp1006 ) through stsm . the dc current measured in a laboratory @xmath0 can be computed by time - averaging the measured value of the total current @xmath1 from a _ unique device _ during a large ( ideally infinite ) period of time @xmath2 as mentioned in eq . ( [ noise1 ] ) . if we can justify the ergodicity of electronic devices , we can alternatively compute @xmath0 from an ensemble - average of all possible values of the current @xmath30 measured , at one particular time @xmath19 , over an _ ensemble of ( identical ) devices _ as seen in eq . ( [ noise7 ] ) . for dc quantum transport computations , eq . ( [ noise7 ] ) is greatly preferred because it deals directly with the probabilistic interpretation of the wave function . it is important to realize that while eq . ( [ noise1 ] ) implies measuring the quantum current many times , eq . ( [ noise7 ] ) involves only one measurement . we do not need to worry about the evolution of the wave function after the measurement when using eq . ( [ noise7 ] ) . let us discuss this point in more detail . we can define the eigenstates @xmath179 of a particular operator @xmath180 , as those vectors that satisfy the equation @xmath181 . the eigenvalue @xmath30 is one of the @xmath79 possible measured values in eq . ( [ noise7 ] ) . since the entire set of eigenstates form a basis of the hilbert space , the wave function can be decomposed as @xmath182 , with @xmath183 . then , we can rediscover eq . ( [ noise7 ] ) as follow : @xmath184 where we have used the orthonormal property of the eigenstates @xmath185 and the definition of the ( born ) probability @xmath186 . we emphasize that @xmath187 does not require the explicit knowledge of the eigenstates . only the free evolution of the state @xmath188 and the _ measuring _ operator @xmath180 are needed . at this point , it is mandatory to provide some discussion about the use of the ergodic theorem . strictly speaking , no ergodic theorem exists for an out of equilibrium system @xcite . indeed , an out of equilibrium system is represented by a distribution function , or probability function , that is different from that in equilibrium and arises from a balance between the driving forces and the dissipative forces . the applied bias used to measure the dc current of any device implies that the device is quite likely in a far from equilibrium state . therefore , the ergodic connection between eq . ( [ noise1 ] ) and eq . ( [ noise7 ] ) has to be considered as only a very reasonable approximation for dc transport , but not as an exact result @xcite . bohmian mechanics is a version of quantum theory whose basic elements are waves and point - like particles . the many - particle wave function evolves according to the schrdinger equation ( [ tdse ] ) while particles have definite position at any time with a law given by eq . ( [ eq - guidance ] ) , therefore being a fully deterministic theory . the configuration of the particles , say at time @xmath131 , is chosen randomly according to @xmath189 at the initial time , known as _ quantum equilibrium hypothesis _ @xcite . thanks to the continuity equation where @xmath191 and @xmath192 the bohmian vector field . an important consequence of the quantum equilibrium hypothesis and equivariance is the empirical equivalence between bohmian mechanics and orthodox quantum theory for any kind of non - relativistic quantum experiments . consider a quantum system of @xmath49 particles and a partition of it in such a way that its spatial coordinates can be split as @xmath193 . where we denote with @xmath194 the position in @xmath195 space of the electron @xmath118 , while with @xmath196 the positions of the rest of the electrons in a @xmath197 space . the actual particle trajectories are accordingly denoted by @xmath198 . _ how can one assign a wave function to the electron @xmath118 ? _ in general this is not possible if the two subsystems are entangled , i.e. the total wave function can not be written as a product @xmath199 . however , we can modify our question and ask what is the wave function of the electron @xmath118 that provides the exact velocity @xmath200 given a particular configuration @xmath201 for the rest of the particles . the answer given by bohmian mechanics is the so called _ conditional wave function _ @xcite : @xmath202 which constitutes a slice of the whole multi - dimensional wave function . the wave function constructed in such a way gives exactly the same bohmian velocity if needed , bohmian mechanics can make use of operators , but only as a mathematical trick . without any physical or fundamental role in the operator . we briefly explain how it is possible to calculate the mean value of a general hermitian operator with bohmian trajectories . the _ quantum equilibrium hypothesis _ at the initial time @xmath131 can be expressed in terms of the trajectories as follows where the superindex @xmath155 takes into account the uncertainty in the initial position of the particles . it can be easily demonstrated @xcite that the evolution of the above infinite set of quantum trajectories @xmath205 reproduce at any time @xmath19 the probability distribution , @xmath65 . + for computing the mean value of an operator @xmath57 it can be demonstrated @xcite that de jong , m.j.m . , beenakker , c.w.j . : shot noise in mesoscopic systems . in : sohn , l.l . , kouwenhoven , l.p . , schn , g. ( eds . ) nato asi series e vol . 225 - 262 . kluwer academic publishing , dordrecht ( 1997 ) marian , d. , zangh , n. , oriols , x. : on the back - action of thz measurement on the total current of quantum devices . computational electronics ( iwce ) , 2014 international workshop on ( 2014 ) . doi : 10.1109/iwce.2014.6865844 traversa , f.l . , et al . : time - dependent many - particle simulation for resonant tunneling diodes : interpretation of an analytical small - signal equivalent circuit . ieee trans . * 58 * , 2104 - 2112 ( 2011 ) p. a. m. dirac , wrote in 1929 : _ `` the general theory of quantum mechanics is now almost complete . the underlying physical laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known , and the difficulty is only that the exact application of these laws leads to equations much too complicated to be soluble . '' _ @xcite alarcn , a. , cartoix , x. , oriols , x. : towards the explicit computation of bohm velocities associated to n - electron wavefunctions with arbitrary spin - orientations . status solidi . * 7 * , 2636 - 2639 ( 2010 ) alarcn , a. , yaro , s. , cartoix , x. , oriols , x. : computation of many - particle quantum trajectories with exchange interaction : application to the simulation of nanoelectronic devices . j. phys . : condens . matter . * 25 * , 325601 ( 2013 ) cavalcanti , e.g. , wiseman , h.m . : bell nonlocality , signal locality and unpredictability ( or what bohr could have told einstein at solvay had he known about bell experiments ) . * 42 * , 13291338 ( 2012 )

we consider the assouad dimensions of orthogonal projections of planar sets onto lines . our investigation covers both general and self - similar sets . for general sets , the main result is the following : if a set in the plane has assouad dimension $ ] , then the projections have assouad dimension at least almost surely . compared to the famous analogue for hausdorff dimension namely _ marstrand s projection theorem _ a striking difference is that the words ` at least ' can not be dispensed with : in fact , for many planar self - similar sets of dimension , we prove that the assouad dimension of projections can attain both values and for a set of directions of positive measure . for self - similar sets , our investigation splits naturally into two cases : when the group of rotations is discrete , and when it is dense . in the ` discrete rotations ' case we prove the following dichotomy for any given projection : either the hausdorff measure is positive in the hausdorff dimension , in which case the hausdorff and assouad dimensions coincide ; or the hausdorff measure is zero in the hausdorff dimension , in which case the assouad dimension is equal to 1 . in the ` dense rotations ' case we prove that every projection has assouad dimension equal to one , assuming that the planar set is not a singleton . as another application of our results , we show that there is no _ falconer s theorem _ for assouad dimension . more precisely , the assouad dimension of a self - similar ( or self - affine ) set is not in general almost surely constant when one randomises the translation vectors . + + _ mathematics subject classification _ 2010 : 28a80 , 28a78 . + _ key words and phrases _ : assouad dimension , projection , self - similar set .

1509.01128

one of the most fundamental dimension theoretic questions in geometric measure theory is : _ how does dimension behave under orthogonal projection _ ? this line of research began with the seminal paper of marstrand from 1954 @xcite , influenced by some earlier work of besicovitch . one version of marstrand s projection theorem states that if @xmath5 is an analytic set with hausdorff dimension @xmath6 $ ] , then the hausdorff dimension of the orthogonal projection of @xmath7 in almost every direction is @xmath8 , which is as big as it can be . here ` almost every ' refers to lebesgue measure on the interval @xmath9 , with projections parameterised in the obvious way . for the purposes of this paper , the important thing about this projection theorem is that , no matter what analytic set @xmath5 one considers , the hausdorff dimension of the projections @xmath10 is _ almost surely constant_. other important notions of dimension include the packing dimension and upper and lower box - counting dimension , see @xcite . the behaviour of these dimensions under projection is rather more subtle than for the hausdorff dimension , as first demonstrated by jrvenp @xcite . however , in the mid-1990s it became clear that each of these dimensions is at least _ almost surely constant _ under projection , in the same sense that the hausdorff dimension is . the precise value of the constant is more complicated than simply @xmath8 , but can be stated in terms of _ dimension profiles _ , see @xcite . there are also natural higher dimensional analogues of these projection results . in particular , the higher dimensional analogue of marstrand s theorem , where one considers projections of analytic sets @xmath11 to @xmath12-planes , was proved by mattila in 1975 @xcite . in this setting the almost sure value of the hausdorff dimension is given by @xmath13 and ` almost sure ' refers to the natural invariant measure on the grassmannian manifold @xmath14 , which consists of all @xmath12 dimensional subspaces of @xmath15 . for more information on the rich and fascinating topic of projections of fractal sets and measures , see the recent survey papers @xcite and the references therein . the assouad dimension is another notion of dimension , which has been very useful as a tool in several disparate areas of mathematics . in recent years it has been gaining more attention in the setting of fractal geometry and geometric measure theory . as such , it is natural to consider the fundamental geometric properties of the assouad dimension , such as its behaviour under orthogonal projections . we recall the definition here , but refer the reader to @xcite for more details . for any non - empty subset @xmath16 and @xmath17 , let @xmath18 be the smallest number of open sets with diameter less than or equal to @xmath19 required to cover @xmath20 . the _ assouad dimension _ of a non - empty set @xmath21 is then given by @xmath22 where @xmath23 denotes the open euclidean ball centred at @xmath24 with radius @xmath25 . it is well - known that the assouad dimension is always an upper bound for the hausdorff dimension . in this paper we prove that , unlike the hausdorff , packing and box dimensions discussed above , the assouad dimension of orthogonal projections of a compact set @xmath26 need _ not _ be almost surely constant as a function of the projection angle . however , in analogy with the result for hausdorff dimension , the essential infimum of the said function is at least @xmath27 ; even if @xmath7 is not compact , or even analytic . these results are discussed in section [ maingeneral ] . we establish the non - constancy result via a detailed study of the assouad dimensions of projections of planar self - similar sets . the hausdorff dimension of projections of self - similar sets has attracted a lot of attention in recent years , see section [ sssection ] , and thus it is natural to consider the analogous questions for assouad dimension . our results for self - similar sets are discussed in detail in section [ mainresults ] , and the applications concerning non - constancy and falconer s theorem ( mentioned in the abstract ) will be presented in sections [ nomarstrand ] and [ nofalconer ] respectively . on route to proving our main result for self - similar sets , we obtain new information about the assouad dimension of graph - directed self - similar sets in the line with overlaps , theorem [ gdassouad ] , which is the natural extension of ( * ? ? ? * theorem 1.3 ) to the graph - directed setting and is of independent interest . many questions remain unanswered by the results in this paper , and we pose some of them in sections [ maingeneral ] through [ selfaffinesets ] . in particular , very little is known for self - affine sets , and in higher dimensions . self - similar sets are arguably the most fundamental class of fractal set and have been studied extensively , see @xcite . let @xmath28 be a finite collection of contracting similarities mapping @xmath29^d$ ] into itself . by _ , we mean that for each @xmath30 , there exists a similarity ratio @xmath31 such that for all @xmath32^d$ ] we have @xmath33 which means that the contractions @xmath34 scale uniformly by @xmath35 in every direction . as such we may decompose each @xmath34 uniquely as @xmath36 where @xmath37 is a @xmath38 orthogonal matrix and @xmath39 is a translation . a fundamental result of hutchinson @xcite states that there is a unique attractor of the iterated function system ( ifs ) @xmath28 , that is , a unique non - empty compact set @xmath40^d$ ] satisfying @xmath41 the set @xmath7 is called self - similar and often has a rich fractal structure . a self - similar set satisfies the open set condition ( osc ) ( for a given ifs defining it ) if there exists a non - empty open set @xmath42^d$ ] such that @xmath43 with the sets @xmath44 pairwise disjoint . if @xmath7 satisfies the osc , then its hausdorff and assouad dimensions are given by @xmath45 where @xmath3 is the _ similarity dimension _ given by solving the equation @xmath46 if the osc is not satisfied , then the dimensions are more difficult to get hold of and indeed the assouad and hausdorff dimensions may be distinct @xcite . one still expects the hausdorff dimension to be given by @xmath45 , unless there is a good reason for it not to be , such as exact overlaps in the construction , see ( * ? ? ? * question 2.6 ) . recent major advances were made in this area by hochman @xcite . there has also been intense interest in the dimension theory of the projections of self - similar sets in recent years , see the survey @xcite . in particular , we have the following theorem : [ hausdorffprojection ] let @xmath47^d$ ] be a self - similar set containing at least two points , suppose that the group generated by @xmath48 is dense in @xmath49 or @xmath50 and fix @xmath51 less than @xmath52 . then @xmath53 for _ * all * _ @xmath54 . the particular interest of this result is that ` dense rotations ' guarantees that there are _ no _ exceptions to marstrand s theorem . this result is essentially due to hochman and shmerkin s breakthrough work @xcite , although they stated the result assuming some separation conditions . these conditions were explicitly removed by farkas @xcite and falconer - jin @xcite . the result in the planar case was obtained earlier by peres and shmerkin @xcite . we parameterise orthogonal projections onto lines in @xmath55 by @xmath56 in the natural way , by letting @xmath57 be the projection onto the line @xmath58 passing through the origin and forming an angle @xmath59 with the positive @xmath24-axis . with this notation , our main result for general sets is the following : [ maintheoremforgeneralsets ] assume that @xmath26 . then , for almost all @xmath56 , @xmath60 theorem [ maintheoremforgeneralsets ] will be proved in section [ proofforgeneralsets ] . the lower bound is obviously sharp , and in section [ nomarstrand ] we demonstrate by example that the inequality can not be replaced by an equality : the function @xmath61 need not be almost surely constant , and it can attain both the values @xmath62 and @xmath4 for a set of @xmath59 s with positive measure . we do not know if other values are possible here , or if there can be three distinct values : given a set @xmath26 , how many distinct values can the assouad dimension of @xmath63 assume for a set of @xmath59 s with positive measure ? if there are only two such values , are they always @xmath62 and @xmath4 ? in this section we state our main results for self - similar sets , which are rather complete in the context of assouad dimension . in the @xmath64-dimensional setting the groups @xmath65 and @xmath66 are particularly simple : @xmath65 consists of counterclockwise rotations by angles @xmath67 and the corresponding reflections with orientation reversed and @xmath66 just consists of the rotations . as such , the group generated by @xmath48 will be dense if and only if one of the @xmath68 rotates by an irrational multiple of @xmath69 and otherwise it will be discrete ( in fact finite ) . here is the main result for self - similar sets : [ assouadprojection ] let @xmath70 be self - similar set containing at least two points , and first suppose that the group generated by @xmath48 is discrete . then , for a given @xmath56 , we have : 1 . if @xmath71 , then @xmath72 2 . if @xmath73 , then @xmath74 . secondly , suppose that the group generated by @xmath48 is dense in @xmath65 or @xmath66 . then @xmath75 for * all * @xmath56 . the discrete rotations case of theorem [ assouadprojection ] will be proved in section [ discreterotationsproof ] and the dense rotations case will be proved in section [ denserotationsproof ] . we emphasise that we assume no separation conditions for @xmath7 , in particular the osc may fail and the hausdorff dimension of @xmath7 may be strictly smaller than the similarity dimension . by marstrand s theorem , we have that , for almost all @xmath76 , @xmath77 , but farkas ( * ? ? ? * theorem 1.2 ) showed that in the discrete rotations case there is always at least one direction @xmath76 where the dimension drops , i.e. @xmath78 , provided the hausdorff dimension of @xmath7 is given by the similarity dimension and this value is less than or equal to 1 . in the dense rotations case erolu @xcite and farkas ( * theorem 1.5 ) proved that @xmath79 for all @xmath56 . so , the dichotomy seen in the ` discrete rotations ' part persists for dense rotations , but case _ 1 . _ never occurs . in the dense rotations case , the assouad dimension of @xmath63 is constant and independent of the dimension of @xmath7 . neither of these phenomena are generally manifest in the discrete case , but one can always find _ at least one _ direction @xmath59 for which the assouad dimension of @xmath63 attains the maximal value of 1 , independent of the dimensions of @xmath7 , provided @xmath7 is not contained in a line . [ assouadprojectionexamples ] let @xmath47 ^ 2 $ ] be a self - similar set , which is not contained in a line . then there exists @xmath56 such that @xmath80 we will prove theorem [ assouadprojectionexamples ] in section [ assouadprojectionexamplesproof ] . our key application of theorem [ assouadprojection ] is that , unlike the hausdorff , upper and lower box , and packing dimensions , the assouad dimensions of orthogonal projections of a ( compact ) set are not almost surely constant in general . thus , we do not have a ` marstrand s projection theorem for assouad dimension ' . [ mainex ] for any @xmath3 satisfying @xmath81 , there exists a compact set @xmath5 with hausdorff and assouad dimension equal to @xmath3 for which there are two non - empty disjoint intervals @xmath82 such that @xmath83 in particular , @xmath84 is not an almost surely constant function . the restriction to @xmath85 in theorem [ mainex ] may well be an artefact of our method , but at present we are unaware how to construct lower - dimensional examples : given any @xmath86 , in particular @xmath87 , can one construct a compact planar set with hausdorff dimension @xmath3 , for which the assouad dimension of the projections is not almost surely constant ? also , we do not know if the words ` almost all @xmath88 ' could be strengthened to ` all @xmath88 ' with a different construction : can one construct a compact planar set for which the assouad dimension of the projection takes different values on two sets with non - empty interior ? note that theorem [ mainex ] demonstrates the apparently strange property that assouad dimension can increase under projection ( a lipschitz map ) , which can not happen for the dimensions discussed in section 1 . this peculiarity of the assouad dimension was observed previously in ( * ? ? ? * section 3.1 ) . the rest of this section will be dedicated to constructing an example with the properties required by theorem [ mainex ] . the set @xmath7 will actually be very simple : it will be a self - similar modification of the sierpiski triangle , where the contraction ratios are equal to @xmath89 , see figure 1 . fix @xmath89 and let @xmath90 be the self - similar attractor of the ifs on @xmath29 ^ 2 $ ] given by @xmath91 . observe that @xmath90 satisfies the _ open set condition _ ( osc ) and so @xmath92 . it follows from theorem [ assouadprojection ] that 1 . @xmath93 if and only if @xmath94 2 . @xmath95 if and only if @xmath96 . in light of this dichotomy , and the fact that @xmath97 almost surely by marstrand s theorem , in order to complete the proof it is sufficient to show that there is a non - empty open interval of @xmath59 s for which @xmath98 , and a non - empty open interval of @xmath59 s within which @xmath99 almost surely . the first of these problems is straightforward , because one can easily find an open interval where the induced ifs defining the self - similar set @xmath63 satisfies the open set condition and it then follows from standard results that @xmath100 and @xmath98 , see ( * ? ? ? * chapter 9 ) . the existence of such an interval relies on the assumption @xmath101 . the second problem is more delicate , but fortunately has already been solved by peres , simon and solomyak @xcite . they defined the set of _ intersection parameters _ @xmath102 and proved that @xmath103 contains a non - empty interval provided @xmath89 and , moreover , for almost every @xmath104 we have @xmath105 . this can be found in ( * ? ? ? * theorem 1.2(i ) and example 2.8 ) . we could also have used the 4-corner cantor set with contraction parameter in the interval @xmath106 , which was discussed in @xcite , but chose the sierpiski triangle because it yielded the least restrictive conditions on the dimension of @xmath7 . ( with @xmath107 ) and two typical projections . there is a small interval of projections @xmath108 for which the 3 pieces of @xmath90 project into pairwise disjoint intervals , meaning that projections in these directions satisfy the open set condition . there is also an interval @xmath109 within which the projection map is not injective . , width=340 ] in 1988 falconer proved a seminal result in the dimension theory of self - affine sets , see @xcite . self - affine sets are closely related to self - similar sets , but the contractions in the defining ifs can be affine , i.e. @xmath110 can be replaced with any contracting invertible @xmath38 matrix . this means that the scaled copies of the attractor can scale by different amounts in different directions , as well as being skewed or sheared , which makes them much more difficult to study . even if the osc is satisfied , the dimensions may be hard to compute and the assouad , box and hausdorff dimensions may all be distinct . despite this , falconer proved that the hausdorff and box dimensions are ` generically ' equal to the _ affinity dimension _ ; the self - affine analogue of the similarity dimension . the affinity dimension depends only on the linear parts of the defining maps ( as with the similarity dimension ) and falconer proved that for lebesgue almost all choices of translation vectors @xmath111 , the box and hausdorff dimensions of the corresponding self - affine set are equal to the affinity dimension , provided the spectral norms of the matrices were all strictly less than 1/2 . in fact , falconer s original proof required 1/3 here , but solomyak relaxed this assumption to 1/2 and pointed out that this was optimal @xcite . also the affinity dimension is always an upper bound for the hausdorff and box dimensions , but this is not true for the assouad dimension @xcite . specialising to the case of self - similar subsets of the line ( which is a very restrictive class of self - affine sets ) , the assumption on norms is not required . in particular , we have the following result due to simon and solomyak @xcite : [ ssfalconer ] fix a set @xmath112 with each @xmath113 and let @xmath3 be the corresponding similarity ( affinity ) dimension given by @xmath114 for a given set of translations @xmath115 with @xmath116 , let @xmath117 denote the self - similar attractor of the ifs @xmath118 . then for lebesgue almost all @xmath119 , one has @xmath120 it is natural to ask if such a theorem exists for the assouad dimension , for example , is the assouad dimension of a self - affine or self - similar set almost surely constant upon randomising the translations in the above manner . we point out that the answer to this question is no , once again due to the example of peres , simon and solomyak @xcite discussed in section [ nomarstrand ] . [ mainex2 ] fix @xmath89 and let @xmath121 . for a given set of translations @xmath122 let @xmath117 denote the self - similar attractor of the ifs @xmath123 . then there exists two non - empty disjoint open sets @xmath124 such that @xmath125 in particular , @xmath126 is not an almost surely constant function . similar to section [ nomarstrand ] , the open set @xmath127 is easy to find : choose @xmath128 such that the osc is satisfied ( which can be done since @xmath129 ) . then observe that the osc is still satisfied for all @xmath130 in some open neighbourhood of @xmath131 in @xmath132 . again , the second set @xmath133 is more subtle , but dealt with by simon and solomyak @xcite after re - parameterisation . observe that for all @xmath134 , the associated self - similar set is equal to the self - similar set associated to @xmath135 for some @xmath136 $ ] , following translating and re - scaling . more precisely , consider the open subset of @xmath132 given by @xmath137 and the map @xmath138 defined by @xmath139 which is easily seen to be a continuous surjection satisfying @xmath140 where @xmath141 is @xmath142-dimensional lebesgue measure restricted to @xmath143 and @xmath144 is @xmath4-dimensional lebesgue measure restricted to @xmath145 . the attractor @xmath117 is affinely equivalent to the attractor corresponding to @xmath146 . here , the affine rescaling is translation by @xmath147 followed by rescaling by @xmath148 . moreover , the sets @xmath149 for @xmath150 are just a smooth bijective reparamerisation of the sets @xmath63 for @xmath151 where @xmath7 is the modification of the sierpiski triangle from the previous section . it follows by the example of peres , simon and solomyak @xcite that there exists a non - empty open interval @xmath152 such that for almost all @xmath153 , @xmath154 . it follows that the set @xmath155 is open and , by ( [ abscont ] ) , for lebesgue almost all @xmath156 we have @xmath157 . moreover , theorem [ ssfalconer ] implies that for lebesgue almost all @xmath156 we have @xmath158 and so by ( * ? ? ? * theorems 1.3 ) and ( * ? ? * corollary 3.2 ) , for lebesgue almost all @xmath156 , we have @xmath159 . this final implication also follows from our theorem [ gdassouad ] , stated in section [ gdsection ] . the higher dimensional variants of our results remain mostly open . in the dense rotations case , it is quite simple to show that the assouad dimension is constant , irregardless of the ambient dimension : [ assouadconstant ] let @xmath47^d$ ] be self - similar , suppose that the group generated by @xmath48 is dense in @xmath49 or @xmath50 and fix @xmath51 less than @xmath52 . then @xmath160 takes the same value for * all * @xmath54 . we will prove theorem [ assouadconstant ] in section [ assouadconstantproof ] . viewing the proof of the planar case in section [ denserotationsproof ] , it seems likely that this constant is always as large as possible : let @xmath47^d$ ] be a self - similar set containing at least two points , suppose that the group generated by @xmath48 is dense in @xmath49 or @xmath50 and fix @xmath51 less than @xmath52 . then is it true that @xmath161 for * all * @xmath54 ? the ` non - dense case ' is more complicated in higher dimensions , as ` non - dense ' no longer implies ` discrete ' . let @xmath47^d$ ] be self - similar and fix @xmath51 less than @xmath52 . then is it true that @xmath160 is almost surely equal to either @xmath12 or @xmath162 ? in the ` discrete ' case , in all ambient dimensions , the projections are still graph - directed self - similar sets , but in higher dimensions the assouad dimension of such sets is more complicated and can take values other than @xmath12 or @xmath163 for particular directions , see ( * ? ? ? * section 4.1 ) . the examples in section [ nofalconer ] show that there is no ` falconer s theorem for assouad dimension ' . in light of ( * theorem 1.3 ) we know that for self - similar sets in the line with random translations , the situation is still relatively simple . in particular , there are at most two values which @xmath164 can take for a set of @xmath130 with positive measure : the similarity dimension , or 1 . the situation in higher dimensions and for self-_affine _ sets is still unclear however . for @xmath165 and a finite set of non - singular contracting @xmath38 matrices @xmath166 , how many values can the assouad dimension of the attractor of the ifs @xmath167 take , for a set of translates @xmath168 with positive measure ? starting from this section , we will often use the notation @xmath169 to signify that there exists a constant @xmath170 depending only on @xmath171 such that @xmath172 . let @xmath173 be a very small parameter to be specified later . we use the following notion of _ @xmath174-sets _ : let @xmath175 . a finite set @xmath176 is called a @xmath174-set , if the points in @xmath177 are @xmath178-separated , and @xmath179 here , and below , @xmath180 stands for cardinality . the following estimate was obtained by the second author in ( * ? ? ? * proposition 4.10 ) : [ discretemarstrand ] let @xmath181 be a @xmath182-set with @xmath183 points , and let @xmath184 be a @xmath178-separated collection of vectors such that @xmath185 then @xmath186 for some constant @xmath187 . a frostman type argument shows that if @xmath188 , then one can find large @xmath182-sets inside large @xmath174-sets : [ frostman ] assume that @xmath176 is a @xmath174-set with @xmath189 . then , there exists a @xmath182-set @xmath190 with @xmath191 . the proof is fairly standard , but we give it at the end of the section for completeness . the lemma obviously fails for for @xmath192 , but the substitute will be the trivial observation that every @xmath174-set is automatically a @xmath182-set by definition . we are ready to prove theorem [ maintheoremforgeneralsets ] . write @xmath193 $ ] . by the definition of assouad dimension , two things hold : * we may find two sequences @xmath194 and @xmath195 of positive reals such that @xmath196 , and @xmath197 , and @xmath198 for some @xmath199 . * for any @xmath200 and @xmath201 , we have @xmath202 . fix @xmath196 as in ( i ) , write @xmath203 , and let @xmath204 be an @xmath205-separated set with @xmath206 . then , if @xmath207 is the simplest possible affine mapping taking @xmath208 to @xmath209 , we find that @xmath210 is a @xmath211-separated set with cardinality @xmath212 . furthermore , @xmath213 is a @xmath214-set : if @xmath215 , then @xmath216 , and so ( ii ) gives @xmath217 if @xmath218 , we use lemma [ frostman ] to find a @xmath182-set @xmath219 such that @xmath220 . if @xmath192 , we just repeat the observation that @xmath221 is a @xmath222-set . these alternatives will lead to the generic lower bounds @xmath223 or @xmath224 , respectively : their proofs are so similar that we only record explicitly the case @xmath192 . to reach a contradiction , assume that there is a @xmath225 , and a positive measure set of directions @xmath226 such that @xmath227 for all @xmath228 . at this point , we fix @xmath173 so small that @xmath229 , where @xmath187 is the constant appearing in proposition [ discretemarstrand ] . for each direction in @xmath230 , one should be able to find infinitely many values of @xmath231 such that @xmath232 for all @xmath233 ; otherwise clearly @xmath234 . with the ( easier ) borel - cantelli lemma in mind , we define the sets @xmath235 , @xmath231 , by @xmath236 according to the preceding discussion , every point of @xmath230 should lie in @xmath235 for infinitely many values of @xmath231 . so , by the borel - cantelli lemma , we wish to show that @xmath237 this forces @xmath238 and so brings the desired contradiction . to estimate the measure of @xmath235 , it suffices to estimate the maximum number of @xmath211-separated points in @xmath235 , where @xmath203 : if this number is @xmath239 , we have @xmath240 . since @xmath241 , lemma [ discretemarstrand ] implies that there are @xmath242 angles @xmath56 , which are @xmath211-separated and such that @xmath243 moreover , if @xmath59 satisfies the converse inequality , then @xmath244 . indeed , if @xmath245 , then , for a certain @xmath246 ( depending on the mapping @xmath207 ) we have @xmath247 and so @xmath244 . these observations show that @xmath248 , and hence @xmath249 . if @xmath250 fast enough ( as we may assume ) , this proves and the theorem . it remains to prove lemma [ frostman ] , or in fact a slightly more general version of it . the proof is essentially the same as that of ( * ? ? ? * proposition a.1 ) . let @xmath251 and @xmath252 . assume that @xmath253 is a @xmath178-separated set with @xmath254 and satisfying @xmath255 for all @xmath256 and @xmath257 , where @xmath258 are constants . then , there exists a subset @xmath259 with cardinality @xmath260 satisfying @xmath261 for all @xmath256 and @xmath257 . the implicit constants in @xmath262 and @xmath263 are independent of @xmath264 and @xmath187 . without loss of generality , assume that @xmath265 for some @xmath51 and @xmath266^{d}$ ] . denote by @xmath267 the dyadic cubes in @xmath268 of side - length @xmath269 . for a particular cube @xmath270 we will write @xmath271 for its diameter and @xmath272 for the common side - length . first , find all the dyadic cubes @xmath273 which intersect @xmath274 , and choose a single point @xmath275 for each @xmath276 . the finite set so obtained is denoted by @xmath277 . next , modify @xmath277 as follows . consider the cubes in @xmath278 . if one of these , say @xmath279 , satisfies @xmath280 remove points from @xmath281 , until the reduced set @xmath282 satisfies @xmath283 repeat this for all cubes @xmath284 to obtain @xmath285 . then , repeat the procedure at all dyadic scales up from @xmath178 , one scale at a time : whenever @xmath286 has been defined , and there is a cube @xmath287 such that @xmath288 remove points from @xmath289 , until the reduced set @xmath290 satisfies @xmath291 stop the process when the remaining set of points , denoted by @xmath177 , is entirely contained in some dyadic cube @xmath292^{d}$ ] . now , we claim that for every point @xmath293 there exists a unique maximal dyadic cube @xmath294 such that @xmath295 and @xmath296 we only need to show that there exists * at least one * cube @xmath297 satisfying ; the rest follows automatically from the dyadic structure . if @xmath298 , we have for the dyadic cube @xmath299 containing @xmath24 . on the other hand , if @xmath300 , the point @xmath24 was deleted from @xmath277 at some stage . then , it makes sense to define @xmath301 as the dyadic cube containing @xmath24 , where the ` last deletion of points ' occurred . if this happened while defining @xmath302 , say , we have with @xmath303 . but since this was the last cube containing @xmath24 , where * any * deletion of points occurred , we see that that @xmath304 . this gives . now , observe that the cubes @xmath305 , * cover @xmath274 , because they cover every cube in @xmath267 containing a point in @xmath274 , * are disjoint , hence partition the set @xmath177 . these facts and yield the lower bound @xmath306 it remains to prove that @xmath261 for all balls @xmath307 with @xmath257 . for dyadic cubes @xmath308 with @xmath309 it follows immediately from the construction of @xmath177 , in particular the right hand side of , that @xmath310 the statement for balls follows by observing that any intersection @xmath311 can be covered by @xmath312 intersections @xmath313 , where @xmath270 is a dyadic cube with @xmath314 the discrete rotations case will be proved by extending the work of @xcite on self - similar sets in the line to graph - directed self - similar sets in the line , and then recalling that projections of planar self - similar sets with discrete rotations onto lines are precisely graph - directed self - similar sets , see section [ gdequalsprojections ] . graph - directed self - similar sets are an important and natural generalisation of self - similar sets . first considered by mauldin and williams @xcite , roughly speaking one has a family of sets rather than a single set ( as in the self - similar case ) and each member of the family is made up of scaled copies of other sets in the family . more precisely , let @xmath315 be a finite connected directed multigraph , with finite vertex set @xmath316 and finite multiset of directed edges @xmath317 . to each @xmath318 , associate a contracting similarity map @xmath319 \to [ 0,1]$ ] with contraction ratio @xmath320 . for @xmath321 , let @xmath322 be the multiset of edges from @xmath323 to @xmath324 . then there exists a unique @xmath325-tuple of compact nonempty sets @xmath326 , each contained in @xmath29 $ ] , satisfying @xmath327 the family @xmath326 is the _ family of graph - directed self - similar sets_. since the directed graph @xmath328 is connected , it follows that the sets @xmath329 have a common hausdorff dimension . here we only consider graph - directed self - similar sets in the line , but one can consider more general models in the same way where , for example , one works with more general maps or in higher dimensions . zerner @xcite , following lau - ngai @xcite , defined the _ weak separation property _ ( wsp ) for self - similar sets . this is weaker than the open set condition ( osc ) but in many cases plays a similar role in that if the wsp is satisfied , then the overlaps in the construction are controllable and the attractor shares many properties with attractors in the osc case . das and edgar @xcite generalised the wsp to the graph - directed setting and we will use their condition here . for @xmath321 , let @xmath330 where @xmath331 denotes the set of all finite directed paths in the graph going from @xmath323 to @xmath324 and we write @xmath332 for the similarity defined by traversing @xmath333 and composing the similarity maps corresponding to each edge in the appropriate order . also @xmath334 will denote the similarity ratio of @xmath332 . the graph @xmath328 and associated mappings satisfy the _ graph - directed weak separation property ( gdwsp ) _ if for any ( or equivalently all ) @xmath335 , the identity is an isolated point of @xmath336 . das and edgar gave many equivalent formulations of the gdwsp @xcite which parallel zerner s list of equivalent definitions of the wsp in the self - similar ( or 1-vertex ) case ( * ? ? ? * theorem 1 ) . das and edgar then went on to show that many of the properties of self - similar sets satisfying the wsp generalise to graph - directed self - similar sets satisfying the gdwsp . our main result for graph - directed self - similar sets is the following . [ gdassouad ] let @xmath326 be a family of graph - directed self - similar sets in @xmath29 $ ] with common hausdorff dimension @xmath337 . then 1 . if the gdwsp is satisfied , then for all @xmath335 we have @xmath338 and @xmath339 2 . if the gdwsp is not satisfied , then for all @xmath335 we have @xmath340 and @xmath341 . this result can be seen as a generalisation of ( * ? ? ? * theorem 1.3 ) and ( * ? ? ? * corollary 3.2 ) to the graph - directed setting . in particular , it gives a precise dichotomy for the assouad dimension of graph - directed self - similar sets on the real line and proves that if the hausdorff dimension is strictly less than 1 , then the gdwsp is equivalent to , for example , positivity of the hausdorff measure in the hausdorff dimension . this shows that the gdwsp can be viewed as a property of the _ sets _ @xmath329 , rather than the defining graph . the proof of this theorem is divided into two parts , proving _ 1 . _ ( concerning weak separation ) and _ 2 . _ ( concerning lack of weak separation ) respectively . this result is proved by combining previous work of das and edgar on the gdwsp @xcite with falconer s implicit theorems @xcite and recent work of farkas and fraser on ahlfors regularity and hausdorff measure @xcite . let @xmath3 be the common hausdorff dimension of the graph - directed family @xmath326 and fix @xmath335 . since the gdwsp is satisfied , it follows from ( * ? ? ? * ( 4b ) ) that there exists a uniform constant @xmath342 such that for all sets @xmath343 and @xmath344 , @xmath345 it then follows from ( * ? ? ? * theorem 2 ) that @xmath346 we note that this observation was made explicitly in the self - similar case by zerner ( * ? ? ? * corollary p 3535 ) . now that we have established positivity of the hausdorff measure , the equality of the hausdorff and assouad dimensions is an immediate consequence of ( * ? ? ? * corollary 3.1 ) , which proved that for a graph - directed self - similar set @xmath20 , the hausdorff measure of @xmath20 is positive in the hausdorff dimension if and only if @xmath20 is ahlfors regular . we note that ( * ? ? ? * corollary 3.1 ) is stated for irreducible subshifts of finite type rather than graph - directed sets , but it is well - known that these two notions are equivalent . in particular , see the discussion following ( * ? ? ? * corollary 3.1 ) and ( * ? ? ? * proposition 2.52.6 ) . alternatively , see ( * ? ? ? * propositions 2.2.6 and 2.3.9 ) . it is also well - known and straightforward to prove that ahlfors regular sets have equal hausdorff and assouad dimensions . this part follows the proof of ( * ? ? ? * theorem 3.1 ) . indeed , one just has to check that the proof there extends to the graph - directed setting . for completeness we include the argument , but in a slightly streamlined form . since the gdwsp is not satisfied , the identity , @xmath108 , is not an isolated point of @xmath336 . this means we may find a sequence @xmath347 such that @xmath348 as @xmath349 . moreover , we may assume that for all @xmath12 the maps @xmath350 and @xmath351 have trivial reflectional components . if this was not the case , then there must be an edge @xmath352 which also contains a reflection and then , whenever @xmath350 and @xmath351 both contain reflections , one may replace them by @xmath353 and @xmath354 in the sequence . both of these maps have trivial reflectional component and @xmath355 and so the convergence is unaffected . let @xmath356 , which is either a similarity or a non - zero constant function . we may choose a point @xmath357 , an edge @xmath358 , and a small radius @xmath17 such that @xmath359 and for all @xmath12 , we have @xmath360 this can be achieved by choosing two such fixed points @xmath361 and @xmath362 and then observing that for all @xmath12 either @xmath363 or @xmath364 does not include zero and so lies to the left or right of zero . then by choosing a subsequence ( and flipping the axes if necessary ) we can achieve this using @xmath365 or @xmath366 for all @xmath12 . moreover , by an affine change of coordinates we may assume without loss of generality that @xmath367 . finally we may assume that @xmath368 does not contain a reflection , since if it did we could replace it by @xmath369 . denote the map @xmath368 by @xmath370 and let @xmath371 . for @xmath51 , let @xmath372 observing that @xmath373 as @xmath349 . choose @xmath374 large enough to ensure that @xmath375 which we may do since @xmath370 is a contraction and fixes @xmath376 . observe that for any @xmath183 @xmath377 and so for @xmath378 and any @xmath379 we have @xmath380 observing that @xmath381 since @xmath382 is a similarity and @xmath383 by assumption . note that the definition of @xmath384 used @xmath385 here instead of @xmath19 . let @xmath386 and choose @xmath387 and maps @xmath388 by induction on @xmath389 as follows . begin by choosing @xmath390 large enough that @xmath391 and then choose @xmath392 such that @xmath393 also define @xmath394 for @xmath395 , similar to above choose @xmath396 such that @xmath397 where @xmath398 is the similarity ratio of @xmath399 , and define @xmath400 it follows that for all @xmath389 and all @xmath379 @xmath401 which , by ( [ bounds333 ] ) , implies that @xmath402 we will now prove by induction that for all @xmath403 we have @xmath404 for @xmath405 , this is evident . given the claim for @xmath406 , choose @xmath407 such that @xmath408 and so , in particular , @xmath409 is the identity . observe that @xmath410 finally , the missing point , @xmath411 is clearly in @xmath329 which completes the inductive argument . for @xmath412 , consider the ball centered at @xmath413 with radius @xmath414 , which is very small for large @xmath415 . by ( [ separation ] ) and ( [ contains ] ) we have @xmath416 and the @xmath415 points on the right hand side are all separated by at least @xmath417 . setting @xmath418 , this means that @xmath419 and this yields @xmath420 as required . all that remains to complete the proof is to show that the hausdorff measure of @xmath329 is zero in the hausdorff dimension , but this follows again by ( * ? ? ? * corollary 3.1 ) since if the hausdorff measure was positive , then @xmath329 would be ahlfors regular and thus have assouad dimension equal to @xmath337 . the discrete rotations case in theorem [ assouadprojection ] follows immediately from theorem [ gdassouad ] . the reason for this is that projections @xmath63 are graph - directed self - similar subsets of @xmath29 $ ] ( following appropriate rescaling and translating ) . for example , if the group generated by @xmath48 is a discrete subgroup of @xmath66 , then it is isomorphic to the finite cyclic group of order @xmath415 for some @xmath403 . it follows that , for a given @xmath56 , the family @xmath421 is a family of graph - directed self - similar sets with associated graph and ifs inherited from the ifs @xmath28 in the natural way . if the group contains orientation reversing maps , then the situation is not much more complicated . for the details , the reader is referred to ( * ? ? ? * theorem 1.1 ) , which also handles the higher dimensional setting . we begin by reducing the proof of the dense rotations case of theorem [ assouadprojection ] to studying a very simple class of ifs . we will write @xmath422 for the set of finite words over @xmath423 and for @xmath424 , we write @xmath425 for the similarity map defined by the appropriate compositions of the maps @xmath34 . we write @xmath426 for the contraction ratio of @xmath425 . [ reduction ] let @xmath40 ^ 2 $ ] be a self - similar set such that the group generated by @xmath48 is dense in @xmath65 or @xmath66 . then @xmath7 contains a self - similar set @xmath20 which is the attractor of an ifs consisting of two maps @xmath427 , both of which have the same contraction ratio @xmath428 , the same orthogonal component @xmath429 corresponding to anti - clockwise rotation by an irrational multiple of @xmath69 , and such that @xmath430 . this lemma is almost trivial since we are not concerned with losing any dimension . since we are in the dense rotations case , we may choose finite words @xmath431 such that the orthogonal parts @xmath432 are orientation preserving and correspond to anti - clockwise rotations by angles @xmath433 respectively , where @xmath434 and at least one of which is irrational . moreover , since @xmath7 is assumed not to be a single point we can guarantee that @xmath435 . assume without loss of generality that @xmath436 is irrational and let @xmath437 be such that @xmath438 is irrational . this can be done irrespective of @xmath439 since , for example , if @xmath440 is irrational then @xmath441 will do and if @xmath442 is rational , then @xmath443 suffices . finally , the maps @xmath444 satisfy the requirements of the lemma , both rotating by @xmath445 and contracting by @xmath446 . in light of this lemma , and the fact that assouad dimension is monotone , it suffices to prove the dense rotations case of theorem [ assouadprojection ] for @xmath7 generated by similitudes of the form given in the lemma . as usual , we assume @xmath40^{2}$ ] . we wish to reserve @xmath264 " for a small constant and @xmath447 " for a real number , so for the mappings @xmath448 we write @xmath449 where @xmath450 is rotation by a fixed @xmath451 , @xmath452 , and @xmath453 . we will use the following special case of a lemma of erolu , see ( * ? ? ? * lemma 2.6 ) : [ eroglu ] given @xmath454 and @xmath173 , there exist @xmath455 distinct finite words @xmath456 of ( common ) length @xmath51 such that the following assertions hold : * all the rotational components of the similitudes @xmath457 are within @xmath458 of each other . * for some @xmath56 , and for each pair @xmath459 , there exists a point @xmath460 such that @xmath461 . two rotational components are said to be at distance @xmath458 from each other , if the corresponding angles defining the rotations are . in this section , it will be convenient to assume that @xmath462 lies in @xmath463 ( instead of @xmath464 ) for @xmath56 and @xmath465 , so we write @xmath466 the previous lemma self - improves to the following corollary , where the user " may specify the direction @xmath59 , the rotational components of the mappings @xmath467 , and the point @xmath24 : [ ercor ] given @xmath454 , @xmath173 and @xmath56 , there exists some @xmath51 and distinct finite words @xmath456 such that the following assertions hold : * all the rotational components of the similitudes @xmath457 are within @xmath458 of the identity . * for each pair @xmath459 , we have @xmath468 . apply erolu s lemma with parameters @xmath469 and @xmath251 ( to be specified later ) to find @xmath470 and @xmath471 satisfying ( a ) and ( b ) of lemma [ eroglu ] for some direction @xmath472 . thus , the rotational components of the mappings @xmath467 are within @xmath178 of each other , and for each pair @xmath473 , there exists a point @xmath460 such that @xmath474 . we first wish to replace @xmath475 by @xmath59 , which is very easy : by the irrationality of @xmath476 , a suitable choice of @xmath477 , with @xmath478 , ensures that @xmath479 since all the rotational components of @xmath467 were @xmath178-close to each other , the same clearly holds for @xmath480 . next , we wish to replace @xmath480 by something with rotational component close to the identity , and @xmath24 by @xmath376 . to this end , observe that all the projections @xmath481 , @xmath482 , are contained in a single interval of length @xmath483 , since each individual such projection is contained in an interval of length @xmath484 , and these intervals are at distance @xmath485 from each other . now , with @xmath486 , consider the points @xmath487 there are @xmath488 such points , all contained in a single interval of length @xmath483 . thus , if @xmath488 is large enough , depending only on @xmath178 , we can find a subset of cardinality @xmath455 , which is contained in a single interval of length @xmath489 . without loss of generality , assume that this subset is @xmath490 . now , if @xmath491 is chosen suitably , with @xmath492 , the rotational components of the similitudes @xmath493 are within @xmath458 of identity . moreover , @xmath494 for all @xmath495 , if @xmath178 was chosen to be smaller than @xmath496 to begin with . this completes the proof of the corollary . we also need the following simple and well - known geometric fact : [ geolemma ] for @xmath465 , and any angles @xmath497 , we have @xmath498 moreover , if @xmath499 for @xmath500 , then also @xmath501 assume without loss of generality that @xmath502 . then @xmath503 by the mean - value theorem . if @xmath504 , then @xmath505 are rather close to zero , and the inequality can be reversed up to a multiplicative constant ; @xmath506 is certainly on the safe side . we define a continuous strictly increasing auxiliary function @xmath507 by @xmath508 , where @xmath509 are chosen so that @xmath510 , @xmath511 for all @xmath512 , and @xmath513 as @xmath514 . the main chore on route to our theorem is to establish the following lemma by induction : [ indclaim ] given any @xmath515 $ ] and @xmath374 we can find a direction @xmath516 with the following property . for some @xmath517 , the projection @xmath518 contains a subset @xmath519 such that @xmath520 the case @xmath521 is clear , so assume that the lemma has been proven for some @xmath522 , and all @xmath523 . fix @xmath523 , and choose @xmath472 such that the projection @xmath524 contains @xmath525 with @xmath526 for some @xmath517 and all @xmath527 . find @xmath491 points @xmath528 such that @xmath529 . thus , @xmath530 for all @xmath531 . without loss of generality , we may assume that @xmath532 , so also @xmath533 for all finite words @xmath534 . next , apply corollary [ ercor ] with @xmath535 , @xmath536 and @xmath537 , where @xmath170 and @xmath538 are large and small constants , respectively , depending on @xmath491 and to be specified later . we obtain finite words @xmath539 of equal length @xmath51 such that the rotational components of @xmath467 are within @xmath458 of the identity , and @xmath540 for @xmath495 . it follows that the mutual distance between two of the sets @xmath541 is at least @xmath542 , and , relabeling the sets if necessary , we assume that @xmath543 let @xmath56 be an angle close to @xmath475 , to be specified later , and consider the @xmath544-element set @xmath545 see figure [ fig1 ] . clearly @xmath546 , so it suffices to verify that , for a suitable choice of @xmath59 , the elements in @xmath547 can be ordered so that any consecutive points have distance between @xmath548 to this end , we first consider the points @xmath549 , for @xmath550 . write @xmath551 , where @xmath552 is within @xmath537 of the identity by assumption . then , write @xmath553 and @xmath554 , so that @xmath555 . with this notation , @xmath556 . \end{aligned}\ ] ] by , for all @xmath531 . in fact , the same holds if we replace @xmath557 by any angle @xmath56 satisfying @xmath558 : just observe that @xmath559 , and then infer from lemma [ geolemma ] that @xmath560 if @xmath538 is small enough , depending on the difference @xmath561 , combined with this proves that @xmath562 for @xmath550 , as long as @xmath558 . now we have shown that @xmath547 contains at least @xmath491 elements , which lie in the same order as the points @xmath563 , and such that the distance between consecutive points is within the correct range . to increase this number to @xmath564 , we need to choose @xmath59 so that that @xmath565 can act as the @xmath566 element . there is not much choice : either @xmath565 should lie at distance @xmath567 left from @xmath568 , or at distance @xmath569 right from @xmath570 . we make the first choice , which means the question becomes : can we guarantee that @xmath558 ( in order to maintain ) ? to answer this affirmatively , observe that implies @xmath571 , and furthermore @xmath572 by the choice of @xmath573 , and . now , let @xmath56 be an angle such that @xmath574 by lemma [ geolemma ] and the triangle inequality , @xmath575 recalling that @xmath536 for a large constant @xmath187 , we can force @xmath558 , as required . thus , the set @xmath547 has the desired properties , and the inductive proof of lemma [ indclaim ] is complete . the next simple observation says that lemma [ indclaim ] self - improves : the existence of one @xmath59 implies a similar statement for every @xmath59 . [ improvedclaim ] fix @xmath56 , @xmath515 $ ] and @xmath374 . then , for some @xmath517 , the projection @xmath576 contains a set @xmath519 such that @xmath577 for all @xmath550 . according to lemma [ indclaim ] , for a suitable angle @xmath472 , the projection @xmath524 contains a subset @xmath578 such that @xmath579 for @xmath550 . write @xmath477 . then , for a certain @xmath580 depending on the rotation parameter @xmath436 and @xmath581 , it follows that the projection @xmath582 contains @xmath491 points @xmath583 such that @xmath584 . the sequence @xmath585 is asymptotically dense in @xmath9 , so @xmath586 for some @xmath581 . then , writing @xmath587 for some @xmath588 , it follows from lemma [ geolemma ] that @xmath589 for @xmath531 . this proves the result for the points @xmath590 . the dense rotations case of theorem [ assouadprojection ] now follows immediately . fix @xmath56 and observe that , by corollary [ improvedclaim ] , we can find arbitrarily large @xmath374 such that @xmath591 for some @xmath225 and @xmath592 . this yields @xmath593 , as required . this theorem is proved using a simple subsystem trick and adapting an example of bandt and graf ( * ? ? ? * section 2 ( 5 ) ) . first observe that the dense rotations case is taken care of by theorem [ assouadprojection ] and so we only need to deal with the discrete rotations case . let @xmath47 ^ 2 $ ] be a self - similar set which is not contained in a line and suppose that the group generated by @xmath48 is discrete . then @xmath7 contains a self - similar set @xmath20 which is the attractor of an ifs consisting of three similarities , all of which have the same contraction ratio , no rotational or reflectional component ( i.e. trivial orthogonal part ) and such that the three fixed points are not collinear . again , this lemma is almost trivial since we are not concerned with losing any dimension . since @xmath7 is not contained in a line , we may choose finite words @xmath594 such that the fixed points of @xmath595 are not collinear . each of these maps has an orthogonal component with finite order and so taking @xmath12 to be the lowest common multiple of these orders , the maps @xmath596 all have trivial orthogonal component . since the fixed points of these three maps are not collinear , we may choose @xmath597 sufficiently large to guarantee that there are no triples @xmath598 such that @xmath599 are collinear . since similarity ratios are multiplicative , the ifs consisting of the three maps @xmath600 all have the same contraction ratios , trivial orthogonal components , and their fixed points are not collinear . this completes the proof . since assouad dimension is monotone , we only need to prove the result for self - similar sets of the same form as @xmath20 from the above lemma . let @xmath428 be the common similarity ratio and observe that for any @xmath601 we can choose @xmath56 such that @xmath602 is an affinely scaled copy of the attractor @xmath603 of the ifs consisting of the maps @xmath604 acting on the line defined by @xmath605 thus to complete the proof we need to prove that for some @xmath601 , this ifs fails the wsp . this is a simple adaptation of the example considered by bandt and graf in ( * ? ? ? * section 2 ( 5 ) ) , but we include the details for completeness . choose @xmath606 and observe that the maps @xmath607 where @xmath608 are precisely maps of the form @xmath609 for any sequence @xmath610 over @xmath611 . let @xmath612 for some large @xmath183 and choose the sequence @xmath613 by applying the rule : @xmath614 this means that for any @xmath597 we can find maps of the form @xmath607 equal to @xmath615 since @xmath616 as @xmath617 , the ifs fails the wsp . we may then apply ( * ? ? ? * theorem 1.3 ) , or theorem [ gdassouad ] , to deduce that @xmath618 and thus there is some @xmath59 for which @xmath619 . one of the most effective ways to bound the assouad dimension of a set from below is to construct _ weak tangents_. this approach was introduced by mackay and tyson @xcite , but the minor adaptation we state and use here was proved in ( * ? ? ? * proposition 3.7 ) . first we need a suitable notion of convergence for compact sets , which is given to us by the hausdorff metric . let @xmath620 denote the set of all compact subsets of @xmath15 , which is a complete metric space when equipped with the hausdorff metric @xmath621 defined by @xmath622 where @xmath623 is the hemimetric defined by @xmath624 [ pseudoweaktangent ] let @xmath625 and suppose there exists a sequence of similarity maps @xmath626 such that @xmath627 as @xmath349 . then @xmath628 is called a _ weak pseudo tangent _ to @xmath7 and , moreover , @xmath629 . note that the insertion of the word ` pseudo ' in this definition refers to the fact we use the hemimetric @xmath630 instead of the hausdorff metric used by mackay and tyson . the advantage of this approach is that one only needs a subset of @xmath631 to get close to @xmath628 . this is useful when dealing with overlaps , as we will need to do here . fix @xmath634 and let @xmath386 . also , let @xmath637 denote the natural metric on @xmath14 induced by an appropriate operator norm . observe that the map from @xmath638 to @xmath639 defined by @xmath640 is continuous and so we may choose @xmath641 such that for @xmath54 @xmath642 since the group generated by @xmath48 is dense in @xmath50 or @xmath49 , we may choose a finite word @xmath643 such that @xmath644 and @xmath645 here we consider the right action of @xmath50 on @xmath14 defined by @xmath646 . consider the set @xmath647 and the projection @xmath648 . let @xmath649 denote the unique fixed point of @xmath425 and let @xmath650 be defined by @xmath651 which blows up by the reciprocal of the contraction ratio of @xmath425 around the point @xmath652 . the similarity @xmath370 was defined in this way to ensure that @xmath653 which , by ( [ 111 ] ) and ( [ 222 ] ) , yields @xmath654 this in turn implies that @xmath655 which means that by taking a sequence of @xmath458 s tending to zero and choosing @xmath370 in this way , we obtain a sequence of similarity maps @xmath656 , such that @xmath657 as @xmath349 . this proves that @xmath658 is a weak pseudo tangent to @xmath659 and completes the proof . jmf is grateful to bel farkas , alexander henderson , eric olson and james robinson for many fruitful and interesting discussions on topics related to this work , in particular during the writing of @xcite . he also thanks sascha troscheit for earlier discussions about the possibility of a ` falconer theorem for assouad dimension ' . to is supported by the academy of finland through the grant _ restricted families of projections and connections to kakeya type problems . _ k. j. falconer , j. m. fraser and x. jin . sixty years of fractal projections , _ fractal geometry and stochastics v , ( eds . c. bandt , k. j. falconer and m. zhle ) _ , 2014 , available at http://arxiv.org/abs/1411.3156 . p. mattila . recent progress on dimensions of projections , in _ geometry and analysis of fractals _ , d .- j . feng and k .- s . lau ( eds . ) , pp 283301 , _ springer proceedings in mathematics & statistics . _ * 88 * , springer - verlag , berlin heidelberg , 2014 . y. peres and b. solomyak problems on self - similar sets and self - affine sets : an update . _ fractal geometry and stochastics , ii _ ( greifswald / koserow , 1998 ) , 95106 , _ progr . _ , 46 , birkhuser , basel , 2000 .

we perform a study of the nonlinear clustering of matter in the late - forming dark matter ( lfdm ) scenario in which dark matter results from the transition of a nonminimally coupled scalar field from radiation to collisionless matter . a distinct feature of this model is the presence of a damped oscillatory cutoff in the linear matter power spectrum at small scales . we use a suite of high - resolution n - body simulations to study the imprints of lfdm on the nonlinear matter power spectrum , the halo mass and velocity functions and the halo density profiles . the model largely satisfies high - redshift matter power spectrum constraints from lyman- forest measurements , while it predicts suppressed abundance of low - mass halos ( h m ) at all redshifts compared to a vanillacdm model . the analysis of the lfdm halo velocity function shows a better agreement than thecdm prediction with the observed abundance of low - velocity galaxies in the local volume . halos with mass h m show minor departures of the density profiles fromcdm expectations , while smaller - mass halos are less dense , consistent with the fact that they form later than theircdm counterparts .

1412.1103

the cold dark matter ( cdm ) scenario has been tremendously successful in reproducing observations of the distribution of matter on cosmic scales @xcite . in spite of this remarkable success the origin of this invisible component is still not known . this is because cosmological observations shed no light on the particle physics nature of dm and only suggest that dm consists of an approximately pressureless component that clusters gravitationally since a few e - foldings before matter - radiation equality . the leading high - energy physics candidates to dm are weakly interacting massive particles ( wimps ) . in principle these can be detected in underground laboratory experiments with a signal characterized by an annual modulation pattern . this has been thought to be a distinctive signature of dm direct detection against possible contamination from radioactive background noise . in fact the annual modulation is commonly considered a consequence of the orientation of the earth s orbit along or against the flow of dm particles in the galactic halo during the annual motion of the earth around the sun ( for a review see ref . a number of underground experiments such as dama / libra @xcite , cogent @xcite and cressst - ii @xcite have claimed the detection of dm particles with an annual modulation pattern . in contrast , the cdms - ii @xcite , xenon10 @xcite and more recently lux @xcite collaborations have reported none . searches for electroweak wimps at the large hadron collider have also given negative results so far . more puzzling is the fact that the detections from the underground experiments constrain different regions of the wimps parameter space @xcite . thus , it is possible that such contrasting outcomes may be hiding a much richer physics in the weak scale dark sector than previously thought or indicate the need of a completely new dark matter paradigm beyond weak scale wimps . the small - scale clustering of matter in the universe has emerged as the new arena to test the nature of dm . the discovery of a number of anomalies at small scales has cast doubts on the validity of the cdm hypothesis . the core - vs - cusp problem @xcite and the missing satellite problem ( see e.g. @xcite ) have motivated the study of alternative scenarios beyond the cdm paradigm . moreover , recent studies of the dynamical properties of the most luminous milky way dwarf spheroidal satellite galaxies have pointed to a new anomaly , the so - called too - big - to - fail problem @xcite . baryonic processes in galaxy formation have been invoked as the natural solution to these discrepancies ( see e.g. refs . @xcite ) . as an example , baryon feedback and observational incompleteness can account for the missing satellite problem ( see e.g. refs . similarly , statistical variations in the predictions of subhalo abundances from n - body simulations combined with the uncertain value of the milky way virial mass may considerably alleviate the too - big - to - fail problem @xcite . however , it is still unclear whether baryonic feedback models can provide a unique self - consistent explanation to the entirety of dm anomalies ( see e.g. refs . furthermore , analyses of dwarf galaxies in the local field have shown that their abundance and properties differ significantly from the @xmath4cdm predictions @xcite . this is more difficult to reconcile with the cdm paradigm in terms of baryon feedback models given the fact that such isolated systems , contrary to satellites , undergo less complex processes . therefore , it can not be _ a priori _ excluded that such anomalies in the small - scale clustering of matter are ( also ) related to the unknown nature of dm ( for a review see ref . @xcite ) or a hint of broken scale invariance of the inflationary power spectrum at small scales @xcite . among the scenarios alternative to cdm , the existence of a warm dark matter ( wdm ) component has been motivated by particle physics models of sterile neutrinos ( see e.g. refs . @xcite ) . differently from the cdm , wdm particles with mass of a few kev free - stream on a scale @xmath6 kpc due to velocity dispersion @xcite . this causes a characteristic suppression of the linear matter density power spectrum at small scales and alters the properties of dm halos . comparison with observations thus provides bounds on the dm particle mass . as an example , it was pointed out in ref . @xcite that the solution to the core - vs - cusp problem requires wdm particles with mass @xmath7 kev . the authors of refs . @xcite have shown that wdm particles with mass in the range @xmath8\lesssim 2 $ ] can solve the too - big - to - fail problem . this also happens to be the optimal range for realistic sterile neutrino models @xcite . the effective thermal wdm mass range is severely constrained by the determination of the high - redshift matter power spectrum inferred from lyman-@xmath0 forest measurements @xcite . for instance , the recent analysis by viel _ et al . _ @xcite indicates a lower limit @xmath9 kev at a @xmath10 confidence level . even assuming a particle mass within such a bound ( e.g. @xmath11 kev ) the authors of ref . @xcite have shown that the dynamical properties of galactic wdm subhalos are indistinguishable from those of cdm , thus leaving the too - big - to - fail problem unsolved . indeed , if the small - scale anomalies are entirely due to the particle physics nature of dark matter , then requiring their simultaneous solution poses strong constraints on dm models ( see e.g. ref . @xcite ) . another class of models alternative to the standard cdm hypothesis has developed around the possibility that dm particles have significant self - interactions @xcite . this scenario , also known as self - interacting dark matter ( sidm ) , arises in the context of particle physics models of the dm sector ( see e.g. refs . @xcite ) . a key prediction of these models is the fact that dm particles have velocity - dependent self - scattering cross sections . these self - interactions , which can be assimilated to gravitational scalar forces , alter the nonlinear gravitational collapse , leaving characteristic signatures in the small - scale clustering of matter . n - body simulations of sidm models have been presented in numerous studies . recent analysis ( see e.g. @xcite ) has shown that in a narrow range of the sidm model parameter space the dynamics of subhalos in parent halos with mass about that of the milky way reproduces that of the dwarf - spheroidal galaxies in the milky way . at the same time , the subhalos possess core profiles @xmath12 , thus simultaneously solving the core - cusp problem and the too - big - to - fail problem . on the other hand , in sidm models the subhalo mass function remains identical to that of cdm , thus leaving the missing satellite problem unsolved . it has been shown in refs . @xcite that such a class of models can provide a successful solution if an additional interaction of the dm component with radiation is considered . a different family of models consists of scenarios in which dm is the result of a decay or a phase transition process . for instance , the authors of ref . @xcite have explored the possibility that neutral dark matter particles are the result of the decay of charged particles coupled to the photon - baryon plasma before recombination , while ref . @xcite has proposed a scenario in which early dark matter particles decay into a less massive species and a massless noninteracting one . in the latter case the resulting species eventually disrupt the formation of low - mass halos as investigated in refs . @xcite , while still satisfying bounds from lyman-@xmath0 observations @xcite . here , we consider a scenario in which the dm is consequence of a phase transition in the dynamics of a scalar field . in particular , we focus on _ late - forming dark matter _ ( lfdm ) @xcite , inspired by particle physics models of neutrino dark energy @xcite which aim to provide a unified description of dark matter and dark energy . in lfdm models , dm particles are the result of a phase transition in the equation of state of a scalar field from radiation ( @xmath13 ) to matter ( @xmath14 ) ( see ref . @xcite for a different realization of this scenario ) . a distinctive characteristic of lfdm models is the presence of a damped oscillatory tail at the small scales of the linear matter power spectrum . however , contrary to other nonstandard dm models , the range of scales where such a distinctive feature occurs is set by the epoch of the phase transition ( rather than the value of the dm particle mass or the amplitude of the self - interaction cross section ) . the earlier the transition , the smaller the scale where the suppression of power occurs . in this regard , models of ultralight axion ( ula ) dark matter can also be seen as a form of late - forming dark matter where the axion field transitions from a vacuum state ( @xmath15 ) to matter ( @xmath14 ) when the field mass becomes comparable to the hubble scale ( see e.g. ref . ula dark matter alters the linear cosmic structure formation ; thus it is constrained by observations of the large - scale structures as recently shown in @xcite . this occurs in lfdm only if the transition to dm state occurs after recombination , otherwise lfdm exclusively affects the linear matter power spectrum at small scales , thus leaving potentially observable features in the nonlinear structure formation . given the similarities between the ula and lfdm models , it is reasonable to expect that in some viable region of the ula parameter space , the two scenarios may share the same phenomenology of the matter clustering at small scales , which is yet to be studied . in this paper we present a first study of the nonlinear structure formation of lfdm and assess the viability of this scenario using a series of high - resolution n - body simulations . the paper is organized as follows : in sec . [ lfdm_cosmo ] we recall the main features of the lfdm model . in sec . [ simulation_code ] we describe the numerical methods , the generation of initial conditions and n - body simulation characteristics . in sec . [ pow ] we analyze the nonlinear matter power spectrum . in sec . [ hmf ] we describe the imprints on the halo mass function . in sec . [ profiles ] we present the results from the halo density profiles . in sec . [ mcvf ] we probe the circular velocity distribution , and we conclude in sec . [ conclu ] . in the following , we will briefly review the main features of the lfdm scenario that are relevant to the study of cosmic structure formation . we refer the reader to the original paper by das and weiner @xcite for a detailed description of the model . in this scenario , dm is the result of the dynamical evolution of a scalar field coupled to a thermal bath of relativistic particles ( e.g. ev sterile neutrinos ) . at high temperature ( early times ) the field is trapped in a metastable state and behaves as a dark radiation fluid with an equation of state @xmath13 . as temperature drops due to the cosmic expansion , a new lower energy minimum appears and the field rolls into the lower energy state while oscillating around the true minimum of the scalar potential , thus behaving as a collisionless dm component ( @xmath14 ) . because of the coupling to the thermal bath of relativistic particles , deep in the radiation era ( before phase transition ) the scalar field density fluctuates through damped oscillations ( similarly to the baryon acoustic oscillations in the photon - baryon plasma during the tight - coupling regime ) . after the transition , lfdm behaves as standard cdm , and the damped oscillations remain imprinted in the density fluctuation power spectrum . this damped oscillatory pattern is characteristic of other alternative dm scenarios in which dm is coupled to radiation ( see e.g. refs . . however , in the lfdm model the range of scales carrying the imprint of damped oscillation depends on the redshift of the phase transition , @xmath16 . the later the phase transition , the larger the scales where the power spectrum exhibits a damped oscillatory tail . the absence of these features in the large - scale power spectrum of galaxies imposes constraints on the redshift of the transition . in particular , a recent analysis indicates that @xmath17 @xcite . since after the phase transition lfdm behaves as a collisionless component , the linear cosmic structure formation is indistinguishable from that of the standard @xmath4cdm model . however , in order to produce sufficient dark matter abundance , an excess of scalar dark radiation ( in addition to that of neutrinos and photons ) is needed at early times . as shown in ref . @xcite , a small excess contributing to @xmath18 , thus well within the cmb constraints @xcite , is sufficient to give the right amount of dark matter particles . as an example , in the left panel of fig . [ fig1 ] we plot the cosmic microwave background ( cmb ) temperature anisotropy power spectrum for a flat @xmath4-lfdm model with cosmological constant against a @xmath4cdm model best fit to planck data @xcite . in this specific lfdm model the phase transition occurs at @xmath19 , and the cosmological parameters are set to the values of the standard @xmath4cdm model . we have computed the cmb and matter power spectrum of the lfdm model using a modified version of the camb code @xcite to solve the dynamics of the coupled linear perturbation equations associated with the lfdm component . we may notice that even without a model parameter optimization requiring a monte carlo likelihood analysis of available data , the choice of the @xmath4-lfdm model parameters provides a good fit to the planck measurements ( the small differences arise essentially from the small excess of dark radiation density before matter - radiation equality ) . in the right panel of fig . [ fig1 ] , we plot the corresponding linear matter power spectra at redshift @xmath20 against the luminous red galaxy ( lrg ) power spectrum from the sloan digital sky survey seventh data release ( sdss dr7 ) @xcite . as we can see the spectra are identical on the large scales , while differences are only present at @xmath21 h mpc@xmath2 due to lfdm damped oscillations on nonlinear scales . the earlier the phase transition ( i.e. higher @xmath16 ) , the smaller the power spectrum cutoff scale , and the damped part of the spectrum shifts towards larger wave numbers . it is worth mentioning that the linear lfdm power spectrum considered here resembles that of the models studied in refs . nonetheless , there are a few noticeable differences . in ref . @xcite the authors have considered nonstandard models with dm - photon interactions characterized by a linear power spectrum that in the range @xmath22<100 $ ] has a damped oscillatory pattern with a slope different from that of the lfdm case ( see fig . 1 in ref . @xcite ) , while in ref . @xcite the amplitude of the damped oscillations is significantly more suppressed than in our model realization ( see fig . 1 in ref . certain realizations of mixed cold and warm dark matter models such as those considered in refs . @xcite also have spectra that approximate the envelope of the lfdm spectrum ; however , they have different power distribution in the range @xmath22<100 $ ] . indeed , the slope of the suppressed part of the spectrum is a key aspect that differentiates the nonlinear structure formation of scenarios whose phenomenology depends solely on the presence of a cutoff in the linear matter power spectrum . for instance , in the right panel of fig . [ fig1 ] we plot the linear spectra of a wdm thermal relic particle with mass @xmath23 kev characterized by a cutoff scale similar to that of lfdm , and of a broken - scale - invariance ( bsi ) inflationary model @xcite as the one considered in ref . @xcite with highly tuned parameters . in refs . @xcite it was assumed @xmath24 , while to closely reproduce the lfdm spectrum we set @xmath25 . ] such as to have the same cutoff scale and a slope of power suppression similar to that of the lfdm model . we can see that the spectra still carry differences that are likely to alter the nonlinear structure formation . in particular , in the case of the bsi spectrum , the model predicts an excess of power compared to the @xmath4cdm case before the cutoff ; thus it is likely to be highly constrained by high - redshift lyman-@xmath0 measurements . this is not the case for the wdm model , which differs from the lfdm beyond the cutoff scale , where the power exponentially drops several orders of magnitude below the envelope of lfdm damped oscillations . even considering a slightly larger thermal mass particle , this will only shift the exponential tail of the wdm spectrum to higher wave numbers , while still remaining more suppressed than the lfdm prediction due to the shallower slope of the lfdm envelope . hence , it is reasonable to expect that such differences may manifest with different predictions of the abundance and properties of low - mass halos . it is for these very reasons that , despite the numerous numerical studies of the nonlinear clustering of wdm models , it is opportune to investigate the phenomenology of the lfdm scenario and the like . we run a series of high - resolution n - body simulations using ramses @xcite , an adaptive mesh refinement ( amr ) code with a tree - based data structure in which particles are evolved using a particle - mesh solver and the poisson equation is solved using a multigrid method @xcite . we generate the initial conditions with the code mpgrafic @xcite , which uses the zeldovich approximation . we set the initial redshift such that the standard deviation of the initial density field smoothed on the scale of the coarse grid @xmath26 is @xmath27 . with this choice , the initial redshift of the simulations is sufficiently large to suppress spurious effects due to transients @xcite . we assume a flat @xmath4-lfdm model with phase transition redshift @xmath28 and characterized by the linear power spectrum shown in fig . the cosmological parameters are set to the following values : @xmath29 , h @xmath30 , @xmath31 , @xmath32 and @xmath33 . in addition , we also consider a flat @xmath4cdm model with identical parameters . the simulations of the @xmath4-lfdm model consist of a box of @xmath34 volume with @xmath35 particles , a similar volume with @xmath36 particles , an intermediate box of @xmath37 with @xmath35 particles and a large box of @xmath38 volume with @xmath39 particles . this simulation suite enables us to control numerical systematic errors due to mass resolution and volume effects . the characteristics of the simulations are summarized in table [ table_sim ] . we can see that for the highest mass resolution runs the spatial resolution of the simulations at the level of the coarse grid varies from @xmath40 to @xmath41 . however , the actual resolution of the simulations is much higher due to the amr scheme used in the ramses code . for instance , in the case of the @xmath34 volume simulation with @xmath36 particles we have up to six refinement levels of the coarse grid which are triggered during the run ; thus the densest objects are resolved with a spatial resolution of @xmath42 . in addition to the lfdm model simulations , we have run a @xmath34 volume with @xmath35 particles and one with @xmath36 particles for a flat @xmath4cdm model with identical cosmological parameters ( with @xmath43 and @xmath44 , respectively ) and the same phase of the initial conditions , which we use to evaluate differences with respect to the small - scale clustering of @xmath4-lfdm . the simulations were run on the ada supercomputer of the institute for development and resources in intensive scientific computing ( idris ) . in particular , the lfdm simulation of @xmath38 volume with @xmath39 particles was run on 3000 intel sandy bridge e5 - 4650 processors for a total running time of @xmath45 hours . .@xmath4-lfdm n - body simulation characteristics . l is the simulation box length , @xmath46 is the number of n - body particles , @xmath47 is the initial redshift of the simulation , @xmath48 the mass resolution , and @xmath26 is the spatial resolution of the coarse grid . [ cols="^,^,^,^,^",options="header " , ] we compute the matter power spectrum with the code powergrid @xcite . this computes the power spectrum by performing a fourier transform of the density field in band powers @xmath49 , where @xmath50 is the simulation box length . we correct for the smoothing effect due to the cloud - in - cell ( cic ) algorithm used to estimate the density field from the particle distribution . to be conservative , we restrict the scope to wave numbers below the nyquist frequency of the coarse grid . in the left panel of fig . [ fig2 ] , we plot the nonlinear power spectra at @xmath51 and @xmath52 from the four @xmath4-lfdm simulations which differ for volume size and mass resolution . for clarity , it is convenient to focus on the @xmath20 case , with similar trend at higher redshifts . first , we may notice that the spectra of the large ( solid line ) and intermediate ( dotted line ) volumes are fairly in agreement at @xmath53 h mpc@xmath2 , while in the same range the amplitude of the spectrum of the smaller simulation box ( dashed line ) is slightly lower . this is due to finite volume effects as well as the choice of the initial phase ( see e.g. refs . volume effects are negligible at large @xmath54 ; in particular for @xmath55 h mpc@xmath2 , the spectrum of the @xmath34 volume simulation with @xmath35 particles and that of the @xmath38 box differ by less than @xmath56 . on the other hand , at larger wave numbers ( near the nyquist frequency of the simulations ) , we can see the systematic suppression of power due to mass resolution errors for the lower resolution runs compared to the higher one . at small scales , this is the dominant source of numerical uncertainty ; in particular , in the range @xmath57<50 $ ] , the spectra of the @xmath34 volume simulation with @xmath35 and that with @xmath36 particles differ by a few percent in the low - end limit and up to @xmath58 in the high end . however , since we are interested in the relative difference between the lfdm and @xmath4cdm spectra , we expect ( and show in the right panel of fig . [ fig2 ] ) the ratio to be less affected by mass resolution errors . we plot in the right panel of fig . [ fig2 ] the relative difference of the nonlinear matter power spectrum of the @xmath4-lfdm model with respect to the @xmath4cdm case for @xmath21 h mpc@xmath2 at @xmath59 and @xmath60 from the @xmath34 volume simulations . at @xmath61 , the runs with @xmath35 and @xmath36 particles differ from less than a percent at @xmath62 h mpc@xmath2 and increase up to @xmath63 at the high end of the interval . these differences due to mass resolution are clearly subdominant compared to the relative differences between the models . volume effects in this range of wave numbers are expected to be negligible . hence , we are confident the differences seen between @xmath4-lfdm and @xmath4cdm are not due to numerical artifacts . notice that any signature of the lfdm damped oscillations in the initial linear power spectrum has been completely erased . constraints on the high - redshift matter power spectrum ( @xmath64 ) inferred from the recent lyman-@xmath0 measurements @xcite indicate that the nonlinear power spectrum for wdm thermal relic particles of mass @xmath65 kev ( at @xmath66 confidence level ) may deviate from @xmath4cdm by no more than @xmath67 at @xmath62 h mpc@xmath2 ( as can be inferred from fig . 1 in ref . @xcite ) . as shown in the right panel of fig . [ fig2 ] , the @xmath4-lfdm model considered here is consistent with these bounds for @xmath68 , while models with slightly higher @xmath16 will largely evade these constraints in the entire high - redshift range . we detect dm halos using the code pfof @xcite based on the friend - of - friend algorithm @xcite , which identifies halos as group of particles characterized by an intraparticle distance smaller than a given linking length parameter @xmath69 . in our analysis we set this to the standard value @xmath70 and consider only halos with more than @xmath71 particles unless specified otherwise . spurious low - mass halos due to artificial fragmentation of the dm density field occur in models with a sharp cutoff in the initial power spectrum ( see e.g. refs . @xcite ) . these alter the predictions of the low - mass end of the mass function and contribute as a dominant source of systematic errors in the mass range where differences among dm scenario are mostly relevant to observational tests . we have taken fragmentation into account in analysis of the @xmath4-lfdm model presented here and refer the reader to ref . @xcite for a dedicated study on the removal of spurious halos . our spurious halo identification method relies on the fact that since spurious halos originate from the fragmentation of the density field , these are highly nonspherical groups of particles characterized by extreme values of the spin and shape parameters while exhibiting large deviation from the virial theorem . thus , in the range of masses where mass resolution effects are subdominant , spurious halos can be removed by simply retaining those halos that approximately satisfy the virial condition and checking that the distribution of the spin and shape parameters of the remaining halos have well - behaved tails . we compute the mass function as @xmath72 where @xmath50 is the simulation box length and @xmath73 is the number of halos in a mass bin of size @xmath74 . throughout this paper , @xmath75 and @xmath76 denote base-10 and base-@xmath77 logarithms , respectively . in fig . [ fig3 ] we plot the mass function of the @xmath4-lfdm model at @xmath20 from simulations with volumes of @xmath38 with @xmath39 particles ( blue filled squares ) , @xmath37 with @xmath35 particles ( brown open circles ) , @xmath34 with @xmath35 ( green open squares ) and @xmath36 ( red filled circles ) particles in bins of size @xmath78 . in all four cases we can see that the high - mass end of the mass function is characterized by a large level of scatter due to finite volume effects . mass resolution effects are relevant in the low - mass end ; this can be seen in the case of the intermediate simulation box . for which the mass function deviates at more than a @xmath79 level for @xmath80 h@xmath2 m@xmath3 from the higher - resolution runs . likewise , the comparison between the mass functions of the @xmath34 volume simulation with @xmath35 particles and that with @xmath36 shows deviation at more than a @xmath79 level for @xmath81 h@xmath2 m@xmath3 . it is worth noticing the drop of the mass function from the @xmath34 volume run with @xmath36 at low masses ; this is consistent with the small - scale cutoff in the linear matter power spectrum . overall , we find convergence at the @xmath79 level in the mass range @xmath82\lesssim 10^{11}$ ] for the mass functions from the @xmath34 and @xmath38 volume simulations . -lfdm model at @xmath20 in bins of size @xmath78 from simulations with volumes of @xmath38 with @xmath39 particles ( blue filled squares ) , @xmath37 with @xmath35 particles ( brown open circles ) , and @xmath34 with @xmath35 ( green open squares ) and @xmath36 ( red filled circles ) particles for halos with at least @xmath71 particles . error bars are given by poisson errors . ] -lfdm halo mass function at @xmath20 ( blue open squares ) , @xmath83 ( red solid circles ) , @xmath52 ( magenta open triangles ) , @xmath60 ( green stars ) and @xmath84 ( cyan asterisks ) from the @xmath38 volume simulation . error bars are given by poisson errors . the solid lines show the mass function assuming the sheth - tormen multiplicity function with parameters calibrated at @xmath20 . ] in fig . [ fig4 ] we plot the redshift evolution of the @xmath4-lfdm mass function from the @xmath38 volume simulation at @xmath85 and @xmath84 . here , to be conservative , we limit the mass range to halos with at least @xmath86 particles . for comparison , we also plot the sheth - tormen ( st ) @xcite mass function calibrated at @xmath20 and extrapolated to higher redshifts ( see the appendix ) . we can see that despite the suppressed amplitude of the linear matter power spectrum at small scales ( see right panel in fig . [ fig1 ] ) , the halo formation proceeds as in a hierarchical bottom - up scenario , and the abundance of halos increases from high to low redshifts as a function of halo mass . at @xmath20 , the st mass function provides a good fit to the n - body measurements to better than @xmath87 in the mass range @xmath88<2\times 10^{12}$ ] . for larger masses , deviations from the st fit are dominated by finite volume errors . in contrast , at lower masses , the st overestimates the abundance of low - mass halos with deviations up to @xmath89 . at redshifts @xmath90 and @xmath91 , such deviations increase and extend over the entire range of masses with deviations as large as @xmath58 in the range where volume effects are negligible . at @xmath92 , the low - mass end is still overestimated , while the abundance of halos with mass @xmath93 is underestimated . this trend is a manifestation of departure from the universality of the halo mass function already found in standard cdm cosmologies ( see e.g. ref . nevertheless , the fact that st systematically overestimates the abundance of low - mass halos ( @xmath94 ) suggests a substantial departure of the nonlinear gravitational dynamics from that of the standard cdm ellipsoidal collapse model encoded in the st multiplicity function ( see the appendix ) . to see this more clearly we compare the multiplicity function of the @xmath4-lfdm model to that of the @xmath4cdm . for this we use data from the @xmath34 volume simulation with @xmath35 particles . as with the case of the relative difference between the matter power spectra presented earlier , the ratio of the multiplicity functions is less affected by mass resolution errors . in the left panel of fig . [ fig5 ] , we plot the ratio of the multiplicity functions at @xmath85 and @xmath84 in mass bins of size @xmath95 containing at least @xmath71 halos . this limits the mass range of interest to an interval where volume effects remain negligible , and thus it reduces from @xmath20 to @xmath92 , as volume effects are largest at the high - mass end . we can see that independently of the redshift , for @xmath93 the multiplicity functions of the two simulated cosmologies are identical to numerical precision . we may also notice that in this mass range the ratio of the st multiplicity functions calibrated to the data at @xmath20 for the two cosmological models reproduces fairly well the numerical results . this is not the case at lower masses , where st predictions largely deviate from the n - body results . this suggests that the use of standard mass function formulas developed in the framework of the cdm paradigm to infer predictions for nonstandard cdm models characterized by a cutoff in the linear matter power spectrum at small scales may lead to significant misestimates of the halo abundances at low mass . let us now turn to the comparison of the halo abundance between @xmath4cdm and @xmath4-lfdm . in the right panel of fig . [ fig5 ] , we plot the mass function of @xmath4cdm halos relative to that of the @xmath4-lfdm at @xmath85 and @xmath84 in mass bins of size @xmath95 containing at least @xmath71 halos . again , the ratio is of order unity at all simulated redshifts for @xmath93 which is consistent with the fact that in this range the multiplicity functions of the two models are identical . furthermore , in this range also the variance of the smoothed linear density field predicted by the two cosmologies is nearly identical , thus leading to a similar halo abundance well reproduced by the st mass function . in contrast , we can see that for @xmath94 , halos in @xmath4cdm are up to a factor of @xmath52 more abundant than in @xmath4-lfdm . this is due to the differences in the multiplicity function shown in the left panel of fig . [ fig5 ] and amplified by the fact that on these scales the smoothed linear density field of the @xmath4-lfdm model is suppressed compared to the @xmath4cdm case . also notice that there is no redshift evolution of the relative abundance between @xmath96 and @xmath97 , while a modest evolution occurs at higher redshifts for halos with mass @xmath98 . this is consistent with the fact that in @xmath4-lfdm , small - mass halos form later than in @xmath4cdm , thus leading to differences that can be tested through high - redshift universe observations as already explored in the case of wdm models in ref . the lower abundance of low - mass halos in lfdm models compared to @xmath4cdm is a direct consequence of the suppression of power at small scale in the linear matter power spectrum beyond the cutoff scale . since in the lfdm scenario the location of the cutoff depends on the epoch of dark matter formation , we can expect that dm forming earlier than that in the model considered here will decrease the discrepancy with respect to the @xmath4cdm model , while a later formation will increase it . thus , measurements of the abundance of dwarf galaxies may provide direct constraints on the epoch of dark matter formation in lfdm models . we now focus on the density profile of dm halos . in fig . [ fig6 ] we plot the relative difference of the @xmath4cdm averaged halo density profile at @xmath20 with respect to that of the @xmath4-lfdm case for halos in three different mass bins . within each mass bin , individual halo profiles are stacked to obtain the averaged profile . to facilitate stacking , radius is normalized by @xmath99 corresponding to an enclosed overdensity @xmath100 relative to the cosmic mean matter density . to control numerical resolution effects , we plot results from the @xmath34 box simulations with @xmath35 ( blue lines ) and @xmath36 ( red lines ) particles . moreover , we limit the curves up to the physical scale associated with the size of the most refined cell of the simulations . we can see that the lower - resolution simulations systematically underestimate the profile differences with respect to the higher - resolution runs with deviations up to @xmath101 level . we find that density profiles for halos with mass @xmath102 are nearly identical in the two cosmologies for @xmath103 . at smaller radii , deviations do not exceed the @xmath87 , level with @xmath4cdm halos being on average denser than in @xmath4-lfdm . in the case of halos with mass in the range @xmath104 , we find larger deviations . in particular , in the cdm case the averaged profile at radii @xmath105 is @xmath106 denser than the lfdm counterpart . as already noticed in ref . @xcite , this is consistent with the fact that the two models only differ for the amplitude of the initial power spectrum at small scales . in fact , as shown by the analysis of the mass function , the suppression of power at small scales in the @xmath4-lfdm model causes small - mass halos to form later than in @xmath4cdm , thus resulting in lower density profiles . halos of higher mass , on the other hand , assemble nearly at the same time in the two cosmologies , thus leading to nearly identical profiles . if the trend shown in fig . [ fig6 ] at small radii is extrapolated to lower - mass halos , then these results do not exclude the possibility that halos with mass @xmath107 may have cored profiles ; however , we are not able to address this point at the moment , since our simulations do not possess the required resolution . in @xmath4cdm with respect to the @xmath4-lfdm model , as a function of radius normalized by @xmath99 . the averaged profiles are obtained by stacking the halo profiles within mass bins @xmath108 h@xmath2 m@xmath3 ( dashed line ) , @xmath109 h@xmath2 m@xmath3 ( dotted line ) and @xmath110 h@xmath2 m@xmath3 ( solid line ) for the @xmath34 box simulations with @xmath35 ( blue lines ) and @xmath36 ( red lines ) particles . ] the abundance of dark matter halos can be probed through measurements of the galaxy luminosity function provided prior knowledge of the relation between galaxy luminosity and halo mass exists . however , this is hard to predict , since it depends on the galaxy formation process itself . in contrast , the velocity function , i.e. the number density of galaxies with a given circular velocity , can be predicted more reliably and tested against observations . it is indeed the comparison of the circular velocity of the satellites of the local group against n - body simulation results that originally pointed to the missing satellite problem @xcite . another advantage of the velocity function is that it probes the abundance of all virialized structures . in the past few years , analyses of the velocity function inferred from a number of surveys of the local volume have pointed to suppressed abundances of low - velocity field galaxies compared to expectations from @xmath4cdm simulations @xcite . these findings have been recently confirmed by the measurements of the circular velocity distribution in galaxies within @xmath111 mpc from the local volume catalog @xcite . the circular velocity distribution is indeed a promising tool to constrain dm properties ; however , it is worth mentioning that any analysis using dm - only simulations can introduce a mass - dependent bias , since the measured circular velocity is that of the baryon component inside dark matter halos . here , we compare the measurements of the circular velocity distribution from ref . @xcite against estimates from our @xmath4cdm and lfdm halo catalogs of the @xmath34 volume simulations with @xmath36 particles . to derive predictions from dm - only simulations that can be compared to observations , we follow the procedure described in ref . more specifically , we correct the estimated halo circular velocities to account for the effect of baryons using the prescription described in ref . @xcite and based on the analysis by ref . then , we multiply the estimated velocity functions by a factor @xmath112 to account for the contribution of subhalos not included in our numerical catalogs . as pointed out in ref . @xcite , this factor corresponds to the fraction of subhalos in @xmath4cdm for velocities @xmath113 km s@xmath2 . in the lfdm scenario , the fraction of subhalos may be even smaller due to the power suppression at small scales ; thus the choice adopted here is a rather conservative one . in fig . [ fig7 ] , we plot the differential circular velocity function @xmath114 for the @xmath4cdm ( red triangles ) and @xmath4-lfdm ( blue squares ) models . the dotted straight line is the fitting function to the @xmath4cdm simulation given by eq . ( 5 ) in ref . @xcite , while the dashed line includes the baryonic corrections . since the cosmological parameters considered here are marginally different from ref . @xcite , we normalize these fitting functions to the circular velocity function obtained from our @xmath4cdm simulation . the cyan solid lines enclose the region of the observed velocity function ( with a fitting function given by eq . ( 12 ) in ref . @xcite ) to within @xmath115 error . as we can see , the lfdm model predicts a lower abundance of low - velocity galaxies in better agreement with observations than @xmath4cdm . furthermore , since the suppression is related to the transition redshift @xmath16 of dm formation ( the later the transition , the larger the suppression of low - mass halo abundances ) , it is possible that setting @xmath16 to a value slightly smaller than that assumed here may reproduce the observed velocity distribution up to statistical errors . box simulations with @xmath36 particles for the @xmath4cdm ( red triangles ) and @xmath4-lfdm ( blue squares ) models . error bars are given by poisson errors . the dotted straight line is the fitting function to the @xmath4cdm simulation given by eq . ( 5 ) in ref . @xcite , while the dashed line includes the baryonic corrections . the cyan solid lines denote the region of the observed velocity function to within @xmath115 error from ref . @xcite . ] the distribution of matter at small scales may provide insights on the nature of dark matter particles and potentially constrain scenarios alternative to the standard cold dark matter paradigm . in this work we have performed a first study of the nonlinear clustering of late - forming dark matter using a suite of high - resolution n - body simulations . the main phenomenological feature of this scenario is the presence of damped acoustic oscillations at large wave numbers in the linear matter power spectrum . the scale where such a power suppression occurs depends on the epoch of the phase transition of a scalar field coupled to relativistic particles that generates the dm particles . without requiring an extreme fine - tuning of the microscopic model parameters , this may occur before matter - radiation equality and imprint a pattern of damped oscillations at @xmath21 h mpc@xmath2 in the initial conditions . because of this , lfdm models are indistinguishable from standard cdm on large scales , while they are constrained by probes of the small - scale clustering of matter . from the analysis of n - body simulations , we have shown that lfdm could be a viable alternative to cdm , as it provides predictions of the nonlinear matter power spectrum largely consistent with bounds from high - redshift lyman-@xmath0 power spectrum measurements . it should be noted that the lyman-@xmath0 constraints have so far been obtained for a restricted class of dm models ( such as wdm ) with a sharp cutoff in the linear power spectrum . as lfdm models have spectra with considerably shallower slopes than wdm , it is very likely that lfdm models that resolve @xmath4cdm inconsistencies on galactic and subgalactic scales would also be in better agreement with lyman-@xmath0 measurements . we have evaluated the halo abundance ; comparison with predictions from the standard @xmath4cdm model shows that small - mass lfdm halos ( @xmath116<10^{10}$ ] ) are less abundant than their cdm counterparts . we have also evaluated the corresponding differences in the circular velocity function at @xmath20 and shown that the lfdm scenario is in better agreement with observations of the number density of low - velocity galaxies than @xmath4cdm . the number density of low - mass halos varies with redshift , with low - mass halo abundance up to @xmath117 smaller than @xmath4cdm at @xmath118 . this indicates that halos in this mass range form later than in @xmath4cdm , thus leading to lower densities in the internal part of the halos as confirmed by the study of density profiles . on the other hand , the assembly of more massive halos occurs as in @xmath4cdm , and we find no significant differences in their density profiles between the two cosmologies . these results confirm previous numerical studies of nonstandard dm models characterized by initial linear matter power spectra similar to that of lfdm . since lfdm behaves as a collisionless component , the lack of cored profiles in milky way like halos seems inherently related to the fact that in order to satisfy the lyman-@xmath0 constraints , the suppression of power in the initial power spectrum must occur on wave number @xmath119 h mpc@xmath2 . this suggests that only by including dm self - interactions would it be possible to develop cored profiles . on the other hand , one should note that baryonic feedback does play a role at these scales . given the differences in the abundance and assembly of small - mass halos in @xmath4-lfdm and @xmath4cdm , it will be of interest to investigate the dynamics of baryons in the lfdm scenario . this may leave distinct observational features in the formation of stars and galaxies which warrant further investigation . we would like to thank matteo viel and jess zavala for useful comments and suggestions . this work was granted access to hpc resources of idris through time allocations made by genci ( grand quipement national de calcul intensif ) on the machine ada no . we acknowledge support from the dim acav of the region le - de - france . the research leading to these results has received funding from the european research council under the european community s seventh framework programme ( fp7/2007 - 2013 grant agreement no . 279954 ) . in the press - schechter approach @xcite the halo mass function can be written as @xmath120 where @xmath121 is the mean cosmic matter density and @xmath122 is the root - mean - square fluctuation of the linear matter density field smoothed on a scale @xmath123 enclosing a mass @xmath124 with @xmath125,\ ] ] where @xmath126 is the linear dm power spectrum and @xmath127 is the fourier transform of the smoothing function in real space ( here we consider a top hat filter ) . the function @xmath128 is the so - called `` multiplicity function '' which carries all information on the nonlinear gravitational processes that lead to the formation of halos . in the framework of the excursion set theory @xcite , this can be computed from the distribution of random walks first crossing a collapse density threshold that encodes the nonlinear gravitational dynamics of matter collapse . it was shown in ref . @xcite that the multiplicity function for uncorrelated walks with a threshold barrier motivated by the ellipsoidal collapse model is well approximated by a the sheth - tormen function ( apart from a fudge factor coefficient necessary to recover agreement with the n - body results ) given by @xcite @xmath129e^{-a\delta_c^2/2\sigma^2},\ ] ] where @xmath130 is the linearly extrapolated spherical collapse density threshold . since after the transition lfdm behaves as a collisionless component , the spherical collapse dynamics is identical to that of cdm . at @xmath20 we have @xmath131 , and we compute the exact redshift dependence @xmath132 by numerically solving the spherical collapse equations as in ref . @xcite . we calibrate the st parameters @xmath133 , @xmath134 and @xmath135 to the numerical mass function at @xmath20 from the @xmath34 volume simulation of the @xmath4cdm and @xmath4-lfdm model . we find @xmath136 and @xmath137 for both models , while @xmath138 for @xmath4cdm and @xmath139 for @xmath4-lfdm . as pointed out in ref . @xcite , the parameter @xmath135 is determined by the shape of the mass function at the low - mass end , which in turn depends on the form of the ellipsoidal collapse threshold ( see also refs . @xcite for an explicit relation between the mass function and the parameters of an ellipsoidal - collapse - inspired barrier ) . hence , the large deviations from unity of the ratio of the multiplicity functions at low masses shown in the left panel of fig . [ fig5 ] are clearly indicative of a departure of the nonlinear collapse dynamics of the @xmath4-lfdm model from that of the @xmath4cdm . 99 g. efstathiou _ et al . _ , mon . not . . soc . * 330 * , l29 ( 2002 ) d. n. spergel _ _ , astrophys . ser . * 148 * , 175 ( 2003 ) m. tegmark _ et al . _ , d. * 69 * , 103501 ( 2004 ) d. clowe _ et al . _ , j. * 648 * , l109 ( 2006 ) p. a. r. ade _ et al . _ ( planck collaboration ) , astron . & astroph . * 536 * , a1 ( 2011 ) k. freese , m. lisanti , c. savage , rev . 85 , 1561 ( 2013 ) r. bernabei _ et al . phys . j. c * 67 * , 39 ( 2010 ) c. e. aalseth _ et al . _ , phys . lett . * 107 * , 141301 ( 2011 ) g. angloher _ et al . _ , j. c * 72 * , 1971 ( 2012 ) z. ahmed _ et al . . lett . * 106 * , 131302 ( 2011 ) j. angle _ et al . lett . * 107 * , 051301 ( 2011 ) d.s . ( lux collaboration ) , phys . lett . * 112 * , 091303 ( 2014 ) d. hooper , j. cosmol . . phys . * 09 * , 035 ( 2013 ) b. moore , nature ( london ) * 370 * , 629 ( 1994 ) r. kuzio de naray , t. kaufmann , mon . not . . soc . * 414 * , 3617 ( 2011 ) a. klypin , a. v. kravtsov , o. valenzuela , f. prada , astrophys . j. * 522 * , 82 ( 1999 ) b. moore _ et al . _ , astrophys . j. * 524 * , l19 ( 1999 ) m. boylan - kolchin , j. s. bullock , m. kaplinghat , mon . not . * 415 * , l40 ( 2011 ) m. boylan - kolchin , j. s. bullock , m. kaplinghat , mon . not . . soc . * 422 * , 1203 ( 2012 ) j. s. bullock , a. v. kravtsov , d. h. weinberg , astrophys . j. * 539 * , 517 ( 2000 ) a. j. benson _ _ , mon . not . . soc . * 333 * , 177 ( 2002 ) r. s. somerville , astrophys . j. * 572 * , l23 ( 2002 ) m. ricotti , n. y. gnedin , m. j. shull , astrophys . j. * 575 * , 49 ( 2002 ) m. ricotti , n. y. gnedin , astrophys . j. * 629 * , 259 ( 2005 ) j. i. read , a. p. pontzen , m. viel , mon . not . . soc . * 371 * , 885 ( 2006 ) n. i. libeskind _ _ , mon . not r. astron . * 374 * , 16 ( 2007 ) a. v. maccio _ et al . _ , mon . not . * 402 * , 1995 ( 2010 ) a. s. font _ _ , mon . not . soc . * 417 * , 1260 ( 2011 ) a. pontzen , f. governato , mon . not . r. astron . * 421 * , 3464 ( 2012 ) r. teyssier , a. pontzen , y. dubois , j. i. read , mon . not . . soc . * 429 * , 3068 ( 2013 ) s. koposov _ et al . _ , j. * 686 * , 279 ( 2008 ) q. guo _ et al . _ , mon . not . . soc . * 413 * , 101 ( 2011 ) j. wang , c. s. frenk , j. f. navarro , l. gao , t. sawala , mon . not . * 424 * , 2715 ( 2012 ) c. w. purcell , a. r. zentner , j. cosmol . * 12 * , 007 ( 2012 ) j. pa@xmath140arrubia , a. pontzen , m. g. walker , s. e. koposov , astrophys . j. * 759 * , l42 ( 2012 ) i. ferrero _ et al . _ , mon . not . . soc . * 425 * , 2817 ( 2012 ) s. garrison - kimmel , m. rocha , m. boylan - kolchin , j. s. bullock , j. lally , mon . not . . soc . * 433 * , 3539 ( 2013 ) j. zavala _ _ , astrophys . j. * 700 * , 1779 ( 2009 ) e. papastergis , a. m. martin , r. giovanelli , m. p. haynes , astrophys . j. * 739 * , 38 ( 2011 ) e. n. kirby , j. s. bullock , m. boylan - kolchin , m. kaplinghat , j. g. cohen , mon . not . . soc . * 439 * , 1015 ( 2014 ) a. klypin , i. karachentsev , d. makarov , o. nasonova , arxiv:1405.4523 s. garrison - kimmel , m. boylan - kolchin , j. s. bullock , e. n. kirby1 , mon . not . . soc . * 444 * , 222 ( 2014 ) e. papastergis , r. giovanelli , m. p. haynes , f. shankar , astron . & astrophys . * 574 * , a113 ( 2015 ) d. h. weinberg _ _ , arxiv:1306.0913 m. kamionkowski , a. r. liddle , phys . lett . * 84 * , 4525 ( 2000 ) a. r. zentner , j. s. bullock , phys . d * 66 * , 043003 ( 2002 ) a. r. zentner , j. s. bullock , astrophys . j. * 598 * , 49 ( 2003 ) s. dodelson , l. m. widrow , phys . * 72 * , 17 ( 1994 ) g. m. fuller , a. kusenko , i. mocioiu , s. pascoli , phys . d * 68 * , 103002 ( 2003 ) k. abazajian , phys . d * 73 * , 063513 ( 2006 ) a. boyarsky , j. lesgourgues , o. ruchayskiy , m. viel , phys . lett . * 102 * , 201304 ( 2009 ) a. boyarsky , j. lesgourgues , o. ruchayskiy , m. viel , j. cosmol . phys . * 05 * , 012 ( 2009 ) d. boyanovsky , h. j. de vega , n. g. sanchez , phys . rev . d * 78 * , 063546 ( 2008 ) a. v. maccio _ et al . _ , mon . not . r. astron . soc . * 424 * , 1105 ( 2012 ) m. r. lovell _ et al . _ , mon . not . soc . * 420 * , 2318 ( 2012 ) m. r. lovell _ et al . _ , mon . not . soc . * 439 * , 300 ( 2014 ) k. n. abazajian , phys . * 112 * , 161303 ( 2014 ) m. viel , j. lesgourgues , m. g. haehnelt , s. matarrese , a. riotto , phys . d * 71 * , 063534 ( 2005 ) u. seljak , a. makarov , p. mcdonald , h. trac , phys . * 97 * , 191303 ( 2006 ) m. viel _ et al . _ , lett . * 100 * , 041304 ( 2008 ) m. viel , g. d. becker , j. s. bolton , m. g. haehnelt , phys . rev . d * 88 * , 043502 ( 2013 ) a. schneider , d. anderhalden , a. v. macci , j. diemand , mon . not . . soc . * 441 * , l6 ( 2014 ) d. anderhalden _ et al . _ , j. cosmol . astropart . phys . * 03 * , 014 ( 2013 ) d. n. spergel , p. j. steinhardt , phys . rev . lett . * 84 * , 3760 ( 2000 ) j. l. feng , m. kaplinghat , h. tu , h .- b . yu , j. cosmol . . phys . * 07 * , 004 ( 2009 ) n. arkani - hamed , d. p. finkbeiner , t. r. slatyer , n. weiner , phys . d * 79 * , 015014 ( 2009 ) j. m. cline , z. liu , g. d. moore , w. xue , phys . d * 90 * , 015023 ( 2014 ) m. vogelsberger , j. zavala , a. loeb , mon . not . . soc . * 423 * , 3740 ( 2012 ) j. zavala , m. vogelsberger , m. g. walker , mon . not . . soc . * 431 * , l20 ( 2013 ) m. vogelsberger , j. zavala , s. christine , a. jenkins , mon . not . . soc . * 444 * , 3684 ( 2014 ) a. kamada , n. yoshida , k. kohri , t. takahashi , j. cosmol . . phys . * 03 * , 008 ( 2013 ) c. bhm _ et al . _ , mon . not . soc . * 445 * , l31 ( 2014 ) m. r. buckley , j. zavala , f. y. cyr - racine , k. sigurdson , m. vogelsberger , phys . d * 90 * , 043524 ( 2014 ) k. sigurdson , m. kamionkowski , phys . lett . * 92 * , 171302 ( 2004 ) p. h. g. annika , c. e. moody , m. kamionkowski , phys . d * 81 * , 103501 ( 2010 ) m .- y . wang , a. r. zentner , phys . d * 85 * , 043514 ( 2012 ) m .- y . _ , mon . not r. astron . soc . * 445 * , 614 ( 2014 ) m .- y . et al . _ , phys . d * 88 * , 123515 ( 2013 ) s. das , n. weiner , phys . d * 84 * , 123511 ( 2011 ) r. fardon , a. e. nelson , n. weiner , j. high energy phys . * 03 * , 042 ( 2006 ) s. das , journ . series * 405 * , 012011 ( 2012 ) d. j. e. marsh , p. g. ferreira , phys . d * 82 * , 103528 ( 2010 ) r. hlozek , d. grin , d. j. e. marsh , p. ferreira , phys . d * 91 * , 103512 ( 2015 ) c. boehm , a. riazuelo , s. h. hansen , r. schaeffer , phys . d * 66 * , 083505 ( 2002 ) f .- y . cyr - racine , k. sigurdson , phys . d * 87 * , 103515 ( 2013 ) r. j. wilkinson , j. lesgourgues and c. boehm , j. cosmol . 04 * , 026 ( 2014 ) a. sarkar , s. das , s.k . sethi , j. cosmol . . phys . * 03 * , 004 ( 2015 ) planck collaboration : p. a. r. ade _ et al . _ , astron . astrophys . * 571 * , a16 ( 2014 ) planck collaboration : p. a. r. ade _ et al . _ , astron . astrophys . * 571 * , a15 ( 2014 ) a. lewis , a. challinor , a. lasenby , astrophys . j. * 538 * 473 ( 2000 ) b. a. reid _ et al . _ , mon . not * 404 * 60 ( 2010 ) a. v. maccio , o. ruchayskiy , a. boyarsky , j. c. mun@xmath141z - cuartas , mon . not . . soc . * 428 * , 882 ( 2013 ) a. a. staboninsky , jetp lett . * 55 * , 489 ( 1992 ) j. lesgourgues , d. polarski , a. a. starobinsky , mon . not . * 297 * , 769 ( 1998 ) r. teyssier , astron . & astrophys . * 385 * , 337 ( 2002 ) t. guillet , r. teyssier , j. comput . phys . * 230 * , 4756 ( 2011 ) s. prunet _ et al . _ , . ser . * 178 * , 179 ( 2008 ) m. crocce , s. pueblas , r. scoccimarro , mon . not . r. astron . soc . * 373 * , 369 ( 2006 ) k. heitmann _ et al . _ , j. * 715 * , 104 ( 2010 ) c. orban , phys . d * 90 * , 023509 ( 2014 ) y. rasera _ et al . _ , mon . not . . soc . * 440 * , 1420 ( 2014 ) f. roy , v. bouillot , y. rasera , astron . & astrophys . * 564 * , a13 ( 2014 ) m. davis , g. efstathiou , c. s. frenk , s. d. m. white , astrophys . j. * 292 * , 371 ( 1985 ) m. gotz , j. sommer - larsen , astrophys . space sci . * 281 * , 415 ( 2002 ) m. gotz , j. sommer - larsen , astrophys . space sci . * 284 * , 341 ( 2003 ) j. wang , s. d. m. white , mon . not . * 380 * , 93 ( 2007 ) s. agarwal , p .- s . corasaniti , phys . d * 91 * , 123509 ( 2015 ) r.k . sheth , g. tormen , mon . not . . soc . * 308 * , 119 ( 1999 ) j. tinker _ et al . _ , j. * 688 * 709 ( 2008 ) m. crocce , p. fosalba , f. j. castander , e. gazta@xmath140aga , mon . not . . soc . * 403 * , 1353 ( 2010 ) j. courtin _ et al . _ , mon . not . . soc . * 410 * , 1911 ( 2011 ) c. schultz , j. o@xmath140orbe , k. n. abazajian , j. s. bullock , mon . not . . soc . * 442 * , 1597 ( 2014 ) s. trujillo - gomez , a. klypin , j. primack , a. j. romanowsky , astrophys . j. * 742 * , 16 ( 2011 ) w.h . press , p. schechter , astrophys . j. * 187 * , 425 ( 1974 ) j. r. bond , s. cole , g. efstathiou and g. kaiser , astrophys . j. * 379 * , 440 ( 1991 ) . r. k. sheth , h. j. mo and g. tormen , mont . not . . soc . * 323 * , 1 ( 2001 ) p.s . corasaniti , i. achitouv , phys . lett . * 106 * , 241302 ( 2011 ) a. lapi , p. salucci , l. danese , astrophys . j. * 772 * , 85 ( 2013 )

to put new constraints on the r - mode instability window , we analyse formation of millisecond pulsars ( msps ) within the recycling scenario , making use of three sets of observations : ( a ) x - ray observations of neutron stars ( nss ) in low - mass x - ray binaries ; ( b ) timing of millisecond pulsars ( msps ) ; and ( c ) x - ray and uv observations of msps . as shown in previous works , r - mode dissipation by shear viscosity is not sufficient to explain observational set ( a ) , and enhanced r - mode dissipation at internal temperatures k is required to stabilize the observed nss . here we argue , that models with enhanced bulk viscosity can hardly lead to a self - consistent explanation of observational set ( a ) due to strong neutrino emission , which is typical for these models ( unrealistically powerful energy source is required to keep nss at the observed temperatures ) . we also demonstrate that the observational set ( b ) requires enhanced r - mode dissipation at low temperatures , k. observational set ( c ) allows us to set an upper limit on the red - shifted internal temperatures of msps , k ( assuming a canonical ns with the accreted crust ) . recycling scenario can produce msps at these temperatures only if r - mode instability is suppressed in the whole msp spin frequency range ( hz ) at temperatures k , providing thus a new constraint on the r - mode instability window . these observational constraints are analysed in more details in application to the resonance uplift scenario of . [ firstpage ]

1610.06380

according to the generally accepted recycling scenario suggested by @xcite , millisecond pulsars ( msps ) are descendants of neutron stars ( nss ) , which were spun up by accretion from roche - lobe filling companion stars in low - mass x - ray binaries ( lmxbs ) . this scenario has been extensively studied in the literature with the main emphasis at the neutron star spin frequency , evolution of the companion star , and parameters of the binary system ( e.g. , @xcite and references therein ) . ns temperature has not been typically considered as an important parameter for neutron star recycling , and is not even mentioned in most of the papers in the field . however , ns temperature is crucial for the instability of r - modes ( similar to rossby waves controlled by the coriolis force ) driven by chandrasekhar - friedman - schutz ( cfs , @xcite ) mechanism due to emission of gravitational waves ( @xcite ) . namely , if a rapidly rotating ns is hot enough , it becomes unstable with respect to excitation of r - modes ( see , e.g. , @xcite ; the corresponding region of ns temperatures and spin frequencies is often referred to as ` the instability window ' ) . this instability is known to modify the ns evolution in lmxbs dramatically ( see section [ sec_minmod ] and , e.g. , @xcite ) . in particular , it can limit rotation frequencies of nss ( @xcite ) . furthermore , we ( @xcite ) argued that the r - mode instability can lead to additional channel of ns recycling : formation of a new class of nss hot widows / hofnars ( from hot and fast non - accreting rotators ) in addition to msps . thus , accurate account for the r - mode instability should be important for the recycling scenario ; conversely , verification of the recycling scenario can put stringent constraints on the r - mode instability and , therefore , on the properties of superdense matter . in this paper we analyse constraints on the r - mode instability within the recycling scenario . we base our analysis on the following observations : * x - ray observations of transiently accreting neutron stars in lmxbs ( section [ sec_xray ] ) ; * timing of msps ( section [ sec_timing ] ) ; * x - ray and uv observations of msps ( section [ sec_xray_msp ] ) . we do not take into consideration observational data on binary system parameters ( orbital period , eccentricity etc . ) , because they seem to be determined by the evolution of binary system ( e.g. , @xcite ) and not affected directly by the r - mode instability . the set of observations ( a ) allows one to estimate the surface temperatures ( in the quiescent state , see , e.g. , @xcite ) and spin frequencies ( e.g. , @xcite ) for a number of accreting nss . it has been shown to be a crucial test of the r - mode instability theory ( @xcite ) . in particular , those references has demonstrated that the ` minimal ' model of r - mode instability window , suggested by @xcite ( ns has a nucleonic core , r - mode damping is associated with shear and bulk viscosities , see section [ sec_minmod ] ) , should be supplemented by additional damping mechanism at redshifted internal temperatures @xmath5 k , rather than by the local temperature @xmath6 , because it is @xmath7 , which is constant throughout the star . ] ( alternatively , saturation amplitude for r - modes should be extremely small , the so called ` tiny r - mode amplitude scenario ' , see section [ sec_tiny ] and , e.g. , @xcite ) . a number of papers is devoted to identification of the required damping mechanism ( see the recent review by @xcite ) . in particular , @xcite argues that finite temperature effects in superfluid ns cores can lead to resonance mode coupling and suppression of the r - mode instability at certain stellar temperatures . as a result , the instability window is splitted up by ` stability peaks ' in the vicinity of these temperatures . @xcite discuss evolution of nss in lmxbs within this model and suggest a ` resonance uplift scenario ' , which allows them to explain the observational set ( a ) . other ideas about the required additional damping include enhanced mutual friction ( e.g. , @xcite ) , crust - core coupling ( e.g. , @xcite ) , and exotic core composition ( e.g. , @xcite ) . concerning the observational set ( a ) we point out that any successful theory should not only stabilize the observed nss in lmxbs , but also explain their temperatures . msp timing [ set of observations ( b ) ; see section [ sec_timing ] ] has also been used to constrain the r - mode instability . @xcite discuss formation of msps assuming bulk viscosity driven suppression of r - mode instability . they predict that spin down of the fastest msps can be affected by r - modes . furthermore , these nss should have internal temperatures @xmath8 k , being the sources of thermal x - rays . @xcite analyse formation of recycled nss within the resonance uplift scenario and concluded that it does not contradict the idea that msps are descendants of nss in lmxbs . they also predict that the most rapidly rotating msps should be rather hot ( with internal temperatures @xmath9 k ) due to r - mode heating . @xcite argue that the minimal instability model requires extremely small r - mode saturation amplitudes to match the observed pulsars spin - down . @xcite discuss msp formation and conclude that compact stars with the cores composed of ` ungapped interacting quark matter ' are consistent with observations of lmxbs and msps ( see a critique of this statement in section [ sec_bulk ] ) . they also confirm that additional damping is required to describe msps by the models with nucleonic core composition . in this paper ( section [ sec_timing ] ) we present an independent argument in favour of enhanced r - mode damping at low temperatures , based on the msp internal heating mechanisms : superfluid vortex creep ( @xcite ) , rotochemical heating ( @xcite ) , and rotation - induced deep crustal heating ( @xcite ) . the x - ray and uv observations [ set ( c ) , section [ sec_xray_msp ] ] attract less attention in the r - mode literature , because the data for most of the observed msps do not require thermal emission from the whole surface ( but only from hot spots , see , e.g. , @xcite ) and the surface temperature is measured only for psr j0437 - 4715 by @xcite . for other msps the surface temperature is thought of as generally unknown ( see , e.g. , @xcite ) or not too high ( e.g. , @xcite ) . in this paper we argue that x - ray and uv observations allow one to put an upper limit on the redshifted surface temperature of msps , @xmath10 , and argue that @xmath11 k can be chosen as a fiducial value , applicable to all msps ( otherwise thermal emission from the whole surface should be observed , see section [ sec_xray_msp ] ) . it corresponds to an upper limit on the internal temperatures of msps : @xmath12 k ( here and below , to estimate internal temperatures of nss , we consider a canonical ns with the mass @xmath13 and @xmath14 km , and use the accreted envelope model from @xcite ) . this constraint is crucial for r - mode instability models , because any successful model should not only stabilize nss at these temperatures ( in the opposite case r - mode instability heats msps up to higher temperatures ) , but also explain how they came into this temperature region . in sections [ sec_tiny]-[sec_gen ] we apply observations ( a - c ) to constrain the r - mode instability parameters . namely , in section [ sec_tiny ] we improve constraints on the tiny r - mode amplitude scenario . in section [ sec_bulk ] we analyse models based on an enhanced bulk viscosity , in particular , the ungapped interacting quark matter model , suggested by @xcite , and conclude that it can hardly explain temperatures of the hottest nss in lmxbs . in section [ sec_reson ] we analyse the resonance uplift scenario . we demonstrate that low - temperature stability peaks at @xmath15 k are required by observations , and those resonances are likely to be broad . these results are extended to a quite general class of r - mode instability models in section [ sec_gen ] . we conclude in section [ sec_consl ] . during the final stage of the preparation of this manuscript we became aware of the work by @xcite , who also use x - ray observations to constrain msp temperatures and derive upper bounds on the r - mode amplitudes for these sources . the results of our section [ sec_tiny ] qualitatively agree with those of @xcite . the preliminary results of this work were presented by one of us ( aic ) at the two conferences : ( i ) ` the modern physics of compact stars and relativistic gravity 2015 ' ( september 30 october 3 , yerevan , armenia , 2015 ) ; and ( ii ) ` the international workshop on quark phase transition in compact objects and multimessenger astronomy : neutrino signals , supernovae and gamma - ray bursts ' ( october 7 october 14 , nizhnij arkhyz & terskol , russia , 2015 ) . in this section we describe the minimal r - mode instability model and confront it with each of the three sets of observations ( a)(c ) from section i. ) on the msp temperature . temperatures and frequencies of nss observed in lmxbs are shown by filled circles , error bars show uncertainties due to ns envelope composition ( section [ sec_xray ] ) . the pulsar b1937 + 21 is shown by star ; its low - temperature error bar is set by the rotation - induced deep crustal heating ( see section [ sec_timing ] ) , the upper error bar is set by the fiducial limit ( [ teffmax ] ) on the msp temperature . ticks on the left side of @xmath16 axis show measured msp frequencies . typical evolution track for a ns in lmxb is shown by thick black line , large filled circle marks the end of lmxb stage . dashed line demonstrates evolution track of a star in the msp stage . ] in the absence of damping the gravitational radiation reaction force leads to exponential growth of the r - mode amplitude on a timescale ( see , e.g. , @xcite ) @xmath17 here ns mass @xmath18 and radius @xmath19 are normalized to the mass and radius of the canonical ns : @xmath20 , @xmath21 , and @xmath22 . here @xmath23 is the ns spin rate . the minimal r - mode instability model suggested by @xcite includes shear and bulk viscosities as damping agents and assumes nucleonic equation of state . the shear viscosity damping time can be estimated as . ] @xmath24 where @xmath25 . the bulk viscosity damping time was calculated by @xcite : @xmath26 the total damping time is @xmath27 , however , the effect of bulk viscosity is negligible for temperatures of msps and nss in lmxbs . the r - mode is stable if @xmath28 ( grey region in fig . [ fig_minmod ] ) and unstable in the opposite case @xmath29 ( white region in fig . [ fig_minmod ] ) . in the unstable region the r - mode amplitude is growing until it reaches saturation , where the nonlinear effects become important . gravitational waves emitted by r - modes reduce stellar angular momentum @xmath30 at a rate @xmath31 the dissipation of r - mode leads to additional heating power ( e.g. , @xcite ) @xmath32 where @xmath33 for saturated and @xmath34 for unsaturated r - mode , respectively , and @xmath35 . general equations that describe evolution of a ns affected by r - mode instability were formulated by @xcite ( see also @xcite ) and the typical evolution track is shown by thick black line in fig.[fig_minmod ] . during the initial phase ns is spinned up by accretion and its temperature @xmath36 is determined by the balance between the cooling [ the corresponding emissivity is @xmath37 and deep crustal heating @xmath38 ( @xcite ) : @xmath39 here the coefficient @xmath40 determines efficiency of deep crustal heating . as shown by @xcite , when a ns enters the instability region due to continuous accretion , the r - mode amplitude starts to grow and its subsequent evolution is very fast ( provided that the saturation amplitude is not too small ) . eventually , the star rapidly heats up and slows down , leaving the instability region . in the stable region it gradually cools down to the temperature @xmath41 , and a new spin up cycle starts again . in principle , during its life a ns in lmxb can complete several such cycles but it is important that the probability to catch it in the instability region is negligible ( see also appendix b in @xcite for a set of analytical estimates ) . after roche - lobe decoupling ( see filled circle in the evolution curve in fig . [ fig_minmod ] ) a star cools down and an msp is formed ( msp evolution track is shown by dashed line ) . msp spin frequency is limited by the r - mode instability curve ( @xcite ) . x - ray observations of some of the transiently accreting nss allow one to measure their surface temperature in the quiescent state ( e.g. , @xcite ) . the internal temperature can be inferred from the surface temperature , assuming thermally relaxed envelope , with the help of analytical fitting formulas presented in @xcite . furthermore , the spin frequency can be measured for accretion- and nuclear - powered msps ( e.g. , @xcite ) . as a result , both spin frequency and internal temperature are available for a rather large number of objects ( @xcite ) . in addition , observations of transiently accreting nss in the active state allow one to estimate average accretion rate and the corresponding heating power associated with nuclear reactions in the accreted crust ( e.g. , @xcite ) . in this paper we employ the data for 20 nss from the table i of @xcite , where a number of misprints found in the previous works has been corrected ( see dots with error bars in fig . [ fig_minmod ] , temperature for hete j1900.1 - 2455 was updated according to @xcite , @xmath42k ) . at least 8 nss among them are predicted to be r - mode unstable within the minimal r - mode instability model . however , a probability to observe even one ns unstable is almost vanishing within the minimal r - mode instability model ( see section [ sec_minmod ] ) . thus , observations of large number of nss , which appears to be unstable , can be treated as evidence of additional ( apart from the shear viscosity ) damping at the temperatures of these objects , @xmath43 k , or r - mode saturation at very small saturation amplitude ( e.g. , @xcite ) . note that this problem can hardly be resolved by enhancing the shear viscosity ( e.g. , to stabilize 4u 1608522 one should enhance it by a factor of 1000 , which seems to be very unlikely ) or by fitting the mass - radius relation ( for example , 4u 1608522 should have a radius @xmath44 km to be stable within the minimal r - mode instability model ) . timing of msps allows one to measure the spin frequency ( see ticks on the left side of @xmath16 axis in fig.[fig_minmod ] ; only pulsars with @xmath45 hz are shown in the figure ) and the spin - down rate . these data were taken from atnf pulsar catalogue ( @xcite ) . within the minimal r - mode instability model msp spin frequencies should be limited by the r - mode instability . thus , origin of the high frequency msps ( especially those with @xmath46 hz ) is unclear in this model , because they should have very low equilibrium temperatures to be stable during the spin up phase in the lmxb stage . thus explanation of msps requires additional damping or use of a tiny saturation amplitude model ( see section [ sec_tiny ] and , e.g. , @xcite ) . a direct constraint on the r - mode instability comes from internal heating mechanisms superfluid vortex creep ( @xcite ) , rotochemical heating ( @xcite ) , and rotation - induced deep crustal heating ( @xcite ) which can be efficient enough to push psr b1937 + 21 ( j1939 + 2134 in j2000 system ) into the r - mode instability window ( see fig.[fig_minmod ] ) . to estimate the internal heating power @xmath47 for msps we rely on the rotation - induced deep crustal heating mechanism , because it is not affected by the uncertainties in the parameters of baryon superfluidity . physics of rotation - induced deep crustal heating is rather simple . msp spin down ( e.g. , by magneto - dipole losses ) results in a gradual compression of stellar matter . within the recycling scenario paradigm , the crust of msps should consist of accreted matter , thus the compression should lead to the same nuclear transformations in the crust as in the case of accreting nss ( i.e. , to deep crustal heating ) . for example , for psr b1937 + 21 which has high frequency ( @xmath48 hz ) and relatively fast spin - down rate ( @xmath49 s@xmath50 ) , the rotation - induced deep crustal heating gives @xmath51 ergs@xmath52 , depending on the equation of state and pulsar mass . to balance this heating power at the lowest possible temperature ( in order to get it stable ) , a powerful cooling process should be active in the core of this source . to get an impression of the possible realistic cooling luminosity , we adopt here a simple phenomenological model , which assumes that the unsuppressed hyperon direct urca process [ see , e.g. , @xcite ] is active in the central region of the core with the radius @xmath53 , where hyperons are present ( see @xcite for a detailed motivation ) . the corresponding luminosity is3658 ( @xmath54 hz ) can indicate that even more violent cooling is possible . the observations put an upper limit on the internal temperature @xmath55 k ( e.g. , @xcite ) , but the deep crustal heating produces @xmath56 ergs@xmath52 for the average accretion rate @xmath57/yr , reported by @xcite . ] @xmath58 for @xmath59 km the equilibrium temperature of psr b1937 + 21 , at which @xmath60 , can be estimated as @xmath61 k. in the absence of direct urca , the equilibrium temperature becomes larger , so that the source is well inside the instability window ( see fig . [ fig_minmod ] , where it is shown by the star symbol ) . note that some other internal heating mechanisms ( see , e.g. , @xcite for a review ) can be competitive with the rotation - induced deep crustal heating , further increasing the temperature of psr b1937@xmath6221 and making it even more unstable . however , as explained in section [ sec_minmod ] , r - modes rapidly heat up an unstable star ( on a timescale of a century ) and spin it down . thus , observations of psr b1937@xmath6221 contradict the minimal r - mode instability model : to explain this source one should stabilize it by introducing an additional damping mechanism efficient at @xmath63 k. the ns temperature is a crucial parameter for the r - mode instability and hence any observational constraint on the stellar temperature can be very important . a large number of msps has been observed in x - rays , however , the thermal emission from them has been either not detected or detected only from pulsars hot spots ( e.g.,@xcite ) . thermal emission from the surface is typically not required to fit the x - ray spectrum . a notable exception is psr j0437@xmath644715 ( @xmath65 hz ) , for which combined analysis of uv and x - ray spectra by @xcite allows to estimate the surface temperature , @xmath66 k. the corresponding internal temperature is @xmath67 k and thus this pulsar is r - mode stable . @xcite have recently suggested a strong upper limit , @xmath68 k , on the surface temperature of psr j1231 - 1411 ( @xmath69hz ) within the blackbody model . as the authors emphasize , they do not account for gravitational redshift and ignore the effects related to stellar atmosphere , which can , in our opinion , significantly affect the actual value of the upper limit ( see , e.g. , @xcite ) . @xcite have also proposed upper limits on the surface temperature of several other msps based either on estimates of their total x - ray thermal luminosity ( available in the literature ) or on the results of @xcite , who estimated maximum possible @xmath70 for a number of sources . here we suggest a general upper limit on the surface temperatures of msps . for several pulsars ( e.g. , for psr 1937 + 21 ) our upper limit is stronger than that suggested by @xcite . the msp surface temperatures are strongly constrained by existing observations : if msps were as hot as transiently accreting nss , whose thermal emission is observed in x - rays even in the quiescent state ( see section [ sec_xray ] and fig . [ fig_minmod ] ) , the thermal emission from the whole msp surface would be detected in x - rays as well . but it is not the case . thus , leaving a detailed analysis of x - ray observations for each msp beyond the scope of this paper , we assume that msps should ( generally ) be colder than the transiently accreting nss with the measured surface temperature . to get an idea of the possible upper limit on the msp surface temperature we rely on observations of mxb 1659@xmath64298 ( assumed distance to the source @xmath71 kpc and x - ray absorption column density @xmath72 cm@xmath50 ) , igr 00291 + 5934 ( @xmath74 kpc and @xmath75 cm@xmath50 ) , and hete j1900.12455 ( @xmath76 kpc and @xmath77 cm@xmath50 ) , reported by @xcite , @xcite , and @xcite . for all of them the effective surface temperature was estimated as @xmath78 k. note that , for these objects both absorption column density and distances are relatively large in comparison to typical fast msps ( with @xmath79 hz ) . : @xmath80 , where @xmath81 is the dispersion measure . thus , msps with @xmath82 should have @xmath83 cm@xmath50 , a requirement , which is satisfied for more than @xmath84 of currently known fast msps ( @xmath79 hz ) , according to atnf pulsar catalogue ( @xcite ) . this catalogue also gives an information on the pulsar distances , and it turns out that 70% of known fast msps ( @xmath79 hz ) are located closer than hete j1900.12455 , but this statistic is not independent because the pulsar distances are estimated using the @xcite dm - based distance model as default . ] thus , it is unlikely that lack of successful observations of thermal emission from the whole surface of fast msps can be explained by strong absorption or large distances to these sources : rather , it is more probable that they are ` cold ' ( colder than mxb 1659@xmath64298 , igr 00291 + 5934 , and hete j1900.12455 ) . therefore , we choose @xmath85 as a fiducial upper limit for the effective surface temperature of msps . this temperature corresponds to the bolometric thermal luminosity @xmath86 ergs@xmath52 [ see equation ( [ l_ph ] ) below ] . the typical x - ray luminosity of msps is @xmath87 ergs@xmath52 ( @xcite ) and should correspond to even lower surface temperatures . the x - ray luminosity of msps with high spin - down power is higher . however , they have power - law x - ray spectrum and high fraction of pulsed emission ( see , e.g. , section 4.2 by @xcite ) . in this paper we are interested in the internal temperatures , which reveal itself in thermal emission from the whole surface , contributing to the unpulsed component of x - ray emission . thus , analyzis of the phase - averaged spectra can lead to overestimation of @xmath88 . for example , for psr b1937 + 21 , which plays an especially important role in this paper , @xmath89 ergs@xmath52 , but it is almost 100% pulsed ( @xcite ) . thus the unpulsed luminosity is much weaker ( numerical upper bound is not reported by @xcite ) , and temperature of this pulsar can hardly exceed the constraint ( [ teffmax ] ) . note that @xcite have likely used total x - ray luminosity ( @xmath90 ergs@xmath52 , quoted in their table 1 ) to obtain an upper limit on the surface temperature of psr b1937 + 21 . it can be a reason why their result ( @xmath91k , according to their figure 1 ) exceeds ( [ teffmax ] ) . high upper bounds for the surface temperature of some msps reported by @xcite ( also quoted in table 2 in @xcite ) correspond to pulsars undetected in x - rays ; thus they can not be considered as evidence of violation of ( [ teffmax ] ) . to further justify our constraint ( [ teffmax ] ) , let us consider the globular cluster 47 tuc as an example . this globular cluster is good for x - ray studies , because it is close ( @xmath92 kpc ) and weakly absorbed ( cluster @xmath93 cm@xmath50 , e.g. , @xcite ) . @xcite reported on the observations of three qlmxbs in this cluster , including the source w17 with @xmath94 k , and confirmed that nss with such surface temperature are observable . @xcite found x - ray counterparts for all 19 msps with precisely known positions in this globular cluster ( msp pairs 47 tuc f and s , and g and i were treated as single sources ) . many msps [ including 47 tuc j ( @xmath95 hz ) , y ( @xmath96 hz ) , and w ( @xmath97 hz ) , which were the fastest msps in this globular cluster known at that moment ] hz . however , its position has not been determined yet , thus we are unable to identify it among the x - ray point sources catalogued by @xcite . ] were well described by a purely thermal spectrum or composite thermal and power - law emission . the size of thermally emitting area was small ( @xmath98 km for the hydrogen atmosphere model ) . x - ray spectra for four relatively slow rotating msps [ namely , 47 tuc d ( @xmath99 hz ) , e ( @xmath100 hz ) , l ( @xmath101 hz ) , and r ( @xmath102 hz ) ] require a two - temperature model . @xcite pointed out , that limited statistics do not allow to put reliable constraints on the emission areas , but x - ray spectra can be fitted with the same two temperatures as used in @xcite for psr j04374715 , with the low ( nonredshifted ) surface temperature @xmath103 k corresponding to the emission radius @xmath104 km . note that this temperature agrees with our upper limit ( [ teffmax ] ) and the emission area is much smaller than the size of a ns . thus , the whole surface for these msps should have even lower temperatures ( otherwise the corresponding emission would be required by the fit ) . @xcite noted that the thermal emission from all msps in 47 tuc can , in fact , consist of two components ; if correct , this means that the effective temperatures and emission radii obtained within a single thermal component model represent a weighted averages of two emitting regions . this may indicate that the thermal emission from the whole msp surface should not exceed the total thermal emission , which was estimated by @xcite as @xmath105 ergs@xmath52 . it suggests that even stronger upper bound can be put on the surface temperature of msps in 47 tuc . however , in this paper we shall adopt the conservative general upper limit ( [ teffmax ] ) and demonstrate that even it constrains the r - mode instability window . it is quite likely that accurate and detailed analysis of x - ray and uv observational data can put even tighter upper bounds on the temperatures of the observed msps , allowing one to apply the analysis of sections [ sec_tiny][sec_gen ] to extract more information about the properties of these sources . for stars with accreted envelope composition ( a reasonable option , since they were recycled by accretion ) , , even if actual layer of light elements is very thin ( see , e.g. , fig . 8 in @xcite ) . ] equation ( [ teffmax ] ) translates into the following condition on the redshifted internal stellar temperature , @xmath106 ( see hatched region in fig . [ fig_minmod ] ) . this condition indicates that the fastest msps ( with @xmath107 hz ) can be unstable within the minimal r - mode instability model . thus , to describe them , an additional damping is likely required in this temperature region . within the tiny r - mode amplitude scenario , nss can be unstable , but saturation amplitudes are chosen small enough to agree with observations . @xcite confronted this model with the observations of nss in lmxbs and concluded that the r - mode amplitudes should range from @xmath108 to @xmath109 . @xcite apply msp timing to constrain r - mode amplitudes in msps . for some msps with the slowest spin - down rate ( e.g. , @xcite ) . [ shklovskii ] ] assuming power - law dependence of relevant parameters on @xmath16 , @xmath7 , @xmath18 , and @xmath19 , they suggest ` universal r - mode spin - down limit ' on gravitational wave signal from msps . however , at least for neutrino luminosity this assumption seems to be oversimplified : simulations of stellar cooling and quiescent temperatures of accreting nss require different cooling mechanisms to be active in nss with different masses leading to very strong ( non - power - law ) dependence of luminosity on the mass and temperature [ see , e.g. , @xcite ] . thus , universal r - mode spin - down limit suggested by @xcite should be treated with a bit of caution . furthermore , as we demonstrate below , upper boundary ( [ teffmax ] ) for the surface temperature of msps can constrain the r - mode amplitudes much stronger . this is in line with @xcite who have recently come to the similar conclusion . for saturated r - mode , the heating power can be estimated from equation ( [ heatrate ] ) . in thermal equilibrium the total heating ( sum of all heating mechanisms : r - mode , rotation induced deep crustal heating , rotochemical heating , etc . ) should be compensated by cooling . if an msp is not massive enough to open direct urca process in its core , the neutrino emission is likely to become negligible [ due to temperature constraint ( [ tmax ] ) ] , so that the total cooling luminosity can be estimated as the thermal emission from the msp surface @xmath110 where @xmath111 is the stefan - boltzmann constant . here and below the red - shift factor , @xmath112 , is assumed to be @xmath113 . this value corresponds to the canonical ns . the upper limit on the r - mode amplitude can be computed assuming that r - modes are the only source of heat [ see equation [ heatrate ] ] . , @xmath14 km , and @xmath114 , where @xmath115 is the unredshifted ( local ) effective temperature . ] @xmath116 for massive nss the direct urca processes can operate in the central region of the core . to estimate the corresponding neutrino luminosity we make use of equation ( [ l_durca ] ) and the relation between the surface and internal temperatures for fully accreted envelope , calculated by @xcite . the result is @xmath117 where @xmath118 is the surface gravity ( for the canonical ns @xmath119 ) . the luminosity ( [ l_durca_teff ] ) results in a generally weaker constraint on the r - mode amplitude , @xmath120 if one allows for both cooling processes ( photon and direct urca ) , one finds the following upper limit on the r - mode amplitude , @xmath121 note that the constraint ( [ a_gen ] ) is tighter than the upper bound found by @xcite from the analysis of msp spin - down rate . thus , r - mode can be responsible only for a small fraction of the total spin down rate @xmath122 ( see @xcite for a detailed discussion ) . if we reduce the observational upper limit for @xmath123 , we will obtain a stronger constraint on the r - mode amplitude . for example , for @xmath124 k ( an upper limit for the surface temperature of the pulsar j0437@xmath644715 ) , the photon emission from the stellar surface dominates . thus , if it will be shown that there exists a cold enough rapidly rotating msp , then it would be possible to set an upper limit on the r - mode amplitude for this pulsar _ directly from observations _ [ by using eq . ( [ a_ph ] ) ] , without referencing to rather uncertain ( but negligible at @xmath125 k ) neutrino emission from the star . the upper limit ( [ a_gen ] ) on the r - mode amplitude in msps is too strong to be explained by previously suggested mechanisms of nonlinear saturation of r - modes ( the only exception is the work by @xcite , but it assumes a hybrid star with sharp hadron - quark interface and thus it is inapplicable to hadronic equations of state ) . note also that the saturation amplitude generally increases with temperature decrease ( e.g. , @xcite ) . thus , an upper limit for the low - temperature msps is , in fact , a stronger constraint than numerically the same upper limit for hot nss in lmxbs . hence , explanation of msps in the tiny r - mode scenario seems to be unrealistic . in the end of the section it is worth noting that the estimates obtained above constrain the r - mode amplitude even if it is limited not by saturation , but by the shape of the instability window ( e.g. , within the resonance uplift scenario , see sections [ sec_reson ] and [ sec_gen ] ) as discussed in section [ sec_observ ] , shear viscosity is not sufficient to stabilize r - modes in the observed nss . can this problem be solved by the bulk viscosity ? for nucleonic equations of state the bulk viscosity is very small at @xmath126 k , being negligible in comparison with the shear viscosity . however , for another core composition the bulk viscosity can be much larger . for example , @xcite demonstrates that the model of ungapped interacting quark matter can successfully stabilize nss in lmxbs . this model predicts that the fastest msps ( with @xmath127 hz ) should have internal temperatures @xmath128 k ( see their fig . 1 ) and thus slightly contradicts the internal msp temperature constraint ( [ tmax ] ) . ) for @xmath129 is fulfilled , but this figure corresponds to the ground state composition of the envelope , which does not look realistic in case of msps . for accreted envelope the ungapped interacting quark matter stability curve shifts to higher surface temperatures and violate the constraint ( [ teffmax ] ) . ] probably the more important problem of @xcite is high neutrino luminosity , @xmath130 ergs@xmath52 , inherent to their model . gev , @xmath131 gev . note that these constants should not be confused with the constants denoting nonperturbative scale of qcd ( @xmath132 mev ) and scale of electroweak symmetry breaking @xmath133 gev ( e.g. , @xcite ) , which often have the same notation . ] for example , 4u 1608@xmath64522 is r - mode stable in ungapped interacting quark model . thus , the main heat source for that ns is deep crustal heating . it provides @xmath134 ergs@xmath52 ( for the average accreting rate @xmath135 , see @xcite ) , corresponding to the equilibrium temperature @xmath136 k. however , this temperature is strongly below the observational estimate , @xmath137 k ( see , e.g. , @xcite ) . to keep 4u 1608@xmath64522 at the observed temperature there should be an additional heat source , compensating the neutrino luminosity @xmath138 ergs@xmath52 . the nature of such source is unclear , especially for stable nss . interpretation of temperatures of other hot nss ( with @xmath0 k ) faces the same problem . the enhanced neutrino emission , leading to problems with explanation of hot nss in lmxbs , should be common for models , which stabilize nss by the bulk viscosity . the reason is simple : both neutrino luminosity and bulk viscosity are associated with the reactions between the particles in the ns core ( see , e.g. , @xcite ) . if all these reactions are accompanied by neutrino production ( as it is the case for nucleonic matter ) , there should be a direct correspondence between the bulk viscosity ( for small - amplitude oscillations ) and neutrino emissivity ( see , for example , equations 1 and 2 in @xcite , written for the modified urca reactions ) . this means that enhancement of bulk viscosity unavoidably results in an enhancement of neutrino luminosity . for more complicated core compositions ( e.g. , for nucleon - hyperon or quark matter ) non - leptonic reactions are possible ( e.g. , @xcite ) that do not involve neutrinos and , being more powerful than neutrino reactions , produce dominant contribution to the bulk viscosity ( e.g. , @xcite , @xcite ) . thus , the ratio of the bulk viscosity energy dissipation rate to luminosity can be enhanced in comparison to the nucleonic matter . however , to stabilize the fastest nss in lmxbs , like 4u 1608 - 522 , bulk viscosity should be enhanced at least by 13 orders of magnitude ( cf . equations [ tau_gw ] and [ tau_b ] ) , which seems to be hardly possible without substantial increase of the neutrino luminosity . at the same time , if all nss in lmxbs are r - mode stable , the quiescent temperatures of the hottest of them can be explained _ only _ if neutrino luminosity is not too strong , i.e. , it should be on the level of the modified urca luminosity for nucleonic cores ( see , e.g. , @xcite ) . the above arguments are , of course , not rigorous enough , but we want to stress that each particular model with r - mode stabilization by enhanced bulk viscosity should be checked for consistency with the observed quiescent temperatures of the hottest nss in lmxbs ( in particular , 4u 1608 - 522 ) . resonance uplift scenario was suggested by @xcite to explain observations of nss in lmxbs . it is based on the proper account of finite temperature effects on the dynamics of superfluid liquid in the core and predicts an enhanced ns stability associated with the resonance interaction of oscillation modes in the vicinity of certain ( resonance ) temperatures ( see fig.[fig_uplift ] , where the instability window is splitted by stability peaks ) . as discussed by @xcite , this picture agrees with detailed calculations of frequencies and damping times for : ( i ) r - modes and inertial modes of superfluid nss at zero temperature ( @xcite ) ; and ( ii ) f- and p - modes in nonrotating superfluid nss at finite temperatures ( @xcite ) . furthermore , @xcite confirm the main features of the resonance uplift scenario by calculating the r - mode spectrum and damping times for a finite - temperature superfluid ns within a simplified model with vanishing entrainment . however , the form of the stability peaks depends on many parameters ( e.g. , on the critical temperature profiles , equation of state etc . ) and remains rather uncertain . here we shall try to constrain it by making use of the observations discussed in section [ sec_observ ] . as argued in @xcite , the fastest msps should be attached to the left slopes of some of the existing stability peaks . note that those peaks ( more accurately , their left slopes ) should appear at @xmath139 in order to match the observational upper limit on msp temperature ( [ tmax ] ) . here , within the phenomenological approach , we discuss two models : ( a ) a model with _ narrow _ low - temperature stability peaks , for which the cooling luminosity @xmath140 does not change significantly inside the peak [ see panel ( a ) in fig . [ fig_uplift ] and sec . [ sec_narrow ] ] and ( b ) a model with _ broad _ low - temperature stability peaks [ see panel ( b ) and section [ sec_broad ] ] . formally , both models can explain the current state of the observed rapidly rotating nss : each ns in an lmxb is either attached to the left slope of some stability peak ( e.g. , 4u 1608 - 522 ) or is stable with respect to r - modes ( e.g. , xte j1814 - 338 ) , as it should be in the resonance uplift scenario ( e.g. , @xcite ) . high - frequency msps [ with @xmath141 hz for model ( a ) and @xmath127 hz for model ( b ) ] can be explained as moving along the left slope of the low - temperature stability peak , other msps should be stable . in both cases , msp temperatures agree with observations ( @xmath142 ) . spin frequency evolution is almost unaffected by r - modes even for high - frequency msps ( see a discussion after equation [ a_gen ] ) . however , the recycling scenario requires that msps should be formed in lmxbs . in the following subsection we analyse constraints on the models ( a ) and ( b ) set by this requirement . note that , generally , the resonance temperatures can depend not only on the microphysics input ( i.e. , on the equation of state , critical temperature profile , etc . ) , but also on the stellar mass ( e.g. , the appearance of hyperons in sufficiently massive nss can change the form of the instability window dramatically ) . thus , we can not exclude that both possibilities ( a ) and ( b ) are realized for different ns masses . typical evolution path , which should lead to formation of low - temperature msps within the model ( a ) , is shown by solid line in panel ( a ) of fig . [ fig_uplift ] . according to the resonant uplift scenario ( see @xcite for details ) , at the initial stage of recycling , an ns is stable and spun up by accretion ; its temperature is determined by the balance between the deep crustal heating and cooling processes [ @xmath36 , where @xmath41 is the solution to equation ( [ teq ] ) ] . if spin up is strong enough , ns can cross the boundary of the stability region . this leads to excitation of r - mode and accompanying heating of a star . as a result , its temperature increases however , growth of the temperature is limited by the boundary of the stability peak , because inside the peak the star is stable and does not have a heat source to further increase @xmath143 . as a consequence , subsequent ns evolution takes place along the boundary of the stability peak . the r - mode amplitude is determined by the heating power required by the thermal equilibrium condition ( see @xcite for details ) : @xmath144 to be associated with the low - temperature stability peak , located at @xmath145 [ @xmath146 is the ` resonance ' temperature , see fig . [ fig_uplift](a ) ] , the equilibrium temperature should be @xmath147 ( in the opposite case the ns evolution in the unstable region will be associated with next stability peak at @xmath148 ) . at the same time , to match the upper limit ( [ tmax ] ) on the msp temperature , one should have @xmath149 k. in this way we obtain the first important constraint for the model ( a ) [ see equation ( [ teq ] ) ] : @xmath150 accretion can not spin a star up above the equilibrium spin frequency @xmath151 , where the net torque acting on the star is zero ( for @xmath152 the torque is negative ) . the exact numerical value of the equilibrium spin rate should be determined from the accretion theory , but it is still a subject of debates ( e.g. , @xcite and references therein ) . the standard model assumes that the spin equilibrium takes place when keplerian disc corotates with ns at the magnetosphere boundary ( see , e.g. , @xcite for details ) @xmath153 here @xmath154 is the dipolar magnetic moment of the star ; @xmath155 and @xmath156 are the numerical coefficients , which describe radius of the magnetosphere and spin - up efficiency . @xmath157 for @xmath158 gives an upper limit on the equilibrium frequency , because , as noted by @xcite , for higher spin frequency the centrifugal force prevents accretion ( the so called `` propeller regime '' ) so that further spin up of a ns is unlikely . note that , if accreting ns is attached to the peak and accretion rate is equal to @xmath159 , the equilibrium r - mode amplitude ( and the corresponding heating power and breaking torque ) are all vanish [ see equation [ thermequl ] ] . thus , at such accretion rate the r - mode instability does not affect the equilibrium frequency @xmath160 . for lower accretion rate , @xmath161 is smaller and the r - mode braking torque can only further decrease it . thus , the maximal equilibrium frequency can be estimated as @xmath162 . in real lmxbs accretion rate is evolving in time , leading to the evolution of the equilibrium spin rate ( e.g. , @xcite ) . the ns spin frequency @xmath163 can not exceed maximum equilibrium spin frequency @xmath157 , which was achieved during the evolution . as far as the accretion rate can not exceed @xmath159 at any moment of the evolution , the upper bound on the ns spin frequency in lmxb is @xmath164 . thus , the frequency of a newly born msp , @xmath165 , also can not exceed @xmath164 . in fact , as it was shown by @xcite , @xmath165 can be even lower because msp can lose more than a half of its rotational energy during the roche lobe decoupling phase . since at the msp stage there are no accretion and any spin up torques , the stellar spin frequency can only decrease , and , as a result , it should be a subject of our second constraint , @xmath166 where the numerical parameter @xmath167 describes spin down of msp during the roche lobe decoupling phase . the third constraint comes from the requirement that the spin - down power , associated with r - modes , @xmath168 , should not exceed the total spin - down power of msps , @xmath169 . consider a ns evolving along the boundary of stability peak . then the r - mode heating power is related to @xmath170 by the formula @xmath171 which follows from equations ( [ jgr ] ) and ( [ heatrate ] ) , applied to the boundary of instability window with @xmath172 . in addition , such star should be in thermal equilibrium , which means that its luminosity at the peak temperature @xmath146 should be equal to @xmath173 , i.e. , in view of equation ( [ wr_spindown ] ) , @xmath174 where @xmath175 is ( unknown ) parameter , which determines fractional contribution of r - mode spin down at the present moment . here we neglect other mechanisms of internal heating , which can contribute to the heating power only at a level of @xmath176 ( e.g. , @xcite ) , becoming comparable to the r - mode heating only for very low values of @xmath177 . in the remaining part of this section we put a lower limit on @xmath178 following from the three constraints , discussed above , and confront this result with observations . according to @xcite , the spin - down power associated with magnetic field , @xmath179 , can be estimated as @xmath180 here @xmath181 is the magnetic dipole inclination angle and we assume that the only torques acting on an msp are associated with r - modes and magnetic field ( i.e. , @xmath182 ) . one can express @xmath154 from this equation and substitute it into equation ( [ nu_std ] ) along with @xmath183 [ given by equations ( [ dotm_lmxb_constr ] ) and ( [ l_msp ] ) ] , to get @xmath164 . the result is @xmath184 note that @xmath185 is cancelled out in this final equation . it holds true for any accretion model ( in particular , for the recent model of @xcite ) , provided that the equilibrium frequency depends on @xmath186 and @xmath154 only in combination @xmath187 . equation ( [ nu_std_fin ] ) combined with constraint ( [ nu_msp_constr ] ) allows one to put a lower limit on the parameter @xmath178 , @xmath188^{-1}. \label{a_min}\ ] ] of the r - mode spin - down power to the total spin - down power in @xmath189 plane . to plot the figure we choose @xmath190 hz , @xmath191 , and @xmath13 . ] the isolines of @xmath192 are shown in fig.[fig_a_min ] . msp with @xmath13 , @xmath190 hz , and @xmath191 is taken as an example . only in the most efficient spin - up scenario ( @xmath193 , @xmath194 ) @xmath192 can be as low as 0.05 . if at least one of the parameters of the accretion model is not so optimistic , @xmath192 is much larger . the larger value of @xmath195 is favoured for many reasons . in particular , @xcite argue that @xmath196 and @xmath197 for at least some of msps ; the most efficient model ( @xmath198 ) does not account for spin - down both during the roche lobe decoupling phase ( @xcite ) and at the msp stage ; recent model by @xcite suggests that @xmath199 should be very small for small @xmath200 ; finally , smaller angle @xmath181 leads to higher @xmath192 . furthermore , to obtain an upper limit for the spin frequency ( [ nu_msp_constr ] ) we assume accretion rate to be equal to its upper limit , @xmath183 . if real @xmath186 is lower , the equilibrium spin frequency is lower too , leading to a stronger lower bound on @xmath178 . all in all , we believe that a realistic value of @xmath192 should exceed @xmath201 at least by a factor of few . let us confront the constraint ( [ a_min ] ) with observations . unfortunately , direct detection of gravitational wave signal from r - modes is impossible , because ( at least , for high frequency msps , @xmath202 ) it is below the sensitivity level of advanced ligo due to very small r - mode amplitudes allowed in msps ( see section [ sec_tiny ] and , e.g. , @xcite ) . another almost straightforward procedure to measure @xmath178 is associated with the so called @xmath203-oscillations modulations of r - mode amplitude , which are unavoidable for msps evolving along a narrow stability peak ( @xcite ) . this modulation leads to ` anti - glitches ' sudden drops of msp frequency on a time - scale of hours - months . such features have not been found yet in msp timing , thus observations can be used to put an upper limit on @xmath178 . indeed , in the limiting case @xmath204 , almost all msp spin - down should be associated with ` anti - glitch ' events , which clearly contradicts observations . at the same time , our preliminary results indicate that @xmath205 can agree with timing residuals for psr b1937 + 21 , reported in @xcite . we leave the problem of accurate determination of an upper limit for @xmath178 to subsequent studies . constraint ( [ a_min ] ) can also be checked indirectly . according to equation ( [ wr_spindown ] ) , r - mode contribution to the spin down unavoidably leads to ns heating . thus , equation ( [ a_min ] ) determines the minimum heating power for msp evolving along a narrow low - temperature stability peak , @xmath206 this heating can be too strong to be balanced by cooling at low temperature . for example , taking the moment of inertia @xmath207 g@xmath208 , psr b1937 + 21 has @xmath209 ergs@xmath52 . thus , for the lowest possible value of @xmath210 , the minimal r - mode heating power is @xmath211 ergs@xmath52 . even assuming that fully unsuppressed hyperon urca is open in the central region with the radius @xmath59 km , this heating power is sufficient to heat the star up to a temperature @xmath212 k [ see equation ( [ l_durca ] ) ] , which exceeds the upper limit ( [ tmax ] ) by a factor of 1.5 . as noted above , if accretion spin - up is not extremely efficient , @xmath192 should be much larger , substantially complicating explanation of msps with @xmath213 erg s@xmath52 . thus , the model with narrow low - temperature resonance peaks should be rejected ( at least ) for pulsars with high spin - down power . but what about the model ( b ) with broad low - temperature stability peaks ? typical evolution path in case of broad low - temperature peaks [ model ( b ) ] is shown in panel ( b ) of fig . [ fig_uplift ] . during the lmxb stage an ns evolves inside the stability peak . its temperature is determined by the deep crustal heating ( see equation [ teq ] ) and the spin frequency is controlled by accretion . after roche - lobe decoupling [ filled circle in panel ( b ) of fig.[fig_uplift ] ] accretion stops , and the ns becomes an msp , which cools down by photon and neutrino emission . if its frequency is high enough [ @xmath214 hz for the example , shown in fig.[fig_uplift](b ) ] , cooling stops at the left boundary of the stability peak and the r - mode excites up to an amplitude required to keep ns in thermal equilibrium [ see equation ( [ thermequl ] ) with @xmath215 . in the opposite case , msp stays stable ( r - mode is not excited ) and evolves to the left until the thermal equilibrium with other ( less effective ) heating processes ( see section [ sec_timing ] ) is established . in order to explain observations of msps the model ( b ) should satisfy two main requirements : \(a ) temperature of the slope of the low - temperature stability peak should not exceed the observational constraint ( [ tmax ] ) in the whole msp frequency range ; \(b ) accretion should be able to spin the star up to high frequencies _ within _ the low - temperature peak [ in the opposite case , immediately after an ns reaches the high - temperature boundary of the peak , the r - modes will be excited , heating the star up to the next ( high - temperature ) stability peak , so that the subsequent evolution will be associated with that peak ] . the constraint ( [ a_min ] ) is unjustified in case of broad low - temperature resonances , because the cooling luminosity decreases significantly during cooling of a newly born msp inside the stability peak . as a result , the contribution of r - modes to the spin down of msps can be very small . both requirements ( a ) and ( b ) can be satisfied by adjusting the parameters of the low - temperature stability peak . for example , in fig . [ fig_uplift](b ) the requirement ( a ) holds true up to @xmath216 hz . the requirement ( b ) can be satisfied by adjusting the high - temperature slope of stability peak , as it was done in fig . [ fig_uplift](b ) . indeed , the fact that a number of high - frequency accreting nss ( which are progenitors of msps according to the recycling scenario ) are contained inside the broad stability peak , can serve as an indirect evidence that the requirement ( b ) is fulfilled . we can also check the consistency of a broad - peak model with requirement ( b ) for certain accretion model . for example , the standard accretion model predicts the following minimum accretion rate , required to spin up an ( r - mode stable ) msp with the observed @xmath16 and @xmath217 , @xmath218 [ see , e.g. , @xcite ; the same result can also be derived from equations ( [ nu_std ] ) , ( [ nu_msp_constr ] ) , and ( [ p_mag ] ) ] . this accretion rate corresponds to the minimal power of deep crustal heating , @xmath219 where in the last equality we assume @xmath220 , @xmath221 , @xmath222 , @xmath223 , and @xmath224 g@xmath208 . for a ( stable ) ns inside the stability peak this heating should be compensated by cooling , providing a constraint on the _ minimal _ ns temperature at the lmxb stage , @xmath225 , where @xmath226 is found from @xmath227 for example , the strongest ( among the known msps ) @xmath228 ergs@xmath52 corresponds to psr b1937 + 21 ( assuming @xmath229 ) . it can be compensated by unsuppressed hyperon direct urca ( assuming @xmath59 km ) at @xmath230 . thus , the temperature of this ns at lmxb stage should not be less than this value . at @xmath231 k the high frequency boundary of the broad stability peak in fig . [ fig_uplift](b ) corresponds to @xmath232 hz , thus psr b1937 + 21 ( with @xmath48 hz ) could , in principle , be spun up by accretion on the lmxb stage , without leaving out this stability peak . to explain nss observed in lmxbs with higher temperatures , our model ( b ) requires also high - temperature stability peaks . as it is argued in @xcite , nss associated with such peaks evolve not to msps , but to so called ` hot widows / hofnars ' ( see that reference for a detailed argumentation and observational evidences ) . the lower bound on the r - mode contribution to the spin - down ( equation [ a_min ] ) can also be applied to hot widows / hofnars , but it is not restricting , because their spin - down can , in fact , be generated purely by r - modes . @xmath203-oscillations can be suppressed for ns evolving along the broad stability peaks ( @xcite ) . thus , absence of ` anti - glitches ' would not allow one to draw any conclusion about validity of the model with broad stability peak . however , if anti - glitches were observed , it would provide strong constraints on the parameters of the stability peaks . the constraints on the resonance uplift scenario obtained in section [ sec_reson ] appeal only to the shape of the instability window , but not to the actual physics which determines that shape . thus , our results are equally applicable to arbitrary r - mode instability model with similar morphology of the stability region ( a set of humps of enhanced stability located at certain temperatures ) . they can be formulated as follows : ( i ) : : to explain formation of msps in agreement with the upper limit ( [ tmax ] ) on msp temperature , the r - mode instability should be suppressed in the whole msp frequency range ( @xmath233 hz ) at least for temperatures @xmath234 [ cross - hatched region in fig.[fig_uplift](b ) ] . here @xmath226 is given by equation ( [ tmin_lmxb ] ) . @xmath226 is specific for each msp , but the strongest bound ( the largest @xmath226 ) comes from psr b1937 + 21 . in case of the standard accretion model and the direct urca emissivity ( see equation [ l_durca ] ) it corresponds to @xmath235 k ( assuming @xmath59 km ) and becomes higher if cooling and/or accretion spin up is less efficient . ( ii ) : : the r - mode instability can be unsuppressed at low temperatures ( @xmath236 ) and , if it is the case , it can , in principle , be confirmed by observations . namely , if minimum of the instability curve for @xmath236 takes place at a point @xmath237 ( e.g. , in fig . [ fig_uplift](b ) @xmath238 k and @xmath239 hz ) , _ all _ msps with @xmath240 should have @xmath241 and can be affected by anti - glitches , associated with @xmath203-oscillations ( if they are not suppressed ) . ( iii ) : : strictly speaking , current observations do not require to stabilize all nss observed in lmxbs by one hump the r - mode instability can be almost unsuppressed in the region @xmath242 k [ as in fig . [ fig_uplift](b ) ] . however , at larger temperatures instability should be suppressed by additional hump ( may be narrow ) to explain the hottest nss in lmxbs with @xmath0 k ( e.g. , 4u 1608 - 522 ) . such stars should evolve not to msps , but to hot widows / hofnars , as suggested by @xcite . at the end let us note that the constraint ( i ) can be used to quickly check for consistency various r - mode instability models . for example , it excludes some of the models discussed in the recent review by @xcite , namely , the model with strong mutual friction and ` strong ' superfluidity ( left panel of fig . 3 in that paper ) and strange star model ( fig . 4 there ) . we put new constraints on the shape of the r - mode instability window . they follow from the recycling scenario and observations of ( a ) nss in lmxbs ( section [ sec_xray ] ) , ( b ) msp timing ( section [ sec_timing ] ) and ( c ) uv and x - ray observations of msps ( section [ sec_xray_msp ] ) . analysing the observational sets ( a ) and ( b ) , which have already been used to put an upper bound on the r - mode instability by several authors , we emphasize the importance of two points : ( i ) any successful model should not only suppress the instability for nss observed in lmxbs , but also explain their temperatures ( this is especially important for models based on an enhanced bulk viscosity , see section [ sec_bulk ] ) ; ( ii ) observations of psr b1937 + 21 provide a new evidence of additional dissipation of r - modes at low temperatures , because the internal heating mechanisms [ superfluid vortex creep ( @xcite ) , rotochemical heating ( @xcite ) , and rotation - induced deep crustal heating ( @xcite ) ] are likely to be strong enough for psr b1937 + 21 to make it unstable within the minimal r - mode instability model . additional constraint comes from the observational set ( c ) , which allows us to put an upper limit on the internal msp temperature , @xmath243 k , limiting thus the maximum possible r - mode amplitude in msps ( section [ sec_tiny ] ) . these results qualitatively agree with similar conclusions of @xcite . combined with the recycling scenario of msp formation , the temperature upper limit ( [ tmax ] ) allow us to conclude that the r - mode instability should be suppressed in the whole msp frequency range ( @xmath233 hz ) at temperatures @xmath244 k ( see cross - hatched region in fig . [ fig_uplift](b ) and section [ sec_gen ] for details ) . this necessary condition can be used to quickly check for consistency various instability windows available in the literature . in particular , it can be fulfilled in the resonance uplift scenario suggested by @xcite ( see section [ sec_reson ] for a detailed analysis ) . it is worth noting , that constraining the surface temperature of the fastest msps by detailed analysis of x - ray and uv observational data can be a very interesting task . it is very likely that such analysis can put much more stringent upper limits on the surface temperatures of individual sources , than our general upper limit @xmath245 k ( see section [ sec_xray_msp ] , and @xcite ) . this can make the r - mode instability even more tightly constrained . in the opposite case , if one observes an msp with higher temperature , it will be even more important , because it will present an almost direct evidence of r - mode instability in such pulsar : the thermal emission from the surface @xmath246 ergs@xmath52 should have an energy source and the r - mode instability seems to be the most natural ` energy supplier ' in the absence of accretion . ( see @xcite ) , producing @xmath247 ergs@xmath52 even for msps with the highest spin down power , @xmath209ergs@xmath52 . ] we are very grateful to dima zyuzin for suggesting us the paper by @xcite . this study was supported by the russian science foundation ( grant no . 14 - 12 - 00316 ) . e. m. , gusakov m. e. , 2016 , presentation at `` newcompstar meeting on oscillations and instabilities in neutron stars '' , southampton , uk , 13 - 14 september 2016 . http://www.southampton.ac.uk/stag/news/events/2016/09/13-newcompstar-meeting-oscillations-and-instabilities-in-neutron-stars.page ( accessed october 20 , 2016 )

we derive the explicit form of the superpropagators in presence of general boundary conditions ( coisotropic branes ) for the poisson sigma model . this generalizes the results presented in and for kontsevich s angle function used in the deformation quantization program of poisson manifolds . the relevant superpropagators for branes are defined as gauge fixed homotopy operators of a complex of differential forms on sided polygons with particular `` alternating '' boundary conditions . in presence of more than three branes we use first order riemann theta functions with odd singular characteristics on the jacobian variety of a hyperelliptic riemann surface ( canonical setting ) . in genus the superpropagators present zero modes contributions .

0709.0635

the poisson sigma model ( p@xmath3 m ) is a topological field theory defined in terms of a functional on the space of maps from the tangent bundle of a two dimensional oriented surface @xmath4 to the cotangent bundle of a given poisson manifold @xmath5 . when the source surface is the unit disk it has been shown in the celebrated paper @xcite that kontsevich s star product on @xmath6 @xcite can be obtained from feynman s expansions of certain green functions , assuming particularly simple boundary conditions . the same authors have extended the previous calculations with the p@xmath3 m with boundary conditions such that the base map @xmath7 maps the boundary of the unit disk to a coisotropic submanifold of @xmath6 @xcite : it turns out that the coisotropic submanifolds of a poisson manifold label the possible boundary conditions of the p@xmath3 m and its quantization is related to the deformation quantization of the coisotropic submanifold itself . in @xcite it has been shown that even non coisotropic branes are allowed at quantum level ; when the brane is defined by the so called second class constraint , then the perturbative quantization of the p@xmath3 m yields kontsevich s star product associated now to the dirac bracket defined on the brane . in @xcite a unifying approach is proposed with the introduction of pre - poisson submanifolds : given a pre - poisson submanifold @xmath8 of a poisson manifold @xmath6 , then it is always possible to find a presymplectic submanifold @xmath9 of @xmath6 containing @xmath8 as coisotropic submanifold . in this note we construct explicitly the superpropagators for the p@xmath3 m in presence of @xmath10 coisotropic branes generalizing the results in @xcite . in the perturbative expansion every coisotropic submanifold @xmath11 , or brane , of the poisson manifold @xmath5 is defined by the constraint @xmath12 ; considering @xmath13 branes the source manifold for the p@xmath3 m is defined by the couple @xmath14 where @xmath15 is a @xmath0 sided polygon and @xmath16 is a suitable homeomorphism between the compactified complex upper half plane @xmath17 and @xmath1 , depending on the number of branes considered . this definition allows to fix the polygon @xmath1 through @xmath18 and to use the technique of the `` mirror charges '' @xcite to write the explicit formula for the superpropagators with the correct boundary conditions . in presence of @xmath19 branes , @xmath18 is chosen to be a suitable schwarz - christoffel mapping @xcite , while for @xmath20 branes we need to introduce particular sections of hyperelliptic riemann surfaces and their projections to the riemann sphere @xmath21 ( we refer to section 6 for the full construction ) . in the sequel each side @xmath22 of @xmath1 is called brane as well , when confusion does not arise with the corresponding coisotropic submanifold @xmath23 . the fundamental superpropagators in presence of @xmath0 branes are those associated to the boundary conditions expressed by the index sets @xmath24 , @xmath25 for @xmath0 even and @xmath26 , @xmath27 for @xmath0 odd ; we call them relevant . let @xmath28 be the complexes of differential forms on @xmath1 with dirichlet boundary conditions on the even branes for @xmath29 and dirichlet boundary conditions on the odd ones for @xmath30 . the relevant superpropagators are defined as the gauge fixed homotopy operators @xmath31 which give the hodge - kodaira decomposition @xmath32 of @xmath33 , where @xmath34 is a projection onto de rham cohomology , for @xmath35 . this setup is a special case of a strong deformation retract used in the homological perturbation theory contest ( @xcite and references therein ) . the degree minus one operators @xmath36 are given in def.1 ; the paper is devoted to the construction of their integral kernels @xmath37 , @xmath38 , @xmath39 satisfying the boundary conditions imposed by the @xmath40 themselves . with @xmath41 we denote superfields @xcite of the p@xmath3 m in the perturbative expansion . in principle @xmath42 will consist of two different contributions : a generalization of the kontsevich angle map @xcite and an additional term due to the projection @xmath34 onto cohomology @xmath43 . the main results of the paper are collected in the following * superpropagators for the p@xmath3 m with @xmath0 branes * let @xmath31 be the relevant superpropagators for the p@xmath3 m with @xmath0 branes defined by the constraints @xmath12 and @xmath40 defined as above . the integral kernels @xmath44 are given by : * two branes case : @xmath45 where @xmath46 with @xmath47 , @xmath48 , @xmath49 : we identify @xmath50 with the couple @xmath51 . * three branes case : @xmath52 where @xmath53 with @xmath54 , @xmath48 , @xmath49 : we identify @xmath50 with the couple @xmath51 . * @xmath55 , @xmath56 branes case : + @xmath57 where @xmath58 , @xmath59 is the abel - jacobi map for the hyperelliptic riemann surface @xmath60 of genus @xmath2 which realizes the two sheeted branched covering @xmath61 with branching points @xmath62 such that @xmath63 , @xmath38 and @xmath64 . the polygon @xmath1 is defined via @xmath65 while @xmath66 denotes non singular odd half periods on @xmath67 ( see section 6.2 ) and @xmath68 are first order riemann theta functions defined on @xmath67 . in presence of an odd number @xmath0 of branes , @xmath69 , we can reduce to the @xmath70 even case . by construction @xmath71 , @xmath72 for any number of branes . the paper is structured as follows . in section 3 we introduce the p@xmath3 m , its batalin vilkovisky action and the superpropagators as homotopy operators . we define the concept of coisotropic branes in poisson geometry and we show how they naturally arise in the bv quantization of the p@xmath3 m action describing the one brane case . we compute then the dimension of the space of the `` zero modes '' , i.e. the dimension of the de rham cohomology of the differential complexes @xmath73 . section 4 and 5 deal with the two and three branes cases : we want integral kernels which are zero on the boundary of @xmath74 and @xmath75 for particular sets of indices specified by the presence of branes . to determine such kernels is equivalent to find generalizations of kontsevich s angle formula @xcite with the correct boundary conditions imposed by the constraint equations defining the branes . once defined the polygon @xmath1 as image under @xmath18 of @xmath17 then it is possible to express the boundary conditions of the integral kernels as reflections properties relative to the sides of the polygon @xmath1 of maps @xmath76 : the method is a multi brane generalization of the classical `` mirror charges '' formalism . all the problem is then reduced to find explicit @xmath77 satisfying the correct reflections . the presence of branes implies the existence of non zero simple observables even in the trivial poisson structure case ; the complete analysis of the algebra of observables for the p@xmath3 m with branes will appear elsewhere . in section 6 we deal with @xmath78 branes ; we write the @xmath77 maps via first order riemann theta functions with odd non singular characteristics and hyperelliptic curves @xmath60 @xcite : we refer to this constructions as the canonical setting . with more than three branes we have zero modes contributions : we show that their presence is stricly correlated with the boundary conditions the integral kernels must satisfy . section 7 is about conclusions and further developments ; in appendices a and b we put known material on riemann theta functions and the proof of proposition 1 . in appendix c we describe some properties of the superpropagators as homotopy operators . this paper is meant to be an introduction to the p@xmath3 m with @xmath79 branes ; with the explicit formulas for the superpropagators we can study the algebraic structure and the deformation of the associative product of the algebra of observables , relating it to the already known results which show @xmath80 properties for the one brane case . with non trivial poisson structure it is possible to extend the results of @xcite for the deformation quantization of branes in the sense of @xmath81 bimodules . in order to study the uniqueness of the superpropagators one can introduce the laplacian in a suitable metric completion of the differential complexes @xmath73 . this topic and the relations between the superpropagators for the @xmath78 cases and classical kernels on compact riemann surfaces will be discussed in @xcite . the non perturbative study of the p@xmath3 m in presence of general boundary conditions involves the construction of a generalization of the fukaya @xmath81-category : work in this direction is in progress motivated by the constructions of section 6 in terms of riemann surfaces . the idea is to begin producing a local version of the fukaya category applying the hpt tools to the differential complexes @xmath82 to generate an @xmath81 structure on the cohomology , then define an associated @xmath81 category . the adjective local refers to the fact that we are in the perturbative contest , expanding around zero modes , in presence of @xmath83 branes . the idea to use the mirror charges formalism to solve boundary conditions is already contained in @xcite where cattaneo and felder use it to produce the generalized angle maps in the @xmath84 branes case . i would really like to thank g. felder for introducing me this method and for the many inspiring discussions we had throughout the writing of this paper . i thank also a. cattaneo , h. kajiura , c. rossi for helpful comments on the initial drafts of this work . this work has been partially supported by the swiss national science fundation ( grant 200020 - 105450 ) . the poisson sigma model @xcite is a two - dimensional topological sigma theory defined on a two dimensional orientable surface @xmath4 and with a poisson manifold @xmath5 as target . it is defined by a classical functional @xmath85 on the space of bundle maps @xmath86 with base map @xmath87 and @xmath88 , where @xmath85 is given explicitly by : @xmath89=\int_{\sigma } \langle \eta , d x \rangle + \frac{1}{2}\langle \pi \circ x,\eta \wedge \eta \rangle,\ ] ] and @xmath90 is the canonical pairing between vectors and covectors of @xmath6 . the euler - lagrange equations express the condition that the pair @xmath91 is a lie algebroid morphism between @xmath92 and @xmath93 . under the infinitesimal gauge transformations @xmath94 , @xmath95 , where @xmath96 is a section of @xmath97 , the action ( [ poisson ] ) transforms by a boundary term @xmath98 . the commutator of two infinitesimal gauge transformations is a gauge transformation only on shell , that is when euler - lagrange equations are fulfilled for the action @xmath85 . thus the gauge transformations form a lie algebra only when acting on the set of critical points , or classical solutions of @xmath85 : in order to quantize the model we need to implement the batalin - vilkovisky ( bv ) procedure ; before this one defines the brst operator @xmath99 @xcite ( the gauge parameters @xmath100 are promoted to ghost fields ) and the boundary conditions for the model @xcite . varying the action ( [ poisson ] ) we observe the appearance of a boundary term of the form @xmath101 if we assume that @xmath102 with @xmath103 vanishing when contracted with vectors tangent to the boundary , then ( [ bound2 ] ) cancels ; imposing that the infinitesimal parameters @xmath100 vanish on @xmath104 we cancel @xmath105 . we refer to these bc as the `` space filling brane '' case @xcite . we introduce some notation concerning the bv formalism ; we refer to @xcite for the full analysis and to @xcite for a geometrical approach . we introduce the `` antifields '' @xmath106 with complementary ghost number and degree as differential forms on @xmath4 w.r.t . the `` fields '' @xmath107 . then we look for the so called batalin - vilkovisky action @xmath108 $ ] of ghost number zero such that @xmath109 $ ] reduces to the classical action @xmath110 $ ] and @xmath111 , where @xmath112 is the bv antibracket and @xmath113 is the bv laplacian . then we fix the gauge @xmath114 . the hodge operator @xmath115 is explicitly given , in terms on the coordinates on @xmath4 , by @xmath116 and @xmath117 with @xmath118 and the gauge fixing fermion is @xmath119 . in order to express the bv laplacian one can introduce the hodge dual antifields @xmath120 with the rule that they must have the same boundary conditions as the fields . selecting the lagrangian submanifold @xmath121 defined by the equations @xmath122 and adding the antighost term @xmath123 , then the gauge fixed action is : @xmath124 where , in particular , @xmath125 and the lagrange multipliers @xmath126 satisfy dirichlet boundary conditions on @xmath104 . in the perturbative expansion of the space filling brane case we select @xmath127 , @xmath128 unit disk in @xmath129 and we expand around the classical solution @xmath130 , @xmath131 , i.e. @xmath132 , @xmath133 fluctuation field . the kinetic part of the gauge fixed action : @xmath134 we map conformally the disk to the ( compactified ) complex upper half plane @xmath135 and we use the standard complex coordinates @xmath136 on it : introducing @xmath137 we obtain what in @xcite are called the `` superpropagators '' for the space filling brane case : @xmath138 where @xmath139 , @xmath140 is kontsevich s angle function @xcite with @xmath141 and @xmath142 , @xmath143dim@xmath6 . the angle function @xmath144 associates to each pair of distinct points @xmath136 in the upper half plane the angle between the geodesics w.r.t . the poincar metric connecting @xmath145 to @xmath146 and to @xmath147 , measured in the counterclockwise direction . on the boundary @xmath148 we have @xmath149 . a submanifold of the poisson manifold @xmath6 chosen as boundary condition is called brane ; symmetries of the model give strict characterizations of branes in terms of poisson geometry . a bunch of definitions ; the sharp map @xmath150 for the poisson manifold @xmath151 is defined as @xmath152 , with @xmath153 , @xmath154 and @xmath90 denotes the canonical pairing . a submanifold @xmath155 of the poisson manifold @xmath5 is called pre - poisson @xcite if @xmath156 has constant rank along @xmath155 : in @xcite this was called a `` submanifold with strong regular conditions '' . in the symplectic context this condition is equivalent to @xmath155 being presymplectic . in @xcite it is shown that if @xmath155 is pre - poisson then @xmath157 is a lie subalgebroid of the full lie algebroid @xmath93 . a submanifold @xmath158 is called coisotropic if @xmath159 . it follows from the jacobi identity for @xmath160 that the characteristic distribution @xmath161 on the coisotropic submanifold @xmath155 is involutive ; the corresponding foliation is called the characteristic foliation . in the symplectic context , @xmath150 yields an isomorphism between @xmath162 and @xmath163 and we recover the usual definition of coisotropic submanifolds in the symplectic case : @xmath164 , where @xmath163 is the subbundle of @xmath165 of vectors that are symplectic orthogonal to all vectors of @xmath166 . if @xmath155 is coisotropic , then @xmath167 . pre - poisson submanifolds are the most general boundary conditions for p@xmath3 m compatible with symmetries @xcite ; it can be shown ( cfr . @xcite ) that if @xmath155 is pre - poisson in @xmath6 then it is always possible to find a cosymplectic submanifold of @xmath6 which contains @xmath155 as a coisotropic submanifold : then it is enough to consider coisotropic submanifolds as boundary conditions for the p@xmath3 m . this means that , in the one brane case , given a coisotropic submanifold @xmath155 of @xmath6 ( i.e the brane ) , we impose the boundary conditions @xmath168 , @xmath169 , where @xmath170 is the inclusion map @xmath171 and the ghosts on the boundary satisfy @xmath172 . the perturbative analysis in presence of a brane is performed considering the case where @xmath6 is an open subset of @xmath173 with coordinates @xmath174 and the brane @xmath155 is given by the constraint equations @xmath175 with @xmath176 . the tangent space to a point of @xmath155 is spanned by @xmath177 , @xmath178 and the conormal bundle by @xmath179 , @xmath176 . we follow the convention that latin indices run along the brane , that is over @xmath180 , while greek indices are associated to coordinates normal to the brane and run over @xmath181 . the whole boundary of the disk , ( or conformally , the compactified real line in the upper half plane ) , is mapped to the brane @xmath155 : the splitting of the indices into @xmath182 and @xmath183 induces @xmath184 and @xmath185 and the boundary conditions for @xmath186 and @xmath103 become @xmath187 and @xmath188 . consequently the `` superpropagators '' with the correct boundary conditions are @xcite : @xmath189 where @xmath190 is kontsevich s angle function and @xmath191 is the kronecker delta restricted to pairs of indices in @xmath192 . this result allows to study the quantization of the coisotropic brane @xmath155 via path integral , with the introduction in the diagrammatics of straight and wavy lines induced by the presence of the two index sets @xmath193 and @xmath194 . in presence of two or more branes a more general analysis of the boundary conditions is needed . the branes are defined as usual by the constraints @xmath195 and given @xmath10 branes ( [ branes ] ) we define the index sets : @xmath196 let @xmath197 be the source manifold for the p@xmath3 m in presence of @xmath79 branes . @xmath1 is defined as a @xmath0 sided convex polygon with boundary @xmath198 given by the decomposition @xmath199 . the corners of the polygon are the elements of the set @xmath200 . the polygon @xmath1 is fixed by the condition @xmath201 , where @xmath202 is a homeomorphism depending by the number @xmath0 of branes considered in the perturbative analysis and @xmath17 denotes the compactified complex upper half plane . in particular this allows to give an ordering to the sides @xmath22 in a consistent way . all the possible index sets @xmath203 , @xmath204 for the integral kernels @xmath205 with @xmath206 are given by the intersections of the sets @xmath207 defining the @xmath0 branes ( [ branes ] ) . in the sequel we will drop out the @xmath208 dependence in ( [ intker ] ) ; we will reduce to the @xmath209 case to simplify notation . fixing the number of branes @xmath0 , one repeats the same calculations of the preceding subsections , giving new boundary conditions to eliminate the additional terms @xmath210 and to develop a coherent bv formalism . the boundary conditions for the @xmath7 field are given by @xmath211 while we choose @xmath212 , @xmath213 @xmath214 generalizing the one brane case . more explicitly , @xmath215 , @xmath216 where @xmath217 denotes the inclusion . we apply the same notation to all the component fields appearing in the sequel . @xmath218 satisfies the same boundary conditions of @xmath103 as it belongs to the same superfield @xmath219 ( we refer to @xcite for a brief introduction to the superfield formalism ) : in particular this allows to cancels the boundary terms coming from @xmath220 ) . at the same time @xmath221 belongs to the same superfield @xmath222 of @xmath7 . in the sequel we will refer to the sides @xmath22 as branes , whenever confusion does not arise with the corresponding ( under @xmath7 ) coisotropic submanifolds @xmath223 . for simplicity let us discuss the boundary conditions in the @xmath84 branes case explicitly , i.e. @xmath224 , @xmath225 , @xmath226 , @xmath227 , @xmath228 . as @xmath229 then @xmath230 , @xmath231 and @xmath232 . in particular we get @xmath233 i.e. specifing the index sets @xmath40 we have dirichlet boundary conditions for the components fields of @xmath234 only on half of the boundary @xmath235 . a general fact : given @xmath0 number of branes , then the `` non trivial '' integral kernels are those associated to the alternating dirichlet boundary conditions respect to the points @xmath50 of the polygon @xmath1 , with index sets given by @xmath236 . in presence of all the other index sets , it is possible to `` reduce '' to a lower number of branes case to compute the superpropagators . developing the bv formalism we generalize the gauge fixing condition @xmath237 to eliminate the boundary terms @xmath238 imposing the extended gauge fixing @xmath239 and @xmath240 with boundary conditions for the lagrange multipliers given by @xmath241 . the extended gauge fixing also imposes @xmath242 . with the choice ( [ egf ] ) we fix the boundary conditions for the components fields of the superfields @xmath234 on the whole @xmath235 for the index sets @xmath40 , getting alternating dirichlet - neumann boundary conditions . the general @xmath0 branes case follows the same lines of the @xmath84 case ; the dirichlet boundary conditions and the extended gauge fixing are collected in the following table . [ cols=">",options="header " , ] this implies that the integral kernels ( [ intker ] ) satisfy @xmath243 with @xmath40 given by ( [ index ! ] ) . the presence of branes in the perturbative expansion induces the `` alternating '' boundary conditions on the component fields of the superfields @xmath234 as shown in the preceding section . this allows to give the following * relevant superpropagators for the p@xmath3 m in presence of branes * the gauge fixed homotopy operators @xmath31 which realize a hodge - kodaira splitting @xmath244 of the differential complexes @xmath245 are called relevant superpropagators for the p@xmath3 m in presence of @xmath0 branes ( [ branes ] ) . with @xmath34 we denote a projection onto cohomology @xmath246 . the @xmath31 operators act via @xmath247 @xmath248 . the integral kernels ( [ intker ] ) are given explicitly by the sum @xmath249 where @xmath250 is called generalized angle function and @xmath251 is the integral kernel of the projection @xmath34 onto cohomology , @xmath64 . the generalized angle functions satisfy @xmath252 for @xmath253 , where @xmath254 are coordinates of the points @xmath50 . the explicit form of the integral kernels is simply given by a contribution which generalizes the kontsevich angle function due to the boundary conditions imposed by the presence of the branes and a non trivial term due to de rham cohomology @xmath246 . with an explicit choice of metric ( i.e. a hodge star operator ) it is possible to introduce a laplacian operator on a metric completion the differential complexes @xmath255 ( and a gauge fixing for the theory , as we have already seen ) . defining harmonic forms on the metric completion one can study the uniqueness of the relevant superpropagators ; we will discuss this elsewhere @xcite . let @xmath255 @xmath35 be as above ; then the de rham cohomologies @xmath256 are given by : @xmath257 we do the analysis for @xmath29 ; the other case is equivalent . we write the short exact sequence @xmath258 , where @xmath259 , @xmath260 and @xmath261 for @xmath0 even or @xmath262 for @xmath0 odd are the inclusions of the even branes in @xmath1 . this sequence induces a long exact sequence in cohomology ; a standard counting gives the thesis . in presence of two branes @xmath263 and @xmath264 we can select the indices for the superpropagators in the sets @xmath265 , @xmath266 and @xmath267 following ( [ index ! ] ) . the @xmath84 sided polygon @xmath74 is defined as @xmath46 , with the schwarz - christoffel mapping @xmath18 @xcite given by @xmath268 points @xmath269 are represented respectively by a pair of complex numbers @xmath51 in the first quadrant , with @xmath270 @xmath271 . @xmath272 is given by the positive imaginary axis , while @xmath273 is the positive real axis . the boundary conditions imposed by the index sets @xmath40 are @xmath274 , @xmath275 . introducing the maps @xmath276 which satisfy the same boundary conditions of @xmath277 and considering that @xmath278 , we get the integral kernels for the superpropagators @xmath31 in presence of two branes are given by @xmath279 with mirror maps ( [ m2 ] ) . the integral kernels satisfy the additional boundary conditions @xmath280 @xmath281 , i.e. every boundary component of @xmath74 is labelled by a boundary condition for both the variables @xmath51 . by construction @xmath282 , @xmath283 . one can verify that @xmath284 and @xmath285 with @xmath190 kontsevich s angle function . the integral kernels here presented correspond to the generalized angle functions in @xcite ; they are used to construct an explicit quantum deformation of bimodule structures and to define a kontsevich s product ( associative ! ) in presence of two branes . additional assumptions are necessary to take care of the faces produced in the compactification of some configuration spaces in order to guarantee associativity ( see @xcite ) . the problem of finding integral kernels with correct boundary conditions is replaced by the easier task to write the @xmath286 maps satisfying some reflection properties respect to the sides @xmath287 : the @xmath286 are odd respect to these reflections , allowing to determine the correct kernels . the method is nothing but a generalization of the classical `` mirror charges '' formalism , due to the presence of multiple axis of symmetry , or branes . in presence of three branes @xmath288 we can select 8 different index sets . we define as usual @xmath289 , @xmath290 ; the three sided polygon @xmath75 is defined via @xmath291 with @xmath292 where the integral is performed along a smooth path in @xmath17 . @xmath75 is a strip in the first quadrant with two sides parallel to the real axis . the above side is @xmath293 ; the other sides are labelled counterclockwise . the boundary conditions for the integral kernels of the relevant superpropagators are @xmath294 , @xmath295 . using the function @xmath296 we write the mirror maps : @xmath297 so we get ( remembering that @xmath298 ) the integral kernels for the relevant superpropagators @xmath31 in presence of three branes are given by @xmath299 with mirror maps given by ( [ po2 ] ) . the integral kernels satisfy the additional boundary conditions @xmath300 @xmath301 , i.e. every boundary component of @xmath75 is labelled by a boundary condition for both the variables @xmath51 . by construction @xmath282 , @xmath283 . choosing the six `` non relevant '' index sets , we can reduce to a lower number of branes analysis . as we impose boundary conditions respect the same variable on adiacent sides of @xmath75 ( not separated by @xmath302 , except in the @xmath303 and @xmath304 cases ) , then it is simple to see that we can introduce a new `` square root '' homeomorphim @xmath305 and reduce to a two branes case . choosing @xmath306 and repeating the same lines we can reduce to the one brane case : the interesting boundary conditions are those associated to the index sets ( [ index ! ] ) . in presence of @xmath78 branes we deal with the so called hyperelliptic cases . the name comes from the fact that we are going to use hyperelliptic riemann surfaces @xmath60 of genus @xmath2 in order to find the superpropagators . in principle one could write a schwarz - christoffel mapping from the unit disk with boundary partitioned into @xmath78 sectors , or conformally , from @xmath17 into a suitable polygon with @xmath0 sides ; this would permit to avoid riemann surfaces and theta functions formalism . the main problem concerning this formulation is that the reflections to impose on the @xmath77 maps would become particularly complicated and it is not clear a priori which functions one should use to get the correct maps fulfilling the @xmath0 boundary conditions of the integral kernels . the canonical setting in presence of hyperelliptic riemann surfaces is a natural choice , instead . we refer to appendix b for a brief introduction on riemann theta functions . let @xmath60 be the hyperelliptic riemann surface of genus @xmath2 which realizes the two sheeted branched covering @xmath307 such that @xmath308 , @xmath309 and @xmath310 , where @xmath311 denotes the set of the branching points . @xmath60 is the compact riemann surface of the algebraic curve @xmath312 . we represent @xmath60 in figure 1 ; with @xmath313 we denote an explicit choice of canonical homology basis for @xmath314 and with @xmath315 the dual basis of holomorphic abelian differentials . the @xmath55 sided polygon @xmath1 is represented in figure 2 ; the restriction @xmath316 is a homeomorphism between @xmath1 and @xmath317 . let @xmath318 be the abel - jacobi map for @xmath60 @xcite , i.e. @xmath319 , @xmath320 : we explicitly choose the branching point @xmath321 as base point for @xmath318 . the main result of section 6 is the integral kernels for the relevant superpropagators @xmath31 in presence of @xmath78 branes with @xmath55 , are given by the rest of the paper is devoted to the proof of theorem 4 . at first we study the mirror maps in the canonical setting ; then we introduce first order riemann theta functions on @xmath67 ; we discuss then the zero set of the mirror maps and the emersion of additional contributions called zero modes : lemma 1 gives non trivial cohomology for @xmath322 , @xmath78 . let @xmath0 be the number of branes . in the above theorem @xmath0 is even ; in fact if the number of branes @xmath0 is odd with @xmath323 , then we can recover a @xmath70 branes case : for this reason it is sufficient to develop the analysis in presence of @xmath55 branes . more explicitly , in presence of @xmath69 branes , with @xmath0 odd , we should identify a polygon @xmath1 on a hyperelliptic curve given by @xmath324 with real @xmath325 . but we can impose , for example , the transformation @xmath326 , @xmath327 even , to get a hyperelliptic curve of the type @xmath328 : the analysis is then reduced to an even number of branes case . the choice of the homology basis @xmath329 for @xmath60 gives @xmath330 for all @xmath331 . let @xmath332 in the sequel we will use the notation @xmath333 , @xmath334 . as @xmath60 is hyperelliptic we write the elements of the basis of holomorphic abelian differentials as @xmath335 ; from the definition of @xmath329 it follows that the coefficients @xmath336 are real and the elements @xmath337 are purely imaginary @xmath338 . we rewrite ( [ abel ] ) as @xmath339 with @xmath340 , @xmath341 . by definition , for every point @xmath342 there exists a point @xmath343 , @xmath344 such that @xmath345 : the pair @xmath50 projects to the same point @xmath346 on the riemann sphere @xmath21 with @xmath347 belonging to different sheets . this means that selecting @xmath348 then ( [ abel ] ) is well defined . the boundary conditions for the integral kernels @xmath349 are given by ( [ bc ] ) ; we project homeomorphically each point of @xmath1 to the jacobian variety @xmath67 via ( [ abel ] ) and we introduce there the @xmath77 maps . this allows to express explicitly the boundary conditions ( [ bc ] ) for the integral kernels in terms of the variables @xmath51 in the jacobian variety itself . we consider now the reflections for @xmath350 . we define the intervals @xmath351 $ ] , @xmath352 , where @xmath353 for @xmath354 and , as usual @xmath355 , @xmath356 , as @xmath310 . if @xmath357 then @xmath358 and the reflection is @xmath359 . selecting @xmath360 we get @xmath361 ; this implies @xmath362 . with @xmath363 we get @xmath364 and so @xmath365 . moreover if @xmath366 then @xmath367 and @xmath368 . in presence of odd branes ( different from @xmath369 and @xmath370 ) , i.e. for @xmath371 , @xmath372 , we have @xmath373 and the reflections @xmath374 . with even branes ( different from @xmath375 and @xmath376 ) that is for @xmath377 , @xmath372 , we get @xmath378 and the reflections @xmath379 . analogously for @xmath380 we have @xmath381 . direct calculations show that in order to compute the reflections for @xmath382 we can simply consider those for @xmath350 , then formally exchange @xmath18 and @xmath383 in @xmath350 and substitute the subscript @xmath193 with @xmath194 . for example @xmath359 becomes @xmath384 ; @xmath379 goes to @xmath385 and so on . we summarize all the reflection properties for @xmath350 and @xmath382 . @xmath386 @xmath387 given all the other possible choices of index sets we can reduce to a lower number of branes cases by pinching the sides @xmath22 of @xmath1 with the same boundary conditions imposed on @xmath388 ; one repeats the process till a no more reducible case : then the computation of the integral kernels begins specifying the correct homeomorphism @xmath18 . we give the explicit form of the mirror maps in terms of first order riemann theta functions with odd characteristics ( canonical setting ) . increasing the genus @xmath2 of the hyperelliptic surface get more reflections to satisfy ; precisely in presence of @xmath55 branes we have @xmath389 reflections to impose but it is possible to construct mirror maps with a combination of four suitable first order riemann theta functions with odd characteristics independently on the number of reflections . we introduce then odd non integer characteristics @xmath390 through the odd half periods @xmath391 , where @xmath392 , @xmath393 , @xmath394 , @xmath395 with @xmath396 @xmath397 mod 2 , @xmath398 . the basic result is the following : the map @xmath399 satisfies the reflection properties : @xmath400 if @xmath401 see appendix b. so , the most general mirror map satisfying ( [ gh ] ) and written in canonical setting is given by : @xmath402 following the same lines of proposition 1 we can state a similar result for the mirror map @xmath350 , or the map @xmath403 satisfies the reflections @xmath404 if @xmath405 for any vector @xmath406 . if we select odd characteristics through @xmath407 @xmath408 @xmath409 , and @xmath410 fulfilling the hypotesis of proposition 2 we can write the most general mirror map in canonical formalism which satisfies ( [ wer ] ) , or @xmath411 we characterize now the zero set of the mirror maps . we need to introduce some definitions for divisors before . given a compact riemann surface @xmath60 of genus @xmath2 , then an odd half period @xmath66 of its jacobian variety @xmath67 is called non singular if it can be written as @xmath412 with @xmath413 non special integral divisor of degree @xmath414 on @xmath60 , @xmath415 vector of riemann constants and @xmath318 abel - jacobi map . we say that integer odd characteristics @xmath416 are non singular if the corresponding odd half period @xmath417 is non singular . an integral divisor @xmath418 , @xmath419 is special if @xmath420 for riemann - roch theorem . here @xmath421 is the index of specialty and @xmath422 the dimension of the divisor @xmath128 . the existence of non singular odd half periods is a result shown , for example , in @xcite . extending the abel - jacobi map @xmath318 ( [ abel ] ) to arbitrary divisors @xmath423 with @xmath424 , we recall that , with @xmath425 , @xmath426 integral non special divisors of degree @xmath2 such that @xmath427 , then @xmath428 . this result comes from the fact that abel s theorem @xcite implies that @xmath429 , where @xmath430 is the divisor of a meromorphic function @xmath431 on @xmath60 ; by non specialty it follows @xmath432 , that is @xmath433 . let us consider the multivalued map on @xmath60 : @xmath434 ; if it is not identically null , then it has @xmath2 zeros , and the zero divisor @xmath435 satisfies @xmath436 . for more details see appendix b. moreover it can be shown that @xmath434 vanishes identically on @xmath60 iff @xmath437 with @xmath410 integral special divisor of degree @xmath2 . by writing @xmath438 , @xmath439 where @xmath350 and @xmath382 are given by ( [ poipoi],[poi ] ) we can state the following let @xmath60 be the hyperelliptic riemann surface of genus @xmath2 of section 6.1 and let the mirror maps @xmath350 , @xmath382 be given by ( [ poipoi],[poi ] ) with @xmath66 non singular odd half period on the jacobian variety of @xmath60 . choose @xmath440 with @xmath441 , where @xmath442 are the odd branching points of @xmath443 , @xmath415 is the vector of riemann constants ( see appendix b ) and caret means omission . then the zero divisor of @xmath444 and @xmath445 on @xmath446 consists only of the diagonal @xmath447 . first of all we note that @xmath66 can be written as @xmath448 mod @xmath449 , @xmath441 as the vector of riemann constants is given by @xmath450 , for the hyperelliptic curve @xmath60 of section 6.1 once we select the canonical homology basis @xmath329 . in this setting it can be shown @xcite that @xmath451 ; this implies that the divisor @xmath452 is non special and with the choice ( [ aaa ] ) @xmath66 is a not singular odd half period . in order to discuss the zero divisor of @xmath444 we begin by writing @xmath453 . now we proceed to study the zero divisor of the multivalued function @xmath454 with @xmath455 fixed . the thetas are multivalued functions on @xmath60 but the zero divisor is well defined as the multivaluedness generates a multiplicative non vanishing factor . as @xmath456 , then the divisor @xmath457 is not special and the zero divisor of the multivalued holomorphic function on @xmath60 : @xmath458 is given by @xmath457 . analogously @xmath459 has the zero divisor @xmath460 with @xmath461 , that is @xmath85 is the image of @xmath462 under the hyperelliptic involution @xmath463 . explicitly @xmath463 acts as follows : we write @xmath464 if @xmath462 is a branching point for @xmath60 ( it is not our case ) , otherwise @xmath465 with @xmath466 and @xmath467 is the two sheeted branched covering of @xmath468 : for this reason @xmath463 is also called the `` sheet exchange '' . from the definition it follows that @xmath469 . we continue with @xmath470 ; from the definition of the abel - jacobi map we get @xmath471 where @xmath472 is a smooth curve joining @xmath473 and @xmath474 and @xmath475 with @xmath476 , for schwarz reflection principle applied to regions in @xmath17 . whenever we have the + sign , introducing the automorphism on @xmath60 @xmath477 , with @xmath478 , we get @xmath479 whenever we have the - sign ( for example , for @xmath462 such that @xmath480 ) we compose @xmath8 with the sheet exchange @xmath463 . so we get the zero divisor @xmath481 for @xmath470 . the bracket selects only one of the two divisors @xmath482 and @xmath483 , depending on the sign of @xmath476 , as above . all the considerations so far imply that the multivalued function @xmath484 has zero divisor given by @xmath485 . collecting all the zero divisors for the thetas we obtain the divisor of the function @xmath486 : @xmath487 i.e. @xmath486 is null only for @xmath488 on @xmath489 for the injectivity of the abel - jacobi map . as @xmath490 , then we study the zero divisor of @xmath491 with @xmath492 fixed . the analysis follows the same lines here discussed and @xmath493 is null only for @xmath488 on @xmath489 for the injectivity of the abel - jacobi map . also the multivalued map @xmath445 presents the same behaviour . in presence of more than three branes we have the emersion of zero modes contributions as the cohomology of @xmath494 is non trivial ( lemma 1 ) . the contributions denoted with @xmath495 in the definition of the integral kernels ( def.1 ) must absorb the extra terms in the reflections of the mirror maps to get the correct boundary conditions for the @xmath42 . in the elliptic case ( @xmath496 ) the image of @xmath497 in @xmath67 is given by the set @xmath498 , ~\operatorname{im}u \in [ 0,\frac{t}{2}]\}$ ] , where @xmath499 . we know that dim@xmath500 ; explicit basis are given by @xmath501 $ ] and @xmath502 $ ] , respectively . with @xmath503 here we denote the basis of holomorphic abelian differentials dual to the canonical homology basis @xmath329 . motivated by these considerations , we write let @xmath42 be given by ( [ intker ] ) for @xmath55 branes with @xmath504 @xmath505 ; then @xmath42 satisfies ( [ bc ] ) for @xmath506 . apply the boundary conditions ( [ bc ] ) to @xmath507 ; the additional contributions generated by the reflections cancel with those of the mirror maps , as in prop.1 - 2 . we are ready to write down the explicit formulas for the superpropagators in the @xmath78 branes case . we have found the generalized angle functions through theta functions , we have studied thier zero divisor and we have introduced the zero modes terms ; they correct the additional contributions generated in the reflections of the mirror maps . collecting all these results we get the superpropagators ( @xmath508 ) @xmath509 where @xmath510 . we have simply used ( [ abel ] ) to replace @xmath51 with @xmath318 . they fulfill the correct boundary conditions expressed by the index sets @xmath40 with no additional terms . moreover it follows @xmath511 , @xmath512 ; the integral kernels are independent on the choice of odd non singular @xmath66 as in ( [ aaa ] ) and the computation of the additional boundary conditions satisfied by @xmath42 is straightforward : this ends the proof of th.4 . the integral kernels for the @xmath78 branes cases present a `` similarity '' with the expression of the green function for the laplacian operator ( that is second order differential operator ) on compact riemann surfaces in @xcite ; such green function it is a sum of a main part involving the prime form defined on the compact curve of genus @xmath2 and @xmath2 zero modes contributions ; here instead of a single prime form we have to use the product of four first order odd riemann theta functions to fulfill the boundary conditions . a mathematical formulation of this remark , involving schiffer and bergmann kernels on @xmath60 will be given in @xcite . we have written the explicit formulas for the superpropagators of the p@xmath3 m in presence of branes . with two or three branes we used a schwarz - christoffel mapping to produce the integral kernels with the correct boundary conditions . with more than three branes we have introduced hyperelliptic curves @xmath60 of genus @xmath2 and first order riemann theta functions with odd characteristics defined on the jacobian variety @xmath67 . the superpropagators include zero modes contributions involving the inverse of the matrix of periods for @xmath60 . with these formulas it is possible to study the algebraic properties and the deformation of the associative product of the algebra of observables @xmath513 for the p@xmath3 m in presence of branes . this should give a generalization of the @xmath80 structure on @xmath514 for a single brane @xmath8 described in @xcite,@xcite , while the non perturbative analysis necessarily leads to the definition of a fukaya @xmath81 category for the p@xmath3 m with branes . before studying a `` global '' fukaya category inspired by the nonperturbative p@xmath3 m with branes , it is possible to analyse a local version deduced by the tools of homological perturbation theory applied to a suitable differential graded category . the superpropagators here deduced play the role of homotopy operators on the space of morphisms of such category . as the two and three branes cases induce bimodules and morphisms of bimodules @xcite @xcite one could expect @xmath81 bimodules and morphisms for the hyperelliptic cases ( or even a more general structure ) ; moreover with linear poisson structure ( and branes as affine subspaces ) the expression of the superpropagators here obtained gives explicit higher order formulas for the diagrammatics developed in @xcite . we introduce briefly first order riemann theta functions ; the material here presented is standard : the reader interested in proofs and wider expositions should consult , for example , @xcite . we write the translation properties for the thetas we used in section 6 . let @xmath515 denote the siegel upper half spaces of genus @xmath2 , that is the space of complex symmetric @xmath516 matrices with positive imaginary part . we define riemann s theta function by @xmath517 where @xmath518 ( viewed as a column vector ) and @xmath519 and the sum extends over all integer vectors in @xmath520 . the function converges absolutely and uniformly on compact subsets of @xmath521 . in the sequel we will fix the matrix @xmath522 and we will consider @xmath68 as a holomorphic function on @xmath520 . let @xmath523 . then @xmath524 @xmath525 and with @xmath526 the @xmath527 identity matrix . in particular @xmath528 where @xmath529 is the @xmath530-th column of the identity matrix @xmath526 and @xmath531 is the @xmath530-th column of @xmath522 . we continue with let @xmath532 , @xmath533 ; the holomorphic functions on @xmath521 @xmath534 ( z,\omega):= \sigma_{n \in \mathbb{z}^g}e^{2\pi i(\frac{1}{2}\null^t(n+\frac{\epsilon}{2})\omega ( n+\frac{\epsilon}{2})+\null^t(n+\frac{\epsilon}{2})(z+\frac{1}{2}))}\ ] ] are called ( first order ) theta functions with characteristics . when selecting @xmath532 , @xmath535 ( integer characteristics ) it is easy to prove that @xmath536 ( -z,\omega)=e^{2\pi i ( \frac{\null^{t}\epsilon\epsilon'}{2})}\vartheta\left[\begin{array}{c } \epsilon \\ \epsilon ' \end{array}\right ] ( z,\omega)$ ] , i.e. theta functions with integer characteristics are odd if @xmath537 mod 2 , even when @xmath538 mod 2 . moreover it follows @xmath536 ( z,\omega)=e^{2\pi i ( \frac{1}{8}\null^{t}\epsilon \omega \epsilon + \frac{1}{2}\null^{t}\epsilon z + \frac{1}{4}\null^{t}\epsilon\epsilon ' ) } \vartheta(z+ \frac{\null^t\epsilon'}{2}i+ \frac{\null^t\epsilon}{2}\omega,\omega)$ ] , i.e. if @xmath536 ( z,\omega)$ ] is odd then @xmath539 ; the riemann theta function vanishes at the odd points of order two in the lattice generated by the columns of @xmath526 and @xmath522 : they are called odd half periods . we introduce now @xmath536 \circ \varphi$ ] , where @xmath318 is the abel - jacobi map for @xmath60 , compact riemann surface of genus @xmath2 . @xmath536 \circ \varphi$ ] is multivalued but its zeroes are well defined on @xmath60 becouse the multivaluedness is multiplicative with a non vanishing factor . then we have the classical let @xmath60 be a compact riemann surface of genus @xmath540 1 with canonical homology basis @xmath541 . let @xmath536 \circ \varphi$ ] be the first order riemann theta function associated with @xmath542 and let @xmath318 be the abel - jacobi map for @xmath543 . then @xmath536 \circ \varphi$ ] is either identically zero as a function on @xmath60 or else it has precisely @xmath2 zeros on @xmath60 . in this case let @xmath544 be the divisor of zeros . we then have @xmath545 , where @xmath415 is the vector of riemann constants which depends on the canonical homology basis and the base point for the abel - jacobi map . we remind that for every compact riemann surface of genus @xmath56 @xmath415 is a half period in the jacobian variety of @xmath60 ; in particular , selecting the hyperelliptic curve @xmath60 described in section 6.1 and the canonical homology basis @xmath329 , it is not hard to show that @xmath450 , where @xmath546 is the set of odd branching points of @xmath60 , i.e. @xmath547 we write here the proof of proposition 1 of section 6.1 ; it states that the map @xmath399 satisfies the reflection properties : @xmath548 provided @xmath549 imposing the reflections given by ( [ 123 ] ) we get : @xmath550 if a @xmath551 and @xmath552 for some @xmath553 . the aim is to select @xmath554 ( if it is possible ) to get @xmath555 with no additional terms . using the translation properties ( [ translat ] ) we can write @xmath556 if and only if @xmath557 . with these choices @xmath558 and @xmath559 . we continue with : @xmath560 if @xmath561 and @xmath562 for some @xmath563 to be determined . we have used the property of first order riemann theta functions @xmath564 as the matrix @xmath522 is pure imaginary . to get the reflection @xmath565 we write @xmath566 if and only if @xmath567 . this reflection imposes @xmath561 , @xmath568 and @xmath569 . the reflections @xmath570 give the same conditions on @xmath571 as translations respect to the columns of the identity matrix induce no new relations for the odd characteristics . the last reflections we study involves the matrix of periods @xmath522 ; by writing @xmath572 , with @xmath573 and @xmath574 : @xmath575 if @xmath551 and @xmath552 for some @xmath553 . thus we get @xmath576 for @xmath574 if and only if @xmath577 and @xmath578 with @xmath579 . the reflection @xmath580 fixes no new equation for @xmath571 . in summary @xmath581 satisfies @xmath582 if @xmath583 and @xmath584 with @xmath585 and @xmath579 . of course the above conditions are not compatible . to solve this impasse we have to relax some of the reflections on the map @xmath382 . we decide to relax those respect to the columns of the matrix of periods : this approach will be compatible with the analysis of the zero modes contributions . explicitly this amounts to get @xmath586 for @xmath558 , @xmath559 and @xmath587 ; comparing this new set of equations for @xmath571 with those relative to the reflections respect to the identity matrix , @xmath588 and @xmath589 we get the constraints for the odd half periods @xmath590 with reflections for @xmath382 given by ( [ abcd ] ) . in this appendix we show some properties of the relevant superpropagators @xmath31 as gauge fixed homotopy operators and we give the proof of the hodge - kodaira splitting @xmath591 appearing in def.1 . first of all the operators let @xmath596 ; the splitting equation ( [ spl ] ) reduces to @xmath597 , where , as usual , @xmath51 denotes the coordinates of points @xmath50 of the polygon @xmath1 . explicitly we get @xmath598 as in the @xmath29 case integrating by parts we get the sum @xmath599 which vanishes for the boundary conditions of the integral kernels and @xmath190 once we add the null term @xmath600 . in the @xmath30 case , we can add the terms @xmath601 and @xmath600 repeating the same trick of the @xmath29 case to eliminate the contributions over @xmath602 . the orientation of @xmath602 is taken to be counterclockwise . let @xmath603 now ; as the cohomology of the differential complexes in not trivial in degree 1 , then we have to distinguish different cases . we begin with @xmath604 ; the splitting equation and lemma 3 give @xmath605 . all the line integrals are taken counterclockwise . the sum of the integrals over the boundary @xmath606 give precisely the identity on the right hand side of the splitting equation ; in both the @xmath35 cases we eliminate the terms along @xmath602 by adding suitable terms as in the @xmath596 case ; in particular for @xmath30 we get the projection onto cohomology integrating by parts . selecting an exact one form , @xmath607 with @xmath608 we get no projection as in the @xmath29 case @xmath609 which , summed to @xmath610 , gives as usual zero ( the @xmath30 case is immediate ) . if @xmath190 is a linear combination of zero modes we have the following let @xmath613 with @xmath614 the @xmath615 component of the abel map @xmath318 on @xmath60 ; we prove ( [ projeee ] ) for @xmath29 : the other case is analog . we want to use riemann s bilinear relations for holomorphic differentials on the hyperelliptic curve @xmath60 of section 6.1 . let @xmath60 be given as in figure 1 ; the hyperelliptic involution @xmath463 is seen as a rotation by @xmath160 radians about an axis passing through the @xmath389 branching points . let @xmath616 be an oriented curve from @xmath617 to @xmath618 for @xmath441 . we have defined the canonical homology basis @xmath619 where the curve @xmath620 is @xmath616 followed by @xmath621 , i.e. @xmath620 joins @xmath617 to @xmath618 and returns to @xmath617 and the curve @xmath622 joins a point on @xmath623 to a point to @xmath624 , returning to the point on @xmath620 . let @xmath625 be the curve that joins @xmath626 to @xmath618 and returns to @xmath626 for @xmath627 and @xmath624 the one which joins @xmath628 to @xmath629 and returns to @xmath628 ( see figure 3 ) . it follows that @xmath630 , @xmath631 @xmath632 so @xmath633 in homology , @xmath634 for @xmath635 , @xmath636 @xmath637 , @xmath638 @xmath639 . thus up to homology we conclude @xmath640 @xmath637 and @xmath641 . in the following @xmath642 is the basis of holomorphic differentials dual to the canonical homology basis @xmath329 . the @xmath647 are closed differentials ; considering the fundamental polygon @xmath648 with symbol @xmath649 associated to the curve @xmath60 we can write @xmath650 where @xmath651 . then @xmath652=\frac{1}{2}\tau_{jk}$ ] ; repeating the same procedure for the second summand of the r.h.s in ( [ decomp ] ) we get @xmath653 . if @xmath654 , then @xmath655 for @xmath656 ( lemma 1 ) . this implies that we can repeat essentially the calculations of the preceding case : the splitting equation reduces to @xmath657 as we have no projection in degree 2 ; this concludes the proof of proposition 5 . a. s. cattaneo and g. felder , `` a path integral approach to the kontsevich quantization formula , '' commun . math . phys . * 212 * ( 2000 ) 591 arxiv : math.qa/9902090 . m. kontsevich , deformation quantization of poisson manifolds , i , arxiv : hep.th/0101170 . i. calvo and f. falceto , `` star products and branes in poisson - sigma models , '' commun . * 268 * ( 2006 ) 607 arxiv : hep - th/0507050 . i. calvo and f. falceto , `` poisson reduction and branes in poisson - sigma models , '' arxiv : hep - th/0405176v2 a. s. cattaneo , `` from topological field theory to deformation quantization and reduction , '' proceedings of icm 2006 , vol . iii , 339 - 365 ( european mathematical society , 2006 ) , preprint nr . 06 - 2006 ( 2006 ) http://www.math.unizh.ch/fileadmin/math/preprints/icm.pdf

we study the robust nash equilibrium ( rne ) for a class of games in communications systems and networks where the impact of users on each other is an additive function of their strategies . each user measures this impact , which may be corrupted by uncertainty in feedback delays , estimation errors , movements of users , etc . to study the outcome of the game in which such uncertainties are encountered , we utilize the worst - case robust optimization theory . the existence and uniqueness conditions of rne are derived using finite - dimensions variational inequalities . to describe the effect of uncertainty on the performance of the system , we use two criteria measured at the rne and at the equilibrium of the game without uncertainty . the first is the difference between the respective social utility of users and , the second is the differences between the strategies of users at their respective equilibria . these differences are obtained for the case of a unique ne and multiple nes . to reach the rne , we propose a distributed algorithm based on the proximal response map and derive the conditions for its convergence . simulations of the power control game in interference channels , and jackson networks validate our analysis . resource allocation , robust game theory , variational inequality , worst - case robust optimization .

1103.4868

designs for multi - user communications networks and systems have been extensively used during the past decade to implement low - cost , scalable , and limited - message - passing networks . in doing so , transmitter and receiver pairs with local information determine their transmission strategies in an autonomous manner . to deploy such designs , it is essential to know whether they converge to a ( preferably unique ) equilibrium , and evaluate their performances at the emerging equilibrium / equilibria . strategic non - cooperative game theory provides an appropriate framework for analyzing and designing such environments where users ( i.e. , transmitter - receiver pairs ) are rational and self - interested players that aim to maximize their own utilities by choosing their transmission strategies . the notion of nash equilibrium ( ne ) , at which no user can attain a higher utility by unilaterally changing its strategy , is frequently used to analyze the equilibrium point of non - cooperative games . to derive the conditions for ne s existence and uniqueness , different approaches such as fixed point theory , contraction mapping and _ variational inequalities ( vi ) _ @xcite have been widely applied in both wired and wireless communication networks , including applications to flow and congestion control , network routing , and power control in interference channels @xcite . however , there are numerous sources of uncertainty in measured parameter values of communication systems and networks such as joining or leaving new users , delays in the feedback channel , estimation errors and channel variations . therefore , obtaining accurate values of users interactions may not be practical , and considering uncertainty and proposing a robust approach are essential in designing reliable communications systems and networks . to make a ne robust against uncertainties , two distinct approaches have been proposed in the literature : the bayesian approach where the statistics of uncertain parameters are considered and the utility of each user is probabilistically guaranteed , and the worst - case approach where a deterministic closed region , called the uncertainty region , is considered for the distance between the exact and the estimated values of uncertain parameters , and the utility of each user is guaranteed for any realization of uncertainty within the uncertainty region @xcite . both of these approaches have been applied to the power allocation problem in spectrum sharing environments and cognitive radio networks @xcite to study the conditions for ne s existence and uniqueness , where the uncertain parameters are interference levels and channel gains . however , to incorporate robustness in communications systems and networks , there exist multiple challenges such as : 1 ) how to implement robustness in a wider class of problems in communication systems ? 2 ) how to derive the conditions for existence and uniqueness of the robust ne ( rne ) ? 3 ) what is the impact of considering uncertainties on the system s performance at its equilibrium compared to that of the case with no uncertainty ? 4 ) how to design a distributed algorithm for reaching the robust equilibrium ? in this paper , we aim to answer the above questions using the worst - case robust optimization . in doing so , we consider a general class of games where the impact of users on each other is an additive function of their actions , which causes couplings between users . we refer to this class of game as the additively coupled games ( acgs ) . in the acg , we consider that the users observations of such impacts are uncertain due to variations in system parameters and changes in other users strategies . via the worst - case approach , we assume that uncertain observations by each user are bounded in the uncertainty region , and each user aims to maximize its utility for the worst - case condition of error . we refer to an acg that considers uncertainty as a robust acg ( racg ) , and an acg that does not consider uncertainty as the nominal acg ( nacg ) . to study the conditions for existence and uniqueness of the rne , we apply _ vi _ @xcite , and show that with bounded and convex uncertainty , the rne always exists . we also show that the rne is a perturbed solution of _ vi _ , and derive the condition for rne s uniqueness based on the condition for ne s uniqueness . furthermore , we compare the performance of the system at the rne with that at the ne in terms of two measures : 1 ) the difference between the users strategies at the rne and the ne , 2 ) the difference between the social utility at the rne and at the ne . when the rne is unique , we derive the upper bound for the difference between social utilities at the rne and at the ne , and show that the social utility at the rne is always less than that at the ne . however , obtaining these two measures is not straightforward when the ne is not unique . in this case , we demonstrate a condition in which the social utility at a rne is higher than that at the corresponding ne . finally , we apply the proximal response map to propose a distributed algorithm for reaching the rne , and derive the conditions for its convergence . the rest of this paper is organized as follows . in section ii , we summarize the system model of the nacg . in section iii , we introduce the racg and its rne . section iv covers the existence and uniqueness conditions if the rne . in section v , we show that when the utility function is logarithmic , the rne can be obtained via affine _ vi _ ( _ avi _ ) , and its uniqueness condition is simplified . in section vi , we propose distributed algorithm for reaching rne . in section vii , we discuss the effects of robustness for the case of multiple nash equilibria , followed by section viii , where we provide simulation results to illustrate our analytical developments for the power allocation problem and for the jackson networks . finally , conclusions are drawn in section ix . consider a set of communication resources divided into @xmath0 orthogonal dimensions , e.g. , frequency bands , time slots , and routes , which are shared between a set of users denoted by @xmath1 , where each user consists of a transmitter and a receiver . we assume that users do not cooperate with each other , and formulate the resource allocation problem as a strategic non - cooperative game @xmath2 , where @xmath3 is the set of players ( users ) in the game , @xmath4 is the joint strategy space of the game , and @xmath5 is the strategy space of user @xmath6 in which the strategy of each user is limited in each dimension . the sum of strategies of each user over all dimensions is bounded , i.e. , @xmath7\,\qquad \text{and } \qquad\,\sum_{k=1}^{k}a_{n}^{k}\leq a_{n}^{\texttt{max}}\}\ ] ] where @xmath8 and @xmath9 is the minimum and the maximum transmission strategy of each user in each dimension and @xmath10 is the bound on the sum of strategies of user @xmath6 over all dimensions , e.g. , the maximum transmit power of each user . the function @xmath11 is the utility function of user @xmath6 and depends on the chosen strategy vector of all users @xmath12 $ ] , where @xmath13 is the action of user @xmath6 . the vector of actions of all users except user @xmath6 is denoted by @xmath14 where @xmath15 is the strategy space of all users except user @xmath6 . in a non - cooperative strategic game , each user aims to maximize its own utility subject to its strategy space as @xmath16 which is the optimization problem of each user in the game . let us assume that the utility of each user is a function of its action and the impact of the others users actions over all dimensions , i.e. , @xmath17 where @xmath18 $ ] is the @xmath19 vector of the additive impact of other users on user @xmath6 with the following elements @xmath20 where @xmath21 $ ] is the vector of system s parameters for user @xmath6 , in which @xmath22 $ ] , and @xmath23 represents the system s parameters between user @xmath6 and user @xmath24 in dimension @xmath25 , e.g. , the channel gain between user @xmath6 and user @xmath24 in sub - channel @xmath25 ; @xmath26 $ ] is the vector of the effect of system on user @xmath6 , e.g. , noise in sub - channels for user @xmath6 , and @xmath27 is related to the @xmath28 dimension . assuming @xmath29 as a linear function is very practical in communication networks . for example , the interference in each receiver is a linear and additive function of the actions of other users and channel gains plus noise . we also assume that * a1 * ) the utility function of each user is an increasing , twice differentiable , and concave function with respect to @xmath30 and has bounded gradients ; * a2 * ) the utility function of each user is a decreasing , twice differentiable , and convex function with respect to @xmath31 ; * a3 * ) the utility function of each user is twice differentiable over @xmath30 , and @xmath32 . note that * a1 * is a typical utility function in communications networks @xcite ; * a2 * is a practical assumption for sharing resources between non - cooperative users @xcite ; and * a3 * indicates that the utility function is continuously differentiable with respect to @xmath30 and @xmath32 . as such , the above assumptions * a1 * -*a3 * are typical in communications systems and networks . interactions between users are studied at the ne , which corresponds to the strategy profile @xmath33 such that for any other strategy profile , we have @xcite @xmath34 in what follows , we denote the achieved utility of user @xmath6 at the ne by @xmath35 and the social utility at the ne by @xmath36 , derive the conditions for ne s existence and uniqueness conditions via reformulating ne through _ vi _ , and show that in this way , the ne and the rne can be analyzed similarly . * remark 1*. consider @xmath37 , where @xmath38 where @xmath39 denotes the column gradient vector of @xmath40 with respect to @xmath30 . the ne of @xmath41 can be obtained by solving @xmath42 ( proposition 1.4.2 in @xcite ) as @xmath43 since @xmath44 is a continuous and concave function with respect to @xmath45 , @xmath46 is a continuous map . from ( [ an ] ) , the set @xmath47 is convex and compact . therefore , the solution set of @xmath42 is nonempty and compact ( theorem 2.2.1 in @xcite ) . consequently , the ne of acg exists . * remark 2*. let us consider the following definitions of mapping @xmath46 , @xmath48 where @xmath49 and @xmath50 are the @xmath51 jacobian matrices of @xmath52 with respect to @xmath30 and @xmath53 , respectively . now , consider the @xmath54 matrix @xmath55 whose elements are @xmath56_{nm}= \left\{\begin{array}{c } \alpha_{n}^{\text{min } } \,\,\,\,\,\qquad\text{if } \qquad\ , m = n\nonumber \\ -\beta_{nm}^{\text{max } } \,\qquad \text{if } \qquad\ , m\neq n.\nonumber \\ \end{array } \right.\end{aligned}\ ] ] when @xmath55 is a @xmath57-matrix , the mapping @xmath46 is strictly monotone and the ne is unique ( theorem 12.5 in @xcite ) . this setup is applicable to some important problems in communications systems and networks , such as the additively coupled sum constrained game @xcite , which can be used to formulate , e.g. , power control in interference channels , and the jackson network . these two systems are chosen to illustrate the validity of our approach and analysis . table [ tableexample ] contains the parameters of the power control , and the jackson network games . the power control in interference channels can be formulated by a acg , where each player , consisting of a transmitter and receiver pair , competes with other players to maximize its total data rate over @xmath0 distinct sub - channels . the strategy of each user is its transmit power in @xmath0 sub - channels , where @xmath58 is the direct channel gain between the transmitter and receiver pair @xmath6 , and @xmath59 and @xmath60 are the normalized interference channel gain between transmitter @xmath24 and receiver @xmath6 , and the normalized noise power in sub - channel @xmath25 , respectively . for the jackson network in table [ tableexample ] , arriving packets to node @xmath6 are split into @xmath61 $ ] traffic classes , and the input rate and service rate for class @xmath25 are @xmath62 and @xmath63 , respectively . here , @xmath62 is the strategy of player @xmath6 in dimension @xmath25 . the total rate is subject to the minimum rate constraint , i.e. , @xmath64 . a packet of class @xmath25 completing service at node @xmath24 is routed to node @xmath6 with probability @xmath65 , or exit the network with probability @xmath66 . in this scenario , we denote @xmath67_{nm}=r_{nm}^{k}$ ] , @xmath68 , and @xmath69_{nm}$ ] . it can be shown that the user s utility for minimizing @xmath70 queueing delay can be expressed by @xmath71 , where @xmath72 $ ] , and @xmath73$]@xmath74 . the optimization problem can be rewritten by maximizing @xmath75 subject to the minimum data rate constraint for each user . as stated earlier , users may encounter different sources of uncertainty caused by variations in @xmath76 and/or @xmath77 , which cause variations in the utility function of each user , and prevent users from attaining their expected performance . to deal with such issues , we assume that all uncertainties for a given user can be modeled by variations in the user s observation @xmath31 , i.e. , @xmath78 where @xmath79 $ ] , @xmath80 $ ] , and @xmath81 $ ] are the actual observation , the nominal observation , and the error in observation of user @xmath6 , respectively . in the worst - case robust optimization theory , uncertainties are assumed to be bounded to the uncertainty region , stated by @xmath82 where @xmath83{(\sum_{i=1}^{n}t_{i}^{p})}$ ] denotes the linear norm with order @xmath84 . in communication and network systems , the ellipsoid region , i.e. , @xmath85 , has been commonly used to model uncertainty @xcite . we also use the norm with @xmath85 in our robust game , and denote the uncertainty region by @xmath86 so as to indicate that it is an additive function of the actions of other users and system parameters , i.e. , it is not a fix region . the effect of uncertainty in @xmath87 is highlighted by a new variable in the utility function of each user as @xmath88 in such a way that @xmath89 the objective of the worst - case approach is to find the optimal strategy for each user that optimizes its utility under the worst condition of error in the uncertainty region . in this approach , from * a2 * , the optimization problem of each user can be formulated as @xcite @xmath90 where @xmath91 is the achieved utility of user @xmath6 in the worst - case approach . the domain of optimization problem ( [ optrobsut ] ) is defined by @xmath92 which is a function of other users strategy . we represent the racg by @xmath93 where @xmath94 . the solution to ( [ utilityrobsut ] ) for user @xmath6 is a pair @xmath95 that satisfies @xcite @xmath96 which is the saddle point of ( [ utilityrobsut ] ) . using the above , the equilibrium of the robust game @xmath41 is defined below . * definition 2 . * the rne of racg corresponds to the strategy profile @xmath97 if and only if for any other strategy profile @xmath98 we have @xcite @xmath99 we denote the achieved utility of user @xmath6 at the rne by @xmath100 and the social utility at the rne by @xmath101 . now we derive the characteristics of the rne in the racg from the ne in the nacg . for convenience , in what follows , we omit the arguments @xmath76 and @xmath102 in @xmath103 . to analyze the existence of rne , we encounter two problems . first , by considering uncertainty in the utility of each user , the utility may become non - convex , and analyzing rne may become impossible . second , the strategy space of user @xmath6 changes to @xmath104 which is not a fix set and is a function of the other users actions . therefore , convexity of the optimization problem of each user is not a sufficient condition for the existence of rne , meaning that we need to utilize _ vi _ in the sequel . * lemma 1 . * 1 ) for the uncertainty region in ( [ iii-1 ] ) , the strategy of each user is a convex , bounded , and closed set . 2 ) @xmath105 is a concave and continuous differentiable function of @xmath30 for every @xmath76 , where , @xmath106 and @xmath107 where @xmath108 $ ] , @xmath109 $ ] , and @xmath110 is defined as @xmath111 the robust game is @xmath112 . see appendix a. * theorem 1 * : for any set of system parameters and strategy space of users , there always exists an rne for @xmath113 . from part 2 ) in lemma 1 , rne is an instance of the generalized nash equilibrium ( gne ) ( see ( 2 ) in @xcite ) , and @xmath114 is the rne iff it is a solution to @xmath115 , where @xmath116 and @xmath117 . since @xmath118 is a convex set and @xmath105 is a concave and continuous differentiable function with respect to @xmath30 , the necessary convexity assumptions for the existence of a solution to @xmath119 hold ( theorem 1 in @xcite ) , meaning that a rne always exists . since the closed form solution to ( [ optrobsut ] ) can not be obtained , the fixed - point algorithm and the contraction mapping can not be applied as in @xcite to derive the conditions for rne s uniqueness . to overcome these difficulties , we show that the rne can be considered as a perturbed ne of the nacg , and that the condition for rne s uniqueness can be derived without a closed form solution to ( [ optrobsut ] ) . * lemma 2 . * @xmath120 is a perturbed bounded version of mapping @xmath46 . see appendix b. from lemma 1 , @xmath118 is a closed and convex set , and form lemma 2 , @xmath120 is a perturbed bounded mapping @xmath46 . therefore , the rne is a perturbed solution to @xmath121 . consequently , rne s uniqueness condition can be obtained from the perturbed ne s uniqueness condition . * theorem 2 . * when @xmath55 is a @xmath57 matrix , for any bounded value of @xmath122 $ ] : 1 ) @xmath113 has a unique rne ; 2)the social utility at the rne is always less than that at the ne , i.e. , @xmath123 ; 3 ) the distance between the strategy spaces at the rne and at the ne is @xmath124 where @xmath125 is the strong monotonicity constant for mapping @xmath126 . see appendix c. from theorem 2 , racg s performance can be examined and compared to that of the acg through the difference between the upper bound of users strategies in ( [ upperbound of variations ] ) at their equilibriums . also , from ( [ upperbound of variations ] ) , the social utility s reduction at the rne ( compared to ne ) can be approximated . consider @xmath127 , where @xmath128 is a @xmath54 matrix whose elements are @xmath129 in part 4 of appendix c , we show that the difference between social utilities at the rne and the ne is @xmath130 when @xmath55 is a @xmath57 matrix , @xmath131 is the attractor for @xmath132 ( proposition 2.4.10 in @xcite ) , i.e. , @xmath133 meaning that when uncertainty approaches zero , the rne converges to the ne . consider the power allocation problem ( example 1 in table i ) , where the transmit power of user @xmath6 on sub - channel @xmath25 is @xmath134 in ( [ an ] ) . we have @xmath135 . in this case , @xmath136 , and @xmath137 . therefore , @xmath138 for simplicity , assume that the normalized noise power in all sub - channels for each user are the same , and @xmath139 . in such cases , the matrix @xmath55 is a @xmath57 matrix iff @xmath140 which is the uniqueness condition for the power control game . otherwise , the power allocation problem may have multiple nash equilibria . as a numerical example for theorem 1 , we consider two users and two sub - channels , and assume @xmath141 . to ensure that the monotonicity condition ( [ uniqnessconditionsof gamepower control ] ) of the power allocation game holds and @xmath55 is a @xmath57 matrix , the interference channel gain between users is chosen much less than the direct channel gain between transmitters and their receivers , i.e. , @xmath142 . in our simulation , the matrix @xmath55 is @xmath143\ ] ] since ( [ uniqnessconditionsof gamepower control ] ) is satisfied for the above matrix , it is a strictly monotone matrix . figs . [ 1 ] ( a ) and [ 1 ] ( b ) show the mappings @xmath144 and @xmath145 for the allocated power levels in subchannel 1 for users 1 and 2 , respectively . for @xmath146 , the mapping @xmath52 is obtained by considering a fixed transmit power of the other user , and as we show in figs . [ 1 ] ( a ) and [ 1 ] ( b ) , it is monotone . [ 2 ] shows the surface of the social utility of users 1 and 2 for @xmath147 and @xmath148 as their strategies in sub - channel 1 and 2 , respectively . we consider @xmath149 for both users . in this case , our simulations show that : 1 ) both the ne and the rne are unique ; 2 ) the social utility at ne is @xmath150 . the social utility at the rne is @xmath151 , which is less than the social utility at the ne , as expected from part 1 in theorem 1 ; and 3 ) at the ne , the allocated power for users 1 and 2 are @xmath152 and @xmath153 , respectively ; and the rnes for users 1 and 2 are @xmath154 and @xmath155 . in this example , the upper bound in ( [ upperbound of variations ] ) is @xmath156 , and the exact value of @xmath157 is @xmath158 , which is less than its upper bound ( [ upperbound of variations ] ) . consider @xmath159 , where @xmath160 is the conjugate transpose of @xmath161 and @xmath162 is the maximum eigenvalue of the matrix . in our simulation @xmath163 and @xmath164 . from ( [ diffrencebetweenustilitues ] ) , the distance between the social utilities at the rne and at the ne is @xmath165 , and in simulation , this difference is to @xmath166 . note that all of the above results validate theorem 2 . we now consider a special form of uncertainty region and utility function that are pertinent to example 1 in table i , to show that in such cases , _ vi _ becomes very simple , and the strict monotonicity requirement is relaxed to the positive semidefiniteness of an affine mapping . let the utility of each user be @xcite @xmath167 where @xmath168 is the fixed system parameter related to each dimension of each user . assume that for the uncertain parameter of each user , we have @xmath169 where @xmath170 , @xmath171 and @xmath172 are the actual value , the nominal value , and the error , respectively . the uncertainty region for each user is @xmath173 where @xmath174 is the bound on the uncertainty region for user @xmath6 in sub - channel @xmath25 . the above formulation models the power control game when uncertainty emanates from variations in the channel gain between transmitter @xmath24 and receiver @xmath6 . * proposition 1 . * for an acg with utility function ( [ utilitylogconvex ] ) : 1 ) the ne is the solution to an affine _ vi _ denoted by @xmath175 where @xmath176 , and @xmath177 where @xmath178 and @xmath179 . 2 ) consider a @xmath54 matrix @xmath180 whose elements are @xmath181 the ne is unique when @xmath182 see appendix d. from proposition 1 , when the impact of users on each other is sufficiently low , the ne is unique . for the example of power control game , proposition 1 implies that when interference between users is sufficiently low , the ne of nacg is unique @xcite . we now derive the rne uniqueness condition for such cases . * theorem 3 . * for the utility function ( [ utilitylogconvex ] ) : 1 ) the _ avi _ mapping of the rne is @xmath183 , where @xmath184 where @xmath185 and @xmath186 $ ] ; 2 ) when ( [ conditionproposition1 ] ) holds , the rne of @xmath113 is unique for any bounded @xmath187 ; 3 ) when ( [ conditionproposition1 ] ) holds , the total utility of each user at the rne is always less than that at the ne , and the upper bound on the strategy space of each user is @xmath188 , where @xmath189 where @xmath190 is the minimum eigenvalue of matrix @xmath191 , @xmath192 $ ] and @xmath193 is the maximum element of the vector . see appendix e. * remark 3 . * from theorems 2 and 3 , rne s uniqueness condition is not related to the size of the uncertainty region . theorems 2 and 3 show that for a closed , bounded , and convex uncertainty region , rne s uniqueness condition is the same as ne s uniqueness condition . by rearranging _ avi _ of the rne for the utility function ( [ utilitylogconvex ] ) , the best response solution of the racg is @xmath194_{a_{nk}^{\text{min}}}^{a_{nk}^{\text{max}}},\ ] ] which is similar to ( 13 ) in @xcite for the power control game in spectrum sharing environments . from theorem 2 in @xcite , the rne s uniqueness condition is related to the size of the uncertainty region . but if we use _ vi _ to analyze the rne , its uniqueness condition is independent of the size of the uncertainty region . to develop a distributed algorithm for this class of games , we use the proximal - response map @xmath195 . consider @xmath196 $ ] , where @xmath195 is the solution to following optimization problem with respect to @xmath197 @xmath198,\ ] ] where @xmath199 $ ] belongs to @xmath118 the above problem can be decomposed into @xmath200 subproblems as @xmath201 , \quad , \textbf{b}\in \widehat{\mathcal{a}}(\textbf{a}_{-n } ) , \quad \forall n \in \mathcal{n},\ ] ] in this way , a distributed iterative algorithm is developed . consider @xmath202 and @xmath203 as solutions for user @xmath6 in the current and pervious iterations , respectively . when a user is informed about the other users solutions before updating its strategy , it can obtain the solution to ( [ proximalresponsemapn ] ) . the distributed algorithm based on proximal response map is summarized in table [ distributedalgorith ] , where users update their transmit strategies at discrete instances @xmath204 denoted by @xmath205 $ ] , @xmath206 denotes the transmission strategy of user @xmath6 at iteration @xmath204 , and @xmath207 is the observation of user @xmath6 at @xmath204 measured by its receiver and sent to the transmitter . in theorem 4 below , we obtain the conditions for convergence of the iterative algorithm . * theorem 4 . * as @xmath208 , the distributed algorithm in table [ distributedalgorith ] converges to the unique rne from any initial strategy @xmath209 , if @xmath55 is a @xmath57 matrix , and @xmath210 . see appendix f. note that the condition @xmath211 holds for the jackson network . for the power control game , when @xmath55 is a @xmath57 matrix , interference in the system is very low , and consequently , the signal to interference and noise ratio of each usr is high . for this case , @xmath212 , and the utility function of each user is @xmath213 which also meets @xmath211 . as we see in theorem 4 , the distributed algorithm converges to the unique ne when @xmath55 is a @xmath57-matrix irrespective of the size of the uncertainty region , so long as the uncertainty region is closed and convex . * remark 4 . * in lemma 1 , we showed that @xmath214 is concave . also the proximal response map is strictly convex . therefore , the lagrange function can be used to derive the solution to ( [ proximalresponsemapn ] ) for each user at each iteration as @xmath215 where @xmath216 is the lagrange multiplier for user @xmath6 that satisfies @xmath217 the solution to ( [ lagrangeofproximal ] ) with respect to @xmath218 is @xmath219 user @xmath6 solves ( [ solution ] ) to obtain @xmath206 in each iteration . for ( [ utilitylogconvex ] ) , the proximal map s solution is @xmath220 where @xmath221 . for example , the proximal map s solution to the power control game is @xmath222_{a_{nk}^{\text{min}}}^{a_{nk}^{\text{max}}},\ ] ] so far , based on ne s uniqueness condition , we obtained rne s uniqueness condition . now we study the characteristics of rnes when nacg has multiple nes . in general , doing so is not straightforward since the _ vi _ mapping for nacg is non - monotone and non - smooth , which makes it difficult to study the characteristics of the perturbed nes ( rnes ) @xcite . to compare the case of multiple nes with that of a single ne , consider the power control game , when @xmath55 is not a @xmath57 matrix ( e.g. , @xmath223 ) . the mapping @xmath126 is non - monotone for both users . as we see in fig . [ 4 ] , there are multiple local optima on the surface of the utility function that correspond to multiple nes for this game . in this case , at the nominal ne , the convergence points for users 1 and 2 are @xmath224 and @xmath225 , respectively , and @xmath226 . when uncertainty is @xmath227 , the rne converges to @xmath228 and @xmath229 , and @xmath230 . this example points out that introducing uncertainty may increase the social utility at the rne when the nacg has multiple nes , which is in line with simulation results in @xcite . this is because rne relates to one local optima on the surface of the utility function , which is different from the local optima at the ne . also , considering uncertainty results in users interfering less with each other at the rne compared to the ne @xcite . this observation shows the benefit of implementing racg in communication systems which may increase the social utility as compared to that of nacg . but , obtaining the conditions under which the social utility increases is not easy . this is because utility function of each user is a non - linear function with respect to its uncertainty region and the other users uncertainty regions . so , we focus on a special case where the strategy of each user is a decreasing function of the bound of uncertainty region . in proposition 2 below , we obtain the condition for increasing the social utility of the racg as compared to that of the nacg . * proposition 2 . * when @xmath231 is a semi negative definite matrix and @xmath232 for all users , the social utility at the rne is higher than that at the ne . see appendix g. proposition 2 implies that when the reduction in the social utility due to the decrease in user s strategies is less than their increase in the social utility due to the decrease in other usrs strategies , introducing robustness in the game increases the social utility . note that this is one case in which the social utility at rne is higher than the social utility at the corresponding ne , and there may be other cases as well . * remark 5 . * when the solution of affine _ vi _ is a monotone decreasing function of @xmath233 , the _ avi _ mapping is a semi - negative matrix ( see appendix h ) . for the power control game , proposition 2 is simplified to @xmath234 this means that when all interference channel gains are sufficiently greater than the direct channels gains , introducing robustness increases the social utility . this is an opportunistic phenomenon in robust games when the game encounters multiple nes . in order to benefit from this and increase the social utility , we propose an opportunistic distributed algorithm in table [ tableopportunesticalgorithm ] obviously , checking the conditions of proposition 2 in a distributive manner is not easy . in addition the social utility may increase under other conditions . therefore , all users play the game @xmath41 without considering the uncertainty . if @xmath235 for user @xmath6 , i.e. , the nacg has multiple nes , users assume uncertainty in their observations . when their utilities increase , they expand their uncertainty regions . otherwise , they interrupt the algorithm . in this way , all users make an effort to escape from their local optima in a distributed manner by playing the robust game . to implement this algorithm , we assume that users update their transmit strategy at discrete time slots @xmath236 with duration of @xmath237 . the vectors @xmath238 and @xmath239 are the transmit strategy and the observation of each user at the end of the iteration time @xmath240 . besides , users exchange the values of @xmath238 and @xmath241 at the end of each iteration . we use simulations in the two examples in table i to provide an insight into the performance of @xmath113 for different bounds on uncertainty region as compared to that of @xmath41 . in the following simulations , the value of @xmath233 is normalized to the nominal value of @xmath242 , i.e. , @xmath243 , each uncertainty region is considered as a linear norm with order 2 , i.e. , an ellipsoidal region , and uncertainty for all users is assumed to be the same , denoted by @xmath244 . for the power control game , we begin by studying the effect of uncertainty on its performance in both robust and non - robust approaches in terms of utility variations at their equilibria . to do so , we consider @xmath245 users and @xmath246 , and the amount of uncertainty is assumed to be @xmath247 at the rne . after convergence to the rne and to the ne , the system parameter varies uniformly from @xmath248 to @xmath249 , which causes variations in the utility of each user at the ne and at the rne . variations in the social utility are shown in fig . [ 5 ] . note that the social utility varies considerably at the ne of the nominal game for both values of @xmath250 $ ] , meaning that communication is very unreliable from the user s perspective . in contrast , the total system utility at the rne of the robust game is stable for both cases . note that although we assumed @xmath247 in the racg , reliable transmissions is provided even at values higher than @xmath247 , e.g. , up to @xmath249 . next , in fig . [ 6 ] , we compare the effect of uncertainty when theorem 2 holds with that of the case when it does not hold , in terms of the ratio of the social utility at the rne and at the ne for different amounts of uncertainty . simulations are performed for rayleigh fading channels for bounded and uniformly generated errors for each cross sub - channel gain . to satisfy the ne s uniqueness condition , the channel gains are obtained in such a way that @xmath142 , and for multiple nes , we set @xmath223 . the ratio of the social utility in fig . [ 6 ] is obtained by averaging over 100 channel realizations . note that when the ne is unique , the total throughput of the system gradually decreases , but for the case of multiple nes , no uniformity is observed . for example , when @xmath251 , the total achieved utility is higher at the rne as compared to that at the ne ; and when @xmath252 , the ratio falls substantially . the trend is not monotonic for different values of uncertainty , e.g. , the social utility at the rne is higher for @xmath253 as compared to those for @xmath254 and @xmath255 . in fig . [ 7 ] , the throughput of the users versus iteration numbers are shown for the proximal response map when @xmath256 . note that as shown in this figure , the convergence time of the proximal response map is longer than that of the iwfa without any uncertainty for users . finally , in fig . [ 8 ] , we compare the performance of the opportunistic approach with that of the nominal game using @xmath257 where @xmath258 is the achieved utility of user @xmath6 by utilizing the opportunistic algorithm at the end of the algorithm . the value of @xmath259 is obtained by averaging over 1200 channel realizations for @xmath260 $ ] , @xmath261 $ ] for different values of @xmath262 . note that for very high interference levels , i.e. , when @xmath263 , the efficiency of the proposed opportunistic approach is considerably higher as compared to that of moderate interference levels , i.e. , when @xmath264 . thus , the opportunistic algorithm performs significantly better in high interference levels when the number of sub - channels is high . to show the effect of uncertainty on the system performance in jackson networks , we consider a network with @xmath265 nodes and @xmath266 traffic classes 9](a ) and [ 9](b ) show the effect of uncertainty in @xmath62 on the convergence of jacobi scheme and the gradient approach for reaching the ne . in this figures , we show the percentage of increasing the total delay in the network under perturbation as compared to that at the ne , i.e. , @xmath267 , where @xmath268 is the total delay under perturbation and @xmath269 is the delay at the ne . note that when robustness is not applied , neither of the two algorithms converge to the ne , and increasing uncertainty increases the queuing delay . [ 10 ] ( a ) and [ 10 ] ( b ) show the rne for @xmath252 and @xmath270 , respectively , by using the proximal response map . note that in both cases , the rnes converge to the vicinity of the nominal ne as compered to figs . [ 9](a ) and [ 9 ] ( b ) . for example , when @xmath270 , the total delay increases up to @xmath271 as compared to that at the ne , while the total delay is about @xmath272 at the rne . therefore , rnes are stable , and the iterative algorithm converges very fast . note that proximal response map , the convergence rate increases when uncertainty region is expanded . [ 11 ] shows the probability of rne s convergence , versus the total routing probability @xmath273 for different uncertainty regions . note that by increasing uncertainty , the system converges to the rne for a smaller value of @xmath274 as compared to that of the ne ( i.e. , @xmath275 ) . for example , if @xmath276 , only for @xmath277 , the system converges to its equilibrium , while for @xmath278 , the value of @xmath274 can be up to @xmath279 for convergence to the rne . [ 11 ] shows that the effect of uncertainty is more profound in a network with a high probability of packet loss , causing poor performance , i.e. , large delays in the network . therefore , from a practical perspective , when the system encounters a high degree of uncertainty , all nodes should reduce their probability of packet drops and increase their routing probability . we studied the rne for a wide range of problems in communication systems and networks when users impact on each other can be modeled as an additive function of system parameters and users strategies . in this game , since the user s observations of these impacts are uncertain , each user optimizes its utility using worst - case robust optimization . we showed that the theory of finite - dimension _ vi _ and its sensitivity analysis can be used to obtain the conditions for the existence and uniqueness of the rne when there is no closed - form solution to the optimization problem of each entity . we also proposed a distributed algorithm for reaching the rne . in the case of multiple nes , simulations showed that at a rne , the social utility may be higher than that at the ne , and we derived the conditions for observing such a case . to benefit from this , we proposed a distributed algorithm to improve the social utility at the rne . simulations confirmed our analysis for two examples , namely , the power control in interference channels , and delay minimization in jackson networks . \1 ) @xmath280 is a linear function of other users strategies and system parameters . besides , the norm function is a convex function bounded to @xmath233 . therefore , @xmath281 is a convex , bounded and close set . 2 ) to prove the concavity of ( [ psi2 ] ) with respect to @xmath30 , consider @xmath282 , and @xmath283 for any positive value @xmath284 $ ] . we have @xmath285 note that inequality ( [ ps2 ] ) is based on concavity of @xmath286 with respect to @xmath30 . therefore , @xmath283 is a concave function of @xmath30 . based on concavity of @xmath283 , the lagrange dual function of ( [ psi2 ] ) for the uncertainty region is @xmath287 where @xmath288 is the nonnegative lagrange multiplier that satisfies ( [ iii-1 ] ) , i.e. , @xmath289 the solution to ( [ lagrangedualfunction ] ) for @xmath290 can be obtained by the optimality condition of the optimization problem without the constraint @xcite , i.e. , @xmath291 , which is equivalent to @xmath292 considering ( [ solution2 ] ) in ( [ lagrangemultiplier ] ) , the uncertain parameter is @xmath293 , where @xmath108 $ ] , @xmath109 $ ] , and @xmath110 is @xmath294 using @xmath110 in the utility function @xmath295 , we have @xmath296 comparing ( [ psi2 ] ) with @xmath297 indicates that the difference between @xmath298 and the utility function of the nominal game is the extra term @xmath299 . from * a2 * , @xmath299 is continuous . therefore @xmath105 is continuous with respect to @xmath30 . the derivative of @xmath298 with respect to @xmath30 is @xmath300 where @xmath301 is a @xmath19 vector whose elements are equal to one . the last term in ( [ derivativepsi3 ] ) contains @xmath302 . from * a3 * , the term @xmath302 exists . therefore , @xmath105 is differentiable with respect to @xmath30 . now , the optimization problem of each user can be rewritten as @xmath303 . therefore , the game can be reformulated as @xmath304 . for the racg , we have @xmath305 , and @xmath306 , where @xmath307 is obtained by ( [ derivativepsi3 ] ) for user @xmath6 . let @xmath308 denote variations in the system s of parameters and other users strategies for user @xmath6 where @xmath309 and @xmath310 are variations of @xmath311 and @xmath53 , respectively . when @xmath312 , we have @xmath313 and vice versa . from * a1 * and * a2 * , the mapping @xmath314 is continuous and differentiable around the uncertain parameters . we use the taylor series of the uncertain parameter and write @xmath315_{\hat{\textbf{z}}_n=0}+ \varepsilon_{n}[\nabla_{\hat{\textbf{z}}_n } \widetilde{\mathcal{f}}_n(\textbf{a})]_{\hat{\textbf{z}}_n=0 } + [ \sum_{i=2}^{\infty } \frac{1}{i ! } ( \varepsilon_{n})^i(\nabla^i_{\hat{\textbf{z}}_n } \widetilde{\mathcal{f}}_n)]_{\hat{\textbf{z}}_n=0}\ ] ] for @xmath316 , ( [ tailorseriesofmappingf ] ) is equivalent to @xmath317_{(\varepsilon_n=0 ) } \\ \label{tailorseriesmappingf2 } & & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\ ! -\frac{\varepsilon_n}{2}[\nabla_{\textbf{a}_n \boldsymbol{f_n}}^2 u_n(\textbf{a}_n , \textbf{f}_n- \varepsilon_n \boldsymbol{\vartheta}_{n } ) \nabla_{\textbf{z}_n } \boldsymbol{f_n } - \varepsilon_n \nabla_{\textbf{f}_n \textbf{f}_n}^2 u_n(\textbf{a}_n , \textbf{f}_n- \varepsilon_n \boldsymbol{\vartheta}_{n } ) \nabla_{\textbf{z}_n } \boldsymbol{\vartheta}_{n } ] _ { ( \varepsilon_n=0 ) } \\ & & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\ ! \label{tailorseriesmappingf3 } - \frac{\varepsilon^2_n}{3!}[\nabla^3_{\partial \textbf{a}_n \boldsymbol{f_n}^2 } u_n(\textbf{a}_n , \textbf{f}_n- \varepsilon_n \boldsymbol{\vartheta}_{n } ) ( \nabla_{\textbf{z}_n } \boldsymbol{f_n } ) ^2 \times \textbf{1}_k^{{\scriptsize{\textnormal{t } } } } + \nabla_{\textbf{a}_n \boldsymbol{f_n}}^2 u_n(\textbf{a}_n , \textbf{f}_n- \varepsilon_n \boldsymbol{\vartheta}_{n } ) \nabla^2_{\textbf{z}_n \textbf{z}_n } \textbf{f}_n \times \textbf{1}_k^{{\scriptsize{\textnormal{t}}}}]_{(\varepsilon_n=0)}\\ & + & o \nonumber \end{aligned}\ ] ] from ( [ utilityrobust2 ] ) , the first term in ( [ tailorseriesmappingf ] ) is equal to @xmath318 . since @xmath280 is a linear function of system parameters , the last term in ( [ tailorseriesmappingf3 ] ) is equal to zero . from ( [ mappingg ] ) , we have @xmath319-\frac{\varepsilon^2_n}{3!}[\nabla_{\textbf{a}_n \boldsymbol{f_n}^2}^3 v_n(\textbf{a}_n , \textbf{f}_n ) \times ( \nabla_{\textbf{z}_n}\boldsymbol{f_n})^2\times \textbf{1}_k^{{\scriptsize{\textnormal{t } } } } ] + o , \end{aligned}\ ] ] from * a1 * , all the derivatives of @xmath297 are bounded . therefore , the last terms in ( [ perturbedtailorseries ] ) are bounded , and @xmath307 is the perturbed bounded version of @xmath52 . \1 ) consider the bounded perturbation of mappings @xmath46 and @xmath320 caused by variations in system parameters as @xmath321 . since the strategy space of all users in each dimension is bounded as in ( [ an ] ) , and the uncertainty region is bounded and convex , this region is also bounded , i.e. , @xmath322 . any solution to the worst - case robust optimization in ( [ utilityrobsut ] ) corresponds to a realization of @xmath323 , where @xmath324 , and @xmath325 depends on @xmath326 obtained by ( [ utilityrobsutsaddlepoint ] ) for each user , and always @xmath327 . when @xmath46 is continuous and strictly monotone on the closed convex set @xmath47 , meaning that @xmath55 is a @xmath57 matrix , the solution to @xmath328 , denoted by @xmath329 , is a monotone and single - valued mapping on its domain ( exercise 2.9.17 in @xcite ) , i.e. , @xmath330 thus , when @xmath331 , we have @xmath332 , which is single valued on @xmath333 , i.e. , a unique solution for all @xmath334 . this completes the proof of the uniqueness of rne under the @xmath57 property of @xmath55 . 2 ) recall that when @xmath55 is a @xmath57 matrix , @xmath46 is strictly monotone , and the utility is strictly convex . since @xmath47 is convex in @xmath335 , and @xmath336 is a continuous mapping on @xmath47 , the solution to @xmath328 is always a compact and convex set ( corollary 2.6.4 in @xcite ) . also , since @xmath131 is the optimum value of this convex set for @xmath337 , i.e. , @xmath338 , any point in this set is less than @xmath131 , which is a solution to @xmath328 . note that @xmath339 belongs to this set . since @xmath55 is a @xmath57 matrix and @xmath126 is strictly monotone , we have @xmath340 which is also valid for @xmath339 . as such , the utility at the rne is less than that at the ne . 3 ) since @xmath126 is strongly monotone , there is a unique solution denoted by @xmath341 , which can be considered as the worst - case robust solution to @xmath113 for @xmath342 . now , both @xmath131 and @xmath343 must satisfy @xmath344 where @xmath345 and @xmath346 is the @xmath347 all zero vector . by rearranging [ inequality1 ] , we get @xmath348 since @xmath349 is the co - coercive function of @xmath350 ( proposition 2.3.11 in @xcite ) , the left hand side of ( [ inequality3 ] ) is always less than @xmath351 . using schwartz inequality for the right hand side , we have @xmath352 since @xmath353 and @xmath354 correspond to @xmath355 and @xmath339 , respectively , ( [ upperbound of variations ] ) can be obtained . 4 ) since the difference between utility functions of each user at rne and at ne is equal to first term of the taylor series of @xmath356 with respect to all variations in the strategies of user @xmath6 and other users , we have @xmath357 which is equivalent to @xmath358 when @xmath233 is sufficiently small , the derivative of the strategy of each user is approximately equal to @xmath359 by expanding ( [ theoream42 ] ) for all users in the game , we have @xmath360 by replacing ( [ upperbound of variations ] ) into ( [ diffrencebetweenustilituesappecndix ] ) , the approximation ( [ diffrencebetweenustilitues ] ) is obtained . \1 ) from ( [ utilitylogconvex ] ) , the best - response of the nacg is @xmath361_{a_{nk}^{\text{min}}}^{a_{nk}^{\text{max } } } , \ ] ] where the lagrange multiplier @xmath288 for each user is so chosen to satisfy the sum constraint @xmath362 . therefore , the best response of this problem can be written as an _ avi _ , denoted by @xmath363 , where @xmath364 is obtained from ( [ avilog ] ) @xcite . 2 ) for this case , the game has a unique ne when @xmath365 is strongly monotone or when @xmath366 is positive definite ( sections 2.3 and 2.4 in @xcite ) . by some rearrangements , we have @xmath367 where @xmath368_{nm}=\frac{x^{k}_{nm}}{x^{k}_{nn}}$ ] . when all @xmath369 are positive definite matrices , @xmath370 is strongly monotone . consider @xmath180 in ( [ mmax ] ) , and its strictly lower and upper triangle matrix as @xmath371 , and @xmath372 , respectively . let @xmath373 and @xmath374 . when for any @xmath375>0 $ ] , we have @xmath376 all @xmath369 for all @xmath25 are positive definite matrices ( proposition 1 in @xcite ) . by rearranging ( [ proofpro14 ] ) , we obtain ( [ conditionproposition1 ] ) . therefore , when ( [ proofpro14 ] ) holds , _ avi _ has a unique solution and consequently , the ne is unique . \1 ) from lemma 1 , the map of racg is the perturbed map of nacg . since the map of nacg is linear for the utility function ( [ utilitylogconvex ] ) , the perturbed map is @xmath377 where @xmath378 and @xmath379 are the @xmath28 elements of @xmath380 and @xmath381 , respectively . now , ( [ mappingm3 ] ) can be rewritten as @xmath382 therefore , the map at the rne is ( [ avilog ] ) . 2 ) since @xmath383 is bounded in @xmath384 $ ] and the uncertainty region is bounded , the value of @xmath385 is bounded . therefore , for any bounded uncertainty region considered by users in the racg , its _ avi _ can be rewritten as @xmath386 , where @xmath387 and @xmath388 . therefore , the rne is the perturbed solution to @xmath389 . from theorem 4.3.2 in @xcite , when @xmath390 is semicopositive , the _ avi _ has a unique solution for any value of @xmath391 . therefore , the robust game has a unique solution for any bound on the uncertainty region . 3 ) the second part can be obtained the same as in part 2 of theorem 1 . recall that @xmath392 is strongly monotone on @xmath47 when there exist a @xmath125 , such that for all @xmath393 , and @xmath394 we have @xmath395 when @xmath396 , for our linear formulation we have @xmath397_{nm}\|\textbf{e}_{m}\|_2 . \nonumber \label{lastinequality}\end{aligned}\ ] ] therefore , for all @xmath398 we have @xmath399 given ( [ 1 ] ) and summing over @xmath400 , since @xmath191 is positive semi - definite , we have @xmath401 therefore , the lower bound of the strong monotonicity constant in @xmath370 is obtained . by replacing ( [ endoftheoram3 ] ) into ( [ upperbound of variations ] ) , part 3 of theorem 3 can be obtained . when the solution to ( [ proximalresponsemap ] ) is obtained by the non - expansive or contraction mapping , the distributed algorithm for the proximal response map converges . for any vector @xmath402 in ( [ proximalresponsemap ] ) , we have @xmath403\geq 0\\ \label{theorem 42 } ( \textbf{z}- \widehat{\textbf{a}}(\textbf{b}_2 ) ) [ \widehat{\mathcal{f}}(\widehat{\textbf{a}}(\textbf{b}_2 ) , \textbf{b}_2)+ \widehat{\textbf{a}}(\textbf{b}_2)-\textbf{b}_2]\geq 0 \end{aligned}\ ] ] considering @xmath404 in ( [ theorem 41 ] ) and @xmath405 in ( [ theorem 42 ] ) , from the above two inequalities we get @xmath406 + ( \widehat{\textbf{a}}(\textbf{b}^1)- \widehat{\textbf{a}}(\textbf{b}^2 ) ) [ \widehat{\mathcal{f}}(\widehat{\textbf{a}}(\textbf{b}^2 ) , \textbf{b}^2)+ \widehat{\textbf{a}}(\textbf{b}^2)-\textbf{b}^2]\\ \label{theorem 44 } = ( \widehat{\textbf{a}}(\textbf{b}^2)- \widehat{\textbf{a}}(\textbf{b}^1))[\widehat{\mathcal{f}}(\widehat{\textbf{a}}(\textbf{b}^1 ) , \textbf{b}^1)-\widehat{\mathcal{f}}(\widehat{\textbf{a}}(\textbf{b}^2 ) , \textbf{b}^2 ) ] -\|\widehat{\textbf{a}}(\textbf{b}^2)- \widehat{\textbf{a}}(\textbf{b}^1)\|+ ( \widehat{\textbf{a}}(\textbf{b}^2)- \widehat{\textbf{a}}(\textbf{b}^1))^t ( \textbf{b}^1-\textbf{b}^2)\ ] ] recall that @xmath407 , @xmath408 and @xmath409 . when @xmath410 , ( [ theorem 44 ] ) can be written as @xmath411(\widehat{\textbf{a}}(\textbf{b}^2)- \widehat{\textbf{a}}(\textbf{b}^1))^{{\scriptsize{\textnormal{t } } } } \\+ & & ( \widehat{\textbf{a}}(\textbf{b}^1)- \widehat{\textbf{a}}(\textbf{b}^2 ) ) [ \sum_{m \in \mathcal{n } , m \neq n}-\nabla_{\textbf{a}_n \textbf{a}_m}^2 u_{n } ] ( \textbf{b}^1-\textbf{b}^2)^{{\scriptsize{\textnormal{t } } } } \nonumber -\|\widehat{\textbf{a}}(\textbf{b}^2)- \widehat{\textbf{a}}(\textbf{b}^1)\|+ ( \textbf{a}(\textbf{b}^2)- \textbf{a}(\textbf{b}^1 ) ) ( \textbf{b}^1-\textbf{b}^2)^{{\scriptsize{\textnormal{t } } } } \geq 0 \nonumber\end{aligned}\ ] ] let @xmath412 be the smallest eigenvalue of @xmath413 , @xmath414 for @xmath415 , and @xmath416 . from ( [ theoream467 ] ) we get @xmath417 on the other hand , @xmath418 since the utility is a convex function with respect to @xmath30 , the first term in ( [ 90 ] ) is positive and @xmath419 . besides , @xmath420 which leads to @xmath421 . from these two inequalities , ( [ theoream48 ] ) can be written as @xmath422 for ( [ theoream49 ] ) , when @xmath55 is a @xmath57 matrix , the proximal response map in ( [ proximalresponsemapn ] ) is a contraction mapping ( proposition 12.17 in @xcite ) , and converges to a unique rne from any initial point . variation in utility function of each user for variation in the bound of uncertainty region is @xmath423 where @xmath424 is the @xmath51 matrix , @xmath425_{kk}=(x_{nm}^k)^2 $ ] , @xmath426 $ ] , and @xmath427 is the column gradient vector . the last two terms in ( [ utilityrne ] ) are always positive , because of * a3 * and * a2*. the first term in ( [ utilityrne ] ) is always negative because @xmath297 is increasing according to @xmath30 and @xmath428 . the second term in ( [ utilityrne ] ) is always positive because @xmath297 is a decreasing function of @xmath242 and @xmath429 . therefore , the social utility increases when the negative terms of ( [ utilityrne ] ) are less than the positive terms . by some rearrangement and matrix manipulation , the condition for negative semi - definiteness of @xmath231 can be obtained . \1 ) consider @xmath430 , where @xmath431 is a closed convex set , @xmath432 is the monotone map related to @xmath431 , @xmath433 , and @xmath434 is the vector with bounded positive values . let @xmath435 be the solution to @xmath430 . when @xmath432 is strongly monotone , @xmath435 is monotone ( corollary 2.9.17 in [ 31 ] ) . when @xmath435 is a monotone and decreasing function , we have @xmath436 subtracting ( [ h1 ] ) from ( [ h2 ] ) , we get @xmath437 the above inequality leads to @xmath438 . because of the affinity in _ avi _ , ( [ h3 ] ) is @xmath439 where @xmath440 is a negative vector and @xmath441 . since @xmath431 is a convex and closed region , we have @xmath442 which is equivalent to the semi - negative matrix definition . 2 ) from above , when the strategy of each user in the power control game in ( [ propostion1avi ] ) is a decreasing function , @xmath392 is a semi - negative matrix . therefore , the social utility increases when the strategy of each user is reduced obviously , @xmath392 is semi negative when @xmath443 is semi - positive , which leads to ( [ conditionproposition1forpowercontrolgame ] ) . f. meshkati , a. j. goldsmith , h. v. poor , and s. c. schwartz , `` a game - theoretic approach to energy - efficient modulation in cdma networks with delay qos constraints , '' _ ieee journal on selected areas in communications _ , vol . 25 , no . 6 , pp . 10691078 , 2007 . g. scutari , d. p. palomar , and s. barbarossa , `` optimal linear precoding strategies for wideband noncooperative systems based on game theory - part i : nash equilibria , '' _ ieee transactions on signal processing _ , 56 , no . 3 , pp . 12301249 , march 2008 . , `` optimal linear precoding strategies for wideband noncooperative systems based on game theory - part ii : algorithms , '' _ ieee transactions on signal processing _ , vol . 56 , no . 3 , pp . 12501267 , march 2008 . e. a. gharavol , y .- c . liang , and k. mouthaan , `` robust downlink beamforming in multiuser miso cognitive radio networks with imperfect channel - state information , '' _ ieee transactions on vehicular technology _ , vol . 59 , no . 6 , pp . 2852 2860 , july 2010 . g. scutari , d. p. palomar , f. facchinei , and j .- s . pang , `` convex optimization , game theory , and variational inequality theory in multiuser communication systems , '' _ ieee signal processing magazine _ , vol . 27 , no . 3 , pp . 3549 , may 2010 . f. giannessi , a. maugeri , and p. m. pardalos , eds . , _ equilibrium problems : nonsmooth optimization and variational inequality models ( nonconvex optimization and its applications ) _ , 1st ed.1em plus 0.5em minus 0.4emspringer , january 2002 . m. fukushima and g .- h . lin , `` smoothing methods for mathematical programs with equilibrium constraints , '' _ proceedings of the 12th international conference on informatics research for development of knowledge society infrastructure _ , july 2004 , pp .

recent breakthroughs in the field of numerical relativity have led to dramatic progress in understanding the predictions of general relativity for the dynamical interactions of two black holes in the regime of very strong gravitational fields . such black - hole binaries are important astrophysical systems and are a key target of current and developing gravitational - wave detectors . the waveform signature of strong gravitational radiation emitted as the black holes fall together and merge provides a clear observable record of the process . after decades of slow progress , these mergers and the gravitational - wave signals they generate can now be routinely calculated using the methods of numerical relativity . we review recent advances in understanding the predicted physics of events and the consequent radiation , and discuss some of the impacts this new knowledge is having in various areas of astrophysics . black holes , gravitational waves , numerical relativity

1010.2165

it has been nearly a century since albert einstein s profound physical insight revealed our current standard model of gravitational physics , general relativity . among the theory s extraordinary consequences was the predicted existence of black holes , nonlinearly self - gravitating , stable objects in which gravitational forces have completely overcome all other physical interactions . though once considered a mathematical curiosity , general relativity s description of black holes now provides the best explanation for a widespread class of astrophysical objects whose gravitational potential wells power many of the most energetic astronomical phenomena observed . these range from stellar black - holes powering x - ray sources in the neighboring regions of our galaxy , to supermassive monsters with masses @xmath0 , where @xmath1 is the mass of the sun , with far - reaching astrophysical consequences . in the coming decade , anticipated observations of gravitational waves from black hole binaries will open a new window onto the universe . interpreting such observations will richly engage aspects of our theoretical understanding of strong - field gravity which have never before been confronted with empirical observations . the final coalescence of binaries consisting of two comparable - mass black - holes , with mass ratios @xmath2 , is expected to be one of the strongest astrophysical sources of energy the form of gravitational radiation . as the emission of gravitational waves removes energy and momentum from the binary , the orbits shrink and the black holes eventually merge together into a single black hole , producing an intense burst of radiation . with the advent of ground - based gravitational wave detectors such as ligo and virgo ( which will detect mergers of black holes with masses in the range @xmath3 ) , and with planning underway for the space - based lisa ( which will observe mergers of massive black holes , with masses @xmath4 ) , knowledge of black - hole binary gravitational waveforms is urgent . black - hole binary coalescence can be thought of as proceeding in three stages . during the inspiral the holes have wide enough separations that they can be treated as point particles . in this stage , the orbital period is much shorter than the timescale on which the orbital parameters change , and the holes spiral together on quasi - circular orbits . when the holes get so close together that they can no longer be approximated as point particles , they enter the merger phase . in this strong - field , dynamical regime , the black holes plunge together and merge into a single , highly distorted black hole , surrounded by a common horizon . this remnant black hole then `` rings down , '' shedding its nonaxisymmetric modes through gravitational wave emission and settling down into a quiescent rotating black hole . knowledge of the gravitational waveforms from these three stages of black hole coalescence is important for gravitational wave detection and data analysis , as well as astrophysical applications . the inspiral can be calculated analytically using the post - newtonian ( pn ) approximation , which is an expansion of the full equations of general relativity in powers of @xmath5 , where @xmath6 is the characteristic velocity of the source , @xmath7 is its mass , and @xmath8 is its characteristic size . the inspiral gravitational waveform is a `` chirp '' , which is a sinusoid increasing in both frequency and amplitude . the ringdown can also be calculated analytically using techniques of black - hole perturbation theory , and the resulting gravitational waveforms are damped sinusoids . however , the merger stage can only be understood using full numerical simulations of the einstein equations in 3-d , and the resulting gravitational waveforms were unknown until recently . the final merger of comparable - mass black holes is a powerful source of gravitational radiation . the gravitational waveforms emitted by a black - hole binary with mass ratio @xmath9 scale with the total mass @xmath7 of the binary . setting @xmath10 and @xmath11 , we can express both length and time scales in terms of the total mass : @xmath12 . the strong - field merger will produce a burst of gravitational radiation lasting @xmath13 and having a luminosity @xmath14 , which is greater than the combined luminosities of all the stars in the visible universe . for stellar black - hole binaries ( @xmath15 ) this luminosity will last for @xmath16 ms , and for massive black - hole binaries ( @xmath17 ) , for @xmath18 min . understanding the final merger of a black - hole binary requires solving the full einstein equations using the methods of numerical relativity . this endeavor has proved to have many challenging aspects , ranging from providing astrophysically relevant initial data to understanding the structure of the einstein equations , and the solution eluded researchers for many years . recently , a series of dramatic breakthroughs has ignited the field , making robust , stable , and accurate simulations of binary mergers possible for the first time . these models are opening our understanding of strong - field dynamics , impacting gravitational wave searches and other areas of astrophysics . in this article we provide an overview of these exciting developments , highlighting the key physical results that are emerging . computing the strong - field merger of two comparable - mass black holes has a long history , with the first attempt dating back more than 40 years . overall , progress was generally slow and incremental , requiring the interplay among new ideas in black hole modeling , mathematical investigations into the structure of the einstein equations , the development of effective gauge conditions , advances in computational techniques , and the availability of high performance computing resources . here we provide a brief review of key developments in this arena . solving einstein s equations on a computer typically requires slicing 4-d spacetime into a stack of 3-d spacelike hypersurfaces , each labeled by time @xmath19 @xcite . the einstein equations then divide into two sets : constraint and evolution equations . the constraints are a set of time - independent elliptical equations that must hold on each slice . in particular , the constraints are solved first to obtain valid initial data for a black - hole binary simulation . this data is then propagated forward in time using the evolution equations . general relativity has four spacetime coordinate degrees of freedom , which are associated with four freely - specifiable coordinate or gauge functions that govern the future development of both time and the spatial coordinates . these are generally taken to be the lapse function @xmath20 , which gives the lapse of proper time @xmath21 between neighboring slices , and the shift vector @xmath22 , which determines how the spatial coordinates develop from one slice to the next . appropriate choices for the lapse and shift have proven to be crucial for successful black hole evolutions @xcite . for many years , a primary challenge of numerical relativity was simply to decide which equations to solve @xcite . there is no unique formulation of the einstein equations ; rather there are many choices regarding , for example , which variables to use , whether to write the equations as first - order or second - order in time , and which coordinate conditions to impose . these choices are not arbitrary because some formulations turn out to be more `` numerically friendly '' than others . that is , although they are analytically equivalent when constraints and auxiliary variable definitions are assumed exact , some formulations may be unstable in practice . for example , the evolution equations may admit rapidly growing solutions which violate the constraints . in this case , although the initial data may contain only tiny errors , the subsequent evolution may produce violations of einstein s equations which `` blow - up '' . there can also be pathological numerical solutions that do nt represent solutions to the analytic evolution equations at all , but are supported by the discrete numerical grid structure and depend on the details of the finite - differencing operators . once a formulation has been chosen , the einstein equations are solved numerically for various field variables on a grid of discrete points that represents the spacetime domain of interest . there are two general approaches for dealing with the spatial derivatives that appear in these equations . finite differencing interpolates the derivative at a given point from the surrounding points according to a taylor expansion in the grid spacing . spectral methods assume a solution in the form of a summation of orthogonal functions ; once the coefficients are obtained numerically , the derivatives can be found analytically . in both cases , time integration can be handled in a number of ways , most commonly by a runge - kutta algorithm . to date , most numerical relativity solutions have been carried out using finite differencing , with most current work employing a cartesian grid in three spatial dimensions , although results from evolutions using spectral methods are now becoming available . to model astrophysical binaries , initial data must be generated for two black holes moving on quasicircular inspiralling orbits and having a mass - ratio @xmath9 and some configuration of spins . one usually conceptualizes the initial state of the system with a particle - like description of the masses , positions , momenta , and spins of each black hole . since general relativity is a field theory , such a description can only be seen as a first step . the model requires a full description for the initial field configuration which satisfies general relativity s four initial value constraints , and which somehow corresponds to the system we have described in these particle - like terms . this system is evolved for several orbits and then through plunge , merger , and ringdown , for a total duration on the order of several hundred @xmath7 or more . to obtain the gravitational waveforms , the radiation must be extracted from the simulation at large enough distances from the source to be in the `` wave zone '' . since the length scales of the black holes are @xmath23 and the wavelengths of the emitted radiation are @xmath24 , it is clear that some sort of variable resolution such as adaptive mesh refinement must be used to handle the very large computational domains needed . by the late 1990 s numerical relativity had developed to the point that brief simulations of black - hole binaries in three spatial dimensions plus time were possible . these techniques were sufficient for evolving promptly merging `` grazing collisions '' of black holes @xcite . the simulations were not indefinitely stable , but rather would typically crash after @xmath25 , well before any significant portion of a binary orbit could be evolved . in this arena , the lazarus method emerged as a novel approach to obtaining black - hole binary waveforms , combining short numerical relativity simulations with black hole perturbation methods @xcite . since numerical relativity codes were then only able to evolve for a brief period , and since perturbation theory could approximate the late time dynamics of the distorted remnant black hole , the lazarus project aimed to apply numerical relativity to evolve the strong - field approach to merger , thereby providing a hybrid model for a significant part of the problem . starting from quasi - circular initial configurations near the start of the final merger , the black holes were evolved using numerical relativity for @xmath26 until just before the simulation became inaccurate . then , via a complicated interface , data from late in the numerical simulations was interpreted as initial data describing the perturbed final black hole and the emerging radiation . finally , black hole perturbation theory techniques were applied to evolve this data and derive the full waveforms . the lazarus simulations gave the first indication of what might be expected for the terminal burst of radiation from coalescing black - hole binaries . figure [ fig1_lazaruswf ] shows that the dominant @xmath27 spin-(-2)-weighted spherical harmonic component has a brief burst of radiation smoothly joining the damped sinusoidal signal of the ringdown . the waveforms were remarkable for their simplicity , with predominantly circular polarization ( in the @xmath28 mode ) , and a steady evolution of polarization frequency and amplitude . since these characteristics were robust under variations in the model , it was conjectured that the waveforms from a binary starting from a wide separation late in the inspiral would be well characterized by these simple features . subsequently , the lazarus method was applied to study mergers of more generic black - hole binaries @xcite . concurrently with the lazarus investigations , further progress was being made toward full numerical relativity simulations of black - hole binary mergers . one major milestone was the development of a conformal formulation of the einstein equations known as the baumgarte - shapiro - shibata - nakamura or `` bssn '' system @xcite , that overcame some instability problems associated with an earlier formulation in use at the time @xcite . in this approach , the set of evolution equations is written with first - order time derivatives and second - order spatial derivatives , and is strongly hyperbolic @xcite . stable time evolution was accomplished using a coordinate condition that evolves the lapse function dynamically , causing the slices to avoid crashing into the black hole singularities @xcite . however the simulations were still limited to durations @xmath29 of stable evolution , by a failure of the spatial coordinate system known as `` grid stretching , '' in which the coordinates tend to fall into the black holes , and by instabilities related to how the black holes were handled numerically . eliminating grid - stretching required developing appropriate techniques for dynamically controlling the spatial coordinates , which are governed by the evolution of the shift vector . the first long - lasting evolutions of distorted black holes relied on a class of hyperbolic shift evolution schemes , known as @xmath30-freezing conditions , which were inspired by the bssn formulation . with this approach , a single distorted black hole could be evolved indefinitely ( e.g. , for thousands of @xmath7 ) with reasonable accuracy @xcite . these studies allowed the full numerical determination of gravitational waveforms from black hole ringdowns as the distorted black hole settled down to a physically and numerically stable final state , and provided a foundation for future advances in black - hole binary simulations @xcite . however , another aspect of spatial gauge conditions remained a critical issue for long - lasting black - hole binary simulations . as general relativity allows arbitrary coordinate systems , many groups adopted coordinate conditions which did not allow the black holes to move through the computational domain . this simplified the problem of dealing with the black hole singularities , which were handled either by excising the black hole interiors ( within the horizons ) from the computational domain @xcite , or by representing the black holes as punctures @xcite . though progress had been made toward implementing moving excision regions @xcite , it was computationally much simpler to demand that the excision region remain fixed on the grid @xcite . similarly , the puncture treatment was formulated so that the black hole was fixed on the grid and the singularity was factored out and handled analytically in a time - independent manner . however , for binary configurations in which the black holes physically move , the cost of keeping the black holes fixed in coordinate space was paid by the twisting and stretching of the dynamical fields , eventually leading to large computational errors . for inspiraling binary configurations one potential solution was to attempt to untwist the geometric field data by imposing comoving coordinates using a shift vector that is dynamically adjusted during the evolution of the binary to minimize motion of the black hole horizons @xcite . with this approach , brgmann et al . achieved a significant milestone : the first simulation of a full orbital cycle of a black - hole binary system prior to merger @xcite . in this work , the authors applied a dynamically adjusted corotating frame tracking the measured position of the black hole apparent horizons . this allowed a simulation which remained accurate for a little longer than the @xmath31 duration of a complete orbit . however , problems occurred with the large characteristic speeds that resulted near the outer boundaries , limiting the domains of these simulations and preventing gravitational waveforms from being measured effectively . the coordinate control also required fine tuning which eventually failed , causing the simulation to crash as the black holes finally approached merger . in early 2005 pretorius shocked the numerical relativity community by announcing the first complete , robust simulations of an equal - mass black hole merger @xcite . after completing @xmath32 orbit , the black holes plunged and merged to form a single distorted black hole that then rings down . pretorius extracted the gravitational waves to obtain the first inspiraling merger waveform directly from numerical relativity , shown in fig . [ fig2_pretoriuspsi4 ] . pretorius employed several techniques that were very different from most other numerical approaches to the black - hole binary problem @xcite . rather than using the bssn formulation , he applied a generalized harmonic formalism @xcite , directly integrating the spacetime metric with evolution equations of second - order in both space and time . these equations were implemented numerically using adaptive mesh refinement to allow high resolution around the black holes while maintaining a large computational domain . pretorius utilized spatial coordinates compactified to draw spatial infinity into the computational domain , with a choice of gauge - evolution strongly tied to his evolution formalism . following @xcite , he added terms to the evolution equations to specifically damp away any violations of the constraints . in his simulations , the black holes were excised and moved freely across the computational grid . pretorius success with such novel methods quickly raised questions as to whether the struggling , more widely - pursued bssn - based puncture approach might be off course . later that same year , however , a new robust method based on the bssn formulation was announced . the `` moving puncture '' method was discovered simultaneously and independently by the groups at the university of texas at brownsville ( utb ) @xcite and nasa s goddard space flight center ( gsfc ) @xcite . in this approach the black holes are represented as punctures , but are not constrained to remain fixed on the coordinate grid . rather , they are allowed to move freely through the grid using novel coordinate conditions . figure [ fig3_utbprltrack ] shows the trajectories of the puncture black holes as computed by the utb group ; the formation of a common horizon marks the point of merger . the moving puncture method eliminates the analytic representation of the puncture singularities in favor of an approximate numerical treatment within the black hole horizons . the utb and gsfc groups had discovered and applied similar methods to the same problem : evolving an equal - mass nonspinning black - hole binary from near the final orbit , through merger and ringdown , and studying the gravitational waves . the first generation of merger waveforms from pretorius , gsfc , and utb showed the same simple burst of radiation ending in a damped - sinusoidal ringdown , and were qualitatively consistent with the lazarus project results discussed above and shown in fig . [ fig1_lazaruswf ] . the discovery of the moving puncture method ignited the field of black - hole binary evolutions . since it was based on commonly used methods , most researchers in the field were quickly able to achieve accurate and stable evolutions using their existing codes , with the adoption of simple coordinate conditions @xcite . suddenly , the game was on and nearly all the groups were participating . with the advent of successful numerical evolutions of binaries that inspiral and merge , the numerical relativity community s focus changed to investigating the physics of binary mergers . this advancing frontier is enabled by continued technical improvements in areas ranging from initial data prescriptions to more accurate numerical methods . the astrophysical black hole mergers that are the targets of current and future gravitational wave detectors are expected to reach the merger stage after having radiated away any initial eccentricity @xcite and proceeding through a long quasi - circular inspiral . all equal - mass , nonspinning binary merger simulations starting from such orbits in the inspiral should produce the same gravitational waveform , subject only to rescaling with the total mass of the system . for many years , concerns had been raised about the accuracy and realism of black - hole binary initial data sets , including the effects of spurious gravitational radiation and eccentricity @xcite . with numerical relativity now able to simulate the final merger , the next step was to run models with enough orbits before the plunge and merger to get complete and reproducible waveforms starting from the late inspiral . the gsfc group produced the first representation of the definitive waveform for the final stages of a merger of equal - mass , nonspinning black holes @xcite . they carried out a series of four simulations with the holes starting from quasicircular initial conditions at increasingly larger separations . in these runs , the holes completed @xmath33 , and @xmath34 orbits before the formation of a common horizon . to compare the results of these models , they chose the moment of peak gravitational radiation amplitude as the fiducial time @xmath35 . the orbital dynamics of the binaries can be examined by tracking the black hole centers , given by the location of the punctures . the black hole trajectories for each run were oriented so that they superpose at the fiducial @xmath36 . in the early stages of each run , the tracks clearly showed the effects of eccentricity in the initial conditions , with the amount of initial eccentricity decreasing for more widely separated holes . as the holes spiraled together deeper into the strong - field regime , the tracks locked on to a universal trajectory independent of their initial conditions that continued for the last orbit , plunge , and merger . figure [ fig4_qcnwaves ] reveals the corresponding universal gravitational waveform . here , the dominant @xmath37 , @xmath38 quadrupolar component for each run is shown , shifted in time so that the peak radiation amplitude occurs at @xmath35 . starting from @xmath39 the waveforms show nearly perfect agreement , differing from each other at the level of @xmath40 . the signals for the preceding few orbits agree at the level of @xmath41 , except for a brief burst of spurious radiation at the start of each run . note that the merger waveform shows a remarkably simple shape , making a smooth transition from the inspiral chirp to the damped sinusoid of the ringdown . as the merger begins , both the wave amplitude and frequency increase , albeit faster than in the inspiral . the amplitude reaches a peak and then decreases , dropping exponentially through the ringdown . the frequency increases monotonically to a maximum value that remains constant during the ringdown . of course , the black - hole binary merger dynamics and waveforms can be altered by the presence of large amounts of eccentricity @xcite and spurious gravitational radiation @xcite near the time of the plunge . however , the robustness of the merger to modest deviations from astrophysical initial conditions opened the door to studying many cases of interest using relatively short simulations , starting just a few orbits before the start of the plunge . astrophysical black - hole binaries are unlikely to have exactly equal masses . currently , numerical relativists are able to evolve systems with mass ratios up to @xmath42 @xcite . starting from quasicircular orbits the simulations show that the merger phase for nonspinning , unequal mass black hole binaries is robust to modest deviations from these initial conditions and produces a generally simple waveform shape . an important tool for analyzing these mergers is a decomposition into spin - weighted spherical harmonic modes . berti et al . @xcite analyzed a set of unequal - mass nonspinning mergers with mass ratios ranging from @xmath43 to @xmath44 . studying the multipolar distribution of the radiation , they found that the sub - dominant modes ( @xmath45 ) become more important , carrying a larger fraction of the energy , as @xmath9 increases . specifically , for @xmath46 , the @xmath47 mode typically carries @xmath41 of the radiated energy . also , as expected from symmetry considerations , the odd-@xmath48 modes are suppressed in the equal - mass limit . baker et al . @xcite carried out a complementary study of the radiation from nonspinning mergers with mass ratios in the range @xmath49 . the multipolar decomposition clearly shows that the hallmark simplicity of the waveform persists for @xmath50 and extends to each of the spherical harmonic components @xmath51 ; this property has recently been shown to extend to the @xmath52 case @xcite . in the full mode - summed waveform , this simple shape is also seen when viewing along the system s orbital axis , where the quadrupole mode dominates . a somewhat more complex appearance arises by viewing the system off - axis , where higher multipoles contribute more strongly to the waveform at various angles . throughout the entire coalescence , each of the spherical harmonic waveform components is circularly polarized , with steadily varying phase and amplitude @xcite . for each mass ratio @xmath9 , the rotational phase ( and frequency ) of the different @xmath53 components are the same . during the inspiral this is expected , since the waveform phase is equal to the rotational phase multiplied by the mode number @xmath48 . however , for the @xmath54 modes this relationship also holds throughout the merger and into the ringdown . these properties suggest a simple conceptual interpretation in which the radiation is generated by an `` implicit rotating source . '' in this picture , each @xmath53 mode is generated separately by the @xmath53 moment of some implicit source . the fixed relationship for the @xmath54 modes implies that these components of the source rotate rigidly through the entire coalescence . the @xmath55 components are less rigid and peel away from the main @xmath54 trend during the merger @xcite . the mass ratio @xmath9 is a one - dimensional cut into the parameter space of black - hole binaries . the remaining parameter space is dominated by the spin angular momentum of each hole . as spin is a vector quantity , this adds six more dimensions . simulations of spinning black holes first focused on systems whose spins were expected , on the basis of pn arguments , to have the least effect on the orbital motion . binaries with aligned and anti - aligned spins are relatively easy to treat as there is no precession of the orbital plane or the individual spins . when one or both holes has a spin not parallel to the orbital axis , spin - orbit and spin - spin interactions will cause precession that can complicate the evolution and the resulting waveforms @xcite . the first merger evolutions of spinning black holes were carried out by campanelli et al . @xcite . they simulated the mergers of two equal - mass highly spinning black holes with @xmath56 , and both spin vectors either aligned or anti - aligned with the orbital angular momentum . here , @xmath57 is the magnitude of the black hole spin angular momentum per unit mass . they also evolved a nonspinning equal - mass binary for comparison . all three binaries had the same initial orbital angular frequency corresponding to an orbital period @xmath58 , and merged to form a rotating remnant black hole with @xmath59 . however , the aligned system took noticeably more orbits to merge than the others . this behavior is caused by the spin - orbit interaction , which produces an effective force between the black holes , either an attractive or repulsive for the anti - aligned or aligned cases , respectively . all three binaries generate remarkably similar gravitational waveforms having a simple shape , with the aligned case displaying a longer wavetrain and the anti - aligned case a shorter one . the first fully numerical evolutions of strongly precessing spinning systems were carried out by campanelli et al . @xcite , who observed both significant precession of the orbits and a `` spin flip , '' in which the final post - merger black hole spins in the direction opposite to the two initial black holes . more recent work is beginning to probe the effects of spin precession on waveforms @xcite . studies of this most general of ( non - eccentric ) parameter sets are still in the early stages . no systematic study of waveform shapes and polarizations has been carried out yet . a preliminary study of the multipolar structure of gravitational waves from several classes of binaries with equal spins has been performed by berti et al . they considered an equal - mass case with aligned spins ; several @xmath44 binaries with antialigned ( down - down ) spins ; and three unequal - mass binaries with spins initially in the orbital plane and pointing in opposite directions . examining the distribution of gravitational - wave energy emitted by various modes , they find that , as in the case of nonspinning mergers , odd-@xmath60 multipoles are suppressed for @xmath43 and that , as @xmath9 increases , more energy is radiated in higher-@xmath60 multipoles . because the state of the final black hole formed from the coalescence depends primarily on termination of the inspiral and the late burst of radiation in the merger , it can be accurately probed with relatively short simulations of just a few orbits . the merger of two equal - mass nonspinning black holes produces a final black hole with a moderately high spin , @xmath61 @xcite . since the black holes each start out with no spin , and any tidal spin - up is negligible @xcite , the spin of this final black hole arises from the orbital angular momentum of the original binary . simulations show that , for this simplest black hole merger , the final spin is `` universal , '' or independent of the initial black hole separation , for modest deviations from quasicircular initial conditions . for mergers of nonspinning black holes with unequal masses , the spin of the final black hole decreases as the mass ratio @xmath9 increases . simulations with mass ratios up to @xmath42 show that the final spin parameter scales as @xmath62 , where @xmath63 for @xmath44 and @xmath64 for @xmath42 @xcite . the effects of spin - orbit and spin - spin coupling can become important in determining the spin of the final black hole . in the simplest cases , the spin vectors are parallel to the orbital angular momentum . depending on the mass ratio and the black hole spin , the merger can result in a final black hole with a larger ( spun up ) or a smaller ( spun down ) spin than either of the progenitors @xcite . in certain cases , the merger can lead to a spin flip with the final black hole spinning in a direction opposite to the spins of the initial holes ; in particular , it is possible to produce a final nonspinning black hole , @xmath65 , from the merger of two spinning holes @xcite . more general black - hole binaries , with misaligned spins , are further complicated with the effects of precession @xcite . several attempts have been made to produce expressions for the final spin vector using analytic techniques or by fitting to results from numerical simulations @xcite ; see @xcite for a review . of particular interest is the question of whether a black hole merger can produce a maximally spinning hole ( @xmath66 ) or indeed exceed the kerr limit ( @xmath67 ) . current research suggests that this is not possible , even for mergers occurring from hyperbolic encounters @xcite and mergers of highly boosted black holes @xcite . a notable phenomenon arising from asymmetric binary systems is the merger recoil or kick a net movement of the end - state black hole from the system s center of mass , caused by the anisotropic emission of gravitational radiation during the coalescence . fitchett @xcite produced a useful formula for the kick velocity , which has significant applications in astrophysics , using a quasi - newtonian approximation . several authors also calculated the recoil analytically using pn approximations @xcite . however , since the dominant part of the effect builds up in the strong - field regime close to merger , full numerical relativity simulations are needed for accurate calculations of the kick . in 2006 , recoil from a fully numerical binary merger was demonstrated for the first time by herrmann et al . @xcite , for plunging black - hole binaries with mass ratios as large as @xmath68 . this was followed soon afterwards by a full orbit and plunge simulation by baker et al . @xcite , who found a recoil of between 86 and 97 @xmath69for a mass ratio @xmath70 . a more systematic study of recoil from mergers of unequal - mass binaries was produced by gonzalez et al . in 2007 @xcite , who studied the merger of binary systems with mass ratios between @xmath71 and @xmath72 . figure [ fig5_jenakickfit ] shows the resulting range of recoil speeds , together with several earlier numerical and analytical estimates . the authors synthesized these into a single recoil formula , a nonlinear correction to the fitchett formula , yielding a maximum recoil of 175 @xmath69for a mass ratio of @xmath73 . this has recently been tested for the more extreme @xmath52 mass ratio , with general agreement @xcite . calculations of recoil in the much larger parameter space of spinning binaries began with the non - precessing cases of holes with spins aligned ( or anti - aligned ) with the orbital angular momentum . several studies of this region of parameter space @xcite have revealed that the kick velocity has a quadratic dependence on initial spins , with a maximum kick of 448 @xmath69for extremal kerr holes , @xmath74 . for these anti / aligned cases , as well as for nonspinning unequal - mass black - hole mergers , the direction of the kick velocity is always in the orbital plane . meanwhile attention turned to more general black hole spins . campanelli et al . @xcite speculated from pn arguments that optimal spin configurations could give rise to huge `` superkicks , '' with velocities @xmath75 @xmath69 , out of the initial orbital plane . the first such superkick around 2500 @xmath69 was soon observed by gonzalez et al . tichy et al . @xcite have argued that superkicks arise in mergers with general spin orientations , while greater insight into the mechanism of these kicks has been developed @xcite . with such a large parameter space to cover , it seems useful to try to construct a general formula that will describe kicks from arbitrarily spinning binaries . baker et al . @xcite used new and existing aligned - spin results to produce a single unifying formula for in - plane kicks ; campanelli et al . @xcite proposed an extension to this model , with scaling of the superkick out - of - plane contribution motivated by pn theory . the leading dependence in this formula on the angles between spins and linear momenta of the pre - merger holes has strong support from numerical simulations @xcite ; however , the dominant scaling with mass ratio is still in dispute @xcite . many key features of black - hole binary interactions can be modeled usefully with only a small handful of binary orbits before merger . quantities such as radiated energy and momentum are bulked near the merger and have been shown to be robust to the addition of one or two extra orbits ( see section [ sssec : equal - mass - nospin ] ) . however , optimal observational studies of gravitational waveforms ultimately require theoretical predictions for the full waveform , starting at large separations during the inspiral . before the breakthroughs in numerical relativity , most information about the dynamics of compact binaries came from pn theory . this approach supplied the particle trajectories , energy flux , and most importantly for detector scientists the gravitational waveforms themselves . however pn theory fails before the system merges , so the waveforms are necessarily incomplete . with the advent of numerical simulations encompassing many orbits , scientists finally have a way to develop complete information about the waveforms . this requires longer simulations which reveal the last part of the binary inspiral , and allow overlap with pn waveform predictions . simulations for these studies require careful attention to the initial configuration of the black holes . for a significant population of astrophysical binaries , it can be expected that the orbits have circularized through gravitational - wave emission over many orbits prior to merger . numerical simulations try to mirror this expectation by selecting initial momenta consistent with near - zero eccentricity for the relatively small separations at which a full numerical relativity simulation becomes feasible . for equal - mass , nonspinning binaries , two basic approaches have proved effective in reducing spurious eccentricity . the more direct method is to model the observed eccentricity , and then adjust the momenta using a newton - like step to zero it out ; this approach has led to extremely low eccentricities @xcite . an alternative approach , adopted by husa et al . @xcite , is to use the pn hamiltonian equations of motion to model the early evolution of the particle trajectories , starting from large ( @xmath76 ) separations ; the emission of radiation during this process naturally circularizes the orbit , and low - eccentricity momenta can be read off at the desired separation . this approach has recently been tested with spinning binaries as well @xcite . the first long waveform was produced by the gsfc group @xcite for equal - mass , nonspinning black holes starting @xmath77 orbits or @xmath78 gravitational wave cycles before merger . baker et al . estimated their numerical errors in waveform phase from @xcite as a function of frequency , finding that above a certain frequency numerical errors are smaller than internal errors in the pn sequence . in addition , the numerical and pn waveform phases agree to within one radian of phase drift for a little over ten gravitational wave cycles preceding the last orbit before merger , comparable to numerical error estimates . hannam et al . @xcite improved on this work , with phase and amplitude comparison between their low - eccentricity higher - resolution evolutions and pn waveforms . numerical simulation codes based on pseudospectral differencing techniques are particularly well - suited to long waveform studies . the caltech - cornell group adopted the constraint - damped generalized harmonic formalism used by pretorius , an important component in developing a stable spectral evolution code @xcite . also like pretorius , their code handles black holes by excising the black hole interiors . their spectral code also employs a numerical grid that tracks the black holes explicitly . though there can be difficulties with the changing geometry as black holes merge , this approach allows efficient study of the long - lasting inspiral waveforms . recently this approach has provided , several of the longest and most accurate black - hole binary evolutions.@xcite in particular , the caltech - cornell group have used their spectral code to simulate an equal - mass nonspinning binary starting 16 orbits and 32 gravitational wave cycles before merger : see fig . [ fig6_ccwf ] @xcite these much longer waveforms have been used to validate several competing pn models @xcite . with so many pre - merger waveform cycles now available for the equal - mass case , the results can be cross - checked by comparing the `` complete '' waveform inspiral , merger and ringdown between groups . this is important to verify expectations that differences in methodology and residual numerical effects , such as unwanted eccentricity , are unimportant . a recent effort , dubbed the `` samurai project , '' analyzed long waveforms from several groups in the light of detectability criteria for the ligo and virgo ground - based gravitational - wave detectors . they found that the available numerical relativity waveforms are indistinguishable in these detectors for signal - to - noise ratios ( snrs ) @xmath79 @xcite . hannam et al . have investigated the properties of highly spinning orbit - aligned black holes , comparing their phase and amplitude with pn predictions over the last ten waveform cycles @xcite . in the best cases , they find phase agreement within 2.5 rad for 3.5pn , and amplitude agreement to within around 12% with restricted pn . while not at the same level as their nonspinning results @xcite , this may be attributable to the lower orders of accuracy available for spinning black holes in pn theory at the present . the caltech - cornell group has recently conducted long - lasting simulations of black holes with spins aligned and anti - aligned with the orbital angular momentum , calibrating a tunable pn waveform model to match the results @xcite ; see sec . [ ssec : outlook_analyticmodels ] . generic mergers involving non - aligned , and thus precessing , spins adds four new degrees of freedom to the problem . a systematic understanding of waveforms generated by generic mergers will require considerably more study . recent simulations have begun to explore generic examples @xcite . we have described a sampling of the explosion of numerical relativity studies revealing some of the details of black - hole binary physics as implied by general relativity . while more remains to be learned , this new understanding is already making important contributions in planning and interpreting astrophysical black hole observations where einstein s theory is applied as the standard model of gravitational physics . experimental gravitational - wave detectors were first developed more than 40 years ago . although advances in design have increased their sensitivity by many orders of magnitude , the extreme weakness of expected astrophysical signals ( strain amplitudes @xmath80 ) means the observational challenge is still huge . the output of any gravitational wave detector will be a data stream that must be combed through to find real signals . this search requires accurate `` template '' waveforms representing our best picture of the radiation from expected sources ; these templates can then be compared with the data stream through matched filtering . we refer the reader to a review on gravitational - wave astronomy for an overview of these techniques @xcite . crucially , the most important sources of gravitational radiation are expected to include the mergers of black - hole binaries . before the advent of numerical relativity simulations of black hole mergers , the only test waveforms available for use in data analysis studies were based on pn theory . these waveforms were essentially only inspiral chirps and did not include the strong - field merger . the new , richer , information now available from numerical relativity has revolutionized the data analysis picture in several ways . the availability of the plunge - merger - ringdown portion of the signal can greatly increase the snr in the detector . armed with a long numerical merger waveform of acceptable accuracy , we can extend it backwards to cover an arbitrarily long inspiral by matching to a pn waveform . such a `` hybrid '' waveform was first produced by the gsfc group for the equal - mass , nonspinning case @xcite . using this hybrid , we can investigate the total achievable snr , and the related distance reach , for current and future detectors . figure [ fig7_mergerobsaligosnr ] demonstrates the gain in snr from including the merger portion of the waveform for the ground - based advanced ligo detector . contours of snr as function of redshift @xmath81 and total binary mass @xmath7 for the lisa detector are shown in figure [ fig8_mergerobslisasnr ] . full numerical waveforms , and the longer hybrid waveforms generated from them , can also be used to improve existing data - analysis techniques and template sets . since previously developed gravitational wave data analysis algorithms were not based on knowledge of the merger waveforms , an obvious first step is to test how successfully these techniques detect the merger signals predicted by numerical simulations . in 2009 , the ninja project @xcite used direct injection of a broad range of short and long numerical merger waveforms into mock ligo and virgo data streams for this purpose . the result was the most realistic testing ground to date for disparate data - analysis methods , including full and partial waveform template matching , as well as unmodeled burst searches . further studies of detection algorithms with numerical relativity waveforms are continuing . the snr is only a crude guide to the specific detector response to gravitational waveforms , however . the merger portion of the waveform , though short in duration , may contain important new information not present in the inspiral . as a simple example , we expect the time of merger itself to be well localized with the full waveform , whereas it is not well - defined in the inspiral - only signal . more generally , the different @xmath53 modes of a binary s full inspiral - merger - ringdown waveform scale differently with the time to merger . modes that were not significant during inspiral suddenly become more prominent in the merger , and the detailed information they carry becomes available to the observer @xcite after detection of gravitational waves from distant sources , we are most interested in identifying the physical parameters of the sources . each of the seven intrinsic parameters of a black - hole binary ( mass ratio and spin vectors ) will , in general , be imprinted on the gravitational - wave signal , along with some extrinsic parameters such as sky position and distance to the source . for sufficiently strong snrs , expected for massive binary mergers seen by lisa , we can expect to be able to extract some of these parameters at high precision . while these parameters can be partially disentangled using inspiral - only template information , it has recently been found that the full merger waveform can help break parameter degeneracies and hence drive down uncertainties in several important physical parameters . good localization of the source on the sky is especially important for the development of multi - messenger astronomy . the recent parameter estimation studies of non - spinning mergers that include the merger waveforms indicate that lisa will be able to locate sky positions within a few arcminutes for binaries with @xmath82 at cosmological distances ( redshift @xmath83)@xcite . as previously discussed , asymmetries in a black - hole binary system due to unequal masses and/or spins result in the anisotropic emission of gravitational radiation , ultimately imparting a recoil to the merged remnant black hole . numerically it has been found that , for certain configurations of black hole spins , the recoil velocity can exceed the escape velocity of many galaxies . since it is important to determine the theoretical probability that a massive black hole might be ejected from its host galaxy , there have been some preliminary calculations based on simple distributions of spins and mass ratios , and analytic fits of numerical results giving the dependence of the recoil on mass and spin @xcite . although a consensus on the exact probability of galactic ejection has yet to be achieved , there is general agreement that it is non - negligible . such rogue black holes may have already been observed in the form of two rapidly moving quasars @xcite . it has been speculated that these particular quasars originated from the mergers of massive black holes during the coalescence of their host galaxies . however a previous study of quasar data found no indications of such recoil events , suggesting that they are rare @xcite . the effect of recoil may also be observed less directly in its effect on the growth rates of black holes . those ejected into the sparse intergalactic medium are less likely to encounter and merge with other black holes @xcite . even the growth of recoiled black holes remaining within their host galaxies may be affected , as the motion of the black hole can modify the rate at which it accretes matter @xcite . the expected distribution of masses and spins of astrophysical black holes is another topic of considerable interest in astrophysics . in our current understanding , black holes grow from smaller `` seeds '' early in the history of the universe through a combination of mergers and the accretion of gas @xcite . in general , most of the mass growth is believed to come from accretion , with mergers providing a modest increase . however , the gravitational radiation emitted during black hole coalescence carries away energy , reducing the overall system mass by roughly several percent . the bulk of this loss happens quickly , in the final plunge and merger stage of the coalescence . with this rapid mass loss , nearby matter in an accretion disk around the remnant black hole might react to the accompanying sudden change in gravitational potential and produce a visible change in its electromagnetic profile a possible electromagnetic counterpart to the burst of gravitational radiation @xcite . for the final spin the situation is considerably more interesting , and may answer questions about double - jet `` x - shaped '' radio sources @xcite . we have seen in sec . [ ssec : bhb_final_spin ] that the final spin from black holes merging in vacuum depends on the magnitudes of the initial spins and their orientations relative to the orbital axis . the rit group @xcite has recently used their extended mass and spin formulas in a spin - evolution study , obtaining asymptotic spin distributions for bh merger remnants assuming no accretion . however , interaction of the merging holes with surrounding gas may serve to align the binary spins before merger , changing the picture somewhat @xcite . in general , the effects of binary merger and accretion have to be studied together for a coherent picture to emerge @xcite . gravitational wave observatories may be sensitive to hundreds or thousands of wave cycles . analysis of the observed data requires comparisons with model signals representing the full variety of possible mergers . computing so many orbits of pre - merger evolution using numerical relativity would be computationally very expensive . the pn approximation provides accurate representations of the dynamics and waveforms for the long - lasting inspiral portion of the coalescence during which the black holes remain fairly well - separated and their velocities remain relatively low . the most valuable waveform models for gravitational wave data analysis must combine the efficiency of the pn approach , while accurately representing the final merger portion of the which is only understood by numerical simulations . this requires a means of analytically encoding the merger signals . several approaches are currently being explored for constructing these complete signal models . some are based on the analytic `` effective - one - body '' model of binary coalescence @xcite . figure [ fig9_buonannoeobwf ] shows a comparison between an effective - one - body waveform for the merger of a nonspinning black hole binary with mass ratio @xmath44 with a numerically simulated waveform @xcite . for specific cases , such models have been tuned to high accuracy to agree to agree with numerical results @xcite . another approach , which models the phenomenological shape of hybrid waveforms in frequency space @xcite , has been extended for the dominant waveform modes from spinning , but non - precessing , mergers @xcite . further development of these models , and the production of a family of simulated waveform spanning the parameter space , is a current focus of broad - based research collaborations . the initial data models currently used to begin numerical simulations do not perfectly represent the intended astrophysical systems . research continues to improve these models . for example , although some astrophysical black holes are expected to have near - extremal spins , common initial data can not represent holes with @xmath84 @xcite ; novel methods are being developed to go beyond this limit @xcite . also , current initial data models generally do not contain the physically appropriate radiation content for an inspiraling binary @xcite , but rather harbor spurious radiation that is not astrophysical@xcite . efforts are underway to include initial radiation content more consistent with pn predictions @xcite . in addition , in the case of moving punctures , the initial coordinates typically undergo a rapid transition at the start of the evolution as the black holes relax into a more stable solution . while the resulting transient pulse of `` gauge radiation '' does not alter the physics , it does contain fairly high frequencies that can be challenging to resolve and it has motivated construction of analytic initial coordinates that are closer to the numerically evolved coordinates @xcite . more efficient and accurate methods of numerical evolution would make simulations of many binary orbits , particularly with large mass ratios , more computationally practical . for representing spatial derivatives on a computational grid , the highest accuracy of finite differencing stencil yet achieved in the context of numerical relativity is 8th order in the grid spacing ( although the accuracy of interpolation between refinement boundaries is not yet commensurate ) @xcite . spectral methods are generally more accurate but less robust than finite differencing , requiring fine tuning for generic black hole binaries @xcite ; they are also unlikely to handle shocks in accreting matter ( see section [ ssec : outlook_matter ] ) . alternatives to both finite differencing and spectral methods , such as finite element methods , are also being investigated @xcite . meanwhile more efficient time - integration techniques allowing larger step sizes are being explored @xcite . because the physical domain of a typical simulation is finite , gravitational radiation is usually extracted on a sphere of finite radius rather than at infinity . if multiple extraction surfaces are employed in a region of sufficiently high spatial resolution , then the radiation can be extrapolated to spatial infinity . a more accurate method known as `` cauchy - characteristic extrapolation '' extracts the gravitational wave data at a finite radius and then evolves it along null geodesics to future null infinity ; this method is currently under development @xcite . in addition to being gravitationally loud " , black hole mergers may also be electromagnetically visible . massive black holes at the centers of galaxies are typically surrounded by gaseous accretion disks and magnetic fields . when the black holes merge , the dynamics of the gas and magnetic fields may produce electromagnetic signals , counterparts to the emitted gravitational radiation . for example , the inspiral may twist " the fields , resulting in characteristic electromagnetic radiation @xcite as well as heating of a surrounding accretion disk @xcite . in addition , the violent merger dynamics may induce shock waves in accreting matter , in turn generating electromagnetic radiation . the recoil of the merged remnant , in particular , may have such an effect on the accretion disk @xcite . detection of electromagnetic counterparts of gravitational waves would be of great scientific value . current models of the complex merger physics ( e.g. @xcite ) could be directly tested . einstein - maxwell theory , the coupling of gravitational and electromagnetic fields on macroscopic scales , could be verified . in particular , equality of the speed of gravity with the speed of light could be confirmed @xcite . astronomy would also benefit enormously , as the location and characteristics of gravitational wave sources could be corroborated . in addition , electromagnetically visible mergers could serve as standard candles " , beacons by which to measure the accelerating expansion of the universe , while simultaneously playing the role of standard sirens " emitting gravitational radiation . cross - correlating these signals could result in measurement of the cosmological dark energy " to unprecedented accuracy @xcite . there has been some preliminary work on the dynamic effects of the spacetime of a coalescing binary on surrounding matter . modeling the accretion disk as geodesic particles , large collision energies were found as the binary merged @xcite . and hydrodynamically simulating a gas cloud around around the binary , luminosity due to shocks has been calculated @xcite . the generation of electromagnetic radiation by more direct means has also been simulated , via the twisting of a magnetic field anchored in the accretion disk , as it is frame - dragged by the binary @xcite . future efforts will employ magnetohydrodynamic methods , where challenges include adequately resolving shocks and turbulence in a dynamic spacetime , and accurately representing the divergence - free magnetic field on an adaptively refined grid . we hope to have conveyed some of the excitement of recent progress in understanding black - hole binary physics , and the applications of this knowledge in astrophysics . these advances are the result of sustained efforts by a broad scientific community over many years . in a brief review , it is impossible to adequately represent all of the excellent work that has contributed to the current state of knowledge . we have only been able to provide a few of the highlights as seen through the lens of our particular perspective . we encourage interested readers to pursue the subject further . other resources are available on topics including : numerical relativity techniques @xcite , the breakthroughs in black hole merger simulations @xcite , black hole simulations for gravitational wave data analysis @xcite , and gravitational wave science generally @xcite . we acknowledge support from nasa grants 06-befs06 - 19 and 08-atfp08 - 0126 . bjk was supported in part by an appointment to the nasa postdoctoral program at the goddard space flight center , administered by oak ridge associated universities through a contract with nasa . baker jg , campanelli m , lousto co , 65:044001 ( 2002 ) , arxiv : gr - qc/0104063 . baker jg , campanelli m , lousto co , takahashi r , 65:124012 ( 2002 ) , arxiv : astro - ph/0202469 . baker jg , campanelli m , lousto co , takahashi r , 69:027505 ( 2004 ) , arxiv : astro - ph/0305287 . campanelli m , 22:s387 ( 2005 ) , arxiv : astro - ph/0411744 . shibata m , nakamura t , 52:5428 ( 1995 ) . nagy g , ortiz oe , reula oa , 70:044012 ( 2004 ) , arxiv : gr - qc/0402123 . reula oa , 1:251 ( 2004 ) , arxiv : gr - qc/0403007 . anninos p , camarda k , mass j , seidel e , suen wm , towns j , 52:2059 ( 1995 ) , arxiv : gr - qc/9503025 . alcubierre m , brgmann b , diener p , guzmn fs , hawke i , et al . , 72:044004 ( 2005 ) , arxiv : gr - qc/0411149 . brgmann b , tichy w , jansen n , 92:211101 ( 2004 ) , arxiv : gr - qc/0312112 . pretorius f , 95:121101 ( 2005 ) , arxiv : gr - qc/0507014 . pretorius f , 23:s529 ( 2006 ) , arxiv : gr - qc/0602115 . pretorius f , 22:425 ( 2005 ) , arxiv : gr - qc/0407110 . gundlach c , martn - garca jm , calabrese g , hinder i , 22:3767 ( 2005 ) , arxiv : gr - qc/0504114 . campanelli m , lousto co , marronetti p , zlochower y , 96:111101 ( 2006 ) , arxiv : gr - qc/0511048 . baker jg , centrella jm , choi di , koppitz m , van meter jr , 96:111102 ( 2006 ) , arxiv : gr - qc/0511103 . van meter jr , baker jg , koppitz m , choi di , 73:124011 ( 2006 ) , arxiv : gr - qc/0605030 . peters pc , mathews j , 131:435 ( 1963 ) . pfeiffer hp , cook gb , teukolsky sa , 66:024047 ( 2002 ) , arxiv : gr - qc/0203085 . baker jg , centrella jm , choi di , koppitz m , van meter jr , 73:104002 ( 2006 ) , arxiv : gr - qc/0602026 . hinder i , vaishnav b , herrmann f , shoemaker dm , laguna p , 77:081502(r ) ( 2008 ) , arxiv:0710.5167 [ gr - qc ] . scheel ma , boyle m , chu t , kidder le , matthews kd , pfeiffer hp , 79:024003 ( 2009 ) , arxiv:0810.1767 [ gr - qc ] . campanelli m , lousto co , zlochower y , 74:084023 ( 2006 ) , arxiv : astro - ph/0608275 . pollney d , reisswig c , rezzolla l , szilgyi b , ansorg m , et al . , 76:124002 ( 2007 ) , arxiv:0707.2559 [ gr - qc ] . buonanno a , kidder le , lehner l , 77:026004 ( 2008 ) , arxiv:0709.3839 [ astro - ph ] . tichy w , marronetti p , 76:061502(r ) ( 2007 ) , arxiv : gr - qc/0703075 . dain s , lousto co , zlochower y , 78:024039 ( 2008 ) , arxiv:0803.0351 [ gr - qc ] . kesden m , 78:084030 ( 2008 ) , arxiv:0807.3043 [ gr - qc ] . rezzolla l , dorband en , reisswig c , diener p , pollney d , et al . , 679:1422 ( 2008 ) , arxiv:0708.3999 [ gr - qc ] . rezzolla l , diener p , dorband en , pollney d , reisswig c , et al . , 674:l29 ( 2008 ) , arxiv:0710.3345 [ gr - qc ] . rezzolla l , barausse e , dorband en , pollney d , reisswig c , et al . , 78:044002 ( 2008 ) , arxiv:0712.3541 [ gr - qc ] . tichy w , marronetti p , 78:081501(r ) ( 2008 ) , arxiv:0807.2985 [ gr - qc ] . lousto co , campanelli m , zlochower y , 27:114006 ( 2010 ) , arxiv:0904.3541 [ gr - qc ] . favata m , hughes sa , holz de , 607:l5 ( 2004 ) , arxiv : astro - ph/0402056 . blanchet l , qusailah mss , will cmw , 635:508 ( 2005 ) , arxiv : astro - ph/0507692 . damour t , gopakumar a , 73:124006 ( 2006 ) , arxiv : gr - qc/0602117 . wiseman ag , 46:1517 ( 1992 ) . baker jg , centrella jm , choi di , koppitz m , van meter jr , miller mc , 653:l93 ( 2006 ) , arxiv : astro - ph/0603204 . gonzalez ja , sperhake u , brgmann b , hannam md , husa s , 98:091101 ( 2007 ) , arxiv : gr - qc/0610154 . herrmann f , hinder i , shoemaker dm , laguna p , matzner ra , 661:430 ( 2007 ) , arxiv : gr - qc/0701143 . baker jg , van meter jr , mcwilliams st , centrella jm , kelly bj , 99:181101 ( 2007 ) , arxiv : gr - qc/0612024 . baker jg , mcwilliams st , van meter jr , centrella jm , choi di , et al . , 75:124024 ( 2007 ) , arxiv : gr - qc/0612117 . hannam md , husa s , sperhake u , brgmann b , gonzalez ja , 77:044020 ( 2007 ) , arxiv:0706.1305 [ gr - qc ] . pan y , buonanno a , baker jg , centrella jm , kelly bj , et al . , 77:024014 ( 2008 ) , arxiv:0704.1964 [ gr - qc ] . buonanno a , pan y , baker jg , centrella jm , kelly bj , et al . , 76:104049 ( 2007 ) , arxiv:0706.3732 [ gr - qc ] . dain s , lousto co , takahashi r , 65:104038 ( 2002 ) , arxiv : gr - qc/0201062 . pfeiffer hp , , phd thesis , cornell university , 2005 , arxiv : gr - qc/0510016 . hannam md , husa s , brgmann b , gonzalez ja , sperhake u , 24:s15 ( 2007 ) , arxiv : gr - qc/0612001 . milosavljevic m , phinney es , 622:l93 ( 2005 ) , arxiv : astro - ph/0410343 . dotti m , salvaterra r , sesana a , colpi m , haardt f , 372:869 ( 2006 ) , arxiv : astro - ph/0605624 . kocsis b , frei z , haiman z , menou k , 637:27 ( 2006 ) , arxiv : astro - ph/0505394 . megevand m , et al . , 80:024012 ( 2009 ) , arxiv:0905.3390 [ astro-ph.he ] . balbus sa , hawley jf , 376:214 ( 1991 ) . lang rn , hughes sa , 26:094035 ( 2009 ) , arxiv:0810.5125 [ astro - ph ] . kocsis b , haiman z , menou k , frei z , 76:022003 ( 2007 ) , arxiv : astro - ph/0701629 . pretorius f , binary black hole coalescence in _ physics of relativistic objects in compact binaries : from birth to coalescence _ , , pp . 305369 , springer , heidelberg , germany , 2009 , arxiv:0710.1338 [ gr - qc ] . hannam md , 26:114001 ( 2009 ) , arxiv:0901.2931 [ gr - qc ] . component of re(@xmath85 ) , corresponding to one of the two polarization states of the emitted gravitational radiation . curves are plotted for 10 simulations having different initial black hole separations ( designated qc-0 , etc . ) and transition times to perturbative evolution . reprinted from @xcite and copyright 2002 by the american physical society ( http://link.aps.org/abstract/prd/v65/e124012 ) . , width=480 ] gpc is plotted vs. ( redshifted ) mass for the advanced ligo detector . the dashed line shows the snr calculated using pn techniques for the early inspiral part of the waveform , @xmath86 . the dotted line shows the snr using the late inspiral , @xmath87 , which is the transition region from pn to numerical relativity . the snr for the strong - field merger and ringdown , @xmath88 , was calculated using waveforms from a numerical relativity simulation and is shown using a thin solid line . finally , the snr from the entire waveform is given as the thick solid line.x here , @xmath36 marks the time of maximum gravitational radiation amplitude . reprinted from @xcite and copyright 2007 by the american physical society ( http://link.aps.org/abstract/prd/v75/e124024 ) . , width=480 ] and total binary mass @xmath7 are shown for the lisa detector . these have been calculated using a hybrid equal - mass nonspinning waveform : using pn for the early inspiral , matching to a numerical relativity waveform in the late inspiral , and continuing with the numerical waveform through the merger and ringdown . reprinted from @xcite and copyright 2007 by the american physical society ( http://link.aps.org/abstract/prd/v75/e124024).,width=480 ] mass ratio nonspinning black hole binary calculated using the analytic `` effective one - body '' ( eob ) model and using numerical relativity ( nr ) . reprinted with permission from @xcite and copyright 2007 by the american physical society ( http://link.aps.org/abstract/prd/v76/e104049 ) . , scaledwidth=85.0% ]

doping of the band insulator fes with co on the fe site introduces a small density of itinerant carriers and magnetic moments . the lattice constant , ac and dc magnetic susceptibility , magnetization , and specific heat have been measured over the range of co concentration . the variation of the ac susceptibility with hydrostatic pressure has also been measured in a small number of our samples . all of these quantities show systematic variation with including a paramagnetic to disordered ferromagnetic transition at . a detailed analysis of the changes with temperature and magnetic field reveal small power law dependencies at low temperatures for samples near the critical concentration for magnetism , and just above the curie temperature at higher . in addition , the magnetic susceptibility and specific heat are non - analytic around displaying an extraordinarily sharp field dependence in this same temperature range . we interpret this behavior as due to the formation of griffiths phases that result from the quenched disorder inherent in a doped semiconductor .

0912.2960

the desire for ever increasing computational speeds has led to the conception of spintronics technologies that make use of both the charge and spin properties of electrical charge carriers in solids@xcite . realization of these devices for useful technologies requires the discovery and development of materials that have easily controllable electronic and magnetic properties to the same extent that semiconductors allow control over charge properties . this demand has led to an enormous interest in magnetic semiconductors to investigate if traditional semiconductors can be manipulated by chemical substitution of transition metal elements , or other creative means , into offering control over the spin properties of electrons@xcite . because so many of the suggested devices rely on the production and detection of spin polarized currents , magnetic semiconductors with curie temperatures above room temperature that are compatible with silicon technology are highly desired . however , this goal has thus far not been met and it has become increasingly apparent that to make progress , fundamental knowledge and understanding of the electronic and magnetic properties of magnetic semiconductors are necessary . although some headway in this direction has been made , the complexity of disordered materials with strong interactions makes magnetic semiconductors difficult to model and understand@xcite . in a recent paper we presented the results of an exploration into the magnetic and electronic properties of one such magnetic semiconducting compound , fe@xmath5co@xmath6s@xmath0 based on the diamagnetic insulating parent compound iron pyrite ( fes@xmath0)@xcite . we found that the insulator - to - metal transition that occurs for low concentrations of co doping was followed by a transition from a paramagnet to a disordered ferromagnet first apparent for small dopant concentrations , @xmath2 , at very low temperatures . increased co doping led to an increased transition temperature and to a more ordered ferromagnetic phase@xcite . the main conclusion of our paper was that the quenched disorder associated with the chemical substitution has important consequences on the formation of the magnetic state at small @xmath2 and on the finite temperature phase transitions at larger @xmath2 . unusual power law temperature , @xmath7 , and magnetic field , @xmath8 , dependencies of the magnetic susceptibility , magnetization , and specific heat were measured at low-@xmath7 for @xmath2 near the critical concentration for nucleation of a magnetic phase , @xmath9 , as well as at @xmath7 s just above the curie temperature , @xmath10 for @xmath11 . these were attributed to large , rare disorder fluctuations consisting of regions of local magnetic order which fluctuate as a single large magnetic moment and that grow in size as @xmath12 or @xmath13 . fluctuating rare region phenomena associated with phase transitions in systems with quenched disorder are known under the heading of griffiths phases@xcite . theorists have found the subject of griffiths phases fascinating since griffiths discovered the non - intuitive feature in his calculations that , under certain conditions , exponentially rare regions in disordered systems can dominate the thermodynamic response when in proximity to a phase transition@xcite . this is in contrast to many aspects of phase transitions where disorder can be treated perturbatively or simply ignored . these ideas have drawn even more interest as evidence for non - fermi liquid behavior in disordered systems has been discovered while explanations remain tentative@xcite . while it is clear that the disorder and strong electron correlations are responsible for the breakdown of fermi liquid behavior in some cases , the microscopic origins remain poorly understood@xcite . in this context , the unusual power law behavior associated with the formation of griffiths phases have been discussed as one mechanism for disorder driven non - fermi liquid behavior@xcite . experimental evidence that griffiths phases are observable in condensed matter systems has only recently been reported for a small number of systems@xcite , and in our previous publication we presented evidence for the importance of griffiths phase formation near the critical concentration for magnetic ordering in the magnetic semiconducting system fe@xmath5co@xmath6s@xmath0 . the purpose of the present paper is to present a more thorough rendering of our magnetic susceptibility , magnetization , and specific heat experiments and analysis , as well as to present the results of some more recent ac magnetic susceptibility measurements made at ambient and applied pressures of up to 7 kbar . in a separate accompanying article we present the results of our hall effect and resistivity measurements of these same materials@xcite . we chose to investigate co doped fes@xmath0 since fes@xmath0 is a simple diamagnetic band insulator with the same pyrite crystal structure as cos@xmath0 , a itinerant ferromagnet@xcite . the two compounds had been shown earlier to be miscible such as to form the substitutional series fe@xmath5co@xmath6s@xmath0 without observation of second phases or any phase segregation that occurs in many magnetic semiconducting compounds . as such , it allowed the investigation of the transition from diamagnetic insulator to a paramagnetic metal , then to a ferromagnetic metal with co substitution . in addition , band structure calculations have predicted that at large @xmath2 , @xmath14 , fe@xmath5co@xmath6s@xmath0 is half metallic having a fully spin polarized ground state@xcite we begin our discussion by presenting the details of the experimental techniques that we have used to investigate the extrinsic properties of fe@xmath5co@xmath6s@xmath0 . this is followed by a presentation of the magnetic susceptibility , magnetization , and specific heat over a wide range of @xmath2 covering insulating , paramagnetic metallic , and ferromagnetic metallic behaviors . in section [ griffneartc ] we explore in detail the behavior of our crystals nearest @xmath9 at low temperatures where the power - law in @xmath7 and @xmath8 behaviors are found . this is followed by a discussion of the magnetic properties of crystals with @xmath11 in proximity to @xmath10 displaying similar behavior . we conclude the paper with a summary and discussion of our results . single crystals of fe@xmath5co@xmath6si were synthesized from high purity starting materials , including fe powder ( alfa aesar 99.998% ) , co powder ( alfa aesar 99.998% ) , and sulfur ( alfa aesar 99.999% ) . these materials were sealed in a 16 mm diameter quartz tube under vacuum . iodine ( alfa aesar 99.99+% ) was added to the tube as the transport agent at a concentration level of 2 to 5 gm/@xmath15 . the tubes were then placed in a small , single zone horizontal tube furnace ( lindberg / blue model 55035 ) and heated to a constant temperature . insulating plugs were removed from the ends of the furnace to create a natural temperature gradient of @xmath16 @xmath17c . the charge was placed at the center of the furnace ( hot zone ) , and single crystals would begin to form at the cold end of the tube in approximately 5 days , with a maximum yield by no later than 14 days . good crystals were produced for temperatures in the range of 650 to 850 @xmath17c . crystals were etched in hcl to remove any remaining flux . initial characterization included single crystal x - ray diffraction . single crystal fragments were glued to a glass fiber and mounted on the goniometer of a nonius kappa ccd diffractometer equipped with mo - k@xmath18 radiation ( wavelength @xmath19 ) . all x - ray structure determinations are consistent with a pyrite crystal structure with no anomalous intensities that could be interpreted as evidence for a second phase . the model of the structure was refined to a level consistent with a fully solved crystal structure . we observe no evidence for disorder , including that associated with sulfur deficiencies , in the crystal structure of all our crystals . the lattice constants , @xmath20 , shown in fig . [ fig : latcon ] were determined by the x - ray diffraction measurements . as displayed in the figure , @xmath20 increases systematically with co concentration , @xmath2 in fe@xmath5co@xmath6s@xmath0 , beyond the lattice constant for pure fes@xmath0 , @xmath21 nm , and is comparable to the measurements of refs . the increase in lattice constant with @xmath2 is consistent with vegard s law and the idea that co replaces fe within the pyrite crystal structure . energy - dispersive x - ray microanalysis ( edx ) on a jeol scanning electron microscope equipped with a kevex si(li ) detector was performed to check the stoichiometry of our samples . these revealed that the co concentration was about @xmath22 times the nominal co concentration of the starting materials , while the sum of the co and fe densities was within error , @xmath23% , of one half the sulfur density . magnetization , @xmath24 , and magnetic susceptibility ( both ac and dc ) measurements were performed in a quantum design superconducting quantum interference device ( squid ) magnetometer ( mpms ) for temperatures , @xmath25 k and dc magnetic fields , @xmath8 , of up to 5 t. ac susceptibility measurements were carried out over a range of excitation fields from 0.1 to 1.5 oe and excitation frequencies from 1 hz to 1 khz . in addition , ac magnetic susceptibility measurements were performed in a dilution refrigerator above 50 mk at @xmath4 , a frequency of 1 khz , and an excitation field of 1.5 oe . the saturation magnetization was determined by dc magnetization measurements at 1.8 k and high fields ( up to 5 t ) . these measurements were consistent with our edx measurements showing a magnetic moment density of @xmath26 times the nominal co concentration . because @xmath24 measurements are easily performed and highly reproducible , we have subsequently used the saturated magnetization to determine the co concentration of our samples . thus , throughout this manuscript the stoichiometry of the samples noted in the figures and text was determined in this manner . the variations in the saturated magnetization for crystals from the same growth batch was measured to be @xmath27% of the average value . therefore , we report @xmath2 determined from measurements of @xmath24 at high field for the crystals used in each of our measurements . wherever possible , the same crystal was employed for several different measurement types , particularly for the measurement of the ac susceptibility and specific heat below 1 k. the magnetic susceptibility of several of our crystals was measured with applied pressure ranging from ambient to 7 kbar in a beryllium - copper pressure cell@xcite . our pressure cell was based on previously published designs@xcite . the pressure cell was machined from beryllium - copper alloy 25 since this material has a low magnetic susceptibility and superior mechanical properties . after machining , the cell was annealed in air at a temperature of 335 @xmath17c for 2 hours . we designed our pressure cell to have a length of 12 cm to reduce the background seen by the squid pickup coils originating from the ends of the cell . the pistons are made from high purity technox 3000 zirconia rods from dynamic ceramic . this resulted in a small , diamagnetic , ( @xmath28 emu at 50 oe field ) background signal due to the lack of zirconia at the sample position . this background depended slightly on the distance between the ends of the two zirconia rods which , in turn , depended on the sample thickness , increasing for larger separations . the sample was placed inside a teflon container which was sealed with a beryllium - copper cap and retainer . the teflon container is filled with silicon oil ( dow corning 704 ) as the pressure transmission medium ( see for example ref . for more on silicon oil as a pressure transmission medium ) . the pressure at the sample position was determined by placing a small ( 0.5 mg ) ( v@xmath29ti@xmath30)@xmath0o@xmath31 crystal in the teflon container for use as a manometer@xcite . this material makes an excellent manometer as it displays a readily measurable metal - to - insulator transition which varies linearly in @xmath7 with pressure for pressures of up to 10 kbar@xcite . the specific heat was measured using a standard thermal relaxation method in a quantum design physical property measurement system ( ppms ) above 2 k and a dilution refrigerator above 100 mk . for measurements below 2 k single crystals were polished flat and mounted on a sapphire substrate with ge varnish . the opposite side of the 0.25 mm thick sapphire substrate contained a ruo resistance thermometer which was notched to provide a thermometer - heater pair in good thermal contact with the sample . the thermometer was cycled from 300 k to 4.2 k many times to ensure repeatability of the resistance at low temperatures@xcite . the thermometer was calibrated by comparison to a lakeshore calibrated germanium resistance thermometer . four resistive wires provided electrical and thermal contact of the thermometer and heater to the mixing chamber of the dilution refrigerator . we carefully checked at each measurement temperature that the time constant for thermalization of all elements on the sapphire substrate was much smaller than the thermal time constant for thermalization to the cryostat . specific heat measurements were carried out in magnetic fields between 0 and 6 t with all thermometers calibrated at each magnetic field . the specific heat of the addenda was measured separately and carefully subtracted from the data . in order to establish the magnetic ground states and critical temperatures of the fe@xmath5co@xmath6s@xmath0 chemical substitution series we have measured the dc magnetic susceptibility , @xmath32 , and magnetization , @xmath24 , of our crystals . in fig . [ fig : chidc ] we plot @xmath32 of a dozen representative samples with @xmath2 ranging from nominally pure fes@xmath0 to @xmath33 . our @xmath34 crystal has a diamagnetic susceptibility that changes little as a function of temperature . the small size of the impurity related curie - weiss contribution to the susceptibility of this crystal observable below 20 k , an increase of @xmath35 from 20 k down to 2 k , is consistent with an impurity concentration of @xmath36 @xmath37 . not only does this establish the quality of our pyrite crystals , it is compelling because it reaffirms that the parent compound is non - magnetic despite being 1/3 iron@xcite . the low temperature @xmath32 systematically increases with @xmath2 by over 4 orders of magnitude for our @xmath38 to our @xmath33 samples . at the same time , the @xmath7-dependence changes from being almost curie - like at low-@xmath2 , to displaying a broad , almost constant region of @xmath32 below 5 k for @xmath39 . at higher temperatures @xmath32 is less systematic in @xmath2 with some crystals displaying almost constant @xmath32 above 100 k suggestive of a significant pauli susceptibility , while others show a more @xmath7-dependent @xmath32 to higher temperatures crossing through @xmath32 of samples with smaller @xmath2 . because the pauli susceptibility is proportional to the conduction electron density of states , we interpret the variation of our measured @xmath32 above 100 k as revealing a carrier density and , perhaps , a mass that is highly sensitive to the disorder . because of this change from a systematic low - temperature @xmath32 to a less systematic high-@xmath7 @xmath32 , we conclude that in the two temperature ranges @xmath32 is dominated by two different electron populations ; a more local population of electron moments dominates at low-@xmath7 while the more itinerant electrons dominate @xmath32 at @xmath40 k. to gain insight into the character of the magnetic moments responsible for the low-@xmath7 upturn in @xmath32 we have plotted in fig . [ fig : cwpl ] the inverse dc magnetic susceptibility after subtraction of the temperature independent pauli susceptibility , @xmath41 , @xmath42 as a function of @xmath7 . this is standard procedure for comparing the susceptibility data to a curie - weiss form , @xmath43 where @xmath44 is the curie constant and @xmath45 is the weiss temperature . although some nonlinearity is apparent in the data , we have quantified the curie weiss behavior by fitting a linear @xmath7-dependence to the data as shown in the figure . we interpret the intercept of the fits with the @xmath7-axis as @xmath45 and the slope as @xmath46 where @xmath47 with @xmath48 the density of magnetic moments of size @xmath49 , @xmath50 the gyromagnetic factor , @xmath51 boltzmann s constant , and @xmath52 the bohr magneton . both @xmath45 and @xmath44 are observed to increase systematically with @xmath2 as can be seen in fig . [ fig : tcpl ] where we plot the best fit values of the quantity @xmath53 and @xmath45 . it is also interesting to note that for small @xmath2 , @xmath54 , @xmath45 is less than zero revealing a small antiferromagnetic interaction of the moments for small co doping . this is perhaps not expected since at only slightly larger co concentrations a disordered ferromagnetic ground state results . we note that the curie constant increases with @xmath2 beyond what would be expected for paramagnetic spin-1/2 magnetic moments for all samples measured . if we assume that the density of magnetic moments is equal to the co density , and that @xmath55 we would conclude that the magnetic moment increases from just above @xmath56 to between @xmath57 and @xmath58 for @xmath59 . this is the first indication that the magnetic moments likely form strongly interacting clusters over the temperature scale where the curie constants were determined . the measured curie constant , in this case , most likely reflects the fluctuations of a smaller density of much larger composite magnetic moments . the magnetic properties of our samples was further explored by measuring @xmath24 at 1.8 k in fields of up to 5 t. fig . [ fig : magpl]a displays the enormous increase in @xmath24 with @xmath2 for a range of @xmath2 between @xmath60 and 0.085 . in fig . [ fig : magpl ] b we display @xmath24 , in @xmath52 per co dopant , to demonstrate the normalization used to determine the co density of our crystals . the form of the magnetization can be seen to evolve from that of a brillouin function , @xmath61 , at small @xmath2 to that of a ferromagnetically ordered material at larger @xmath2 . the temperature and field dependence of @xmath24 , such as that shown in fig . [ fig : magpl ] , is commonly compared to a mean field form to determine @xmath10 of magnetic materials . in fig . [ fig : arrott ] we display this comparison of @xmath62 for two representative samples . the mean field form for @xmath24 is typically written as @xmath63 with @xmath64 the constant parameterizing the strength of the molecular mean field . from this starting point an expansion for @xmath65 can be written as @xmath66 where @xmath67 is the initial susceptibility ( @xmath68 ) . thus , a plot of @xmath69 vs. @xmath65 should be a straight line for a range of fields such that the higher terms can be ignored@xcite . this is commonly referred to as an arrott plot and is demonstrated by fig . [ fig : arrott ] frames a and b showing the results of such an analysis for two samples near the critical co concentration for ferromagnetism . it is clear that the region of linearity of @xmath69 in @xmath65 is limited , particularly at low temperatures . frame c of the figure displays a compilation of @xmath70 values determined from the linear fits , such as those in frames a and b. we have plotted the values of @xmath70 vs. @xmath71 since the stoner - wohlfarth theory of itinerant magnetism predicts a @xmath71 dependence of @xmath70 near the transition temperature . linear fits to the low-@xmath7 data are used to determine the @xmath71 intercept and this is interpreted as the mean - field @xmath10 . the values for @xmath10 determined in this way are plotted in fig . [ fig : tcpl ] b for comparison to @xmath45 . modified arrott analysis s having different dependencies of @xmath65 on @xmath24 were also performed without significant changes to the critical temperatures determined from the above analysis@xcite . the conclusion from this mean - field analysis is that the curie temperature is less than @xmath72 for these samples reflecting the importance of disorder on the formation of the magnetic state and that there is a critical co concentration for ferromagnetism near @xmath73 . however , a simple mean field analysis is likely to be inaccurate when @xmath10 is close to zero . although the dc susceptibility and magnetization give us a good initial indication of the formation of a ferromagnetic phase with the substitution of co for fe in fes@xmath0 , the very subtle changes that occur near @xmath9 are best explored at very low magnetic fields where good quality dc magnetization signals are difficult to obtain . therefore , we have made very careful ac susceptibility measurements of our crystals using excitation fields of 1.5 oe or less as shown in fig . [ fig : chipl ] . in agreement with our dc results , the ac susceptibility shows a dramatic increase with @xmath2 . however , there are two important differences between our dc and ac susceptibility apparent in the figures . first , the real part of the ac susceptibility , @xmath74 , is noticeably larger than @xmath32 for all @xmath75 crystals measured . second , distinct maxima appear in both @xmath74 and , at slightly higher @xmath7 s , the imaginary part part of the ac susceptibility , @xmath76 , as can be seen in frames a and b. as is common practice , we preliminarily interpret the temperature of the maxima in @xmath74 as the critical temperature , @xmath10 , for the development of a magnetic , ferromagnetic or spin glass , phase . in fig . [ fig : tcpl]b we plot @xmath10 determined in this manner ( max @xmath74 ) along with the temperature of the maximum in @xmath76 ( max @xmath76 ) for comparison to our analysis of @xmath32 and @xmath77 . in contrast to our @xmath32 results , our very low-@xmath7 measurements ( down to 50 mk ) reveal a peak in @xmath74 for all crystals with @xmath78 indicating that @xmath79 . the contrasting results in our ac and dc magnetic susceptibility indicate that the magnetic state of these crystals are extraordinarily sensitive to magnetic field as we explicitly demonstrate in section [ finitetc ] below . in frame c of fig . [ fig : chipl ] @xmath80 is plotted against temperature to display the curie - weiss - like behavior of @xmath74 . here , the general curie - weiss - like trend is displayed although significant nonlinearity and structure are also apparent in these data . linear fits of these data were used to determine the weiss temperatures , @xmath81 , and curie constants displayed in fig . [ fig : tcpl ] . there appears to be good compatibility of both @xmath81 with the values determined from the dc values , however , the curie constant appears generally smaller for our ac analysis than for our dc values for @xmath82 . the curie constants determined from both methods indicate a fluctuating magnetic moment of @xmath83 per co dopant indicative of magnetic cluster formation . despite the general linear behavior of @xmath80 well above @xmath10 , there is a noticeable departure from this behavior at @xmath84 that can best be seen for our @xmath85 sample in frame c of fig . [ fig : chipl ] . fits of @xmath86 for @xmath87 15 k indicate @xmath88 k , however , at @xmath89 k @xmath80 tends toward much smaller values . this tendency for @xmath74 to increase at a rate faster than the curie - weiss behavior indicates local ferromagnetism and further supports our observation that magnetic clusters form at temperatures above @xmath10 . a more complete description of this behavior is presented in section [ finitetc ] of this paper . the effect of hydrostatic pressure , @xmath90 , on the magnetization of our crystals is demonstrated in fig . [ fig : chiprpl ] . while the magnitude of @xmath74 is depressed by the application of pressures of order of a kbar , there is little change to the temperature of the peak in @xmath74 . the changes that occur with @xmath90 are demonstrated in frame b of the figure where the temperature of the peak in @xmath74 , taken as our definition of @xmath91 , is plotted along with the results of fitting a curie weiss form to @xmath92 above 15 k. both the curie constant and the weiss temperature are reduced as @xmath90 is applied showing that @xmath90 reduces both the average size of magnetic clusters as well as the average interaction energies between magnetic moments evident in the @xmath93 measurements . the curie constant shown in fig . [ fig : chiprpl ] is consistent with a reduction of the density of @xmath94 magnetic moments from @xmath95 to @xmath96 @xmath37 as the pressure is increased from ambient to 6 kbar . equivalently we can assume that the density of moments remains equal to the co density and that the changes in the curie constant reflect a reduction of @xmath49 from 1.3 to 0.9 with @xmath90 . our data could also be interpreted as a general trend toward a stronger kondo coupling with @xmath90 as has been shown to be typical in the case of f - electron compounds@xcite . the specific heat , @xmath97 of our crystals was measured to characterize the changes to the thermodynamic properties that occur upon co doping fes@xmath0 . as can be seen in fig . [ fig : sphpl ] , where @xmath97 is plotted for 7 representative crystals , there is little change in @xmath98 with co doping between 50 and 300 k. in frame b of the figure we concentrate our attention on the low temperature specific heat where changes in the electronic contributions are apparent . here we plot the data in the standard manner for characterizing the electronic contributions , @xmath99 vs. @xmath71 , since in standard paramagnetic metals the specific heat is expected to be accurately described by a @xmath100 form in the temperature range displayed . in this equation @xmath101 represents the coefficient of electronic specific heat and is proportional to the density of electronic states , while @xmath102 parameterizes the phonon contribution to the specific heat at temperatures well below the debye temperature . while @xmath103 j / mol k@xmath104 describes the phonon contribution of all our samples well , there are significant changes to the electronic contribution to @xmath98 with @xmath2 . it is clear that @xmath101 of the nominally pure fes@xmath0 crystal is consistent with zero and that @xmath101 increases with @xmath2 . however , there are contributions to the specific heat that are not described by the standard form . for example , at the lowest temperatures an upturn in @xmath99 with decreasing @xmath7 is apparent in all of the co doped samples . several of the data sets also show an increased contribution between 10 and 70 k that is not described by a simple constant . it is most likely that the differences from the standard metallic form are due to the formation of the magnetic ground state with co doping that was apparent in the magnetic susceptibility . we expand the low-@xmath7 @xmath99 data in fig . [ fig : sph2pl ] to highlight the changes that occur to the electronic contributions with increasing @xmath2 . in this figure the changes to @xmath101 with @xmath2 above 1 k are made clearer along with the large increases in @xmath99 that occur below 1 k in the co doped samples . the increases to @xmath101 with @xmath2 are very large considering that hall effect measurements estimate the carrier concentrations are only 10 to 30% of the co concentration in our samples@xcite . such a large density of electronic states is commonly interpreted as a large electronic effective mass . our estimates for the carrier effective mass , @xmath105 , from the @xmath99 data above 5 k range from 4 times the bare electron mass , @xmath106 , to over 40 @xmath106 for samples with @xmath2 just beyond @xmath9 . if we were to consider the large upturn in @xmath99 below 1 k to be electronic in origin , our estimates of @xmath105 increase to values between 120 and 250 @xmath106 at 120 mk . heavy mass carriers are rare in transition metal compounds , however , several fe - based semiconducting , or semimetallic , materials have been reported to have substantial carrier mass enhancement@xcite . crystals with @xmath107 , as determined from the ac magnetic susceptibility , display a @xmath99 that increases continuously down to the lowest measurement temperatures ( 0.1 k ) . those with @xmath11 display a maximum in @xmath99 at a temperature somewhat below @xmath10 as determined from the temperature of the maximum in @xmath74 . in order to characterize the specific heat further , we have measured the magnetic field dependence of @xmath99 as shown in fig . [ fig : sph4pl ] for 4 of our single crystals . for all four of the samples application of magnetic field decreases the low temperature @xmath99 while increasing slightly @xmath99 above 1 k. this behavior is qualitatively similar to the changes that occur upon application of a magnetic field in simple doped semiconductors such as si : p@xcite . the low-@xmath7 specific heat of si : p has been modeled in terms of local moments induced by the doping that are thought to form singlets such that the ground state of the system has zero net moment@xcite . the effect of disorder is to create a broad distribution of interactions between the local moments . as a result , @xmath99 increases as a small power - law at low temperatures@xcite . however , we do not believe such a description to be relevant in the case of fe@xmath5co@xmath6s@xmath0 since our crystals have a nearly ferromagnetic , or ferromagnetically ordered , ground state . instead , we attribute the unusual temperature dependence of @xmath99 as being due to the interaction of a random distribution of local moments together with the interaction of the local moments and conduction electrons . as the magnetic field is increased beyond the level of the interactions , @xmath99 resembles a sum of contributions from a fermi gas of electrons ( a constant @xmath101 ) and a separate contribution from effectively noninteracting magnetic moments . the contribution from non - interacting moments in a magnetic field is commonly referred to as a schottky anomaly@xcite and is demonstrated in the figure by the dashed lines . we note that above 1 k the specific heat in zero field is well described by this schottky form added to a temperature independent @xmath101 and the phonon contribution taken as @xmath99 of the nominally pure fes@xmath0 crystal . we have checked that our conclusion of formation of ferromagnetic cluster of magnetic moments at low-@xmath7 is consistent with our @xmath98 measurements by quantitatively comparing @xmath99 and @xmath32 . as we stated above , the large curie constants seen in fig . [ fig : tcpl ] are likely caused by the fluctuations of clusters of magnetic moments and both the specific heat and the magnetic susceptibility are sensitive to the density and size of the fluctuating moments . therefore , we have fit the specific heat at @xmath108 k as shown in fig . [ fig : sph4pl ] with the schottky form@xcite discussed above for a set of non - interacting magnetic moments subjected to an effective molecular field . this contribution was added to a @xmath7-independent term representing a fermi liquid of conducting electrons and the specific heat of the nominally pure fes@xmath0 crystal as a model of the contribution to @xmath98 from phonons . for this fitting procedure we have held the the quantity @xmath109 constant at the value determined by @xmath110 and calculated the best fit values of the effective molecular field , @xmath111 , @xmath49 , and @xmath48 . in fig . [ fig : mnj]b we present the results , plotting the best fit values of @xmath49 , which varied from less than 3 to 6 across our range of @xmath2 , and the parameter @xmath112 where @xmath113 and @xmath114 is the density of fe@xmath5co@xmath6s@xmath0 formula units . we plot the quantity @xmath112 to check our assumption that each co dopant donates a local magnetic moment of @xmath115 to fes@xmath0 . a value of @xmath116 corresponds to agreement with this assumption and although there is significant scatter in fig . [ fig : mnj]b , there is general agreement . the best fit values of @xmath111 varied between 1.4 and 2.6 t. we conclude that both @xmath74 and @xmath99 above 2 k are consistent with fluctuating magnetic moments whose size is several times larger than the magnetic moment of a single non - interacting co dopant in an fes@xmath0 lattice . the conclusion of magnetic cluster formation was further explored by calculating the entropy , @xmath117 , from @xmath97 according to the relation @xmath118 for the small number of crystals that we have measured below 1 k. we have subtracted the phonon contributions to @xmath98 prior to performing the integrations and the result of this procedure is shown in fig . [ fig : entropy ] . here @xmath119 approaches @xmath120 only for temperatures well above 10 k. thus , there is significant entropy missing at 10 k consistent with the idea that clusters of spins form at temperatures of this order . taking the entropy at 7 k as a benchmark allows us to estimate the average cluster size at this temperature . we make use of the form @xmath121 where @xmath122 is the average cluster size of spin 1/2 moments , presumably localized on co sites , to make the estimate . we find average cluster sizes that range from about 3 to 5 for the samples where data below 1 k are available . if we assume an exponentially decaying probability for clusters containing @xmath123 spin-1/2 co impurities , we estimate that about 10% of the cluster sizes will contain 10 or more spins at this @xmath7 . the reduction of entropy below @xmath124 with cooling evident in fig . [ fig : entropy ] implies long ranged interaction between the local magnetic moments induced by co substitution . for these co concentration levels , and assuming that co ions are substituted randomly for fe atoms in the fes@xmath0 crystal structure , we calculate that the average distance between impurity sites to be at the level of third nearest neighbors . thus we estimate the range of interactions between the magnetic moments to be at least 1 nm in order to produce the reduction of entropy we observe below 10 k , as well as the formation of magnetic clusters apparent in the magnetic susceptibility to somewhat higher @xmath7 . in addition , the changes to the entropy that occur with magnetic field can be viewed in fig . [ fig : entropy2 ] . the sensitivity of @xmath99 and @xmath119 to magnetic fields confirms the magnetic origin of the specific heat in this temperature range . for very small @xmath2 , @xmath125 for example , increases in @xmath119 are observed below 2 k for @xmath126 t , a sign that there is still a significant contribution to the entropy that we are not accessing because it lies below our lowest measurement temperature . thus , for small @xmath2 , there is a significant density of magnetic moments that interact with energy scales of less than 100 mk . at larger @xmath2 , @xmath127 , a suppression of @xmath119 with field is observed from which we conclude that there are very few magnetic moments with such small energies of interaction . this is consistent with our conclusion that there is significant cluster formation in these samples at temperatures above those accessed in fig . [ fig : entropy2 ] and for our @xmath85 sample , a peak in @xmath74 at 0.5 k that we associate with a disordered ferromagnetic transition . as we noted in our previous discussion of both the ac and dc magnetic susceptibility , the curie constants determined from simple fits to the temperature dependence were larger than is to be expected for paramagnetically fluctuating spin-1/2 magnetic moments . that is , the curie constants displayed in fig . [ fig : tcpl ] lie systematically above the solid line representing the curie constant expected when each co dopant donates a fluctuating @xmath94 magnetic moment . if we focus our attention on those samples closest to @xmath9 , we find curie constants 1.5 to 1.75 times larger than the simple estimate based on the expectation that each co dopant adds a single spin-1/2 moment and the curie law , @xmath128 with @xmath129 . this is an indication that fm ordered clusters of magnetic moments are forming below 100 k , where the curie behavior becomes evident for this dopant concentration . with the assumption that the magnetic moments are coupled ferromagnetically , the size of the curie constant would indicate that on average each cluster contained between 2.5 and 3.2 spin-1/2 magnetic moments for @xmath130 . in addition , we have checked that this estimate is consistent with @xmath98 above 2 k where the data can be interpreted in terms of a schottky anomaly associated with entropy of fluctuating magnetic moments . this estimate of the average size of fluctuating magnetic clusters is further supported by estimates made from the entropy determined from the integral of @xmath99 ( fig . [ fig : entropy ] ) . motivated by theoretical predictions about the importance of rare regions in disordered magnetic systems@xcite , we have carefully examined the temperature and field dependent behavior of the magnetization and specific heat of our samples with @xmath2 in proximity to @xmath9 to search for such effects . we are not disappointed in that unusual behavior is discovered . the most clear indication of the importance of rare large clusters of magnetic moments comes from the temperature dependence of the specific heat below 1 k shown in fig . [ fig : sph2pl]b . here @xmath99 is seen to increase without any suggestion of saturation down to 0.1 k for our samples closest to @xmath9 , @xmath131 and 0.007 . as indicated by the quality of the fits of a power - law form , represented by the dashed lines in the figure , as well as the linearity of the data when plotted on a log - log scale in fig . [ fig : sph3pl ] a , these data are well represented by a @xmath132 form with @xmath133 between 0.3 and 0.45 for over a decade in @xmath7 . similar power - law behavior is observed in @xmath74 over the same temperature range for the same crystals as can be seen in fig . [ fig : sph3pl ] b. here @xmath74 is well fit by a @xmath134 form between 0.8 and 10 k with @xmath135 between 0.2 and 0.25 for these samples . we note , however , that @xmath74 , unlike @xmath99 tends to saturate below 0.5 k. we make a direct comparison of the real part of the ac magnetic susceptibility with @xmath99 in fig . [ fig : wrplot ] where the wilson ratio , defined as @xmath136 is shown . in this figure we observe that the two samples with @xmath2 close to @xmath9 have a wilson ratio somewhat larger than 10 . this value is similar to that observed near the critical point for magnetic ordering in cepd@xmath5rh@xmath6 which is thought to enter a novel kondo - cluster - glass phase and has properties compatible with the quantum griffiths phase scenario@xcite . estimates for the average sizes of the fluctuating magnetic moments from the wilson ratio of fe@xmath5co@xmath6s@xmath0 yield about 3 to 4 spins consistent with our other estimates@xcite . we note that the wilson ratio at the lowest temperatures is not constant , but instead retains a slight temperature dependence reflecting the observation that at low-@xmath7 @xmath99 continues to display a power - law temperature dependence whereas @xmath74 tends to saturate . we note that predictions of the wilson ratio in griffiths phases models indicate a weak , logarithmic , temperature dependence@xcite . for our samples with @xmath11 the wilson ratio becomes very large exceeding 1000 near @xmath10 indicating a strong ferromagnetic tendency . in addition , our finite-@xmath8 dc magnetic susceptibility measurements indicate a small power - law temperature dependence below 10 k as can be seen in fig . [ fig : chidcv3 ] . although the temperature range is restricted , for samples with @xmath107 a fit of the form @xmath137 results in small , between 0.4 and 0.05 , values of @xmath138 . we note that these data were taken in significant dc magnetic fields of between 1 kg and 1 t in order to produce reasonable signal sizes for our squid magnetometery measurements . even so , we find similar small power - law behavior with the tendency for @xmath138 to fall toward zero near @xmath9 . a compilation of the power - laws determined from these different measurements is presented in fig . [ fig : chidcpwl ] where some scatter in the values is apparent along with the tendency for larger @xmath139 values to be found in specific heat measurements than in magnetic susceptibility measurements . our data and analysis outlined above have led us to the conclusion that ferromagnetic clusters form at @xmath140 . the sensitivity of our susceptibility measurements to small increases in pressure is one of the consequences of these weakly interacting ferromagnetically coupled clusters . in addition , we observe small power - law dependencies of @xmath99 , @xmath74 , and @xmath32 at low-@xmath7 with changes in temperature as was demonstrated in fig . [ fig : sph3pl ] for @xmath74 , in figs . [ fig : sph2pl ] and [ fig : sph3pl ] for the specific heat , and is apparent in fig . [ fig : chidcv3 ] for @xmath32 . in each case these quantities are observed to vary as @xmath141 with @xmath142 . all of these power - laws have similar values although there is significant scatter apparent . however , the similarities between these power - laws , @xmath143 , @xmath138 , and @xmath133 for individual samples implies that a single physical mechanism may be responsible . a clue as to the identity of the physical mechanism can be found in the dependence of the magnetization on temperature and magnetic field , @xmath144 . as we have noted above , the magnetic susceptibility does not conform to curie - weiss dependence , instead being characterized by a smaller power - law divergence at low-@xmath7 . this @xmath7-dependence suggests a magnetization that is caused by independent fluctuators having a power - law local energy distribution , @xmath145 , where @xmath146 is the local energy scale of the fluctuator . the form of the magnetization for freely fluctuating paramagnetic moments is well known to follow the curie law @xmath147 , where @xmath61 is the brillouin function@xcite and thus scales as @xmath148 . in the case of the power - law energy distribution the magnetization has been shown to scale as @xmath149@xcite . as can be seen in fig . [ fig : vladsc]a , which displays @xmath144 between 1.8 and 10 k for our two crystals with @xmath2 closest to @xmath9 , the the standard scaling form does not scale our data satisfactorily . this is to be expected for a system on the verge of magnetic ordering where magnetic moments have significant interaction . however , by simply allowing for a small value of @xmath150 we can scale all of our @xmath8 and @xmath7 dependent data reasonably well as can be seen in frame b of the figure ( @xmath151 ) . we note that @xmath150 is smaller than the values we found for @xmath133 and @xmath138 , but still distinct from zero . we note that this form for @xmath24 does not hold well for @xmath152 since @xmath77 increases sharply at low fields thus resembling more closely @xmath77 of ferromagnetic materials . in addition to this unusual field and temperature dependence of @xmath24 , our specific heat measurements also reveal a strong sensitivity to magnetic fields as we demonstrated in fig . [ fig : sph4pl ] . although we found that above 2 k the @xmath7 and @xmath8 dependence of @xmath153 appears consistent with a schottky anomaly form , @xmath153 below 1 k is not , even when a 3 t magnetic field is applied . this is emphasized by the magnetic field dependence of the low-@xmath7 specific heat which is explicitly displayed for @xmath154 mk fig . [ fig : sph5pl ] . here moderately sized fields are seen to suppress @xmath99 dramatically . the power - law like decrease in @xmath99 is displayed in the inset where the lines are power - laws in field , @xmath155 with @xmath156 for the @xmath131 crystal and @xmath157 for the @xmath85 crystal . we note that for a schottky anomaly @xmath99 decreases exponentially in @xmath8 for fields where @xmath158 which is inconsistent with our data@xcite . although there are many physical realizations of this scaling form for the magnetization we observe ( independent fluctuators with a power - law energy distribution ) , and small power - law @xmath7 dependent magnetic susceptibilities have been known to describe doped semiconductors very near insulator - to - metal transitions , our identification of magnetic clusters forming in this doped semiconducting system presents an obvious candidate explanation for all of the data presented thus far . we posit that our data lead naturally to the conclusion that griffiths phase formation is the most likely physical mechanism responsible for the unusual properties we measure for samples in proximity to the critical concentration for magnetic ordering . magnetic griffiths phenomena have been used to describe disordered magnetic systems theoretically in the case where the disorder is sufficient to cause inhomogeneity in the formation of the magnetic ground state@xcite . in the vicinity of the quantum critical point , droplets of the magnetically ordered phase are thought to form due to the inhomogeneities inherent to a chemically substituted material . these droplets reside within the paramagnetic phase and can be exponentially rare if the statistics are poissonian so that @xmath159 , where p is the probability of having a droplet of volume @xmath160 and @xmath161 is related to the strength of the disorder@xcite . since fe@xmath5co@xmath6s@xmath0 has a ferromagnetic ground state for @xmath11 , the droplets most likely carry a net magnetic moment simply proportional to the size of the droplet . these magnetic moments will have a finite probability of tunneling to nearby excited magnetization states . the tunneling rate for such a droplet is predicted to be exponential in the number of spins that form the droplet , such that @xmath162 where b is a constant related to the microscopic tunneling mechanism , and @xmath163 is a frequency cutoff@xcite . since the tunneling rate is proportional to the energy splitting of the different magnetization states the distribution of energy splittings can be written as @xmath164 \sim \delta^{\xi -1}$ ] , where @xmath165 , giving us our power - law distribution of energy levels suggested by the scaling of our @xmath144 data . with this form for the energy splittings , power - law forms for the magnetic susceptibility and specific heat follow by considering all clusters with @xmath166 to be frozen such that they do not contribute to the susceptibility or entropy@xcite . thus , a modified curie form @xmath167 , where @xmath168 is the density of clusters with @xmath169 is predicted@xcite . @xmath168 can be easily calculated from @xmath170 as @xmath171 and thus , @xmath172 . in addition , @xmath173 and since @xmath174 we have @xmath175 . therefore , the griffiths phase phenomenology leads to a form of the magnetization , susceptibility , and , @xmath99 similar to that found in our experiments . there has been some discussion in the literature about the ability of magnetic clusters to tunnel into different magnetization states in the presence of itinerant electrons . the presence of the conduction electrons which couple to the local magnetic moments via the kondo effect cause dissipative effects on the tunneling of magnetic clusters . the dissipation limits the size of clusters allowed to tunnel into opposing magnetization states thus cutting off the singular behavior in the models@xcite . millis et al . argue that in heavy fermion metals the the electrons are sufficiently well coupled to local magnetic moments , with kondo temperatures of order 100 k , so that griffiths phases are unlikely to be observed in these systems@xcite . the idea being that the dissipation is relevant at the scale of the kondo temperature , thus the temperature region where griffiths phases may be observed is very small . however , castro neto and jones argue that in the region of magnetic clusters the coupling of conduction electrons to the magnetic moments is not important since in these regions the electron density of states is not renormalized by the kondo effect@xcite . in fe@xmath5co@xmath6s@xmath0 the relevant energy scales are such that griffiths phases may be observed over a much wider temperature scale . here , the kondo coupling scale is only of order 2 k in the region of the critical co concentration for ferromagnetism . the kondo scale is likely so small because the density of electron states , in the approximation of a spherical electron pocket , is proportional to @xmath176 where @xmath177 is the electron density which is more than 2 orders of magnitude smaller , for the @xmath2 range investigated here , than in most heavy fermion metals@xcite . in addition , the disorder inherent in a doped semiconducting system results in a high scattering rate of the electrons further reducing their ability to screen local moments . therefore , we expect the dissipation due to a conducting gas of electrons to be a much smaller effect than in the heavy fermion metals where electron densities are much larger and the strong kondo coupling renormalizes the electron dos to very high levels . in addition , the relevant energy scale for local moment interactions is commonly estimated by the curie temperature of the clean magnetic system without dilution . in this case the clean ferromagnet is cos@xmath0 and this sets the largest energy scale for local moment interactions at 120 k. estimates for the average interaction energy of local moments include the weiss temperature which we have measured via the magnetic susceptibility and is displayed in fig . [ fig : tcpl ] . from the average and maximum local moment interaction scales compared with the kondo temperature scale in our materials , we conclude that dissipative effects of the conduction electron gas should be much smaller in fe@xmath5co@xmath6s@xmath0 than in heavy fermion metals . however , rounding of the transition due to kondo coupling of the local moments and the conducting electrons may be relevant at the lowest temperatures . the figures and discussion above give a general description of fe@xmath5co@xmath6s@xmath0 as it evolves from a diamagnetic insulator into a ferromagnetic metal . these data establish a magnetic transition that can be observed via the ac magnetic susceptibility down to very low temperatures at very small @xmath2 . it is clear from our data that magnetic clusters form above @xmath10 and an inhomogeneous magnetic state below @xmath10 . in this section we explore in more detail the behavior of our fe@xmath5co@xmath6s@xmath0 crystals having a magnetic phase transition , or perhaps a magnetic glass freezing , at finite temperatures . we examine the frequency and magnetic field dependence of the magnetic susceptibility both above and below the ordering to compare with typical spin glasses and disordered fms to gain insight on how such a low density disordered metal establishes a magnetic ground state . to this end we have carefully measured the @xmath178-dependence of @xmath74 and @xmath76 of a few of our samples with @xmath11 to probe the low frequency dynamics of the highly paramagnetic state above @xmath10 as well as the magnetic state below @xmath10 . in fig . [ fig : chifreqtemp ] @xmath74 and @xmath76 of our @xmath179 sample is displayed for frequencies between 1 and 1000 hz . here , we observe significant decreases in the magnitude of both @xmath74 and @xmath76 , very small changes to the temperature of the maximum in @xmath74 ( at most 0.1 k increase ) , and a larger increase of the temperature of the maximum in @xmath76 with measurement frequency . we observe no indications of the onset of significant skin effects and eddy current absorption which cause a steady increase in the out - of - phase susceptibility ( @xmath76 ) with @xmath178 despite the large out - of - phase component of the signal . the frequency dependence we measure is similar to the frequency dependent changes that occur in the archetypal spin glass cumn@xcite with a relative shift of @xmath10 per decade of measurement frequency of less than 1% . however there are with two important differences with typical spin glass behavior that draw our attention . first , we observe much larger changes in the magnitudes of both @xmath74 and @xmath76 with frequency than is typical for a metallic spin glass . second , the temperature range over which we observe a suppression of @xmath74 is from at least 0.5 @xmath10 to @xmath180 , whereas for cumn of similar dopant concentrations there is no measurable @xmath178-dependence of @xmath74 for reduced temperatures , @xmath181 , where @xmath182 is the spin glass freezing temperature@xcite . the frequency dependence of @xmath74 and @xmath76 for several temperatures is explicitly demonstrated in fig . [ fig : chitempfreq ] for temperatures near @xmath10 . while @xmath74 decreases monotonically with frequency @xmath76 is characterized by a broad peak which moves through the frequency window of our measurement from low @xmath178 to high @xmath178 with temperature . at @xmath183 , @xmath184 is well described by a logarithmic dependence , as demonstrated in the figure by the dashed lines , with an extrapolated zero crossing between @xmath185 and @xmath186 hz . this dependence has also been observed in insulating spin glasses such as ising - like liho@xmath6y@xmath5f@xmath187@xcite . in addition , @xmath76 has very little @xmath178-dependence in the same temperature range so that the dynamic response of the system below @xmath10 is much like that of a spin glass where fluctuations occur on all long time scales@xcite . thus , the magnetic dynamics at these low frequencies resemble that of common spin glass systems , with the added feature that the time dependence extends to temperatures far above @xmath10 . in addition to the enhanced sensitivity to measurement frequency near @xmath10 our crystals display an enhanced sensitivity to external magnetic fields in the same temperature range . this is demonstrated in fig . [ fig : chifield3pl ] for an @xmath179 sample consisting of several small crystals . we note that the data displayed in fig . [ fig : chifield3pl]a was previously published@xcite and is reproduced here for completeness and comparison of @xmath74 and @xmath76 . in this experiment we have carefully minimized the dc magnetic field for our @xmath4 scan which reveals not only the @xmath74 maximum near 4.5 k but a shoulder @xmath188 as well . in addition @xmath76 has two maxima , at 4.5 k and at @xmath189 k. the differences we observe in fig . [ fig : chifield3pl ] with application of very small fields , as well as the appearance of structure in the @xmath7-sweeps when the field is carefully zeroed are extraordinary . we interpret this sensitivity as an indication of enormous degeneracy , or near degeneracy , of magnetic moment configurations in these crystals . we observe that dc fields as small as 10 oe have a significant effect , decreasing the peak value of @xmath74 by a factor of 3.5 and @xmath76 by more than a factor of 30 . the reason for the marked difference between the ac and dc magnetic susceptibilities that we noted above is now quite clear ; the dc magnetic susceptibility was measured at fields between 50 and 10@xmath104 oe applied ( dc ) field , whereas @xmath74 shows dramatic changes between 0 and 200 oe in the range @xmath190 for all samples measured with ac susceptibility techniques at finite @xmath8 . therefore , the ac susceptibility data represent a clearer impression of the @xmath4 magnetic state of our fe@xmath5co@xmath6s@xmath0 crystals . in addition , the sensitivity to frequency and magnetic fields indicates a complex energy landscape with nearly degenerate energy minima for the magnetic moments . our main conclusion from the broad temperature and field dependence of @xmath24 and @xmath97 in sections [ dcsusceptibilityandmagnetizationmeasurements ] , [ acsusceptibilitymeasurements ] , and [ thespecificheat ] , that clusters of magnetic moments form at @xmath191 , is consistent with the conclusion of a spin - glass - like sensitivity to measurement @xmath178 and applied magnetic fields over a broad @xmath7-range about @xmath10 . to aid in understanding the changes that occur to the magnetic response of this system just above the ordering temperature , we have compared @xmath80 taken at small dc magnetic fields as a function of @xmath7 to the curie - weiss form as shown in fig . [ fig : ichivt ] for @xmath192 and @xmath179 . what is interesting here is that @xmath80 falls below the curie - weiss behavior established at higher temperatures . that is , @xmath74 diverges more strongly than the curie - weiss behavior for temperatures substantially above @xmath10 . we note that @xmath80 evolves toward the curie - weiss form , @xmath193 , with the application of small fields of order 100 oe for @xmath192 and only 10 oe for @xmath179 . the strong increase of @xmath74 above @xmath10 and @xmath194 is evidence for short ranged ferromagnetic order , or as we have noted above , ferromagnetically aligned clusters of magnetic moment formation@xcite . we conclude from this form of @xmath74 that the ground state of this system is likely disordered ferromagnetic rather than a spin - glass state . in the bottom half of fig . [ fig : ichivt ] we demonstrate that a power - law temperature dependent form@xcite , @xmath195 with @xmath196 , describes @xmath74 for 1.5 to 2 orders of magnitude in reduced temperature above @xmath197 . here @xmath197 is an indication of the temperature scale where ferromagnetic clusters begin to form and is an indication of the maximum interaction strength between magnetic moments . in contrast , @xmath81 is an average , or mean field interaction between magnetic moments . a compilation of @xmath198 and @xmath197 values from fits of eq . [ eq : grifchi ] to @xmath80 for a small number of crystals with @xmath2 varying between 0.012 and 0.045 is included in fig . [ fig : chidcpwl ] . while @xmath197 tends to increase with @xmath2 , @xmath198 takes on values between 0.2 and 0.4 for @xmath11 . beyond the enhanced frequency and magnetic field dependence seen in figs . [ fig : chifreqtemp ] and [ fig : chifield3pl ] , there are features in our data that indicate that the magnetic state of fe@xmath5co@xmath6s@xmath0 is far more interesting than that of a metallic spin glass , particularly for temperatures above @xmath10 . for example , in fig . [ fig : chirivhfd1 ] we plot the non - linear ac susceptibility , @xmath199 and @xmath200 at temperatures near @xmath10 showing explicitly the changes that occur with @xmath8 at constant @xmath7 . although this figure shows a broad , continuous , decrease of both @xmath74 and @xmath76 with @xmath8 for @xmath183 , it also demonstrates a very sharp low field behavior for @xmath201 . a mean - field analysis of the non - linear susceptibility involves the assumption of analyticity around @xmath4 such that a power series expansion in even powers , @xmath202 can be employed to describe @xmath199 . the extraordinary field dependence of @xmath74 seen in fig . [ fig : chirivhfd1 ] can not be described accurately by this form even over a reduced field range of only 10 oe for @xmath140 . thus the randomness due to the chemical substitutional disorder has modified the field response of this system above @xmath10 such that it is no longer analytic near @xmath4 . this form of @xmath199 is similar to that found in ( body - centered tetragonal ) liho@xmath203y@xmath5f@xmath187 where the magnetic moments associated with the ho ions strongly align along the c - axis and couple to each other via the dipolar interaction@xcite . the ground state is ferromagnetic for @xmath204 and a spin - glass for @xmath205@xcite . a magnetic field applied transverse to the moments , @xmath206 , causes a mixing of excited states with the ground state doublet . this suppresses the ordering such that a quantum critical point ( @xmath207 ) has been accessed for a large range of @xmath2 . for @xmath208 refs . find that @xmath209 and @xmath210 , where @xmath211 is along the @xmath161-axis , both have a finite slope , @xmath212 , and thus can not be expressed in a perturbative expansion in @xmath8 . the interpretation was that the disorder induced field distribution contains rare large amplitude random fields that prevent a description of @xmath213 in a mean field way as @xmath214 . the authors conclude that the non - analyticity at @xmath215 and @xmath140 is a manifestation of griffith singularities@xcite . although there are similarities between our system , fe@xmath5co@xmath6s@xmath0 , and liho@xmath6y@xmath5f@xmath187 , both are disordered ferromagnets that are derived by chemical substitution from an insulating parent compound , there are significant differences which are important to point out . the most obvious is that fe@xmath5co@xmath6s@xmath0 is a metal for @xmath216 , while liho@xmath6y@xmath5f@xmath187 remains insulating for all @xmath2 . in addition , liho@xmath6y@xmath5f@xmath187 is strongly ising - like with the magnetic moments preferring to align along the @xmath161-axis . in contrast , fe@xmath5co@xmath6s@xmath0 is cubic and for @xmath217 has been shown to have a very small cubic anisotropy of its magnetization so that it been described as heisenberg - like@xcite . however , we find that both of these disordered magnets contain rare large amplitude fluctuations that are the hallmarks of griffiths phase formation for @xmath140 . thus , not only do we find evidence for such phases in samples on the verge of magnetic ordering at @xmath207 , but also discover evidence for similar rare region effects at larger @xmath2 , @xmath11 above the ordering temperature . in contrast to mean - field forms , the magnetic field dependence of @xmath74 shown in fig . [ fig : chirivhfd1 ] can instead be well described as a simple power - law in @xmath8 at @xmath140 as demonstrated in figs . [ fig : chirivhfd ] and [ fig : chirivh ] . these figures demonstrate that a small power - law form , @xmath218 , with @xmath219 between 0 and 1 describes the data over at least a decade and 1/2 in field for @xmath140 , while for @xmath183 a much more gentle field dependence is seen . the best fit @xmath219 values have been included in fig . [ fig : chidcpwl ] for two samples where we have sufficient data for an accurate determination of this parameter . in addition , @xmath76 can also be described in a power - law form with an exponent between -1.3 and -2.0 . for fields between 50 and 100 oe @xmath76 is seen to undergo a much steeper decrease so that above these fields it is consistent with zero being no larger than the backgrounds in our measurement . the small power - law form for @xmath199 at @xmath140 is consistent with the griffiths phase hypothesis made on the basis of the temperature dependence above . [ fig : chirivh ] displays @xmath74 and @xmath76 at 1.8 k for 5 different samples with @xmath11 and demonstrates several features that are common . first we note that for these co concentrations @xmath220 k and , as discussed above , @xmath74 is not strongly field dependent below about 100 oe at this temperature . this is in contrast to the data shown above @xmath10 , a small subset of which is shown in the figure for comparison . for larger fields , @xmath74 is seen to decrease as @xmath221 while @xmath76 is suppressed to within the background level of zero above this same field scale . it is interesting to note that @xmath74 above @xmath222 oe is less dependent on @xmath2 suggesting that here the magnetic energy level distribution is much more independent of @xmath2 . our presentation in this section has focused on the ac magnetic susceptibility of our fe@xmath5co@xmath6s@xmath0 crystals with @xmath11 which demonstrates that we have discovered several unusual aspects . this includes a highly frequency dependent @xmath74 and @xmath76 which encompasses a temperature range both above and below @xmath10 that is much larger than in prototypical spin - glass systems . we have found an extraordinary non - analytic field dependence to the ac susceptibility near @xmath4 coupled with a small power law form to the temperature dependence . these aspects are indicative of an inhomogeneous magnetic system where rare ordered regions form at temperatures above the long - range ordering , or perhaps a glass - like freezing , temperature of the disordered system . these local ferromagnetically ordered regions are well described by the griffiths phase phenomenology@xcite where the non - universal power - law dependent properties are the result of quantum mechanical tunneling of these rare regions to states with different magnetizations . in this paper we have described a set of experiments exploring the nucleation of a magnetic ground state in a carrier - doped nonmagnetic insulator . for this investigation we chose the relatively simple diamagnetic insulator fes@xmath0 as it allows for co substitution for fe without the formation of second phases@xcite . in addition , end member of the series cos@xmath0 is an itinerant ferromagnet with a curie temperature of 120 k@xcite . we found that the co substitution yields a small number of itinerant charge carriers@xcite and magnetic moments that are likely localized to the co impurity sites . a weakly coupled magnetic ground state , spin glass or disordered ferromagnetic , developed for @xmath223 as evidenced by a peak in the temperature dependence of the magnetic susceptibility . as @xmath2 is increased beyond @xmath9 the magnetic state becomes more robust to temperature and magnetic field so that the fm state is more evident at fields and temperatures where common tests can easily be performed . in the paramagnetic regions of temperature and co concentration , we found fluctuating magnetic moments whose size was significantly larger than that expected for spin-1/2 moments of individual co dopants in an fes@xmath0 background . this led us to the conclusion that clusters of magnetic moments were forming upon cooling below 10 k for samples with @xmath107 and at @xmath140 for sample with @xmath11 . the formation of clusters of average size of 3 to 5 @xmath94 magnetic moments was also found to be consistent with the measured @xmath99 and the entropy derived from it . because of the disorder and assumed random placement of co impurities throughout our samples , it is likely that a distribution of cluster sizes results . this distribution is likely to have a very long tail for large cluster sizes so that there may be rare regions of local magnetic order . such a description fits in well with the sensitivity to magnetic fields that we observed in the susceptibility and magnetization of our crystals . we posit that it is these rare regions of incipient order that cause the unusual temperature and magnetic field dependencies that we measured in the magnetic susceptibility and specific heat . the consistent appearance of small power - laws in our analysis of both the low temperature behavior of samples in proximity to @xmath9 and at temperatures just above @xmath10 for samples with @xmath11 lead us to conclude that griffiths phase physics is the most likely mechanism determining the static and dynamic properties of this system@xcite . in this scenario large rare regions of order appear in disordered magnets at temperatures below the critical temperature of the system without disorder . these ordered regions which act as single large magnetic moments are subject to tunneling events where the magnetic moment reverses . when a distribution of cluster sizes is included , small power - law temperature dependences are predicted . details such as the anisotropies ( ising - like or heisenberg - like moments ) as well as the mechanisms for dissipation are thought to determine the observability conditions for these effects@xcite . fe@xmath5co@xmath6s@xmath0 appears to be a system where such effects may be observable over a wide temperature range as the kondo temperatures we measure are small , of order 1 k@xcite , while the weiss temperatures are typically much larger , so that dissipation by way of the charge carriers may be relevant only at the lowest temperatures . it is no surprise that the disorder inherent to a doped semiconducting system can lead to the formation of an inhomogeneous magnetic state at low temperatures . what we find extraordinary are the unusual power - law temperature and magnetic field dependencies of physical properties that we measure along with the very sharp , non - analytic , field dependence of the magnetic susceptibility near @xmath4 . that a simple model of disordered magnets@xcite which postulates the existence of rare ordered regions can describe the broad features of such a complicated system is both surprising and compelling . we thank i. vekhter and c. capan for discussions . jfd , dpy , and jyc acknowledge support of the nsf under dmr084376 , dmr0449022 , and dmr0756281 . we thank j. m. honig for providing crystals of v@xmath29ti@xmath30o@xmath31 for use as a manometer in our pressure cells .

the spatiotemporal oscillation patterns of the proteins mind and mine are used by the bacterium _ e. coli _ to sense its own geometry . strikingly , both computer simulations and experiments have recently shown that for the same geometry of the reaction volume , different oscillation patterns can be stable , with stochastic switching between them . here we use particle - based brownian dynamics simulations to predict the relative frequency of different oscillation patterns over a large range of three - dimensional compartment geometries , in excellent agreement with experimental results . fourier analyses as well as pattern recognition algorithms are used to automatically identify the different oscillation patterns and the switching rates between them . we also identify novel oscillation patterns in three - dimensional compartments with membrane - covered walls and identify a linear relation between the bound min - protein densities and the volume - to - surface ratio . in general , our work shows how geometry sensing is limited by multistability and stochastic fluctuations .

1610.05235

the bacterial min - proteins are a well studied example of a pattern - forming protein system that gives rise to rich spatiotemporal oscillations . it was discovered as a spatial regulator in bacterial cell division , where it ensures symmetric division by precise localization of the divisome to midcell @xcite . the dynamic nature of this protein system was demonstrated by live cell imaging in _ e. coli _ bacteria , where these proteins oscillate along the longitudinal axis between the cell poles of the rod - shaped bacterium , forming so - called _ polar zones _ @xcite . most bacteria use a cytoskeletal structure , a so - called _ z - ring _ , for the completion of bacterial cytokinesis @xcite . this z - ring self - assembles from filaments of polymerized ftsz - proteins , the prokaryotic homolog of the eukaryotic protein tubulin @xcite , which serve as a scaffold structure for midcell constriction and the eventual septum formation in the midplane . if successful , this process creates two equally sized daughter cells with an identical set of genetic information @xcite . a necessary prerequisite for successful symmetric cell division is hence the targeted assembly of ftsz towards midcell . in _ cells this is mediated by two independent mechanisms , nucleoid occlusion and the dynamic oscillation of the mincde proteins @xcite . while nucleoid occlusion prevents division near the chromosome , the min - system actively keeps the divisome away from the cell poles through the minc - protein acting as ftsz - polymerization inhibitor @xcite . the characteristic pole - to - pole oscillations create a time - averaged concentration gradient with a minimal inhibitor concentration of minc at midcell , suppressing z - ring assembly at the cell poles @xcite . although minc is indispensable for correct division site placement , it acts only as a passenger molecule , passively following the oscillatory dynamics of mind and mine @xcite . on the molecular level , the oscillations emerge from the cycling of the atpase mind between a freely diffusing state in the cytosolic bulk and a membrane - bound state , induced by its activator mine under continuous consumption of chemical energy by atp - hydrolysis , as shown schematically in figure [ cycle ] . in its atp - bound form mind homodimerizes and can subsequently attach to the inner bacterial membrane as atp - bound dimers using a membrane targeting sequence in form of a c - terminal amphipathic helix @xcite . despite the fact that the physicochemical details of mind membrane binding are not yet fully understood , it has been demonstrated that mind membrane binding is a cooperative process @xcite . when being bound , mind diffuses along the membrane . it also recruits mine which in turn triggers the atpase activity of mind , breaking the complex apart and releasing all constituents back into the cytosol . mind then freely diffuses in the bulk and can , after renewed loading of atp and dimerization , rebind to the membrane at a new position . this cycling of mind between two states is the core mechanism for wave propagation of the min - proteins . for a comprehensive overview on the underlying molecular processes , we refer to recent reviews on this topic @xcite . one of the most intriguing aspects of the min - system is the impact of geometry on the spatiotemporal patterns . while initial experiments in wild - type cells showed characteristic _ pole - to - pole oscillations _ , growing _ e. coli _ cells , which roughly double in length before division , can also give rise to stable oscillations in both daughter cells even before full septum closure @xcite . in very long filamentous mutants the pole - to - pole pattern vanishes and several minde localization zones emerge in a stripe - like manner ( _ striped oscillations _ ) , with a characteristic distance of , strongly reminiscent of standing waves @xcite . no stable oscillation patterns emerge in spherical cells where minde localization appears to be random without stable oscillation axes @xcite . strikingly , the min - system can be reconstituted outside the cellular context using purified components on supported lipid bilayers @xcite . using only fluorescently labeled mind and mine and atp as energy source , traveling surface waves were observed in the form of turning spirals and traveling stripes on flat homogeneous substrates , where mind proteins form a moving wave - front , that is consumed by mine at the trailing edge , demonstrating that mind and mine alone are indeed sufficient to induce dynamic patterning @xcite . interestingly , these assays work for different lipid species , demonstrating the robustness of the min - oscillations with respect to the detailed values of the binding rates . combining this reconstitution approach with membrane patterning , it was shown that the min - system is capable of orienting its oscillation axis along the longest path in the patch and hence in principle capable of sensing the surrounding geometry @xcite . more recently the gap between the traveling _ in vitro _ min - waves and the standing min - waves in live cells was closed , using microfabricated pdms compartments mimicking the shape of _ cells @xcite . in these biomimetic compartments , which confine the reaction space in 3d , pole - to - pole oscillations were observed , reminiscent of the paradigmatic _ in vivo _ oscillation mode . later it was shown that the min - oscillations are indeed sufficient to spatially direct ftsz - polymerization to midcell , linking two key elements of bacterial cell division in a synthetic bottom - up approach @xcite . in order to study the effect of geometry in the physiological context of the cell , one can place growing cells in microfabricated chambers of custom shape @xcite . this _ cell sculpting _ approach allowed the authors to systematically analyze the adaptation of the min - oscillations to compartment geometry and demonstrated experimentally that different oscillation patterns can be stable for the same cell geometry . using image processing , it was possible to measure the relative frequency of the different modes for a large range of interesting geometries @xcite . figure [ cycle ] summarizes the different geometries that have been used before in experiments and that are considered here with computer simulations . while geometry a uses a flat membrane patch , similar to flat patterned substrates @xcite , geometry b corresponds to microfabricated chambers with an open upper side @xcite . geometry c corresponds to the cell sculpting approach @xcite . like for other pattern forming systems , the theory of reaction - diffusion processes offers a suitable framework to address the min - oscillations from a theoretical point of view @xcite . many theoretical models have been proposed to unravel the physical principles behind this intriguing self - organizing protein system and to explain the origin of its rich spatiotemporal dynamics . while all of them rely on a reaction - diffusion mechanism similar to the turing model , they differ severely in their details . the first class of mathematical models used an effective one - dimensional pde - approach and relied strongly on phenomenological non - linearities in the reaction terms @xcite . although all of them successfully gave rise to pole - to - pole oscillations , they did not allow a clear interpretation of the underlying biomolecular processes and were not in agreement with all experimental observations , such as mine - ring formation and the dependence of the oscillation frequency on biological parameters . the next advance in model building was the focus on the decisive role of mind aggregation and the relevance of mind being present in two states ( adp- and atp - bound ) @xcite . very importantly , this highlighted the interplay between unhindered diffusion with a nucleotide exchange reaction in the bulk as a delay element for mind reattachment @xcite . subsequent models shared a common core framework but still differed strongly in the functional form of the protein binding kinetics and the transport properties of membrane - bound molecules @xcite . the main difference between the more recent models was the dimensionality , ranging from one dimension @xcite to two @xcite and three @xcite . moreover , the models can be classified as deterministic pde - models or using stochastic simulation frameworks , and whether they neglected membrane diffusion @xcite or not . while some models contained higher than second order non - linearities in concentrations of the reaction terms @xcite , it is the prevailing opinion to rely on at most second order non - linearities , allowing for a clear interpretation in terms of bimolecular reactions . following the same line of thought , a strong effort was made to distill a minimal system that explains the oscillation mechanism without the necessity of spatial templates or prelocalized determinants @xcite and neglecting secondary processes like filament formation @xcite . the most influential minimal model for the min - system has been suggested by huang and coworkers @xcite . it has been further simplified by discarding cooperative mind recruitment by minde complexes on the membrane @xcite , allowing a clear view on the core mechanisms : the cycling of mind between bulk and membrane , cooperativity of mind - recruitment and diffusion in bulk and along the membrane . using the minimal model , it has been shown that the canalized transfer of proteins from one polar zone to the other underlies the robustness of the min - oscillations @xcite . because the deterministic variants of the minimal model @xcite do not allow us to address the role of stochastic fluctuations , a stochastic and fully three - dimensional version has been introduced to study the effect of stochastic fluctuations in patterned environments @xcite . for rectangular patterns of @xmath0 , it was found that the system can be bistable , with transverse pole - to - pole oscillations along the minor and longitudinal striped oscillations along the major axis , respectively . in this early work , it was observed that the stable phase emerged depending on the initial conditions and that sometimes switching occurred , but the statistics was not sufficiently good to observe switching in quantitative detail . indeed such multistability has been observed experimentally in sculptured cells over a large range of cell shapes @xcite and the deterministic minimal model has been used to explain the relative frequency of the different oscillations patterns for a given shape using a perturbation scheme @xcite . however , as a deterministic model , this approach was not able to address the rate with which one pattern stochastically switches into another . here we address this important subject by using particle - based brownian dynamics computer simulations . compared to earlier work along these lines @xcite , we have developed new methods to efficiently simulate and analyze the switching process . we find excellent agreement with experimental data and measure for the first time the switching time of multistable oscillation patterns . we also use our model to confirm that it contains the minimal ingredients for the emergence of min - oscillations . in addition , we use our stochastic model to investigate the three - dimensional concentration profiles in different geometries and in particular the role of edges in membrane - covered compartments . we identify novel oscillation patterns in compartments with membrane - covered walls and find a surprisingly simple ( linear ) relation between the bound min - protein densities and the volume - to - surface ratio , which might be relevant for geometry sensing by _ e. coli _ cells . for the particle - based simulation , we use the reaction scheme of the minimal model for cooperative attachment @xcite . the model uses the following interactions between min - proteins and the inner bacterial membrane ( schematically shown in figure [ cycle ] ) . freely diffusing cytoplasmic @xmath1 can bind to the membrane with a rate constant @xmath2 @xmath3 @xmath1 binds preferably to high density @xmath4 regions ( cooperative mind binding ) @xmath5 membrane - bound mind also recruits cytoplasmic mine to the membrane with rate @xmath6 , creating a @xmath7 complex @xmath8 all membrane - bound proteins diffuse in the plane of the membrane , but with a much smaller diffusion constant than in the bulk . @llllll parameter & set * a * & set * b * & set * c * & unit & description ( reaction type ) + @xmath9 & & & & & bulk diffusion coefficient of mind + @xmath10 & & & & & bulk diffusion coefficient of mine + @xmath11 & & & & & membrane diffusion coefficient + @xmath12 & & & & & first order , unimolecular + @xmath2 & & & & & first order , membrane attachment + @xmath13 & & @xmath14 & & & second order , bimolecular + @xmath15 & & @xmath14 & & & second order , bimolecular + @xmath16 & & & & & first order , unimolecular + @xmath17 & & & & m & total mind concentration + @xmath18 & & & & m & total mine concentration + mine attachment activates atp hydrolysis of mind . the hydrolysis of atp to adp breaks up the membrane - bound complex and releases @xmath19 and mine back into the cytoplasmic bulk with rate constant @xmath20 @xmath21 finally @xmath19 exchanges adp by another atp molecule ( nucleotide exchange ) with the rate @xmath12 @xmath22 this completes the reaction cycle . in table [ tbl : parameterset ] we list our parameter values as set a. for comparison , we also list parameter values used in other studies ( set b in @xcite and set c in @xcite ) . we use custom - written code to simulate the stochastic dynamics of the min - system with very good statistics . for all simulations we use a fixed discrete time step of @xmath23 . during every time step each particle is first propagated in space . thereafter every particle can react according to the previously introduced min reaction scheme [ eq : reactionschemefirst ] [ eq : reactionschemelast ] . the movement of both free and membrane - bound particles is realized through brownian dynamics . individual molecules are treated as point - like particles without orientation . therefore we can monitor the propagation separately for each cartesian coordinate . during a simulation step of @xmath24 the displacements of the diffusing particles with diffusion constant @xmath25 are drawn from a gaussian distribution with standard deviation @xmath26 @xcite such that @xmath27 where @xmath28 is the probability distribution of @xmath29 . the same update step is used for the @xmath30 and @xmath31 direction . free particles in the bulk of the simulated volume undergo three - dimensional diffusion with reflective boundary conditions at the borders of the simulation volume . the membrane - bound particles perform a two - dimensional diffusion on the membrane with a much smaller diffusion constant @xmath32 ( compare table [ tbl : parameterset ] ) . membrane - bound particles are allowed to diffuse between different membrane areas that are in contact with each other . the different reactions in the min reaction scheme [ eq : reactionschemefirst ] [ eq : reactionschemelast ] can be classified into three different types ( more details on the corresponding implementations are given in the supplementary information ) . the first type considered here are first order reactions without explicit spatial dependence . the conversion of @xmath19 to @xmath1 and the unbinding of the @xmath7 complex from the membrane are of this type . such reactions are treated as a simple poisson process . for a reaction rate @xmath33 , the probability to react during a time step @xmath24 is given by @xmath34 the second type is also a first order reaction , but with confinement to a reactive area at a border of the simulated volume . the membrane attachment of @xmath1 proteins is a reaction of this type . for a given reaction rate @xmath35 , we implement these reactions by allowing particles that are closer to the membrane than @xmath36 to attempt membrane attachment with a poisson rate @xmath37 . this results in a reaction probability of @xmath38 the last reaction type is a second order reaction between free and membrane - bound particles . the cooperative recruitment of cytosolic @xmath1 and mine to membrane - bound mind are reactions of this third type . in our simulation we adopt the algorithm implemented in the software package smoldyn , which has been used earlier to simulate the min - system @xcite . this algorithm is based on the smoluchowski framework in which two particles react upon collision @xcite . however , the classical treatment by smoluchowski only considers diffusion - limited reactions and therefore assumes instantaneous reactions upon collision . in order to take finite reaction rates into account , one imposes a radiation boundary condition @xcite . from the diffusion constant @xmath25 , the reaction rate @xmath35 and the simulation time step @xmath24 , a reaction radius @xmath39 is calculated @xcite . whenever a freely diffusing particle comes within the distance of @xmath39 to a membrane - bound particle , the free particle reacts . for intermediate values of @xmath24 ( such as the time step of that we use for the min - system ) the value of @xmath40 is obtained numerically @xcite . those numerical values are taken from the smoldyn software . for example , for parameter set a the reaction radius for the rate @xmath13 is @xmath41 , and for @xmath15 it is @xmath42 . in our simulations we use rectangular reaction compartments . we considered three different membrane setups as illustrated in figure [ cycle ] . to mimic _ in vitro _ experiments , where substrates or open compartments are functionalized with a membrane layer @xcite , we place the reactive membrane at the bottom ( geometry a ) or at the side walls and the bottom of the simulation compartment ( geometry b ) . to simulate rectangular shaped _ e. coli _ cells , inspired by the cell sculpting approach from @xcite , fully membrane - covered volumes are used ( geometry c ) . we refer to the long side of the lateral extension as the major or the @xmath43-axis , and the smaller side as minor or @xmath30-axis , and accordingly consider the compartment height to extend in the @xmath31-direction , aligning the rectangular geometry perpendicular with the coordinate frame . in our simulations we investigate a wide range of compartment dimensions . for a simulation box with a length of , width of and height of , we use @xmath1 particles , @xmath19 particles and mine particles as initial condition @xcite . these particle numbers amount to a total mind concentration of @xmath44 m and a mine concentration of @xmath45 m . for other simulation compartment sizes we scale the particle numbers linear with the volume , since in experiments _ e. coli _ bacteria typically have a constant min - protein concentration @xcite . in our simulations of the min - system different oscillation patterns emerge along the major or minor axis of the simulation compartment . in order to analyze the frequency of different modes and the stability of the oscillations in the large amount of simulation data , an oscillation mode recognition algorithm is needed . therefore we monitor the mind protein densities at the poles of the different axes over time . to determine the axis along which the oscillation takes place , we compare the fourier transformation of the normalized densities over time ( @xmath46 where i denotes the discretized time resolution ) . if an oscillation takes place , there is a dominant peak in the fourier spectrum and the overall maximal amplitude of the fourier spectrum is significantly higher than the one from the non - oscillating axis . the same fourier spectrum is also used to determine the oscillation period @xmath47 . to differentiate between pole - to - pole oscillations and striped oscillations of a given axis in the system , we extract the phase difference between the density oscillation at the poles of the cell . when identifying switching events , one has to be more careful because stochastic fluctuations might lead to temporal changes that might be mistaken to be mode switches . for this purpose , we therefore smoothen the data . in detail , we calculate the convolution @xmath48 between the densities over time and a gaussian time window @xmath49 @xmath50 here @xmath51 is the time between successive density measurements and we set @xmath52 as width of the time window . the current oscillation mode is now determined from the convoluted densities @xmath48 and assigned to the time @xmath53 . in this way , only switches are identified that persist for a sufficiently long time . first we investigated the oscillations that emerge in geometry a with parameter set a using a rectangular simulation volume with dimensions @xmath54 ( @xmath55 ) . with this particular choice the width of the system approximately matches the typical length of wild - type _ e. coli _ cells and the length of the system corresponds to the length of a grown _ e. coli _ cell which can roughly double in length before septum formation and division . as shown by the kymographs in figure [ switch ] and in agreement with the results of hoffmann _ et al . _ @xcite , in our simulations two different oscillation modes occur ( compare also supplemental movie s1 ) . note from the color legend that dark and light colors correspond to low and high concentrations , respectively , as used throughout this work . in the first mode the min - proteins oscillate along the minor @xmath30-axis from one pole to the other ( pole - to - pole oscillation ) . in the second mode the proteins oscillate along the major @xmath43-axis between the poles and the middle of the compartment ( striped oscillation ) . the system stochastically switches between the two modes , sometimes via a short oscillation along the diagonal of the compartment . the mode switching behavior of the min - system in large volumes is in agreement with the experimental results of wu _ et al . _ @xcite and can not be analyzed completely with conventional pde - models of the min - oscillations because they do not account for the noise in the system leading to the stochastic switch . detailed analyses of the pole - to - pole and the striped oscillations from figure [ switch ] are shown in figure [ all ] and [ all ] , respectively ( compare also supplemental movies s2 and s3 , respectively ) . first , we consider the pole - to - pole oscillations in figure [ all ] . in the kymographs of the bound mind and mine proteins ( top row figures ) we see clusters of bound proteins that detach from the membrane beginning in the middle and from there move towards one of the poles of the compartment in an alternating way . the shapes of the bound mind protein density clusters in the kymographs have a triangular form , in contrast to the line - like structures of the bound mine proteins . those density lines in the bound mine kymograph show that the mine proteins form a high density cluster in the middle of the cell which propagates to one of the compartment poles . this behavior is similar to the experimentally observed ringlike structures of mine proteins in _ e. coli _ bacteria that travel from the middle to the poles of the cell , leading to the dissociation of mind proteins from the membrane . the kymographs of the free particles ( middle row in figure [ all ] ) are averaged over all heights and have the inverse shape of the corresponding kymographs of the bound particles ( top row in figure [ all ] ) . where the density of bound particles is high , the density of free particles is low and vice versa . during the simulations the spatial density differences of both mind and mine are higher for membrane - bound particles than for free particles in the bulk . therefore in the two bottom kymographs in figure [ all ] , which are showing total particle densities , the pattern of membrane - bound particles is dominant . using a compartment of @xmath54 as in figures [ switch ] and [ all ] . * b * density change of non - bound mind and mine in time at different heights above the membrane for a @xmath56 bottom area and @xmath57 compartment height . [ cut ] ] , @xmath58 and @xmath59 , respectively . * b * density kymographs along the major axis for @xmath60 , @xmath61 and @xmath62 . * c * time - averaged density profiles for different membrane diffusion - coefficients @xmath63 . * d * time - averaged density profiles for different cooperative mind membrane - recruitment rates @xmath13 . [ sensitivity ] ] the kymographs of the striped oscillation in figure [ all ] have the same structure as the ones of the pole - to - pole oscillations . however , the edges of the bound min - protein clusters in the kymographs , that indicate the traveling min - waves , are curved , in contrast to the straight lines that we see for the pole - to - pole oscillations as shown in figure [ all ] . the time - averaged density profiles of mind proteins for the pole - to - pole and striped oscillations are shown in figure [ all ] and , respectively . as expected the density of the proteins is minimal between the oscillation nodes of the emerging standing wave patterns . ) . below @xmath64 and above @xmath65 the projected density profile along the side walls of the compartment is shown . ] it is highly instructive to compare the time evolution of the mind and mine protein densities . in figure [ cut ] we plot the time evolution of the particle densities of the transverse pole - to - pole oscillation at a fixed position @xmath66 , which is at the edge of the minor axis along which the oscillation takes place . the shape of the transient density profiles is similar to experimentally observed density profiles of traveling min - protein waves on flat membrane surfaces @xcite . the period of both oscillations modes was @xmath67{33.8(1)}{\second}$ ] , which is in agreement with the results of huang _ _ @xcite . to analyze the influence of the bulk volume on the oscillations , we have also monitored how the mind and mine densities changes at different heights above the membrane . for this we again use geometry a , but now with a bottom surface of only @xmath56 , which robustly produces longitudinal pole - to - pole oscillations along the major axis . for a compartment height of the volume was divided in three layers , and for each layer the mean density on one side of the major axis was plotted over time as shown in figure [ cut ] . we see that the largest density changes take place in the layer directly above the membrane , however , the oscillations persist up to the top layer even for the highest compartment with @xmath68 . strikingly , the mind - density is always much higher than the one of mine . furthermore we reduced the compartment height to . interestingly , the pole - to - pole oscillations were still present . this implies that the bulk of the simulation volume has only a mild effect on the oscillations in geometry a , despite the fact that there are appreciable density variations in the bulk . we next checked that our model indeed includes the essential elements for the emergence of min - oscillations . because the mind - switching between bulk and membrane is clearly indispensible , here we consider the relevance of diffusion on the membrane and of cooperative recruitment . all simulations in this section are carried out using geometry a in a @xmath69 compartment geometry . we chose this geometry since we expect it to give rise to stable longitudinal pole - to - pole oscillations and hence can closely monitor any deviations from this default oscillation mode . all other parameters were kept fixed and we used parameter set a for this test . figure [ sensitivity ] shows three kymographs of the system along the major axis . switching off membrane diffusion entirely ( first panel ) leads to a loss of stable oscillation patterns . instead small striped - like mind - patches emerge erratically on the membrane . the second panel in figure [ sensitivity ] shows the default diffusion coefficient of @xmath58 where stable pole - to - pole oscillations emerge along the major axis . by further increasing the membrane - diffusion - coefficient the pole - to - pole oscillations still robustly emerge but increasingly smear out . this behavior is most clearly illustrated by looking at the time - averaged density profiles as shown in figure [ sensitivity ] . the inset in figure [ sensitivity ] shows that without membrane diffusion , no oscillations emerge at all ( square symbol ) , while a slow diffusivity as used in parameter set a seems optimal . in a similar fashion we analyzed the influence of the cooperative membrane recruitment of mind . figure [ sensitivity ] shows three kymographs for no cooperative recruitment ( @xmath70 ) , the default value from parameter set a and a two - fold increase , respectively . without cooperativity in this process no oscillations emerge at all . the second panel shows again the default oscillation mode while already a two - fold increase also leads to unstable behavior without any patterns emerging . this sensitivity with respect to the cooperative recruitment rate @xmath13 is also summarized in figure [ sensitivity ] , which illustrates that only in a vary narrow range stable oscillations emerge . although a complete parameter scan is out of the question at the current stage for reasons of computer time , we conclude that stable min - oscillations emerge only for certain parameter values and that parameter set a performs very well in this respect . as we have seen in the preceding section , the simulation compartment height has little influence on the emerging oscillation modes when the membrane only covers the bottom area of the simulation compartment ( geometry a ) . however , when we extend the membrane to cover bottom and side walls of the compartment ( geometry b ) , we find that the oscillation modes change with increasing wall height . first , we consider a compartment that only exhibits longitudinal pole - to - pole oscillations along the major axis ( wall height of and bottom area of @xmath71 ) . in contrast to the flat membrane geometry a , here the mind protein density decreases in the vicinity of the membrane edges between the side walls and the bottom area , as shown in figure [ densityprofilew ] . this is due to the decreased volume per membrane area ratio in the vicinity of the membrane corners , leading to a decreased density of bound mind proteins in these regions . this effect is clearly visible by comparing the time - averaged density profile in figure [ densityprofilew ] with the one shown in [ all ] . the full oscillation pattern is shown in [ phasediagram ] for the lowest height value ( black label ) and resembles a pole - to - pole oscillation for geometry a , but now between the two opposing walls along the major axis . now we increase the wall height in increments of . the oscillation changes at a wall height of @xmath72 to a new oscillation mode ( green label , compare also supplemental movie s4 ) . there the proteins start to oscillate between the middle of the bottom area and the upper edges of the walls . this newly identified _ edge oscillation _ with one polar zone at the bottom and one top polar zone at each of the four side walls persists until the wall height reaches . thereafter the oscillation mode changes to a striped edge oscillation along the walls ( yellow label ) . in figure [ phasediagram ] we show results for a simulation volume that gives rise to striped oscillations along the major axis ( bottom area of the length of and width of ) . at a wall height of @xmath73 the simulation gives rise to longitudinal striped oscillations along the major axis or transverse pole - to - pole oscillations along the minor axis . this is the same kind of bistability that we already observed in geometry a using a compartment of @xmath54 . after increasing the wall height to @xmath74 the pole - to - pole oscillations along the minor axis disappear and only striped oscillations along the major axis are observed . at the wall height of @xmath75 the oscillation mode changes to a large pole - to - pole oscillation along the minor axis and no edge or striped oscillations were observed . figures [ phasediagram ] and [ phasediagram ] show the kymographs for the membrane - bound proteins corresponding to figures [ phasediagram ] and [ phasediagram ] , respectively . these results demonstrate that in geometry b , both the mode selection and the detailed shape of the polar regions can be controlled by compartment height . for an edge oscillation the free mind protein densities in the bulk of the compartment along one of the bottom area axis ( @xmath30-axis ) and along the walls ( @xmath31-axis ) are shown in figure [ kymograph3d ] . we see that spatial oscillations in the bulk only take place along the @xmath31-axis . on the @xmath30-axis kymograph ( figure [ kymograph3d ] top kymograph ) the change of the density only takes place along the temporal axis . this corresponds to the change of total numbers of free and bound mind proteins during the oscillation and no spatial oscillation takes place along the @xmath30-axis . on the @xmath31-axis kymograph ( figure [ kymograph3d ] bottom kymograph ) we observe a spatial pole - to - pole oscillation . in figure [ kymograph3d ] we show the densities of the free mind proteins in the bulk of the compartment during the large pole - to - pole oscillation for @xmath68 . we see that the bulk proteins oscillate only along one axis ( here the @xmath30-axis ) and no spatial oscillations occur along the walls ( @xmath31-axis ) . from our study of geometry b , we conclude that in non - flat membrane geometries the oscillations of the min - proteins do not only take place along the membrane , but are rather defined by the canalized transfer of the proteins through the bulk . this shows the importance of three - dimensional simulations for non - flat membrane geometries in order to determine the self - organized oscillation modes . ratios of the compartment geometry.,scaledwidth=95.0% ] we have also monitored the amount of membrane - bound mind and mine proteins and compared it to the total amount of proteins in the system . interestingly , we observed that these two quantities have a linear dependence @xmath76 as shown in figure [ ntonb ] . surprisingly , this relation seems to hold for all geometries and dimensions , as indicated in figure [ ntonb ] by the different symbols . since in our simulations we keep density constant , the total amount of proteins scales linearly with the compartment volume ( @xmath77 ) . overall , we conclude the following relation for the mean bound min - protein density @xmath78 where @xmath79 denotes the total reactive membrane area . thus the mean bound min - protein density increases linearly with the volume - to - area ratio , as verified by figure [ ntonb ] . strikingly , we again observe a dramatic difference between mind and mine . although this scaling is the same for both , mine is much closer to being completely bound ( grey lines ) , indicating that once a stable oscillation emerges , mine is almost completely depleted from the bulk . of a spherocylinder as a function of the cell length @xmath80 for various cell widths @xmath81 . * b * a similar relation is observed for an ellipsoidal shape . ] we next assessed the physiological relevance of these observations . in order to investigate how the volume - to - area ratio @xmath82 changes during cell growth , we approximate the shape of an _ e.coli _ bacterium by a spherocylinder of length @xmath80 and width @xmath81 and evaluate @xmath82 analytically . since _ e. coli _ bacteria mainly grow in length and stay constant in width @xcite , we show the evolution of @xmath82 as function of the cell length @xmath80 for various fixed cell widths in figure [ vtoa ] . one clearly sees that @xmath82 remains nearly constant as a function of the cell length @xmath80 . recalling that our previous observation stated that the mean membrane - bound min - protein densities @xmath83 scale linearly with @xmath82 , this would suggest that living bacteria grow in a fashion that keeps both @xmath82 and thus consequently @xmath83 constant , which might be advantageous for the stability and robustness of the min - oscillations and related processes , such as formation of the ftsz - ring . a similar qualitative behavior is observed if we instead of a spherocylinder would assume an ellipsoidal shape as is shown in figure [ vtoa ] . above we have seen that multistability is a recurrent phenomenon in the min - system , both in geometries a and b. we next turn to a systematic investigation of the stochastic switching between two different oscillation modes . an oscillation mode transition of this type occurs frequently in a @xmath84 compartment using geometry b. in this geometry the min - system gives rise to both pole - to - pole and striped oscillations along the same ( major ) axis ( kymograph in figure [ switch2 ] ) . we measure the lifetimes ( oscillation duration before a mode switch occurs ) of the two modes during a long simulation trajectory . the resulting histograms of the lifetimes are shown in figure [ ratoi ] . we can approximate the mode switching as a poisson process by assuming that the switching probability @xmath85 obeys @xmath86 . fitting an exponential to the switching times histograms we obtain the switching rates of these processes . here we neglect the measurements of short lifetimes below @xmath87 since our oscillation mode detection algorithm can miss a mode transition if its lifetime is shorter . the switching rate for the pole - to - pole oscillation is @xmath88 and for the striped oscillation we find @xmath89 , thus the two modes seem to be equally frequent in this case . we have also performed the same analysis with identical compartment geometry for the two other parameter sets ( set b and c ) as presented in table [ tbl : parameterset ] . parameter set b gives rise to stable longitudinal pole - to - pole oscillations along the major axis ( kymograph in figure [ switch2 ] ) in agreement with the results from @xcite , while parameter set c shows a qualitatively similar switching behavior as parameter set a ( data not shown ) with frequent mode transitions . to analyze the effect of the min - protein concentration on the oscillation mode switching , we have performed the same simulation with increased min - protein particle densities using again parameter set a. the resulting fractions of the two oscillation modes during the simulation trajectory are shown in figure [ switchdata ] , and their corresponding switching rates @xmath90 and @xmath91 are given in table [ tbl : rates ] . we note that the fraction of striped oscillations increases monotonously with increasing particle density . in agreement with this , the rate @xmath91 decreases with increasing particle density . in contrast , the rate @xmath90 does not show a systematic change and seems to fluctuate strongly . the shift to the striped oscillations suggests that the min - oscillations can be used not only to sense geometry , but also to sense concentrations . 0.55@cll @xmath92-fold particle number & @xmath93 $ ] & @xmath94 $ ] + @xmath95 & & + @xmath96 & & + @xmath97 & & + @xmath98 & & wu _ et al_. have experimentally measured the fractions of different oscillation modes of min - proteins in rectangular cell geometries of _ e. coli _ bacteria of various sizes with constant height @xcite . in order to address these observations , we now turn to geometry c , which is a rectangular and fully membrane - covered compartment as sketched in figure [ cycle ] . throughout this section we keep the compartment height fixed at @xmath99 , while we vary the length ( major axis ) between and @xmath100 and the width ( minor axis ) of the compartment between and @xmath101 in steps of @xmath102 , respectively . to determine the oscillation mode fractions we calculate an ensemble average of the oscillation modes after simulation time . for each compartment size a sample of 20 independent simulations is used . -axis , * b * the fractions of longitudinal striped oscillations along the @xmath43-axis , and * c * transverse pole - to - pole oscillations along the @xmath30-axis . * d * phase diagram of dominant oscillation modes . when two modes emerged both with a frequency > @xmath103 , we considered both modes as dominant , as indicated by the striped regions in the diagram.,scaledwidth=95.0% ] in the following analysis we only focus on the three main oscillation modes , which here are longitudinal pole - to - pole and striped oscillation ( along the major axis ) and transverse pole - to - pole oscillation ( along the minor axis ) . other oscillation modes as they have been reported in @xcite were also observed using our framework , however , due to their low probability , a much higher sample size would be necessary to analyze their occurrence frequency with sufficient statistics . the results for the three oscillation modes are shown in figure [ fractions ] . in general , each of the three oscillation modes dominates in one region of the phase diagram , but the transitions are fuzzy and therefore bistabilities occur . for compartments with length below @xmath104 and width below @xmath57 , only pole - to - pole oscillations are observed . most of those oscillations occur along the major axis of the system . the transverse pole - to - pole oscillations along the @xmath30-axis emerge most frequently in compartments with quadratic bottom area . increasing the width further increases the fractions of transverse pole - to - pole oscillations at the expense of longitudinal pole - to - pole oscillations . increasing the length for a fixed width shows a sharp transition from longitudinal pole - to - pole to longitudinal striped patterns at around . for compartments with length larger than @xmath104 , the longitudinal pole - to - pole oscillations vanish almost entirely . in the region of both large long sides ( around to in length ) and large short sides ( around to in width ) , the oscillation mode fractions are rather equally shared between longitudinal striped and transverse pole - to - pole oscillations , which is also in line with the bistability that we observed between these two patterns in the @xmath105 compartment of geometry a as shown in section [ sec : geometrya ] . figure [ fractions ] summarizes these findings in a phase diagram that considers only the dominating mode ( expect in the regions of clear bistability ) . overall we find the determined oscillation mode fractions based on parameter set a and as shown in figure [ fractions ] to be in excellent agreement with the experimental results as reported by wu _ using a stochastic particle - based simulation framework , we have investigated the stochastic switching between multistable oscillation modes of the min - system in different three - dimensional compartment geometries . although it is well known that geometrical constraints have a strong impact on the dynamic oscillations of the min - proteins , multistability and mode switching have only recently been investigated in more detail @xcite . our stochastic framework provides a suitable approach to address the question of oscillation mode selection and stability , since it naturally incorporates fluctuations due to finite copy numbers in the system . this allowed us to address the influence of the three - dimensional shape of the compartment and the boundary conditions , with a close match to existing experimental assays ( flat supported bilayers @xcite , functionalized compartments @xcite , and cell sculpting @xcite ) . for example , we addressed the role of compartment height and demonstrated the emergence of new oscillation modes with increasing height , underlining the importance of an explicit three - dimensional representation of the system . we also showed that diffusion long the membrane and cooperative recruitment are essential elements for the emergence of min - oscillations . in general , particle - based stochastic computer simulations are a great tool for explorative research and in the future could be used to explore more details of the different scenarios that have been suggested for the molecular mechanisms shaping the min - oscillations @xcite . our simulations demonstrated several features that might be related to the physiological function of the min - system in _ e. coli_. first we observe that there is a dominating length scale of @xmath106 , which happens to be the natural length of an interphase _ e. coli _ cell . second we found a linear relation between the density of membrane - bound min - proteins and the volume - to - surface ratio , which tends to be constant during growth of _ e. coli_. third we observed that the relative frequency of competing oscillation modes depends on concentration , suggesting that the min - oscillations can be used not only to sense geometry , but also concentration . fourth , we found that stable oscillations strongly deplete mine from the bulk . for the future , it would be interesting to study possible feedback between protein production and min - oscillations . for our simulations , we mainly used the established parameter set a from table [ tbl : parameterset ] and achieved excellent agreement with experimental results regarding the relative frequency of the three main oscillation modes in different cell geometries @xcite . again the critical length scale around plays an important role in transitions between different regimes , which then are the regions of high bistability . however , we also note that different parameter choices lead to different outcomes and that it would be interesting to perform an exhaustive exploration of parameter space to better understand how robustness and multistability depends on kinetic rates , diffusion constants and concentrations . in the future , such a complete scan might become possible by using gpu - code rather than the cpu - code developed here . most importantly , however , our stochastic approach allowed us to measure for the first time the switching rates between different competing oscillation patterns of the min - system . this was done for parameter set a from table [ tbl : parameterset ] . interestingly , parameter set c gave similar results in this respect , while parameter set b results in very stable oscillations without switching , in agreement with the experimental observations for the cell sculpting experiments @xcite . in the future , it would be interesting to investigate more systematically how the effective barriers between two competing oscillation patterns depend on model parameters and compartment geometry . in general , the min - system is an excellent model system to study not only geometry sensing , but also the role of spatiotemporal fluctuations in molecular systems . uss is a member of the interdisciplinary center for scientific computing ( iwr ) and the cellnetworks cluster of excellence at heidelberg university . nds acknowledges support by the bmbf program immunoquant and by the german academic scholarship foundation . the min reaction scheme as introduced in equations ( 1a 1e ) of the main text consists of five molecular reactions , which can be classified into three different types of reactions . * first - order unimolecular reactions of the type @xmath107{\,\ , k_1\,\ , } b$ ] . these kind of reactions have no dependence on the spatial coordinates of the system . in the min reaction scheme the nucleotide exchange reaction ( @xmath12 ) in the bulk and the membrane detachment ( @xmath16 ) are of this type . * simple membrane attachment reactions are also first - order unimolecular reactions of the above type @xmath107{\,\ , k_{1,m}\,\ , } b$ ] , but with the additional constraint that a particle must be in close proximity to a membrane to be able to attach to it . the binding of mind in its atp - bound state to the membrane ( @xmath2 ) is a reaction of this type . * bimolecular membrane attachment reactions . this reaction type is of second - order @xmath108{\,\ , k_{2,m}\,\ , } \text{product(s ) } \end{aligned}\ ] ] and describes the bimolecular association of a particle in the bulk with an already membrane - bound particle . the cooperative mind recruitment ( @xmath13 ) and the mine recruitment ( @xmath6 ) by membrane - bound mind are reactions of this type . to create a consistent particle - based simulation algorithm for the above introduced types of reactions , we compare our reactive brownian dynamics algorithm with the corresponding mean - field partial differential equations , as they are for example studied in @xcite . for this we consider each of the three reaction types individually using only a simplified minimal setting . for the first reaction type the corresponding differential equation is a simple poisson process @xmath110 where @xmath111 denotes the free particle density and @xmath112 is the corresponding first - order reaction rate constant . introducing a concrete system size @xmath113 we can equally convert particle densities into particle numbers @xmath114 and have equivalently @xmath115 with the standard exponential solution @xmath116 in a particle - based simulation framework with a total of @xmath117 particles of a generic species @xmath79 we are interested in the reaction probability @xmath118 , that a particle of type @xmath79 undergoes such a first - order reaction within a fixed , finite time step of length @xmath24 . after a time @xmath119 the solution of the linear degradation equation tells us that @xmath120 particles will have reacted @xmath121 this means that a single particle has a probability of @xmath122 to react during a time step @xmath24 . for this second reaction type we consider a setting with a membrane at the bottom surface at @xmath124 of a rectangular simulation volume @xmath113 to which a generic species @xmath79 can attach . in this scenario the lateral displacements in both the @xmath43- and @xmath30- direction can be integrated out in a straightforward way allowing us to reduce the system to an effective one - dimensional system where @xmath31 is the only spatial dependency that remains . in the resulting one - dimensional domain we have a membrane at @xmath124 and impose a reflecting no - flux boundary condition at @xmath125 . the corresponding differential equation for the particle density @xmath126 now reads @xmath127 subject to @xmath128 . in the particle - based simulation framework we again want to implement this first - order process using a poisson rate @xmath130 to come up with a reaction probability @xmath131 that a particle has attached to the membrane within a fixed , finite time step @xmath24 . since we want only particles that are in close proximity to the membrane to be able to attach to it , we introduce a finite distance @xmath129 to encode this spatial confinement ( see figure [ ib5 ] ) . in this way only particles within a distance @xmath132 to the bottom surface at @xmath124 can attempt this first order attachment reaction . in this picture the actual system boundary at @xmath124 is also treated as a reflective no - flux boundary like the opposite boundary at @xmath125 , since the adsorption step can now happen anywhere in the layer of thickness @xmath129 above the bottom surface . to relate the poisson - rate @xmath130 to the rate constant @xmath35 of the mean - field equation we approximate the @xmath133-functional in equation by a @xmath134-function of width @xmath129 @xmath135 mimicking our scenario in the particle - based framework . using @xmath136 and the approximation from above we obtain @xmath137 integrating once over the full domain we get @xmath138 \ , \mathrm{d}z.\end{aligned}\ ] ] the term on the left hand side of the above expression @xmath139 is the time derivative of the total amount of particles , while the term @xmath140 denotes all particles @xmath141 that are less than @xmath129 away from the membrane at @xmath124 . the term @xmath142 vanishes because of the reflecting boundary conditions ( @xmath143 ) . with this we obtain @xmath144 since in this setting @xmath114 can only change by membrane attachment of the particles @xmath141 , we can identify @xmath145 as the desired poisson rate @xmath146 for the first - order reaction step that we wish to implement . this means that a single particle that is within distance @xmath129 of a reactive surface of the system has a probability of @xmath147 to react during each time step of length @xmath24 . to compare our simulation algorithm with the mean - field equation , we solved the pde numerically using an explicit time - stepping scheme for a finite - difference discretization using the min - system parameter set a as introduced in table 1 of the main text . for this given parameter choice the numerical solutions for the density profiles @xmath126 were approximately homogeneous along @xmath31 , due to fast diffusion in comparison to other competing timescales . therefore we approximated the solution of equation as a spatially homogeneous density @xmath148 allowing us to obtain an analytical solution for this specific parameter set and system size . under this assumption equation simplifies to @xmath149 which after integrating once over the entire domain gives @xmath150 with the standard solution @xmath151 simulation results from the particle - based algorithm are in good agreement with this analytical solution , as depicted in figure [ ib9 ] , where we chose the intrinsic algorithm parameter @xmath152 . this choice is also sufficiently small compared to the full domain size of @xmath153 ( 4% of @xmath153 for @xmath154 ) . the third reaction type we consider here describes the bimolecular association of a freely diffusing bulk particle with already membrane - bound particles . the corresponding mean - field differential equation for this type of reactions reads @xmath156 where @xmath111 denotes the density of free particles of species @xmath79 and @xmath157 is the density of membrane - bound particles . in the particle - based framework we adopt an algorithm used in the software package smoldyn @xcite . the treatment for irreversible bimolecular association reactions follows the spirit of the classical smoluchowski - model for irreversible diffusion - limited reactions @xcite . here two particles instantaneously react upon collision , leading to the diffusion - limited on - rate for bimolecular association of @xmath158 where @xmath25 denotes the mutual diffusion coefficient of the reacting pair and @xmath159 the effective collision distance which in the most naive picture is taken to be the sum of the molecular radii . to also address activation - limited or only partially diffusion - influenced reactions collins and kimball @xcite proposed to impose a radiation - boundary condition which nicely conveys the picture of a finite activation - limited reaction probability once a pair of reactants has met via diffusional encounter . for the sake of simplicity and algorithmic speed , andrews and bray @xcite came up with an algorithm which keeps the spirit of the original idea by smoluchowski by introducing a binding radius @xmath160 and letting particles react instantaneously once they are separated by @xmath161 . to account for activation - limited effects the sum of the molecular radii is in this picture replaced by this effective binding radius @xmath162 . they propose an algorithm which calculates @xmath160 for a given diffusion coefficient @xmath25 , time step @xmath24 and a forward rate constant @xmath163 . in fact the algorithm solves the forward problem of determining the effective simulated forward rate constant @xmath163 for a given time step @xmath24 , diffusion coefficient @xmath25 and binding radius @xmath164 and stores these results in a look - up table , which is then inverted to interpolate a binding radius for a given reaction rate @xmath163 . the forward problem itself is solved numerically by propagating an initial radial distribution function ( rdf ) @xmath165 forward in time , by carrying out alternating diffusion and irreversible reaction steps . the diffusive steps are implemented by convolving the full radial distribution function with a three - dimensional gaussian and in each reaction step the rdf is set to zero in the range @xmath166 to account for the irreversible reactions that have taken place . in each iteration the effective reaction rate is then given by integrating the area under the rdf from @xmath167 $ ] after the diffusive step . this procedure is repeated iteratively until convergence is achieved . by inverting the tabulated relation between the @xmath163 s and the @xmath164 s one solves the inverse problem and can thus obtain @xmath168 as a function of the diffusion coefficient @xmath25 , the time step @xmath24 and the forward rate constant @xmath163 . the radial distribution functions @xmath165 one obtains using this algorithm reduce to the radial distribution function of the smoluchowski model @xmath169 in the limit of small time steps @xmath170 , as to be expected for infinitely detailed brownian motion . for irreversible bimolecular reactions the rdf in the smoluchowski model reads @xmath171 in general however , the rdfs qualitatively resemble the functional form of the radial distribution function according to the collins and kimball @xcite model . for a quantitative test of our algorithm with the corresponding mean - field results we solve equation numerically . in order to do so we make further simplifying assumptions . for arbitrary distributions of the bound particles @xmath172 the differential equation can not be reduced to one dimension as before , however , it is possible to do so for evenly distributed bound particles . we numerically solved this pde using the min - system parameter set a as introduced in table 1 of the main text and assuming evenly distributed bound particles @xmath173 on the bottom surface of a rectangular simulation box of volume @xmath113 . to match this scenario of the mean - field model in the particle - based framework , we consider a molecular species @xmath79 of freely diffusing particles in a simulation volume @xmath113 which can react with membrane - bound particles of another species @xmath174 . for the sake of simplicity we remove particles of species @xmath79 after a successful reaction step with a membrane - bound particle , while the bound - particles are not removed after the reaction step to keep @xmath175 . the results of this comparison using the min - systems parameters and several different values for @xmath157 are shown in figure [ ib10 ] . free particles that are removed upon reaction with evenly distributed bound particles on the bottom surface of a rectangular simulation volume . @xmath114 denotes number of remaining free particles in the system . blue rectangles depict the values from the numerical solution of equation and red triangles represent the results from the particle - based simulation algorithm . the parameters from the main text are used . those are @xmath176 , @xmath154 and @xmath177 . in the four figures different values for the density of the bound particles were used : top left @xmath178 , top right @xmath179 , bottom left @xmath180 and bottom right @xmath181.,title="fig : " ] free particles that are removed upon reaction with evenly distributed bound particles on the bottom surface of a rectangular simulation volume . @xmath114 denotes number of remaining free particles in the system . blue rectangles depict the values from the numerical solution of equation and red triangles represent the results from the particle - based simulation algorithm . the parameters from the main text are used . those are @xmath176 , @xmath154 and @xmath177 . in the four figures different values for the density of the bound particles were used : top left @xmath178 , top right @xmath179 , bottom left @xmath180 and bottom right @xmath181.,title="fig : " ] + free particles that are removed upon reaction with evenly distributed bound particles on the bottom surface of a rectangular simulation volume . @xmath114 denotes number of remaining free particles in the system . blue rectangles depict the values from the numerical solution of equation and red triangles represent the results from the particle - based simulation algorithm . the parameters from the main text are used . those are @xmath176 , @xmath154 and @xmath177 . in the four figures different values for the density of the bound particles were used : top left @xmath178 , top right @xmath179 , bottom left @xmath180 and bottom right @xmath181.,title="fig : " ] free particles that are removed upon reaction with evenly distributed bound particles on the bottom surface of a rectangular simulation volume . @xmath114 denotes number of remaining free particles in the system . blue rectangles depict the values from the numerical solution of equation and red triangles represent the results from the particle - based simulation algorithm . the parameters from the main text are used . those are @xmath176 , @xmath154 and @xmath177 . in the four figures different values for the density of the bound particles were used : top left @xmath178 , top right @xmath179 , bottom left @xmath180 and bottom right @xmath181.,title="fig : " ] 59ifxundefined [ 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 10.1016/0092 - 8674(89)90586 - 2 [ * * , ( ) ] link:\doibase 10.1046/j.1365 - 2958.1999.01575.x [ * * , ( ) ] link:\doibase 10.1073/pnas.96.9.4971 [ * * , ( ) ] http://www.ncbi.nlm.nih.gov/pubmed/10515933 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1016/s1369 - 5274(01)00262 - 4 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1038/nrmicro2671 [ * * , ( ) ] link:\doibase 10.1371/journal.pgen.1004504 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1128/jb.185.3.735 - 749.2003 [ * * , ( ) ] link:\doibase 10.1073/pnas.232590599 [ * * , ( ) ] link:\doibase 10.1074/jbc.m306876200 [ * * , ( ) ] link:\doibase 10.1074/jbc.m112.407817 [ * * , ( ) ] link:\doibase 10.1146/annurev.biochem.75.103004.142652 [ * * , ( ) ] link:\doibase 10.1038/nrmicro2612 [ * * , ( ) ] link:\doibase 10.1146/annurev - biophys-042910 - 155332 [ * * , ( ) ] link:\doibase 10.1111/mmi.12017 [ * * , ( ) ] , link:\doibase 10.3390/life4040915 [ * * , ( ) ] link:\doibase 10.1016/j.ceb.2016.02.005 [ * * , ( ) ] link:\doibase 10.1038/msb.2010.111 [ * * , ( ) ] link:\doibase 10.1093/emboj/21.8.1998 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1038/nsmb.2037 [ * * , ( ) ] link:\doibase 10.1073/pnas.1206953109 [ * * , ( ) ] link:\doibase 10.1002/anie.201207078 [ * * , ( ) ] link:\doibase 10.7554/elife.03949 [ * * , ( ) ] link:\doibase 10.1073/pnas.1120854109 [ * * , ( ) ] link:\doibase 10.1038/nnano.2015.126 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1007/bf00289234 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.65.851 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1073/pnas.251216598 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.278102 [ * * , ( ) ] , link:\doibase 10.1016/s0006 - 3495(02)75426-x [ * * , ( ) ] link:\doibase 10.1088/1478 - 3975/2/2/002 [ * * , ( ) ] , link:\doibase 10.1073/pnas.2135445100 [ * * , ( ) ] link:\doibase 10.1083/jcb.200411122 [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2958.2007.05607.x [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/17/4/043023 [ * * , ( ) ] link:\doibase 10.1073/pnas.0505825102 [ * * , ( ) ] link:\doibase 10.1371/journal.pcbi.0020080 [ * * , ( ) ] link:\doibase 10.1088/1478 - 3975/3/1/001 [ * * , ( ) ] , link:\doibase 10.1103/physreve.73.021904 [ * * , ( ) ] , @noop * * , ( ) link:\doibase 10.1073/pnas.0502037102 [ * * , ( ) ] link:\doibase 10.1016/j.celrep.2012.04.005 [ * * , ( ) ] link:\doibase 10.1371/journal.pcbi.1003347 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.128102 [ * * , ( ) ] , link:\doibase 10.1039/c3sm52251b [ * * , ( ) ] https://books.google.de/books?id=zm1raaaamaaj[__ ] ( , ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1016/0301 - 0104(78)87025 - 6 [ * * , ( ) ] link:\doibase 10.1063/1.458533 [ * * , ( ) ] link:\doibase 10.1088/1478 - 3967/1/3/001 [ * * , ( ) ] link:\doibase 10.1371/journal.pone.0023126 [ * * , ( ) ]

the work is motivated by a result of manin in , which relates the arakelov green function on a compact riemann surface to configurations of geodesics in a 3-dimensional hyperbolic handlebody with schottky uniformization , having the riemann surface as conformal boundary at infinity . a natural question is to what extent the result of manin can be generalized to cases where , instead of dealing with a single riemann surface , one has several riemann surfaces whose union is the boundary of a hyperbolic 3-manifold , uniformized no longer by a schottky group , but by a fuchsian , quasi - fuchsian , or more general kleinian group . we have considered this question in this work and obtained several partial results that contribute towards constructing an analog of manin s result in this more general context .

0912.1731

lets consider that @xmath0 is a compact riemann surface . then we can define a green s function with respect to a metric and a divisor on @xmath0 . suppose that @xmath1 is a hyperbolic 3 - manifold with infinite volume and having compact riemann surfaces @xmath2 as its conformal boundaries at infinity . also choose a geometry on @xmath1 which bears a metric of constant negative curvature . the main new results of this paper express the green s functions on each @xmath3 in terms of the length of some certain geodesics in @xmath1 . manin has done this provided that @xmath1 has one boundary component ( i.e. n=1 ) and is uniformized by a schottky group in @xcite . in this work we will generalize manin s results in general case , @xmath1 has more than one boundary components and is uniformized by a fuchsian , quasi - fuchsian or kleinian group . + in this introduction we state the main definitions , motivation and the plan of doing the work . we start by introducing the definition of a green s function on a compact riemann surface for a divisor and with respect to a normalized volume form on it . let @xmath0 be a compact riemann surface and @xmath4 ( with @xmath5 integer number ) be a divisor on @xmath0 . we show the support of @xmath6 by @xmath7 . also lets consider that @xmath8 is a positive real - analytic 2 - form on @xmath0 . by a green s function on @xmath0 for @xmath6 with respect to @xmath8 we mean a real analytic function @xmath9 satisfying the following conditions : + ( i ) _ laplace equation : _ @xmath10 where @xmath11 , and @xmath12 is the standard @xmath13 - current @xmath14 . + ( ii ) _ singularities : _ let @xmath15 be a complex coordinate in a neighborhood of the point @xmath16 . then @xmath17 is locally real analytic . + a function satisfying these two conditions is uniquely determined up to an additive constant . and the third condition is \(iii ) _ normalization : _ @xmath18 which eliminates the remaining ambiguous constant . @xmath19 is additive on @xmath6 and for @xmath20 , @xmath21 is symmetric , i.e. @xmath22 . + lets consider that @xmath23 is another divisor on @xmath0 that is prime to the divisor @xmath6 . that means , @xmath24 . and put @xmath25 @xmath19 is additive and @xmath21 is symmetric , then @xmath26 is symmetric and biadditive in @xmath27 . + in general , the function @xmath28 depends on the metric @xmath8 , but in the case that both of the divisors are of the degree zero , from the condition ( i ) we see that @xmath29 depends only on @xmath30 . notice that , as a particular case of the general kahler formalism , to choose @xmath8 is the same as to choose a real analytic riemannian metric on @xmath0 compatible with the complex structure . this means that @xmath31 are conformal invariants when both divisors are of degree zero . also in the case that the divisor @xmath6 is principal , i.e. @xmath6 is the divisor of a meromorphic function like @xmath32 , then @xmath33 is a 1 - chain with boundary @xmath34 . the divisors of degree zero on the riemann sphere @xmath35 are principal . then this formula can be directly applied to divisors of degree zero on riemann sphere . + it is well known that the green s function of the degree zero divisors on a riemann surface of arbitrary genus can be expressed exactly via the differential of the third kind @xmath36 with pure imaginary periods and residues @xmath5 at @xmath16 when the divisor is @xmath37(see , @xcite , @xcite ) . then the generalization of the previous formula for arbitrary divisors @xmath27 of degree zero is @xmath38 in general , when the degree of the divisors are not restricted to zero , the basic green s function @xmath39 can be expressed explicitly via theta functions ( as in @xcite ) in the case when @xmath40 is the _ arakelov metric _ constructed with the help of an orthonormal basis of the differentials of the first kind on the riemann surface . the result of manin in @xcite which relates the arakelov green function on a compact riemann surface to configurations of geodesics in a 3-dimensional hyperbolic handlebody with schottky uniformization , having the riemann surface as conformal boundary at infinity , was extremely innovative and influential and had a wide range of consequences in the arithmetic context of arakelov geometry as well as and in other contexts , ranging from p - adic geometry , real hyperbolic 3-manifolds , the holography principle and ads / cft correspondence in string theory , and noncommutative geometry . a natural question is to what extent the result of manin can be generalized to cases where , instead of dealing with a single riemann surface , one has several riemann surfaces whose union is the boundary of a hyperbolic 3-manifold , uniformized no longer by a schottky group , but by a fuchsian , quasi - fuchsian , or more general kleinian group . such a generalization is not only interesting because it is a very natural question to pass from kleinian - schottky groups to more general kleinian groups , but also for its potential applications to arakelov geometry , to the case of curves defined over number fields with several archimedean places , while manin s result was formulated for the case of arithmetic curves defined over the rationales . we have focused on the formula in manin s work , that expresses the arakelov green s function on a compact riemann surface in terms of a basis of holomorphic differentials of the first kind and of differentials of the third kind . in the case of schottky uniformization , when the limit set has hausdorff dimension strictly smaller than one , one can construct such differentials in terms of averages over the schottky group . while the same type of formula no longer holds in the fuchsian or quasi - fuchsian case , we use the canonical covering map relating fuchsian and schottky uniformization and the coding of limit sets for the fuchsian and schottky case , to express the green function in the fuchsian or quasi - fuchsian case in terms of the one in the schottky case . the approach we follow for the more general kleinian case is via a decomposition of the uniformizing group as a free product of quasi - fuchsian and schottky groups and applying the results we obtained for these cases individually . the paper consists of three chapters devoted to the cases that the hyperbolic 3-manifold @xmath1 have 1 , 2 and @xmath41 many boundary components at infinity respectively . first chapter pays to the one boundary component case and includes four subchapter devoted to the foundations and the genera 0,1 and @xmath42 respectively . also this chapter contains fundamental definitions and basic computations that are bases for the computations in the next chapters . the results of this chapter are a summary of the reference @xcite . all hyperbolic 3-manifolds with two boundary components with the same genera are uniformized by quasi - fuchsian groups ; the second chapter is devoted to this manifolds . in third chapter we consider the general case i.e. the case that @xmath1 is uniformized by some kleinian groups and have @xmath41 boundary components . finally , as a remark for the chapter three we show that for manifolds with @xmath43 many boundary components at infinity the situation is like the previous . i am grateful to matilde marcolli who suggested this problem , supported me in max - planck and hausdorff institute in bonn , and supervised me in all process . i also many thank saad varsaie for several useful and encouraging comments . in this chapter , first we gather some fundamental definitions and some useful notations that will be used in all of the paper . next we bring the computation of green s function for riemann sphere @xmath35 , the boundary at infinity of the hyperbolic space . all 3-manifolds with one boundary component that is a compact riemann surface with genus @xmath44 , are uniformized by schottky groups . the rest of the chapter is devoted to these manifolds . the results of this chapter are a summary of the reference @xcite . consider the hyperbolic space @xmath45 with the upper half space model and coordinate @xmath46 that comes from @xmath47 equipped by the hyperbolic distance function corresponding to the metric @xmath48 of constant curvature -1 . the geodesics in this model are vertical half - lines @xmath49 and also vertical half - circles orthogonal to the plane at infinity @xmath50 ( i.e. for @xmath51 ) . + if we consider the end points of the geodesics ( including @xmath43 for the ends of the vertical half - lines ) , then we can consider the riemann sphere @xmath52 ( or @xmath53 in the unit boll model ) as the boundary at infinity of @xmath45 . lets show the geodesic joining @xmath54 to @xmath55 in @xmath56 by @xmath57 ; by @xmath58 the point on the geodesic @xmath59 closest to the point @xmath54 ( i.e , the intersection point of @xmath59 and a geodesic passing through @xmath54 and orthogonal to @xmath59 ) ; by @xmath60 the distance from @xmath61 to the geodesic @xmath57 and by @xmath62 the oriented distance between two points lying on an oriented geodesic in @xmath45 ( figure [ fig1.10 ] ) . also lets show by @xmath63 the angle ( at @xmath61 ) between the semi - geodesics joining @xmath61 to @xmath54 and @xmath61 to @xmath55 , and for @xmath64 in @xmath65 by @xmath66 the oriented angle between the semi - geodesics joining @xmath58 to @xmath54 and @xmath67 to @xmath55 . in order to calculate @xmath66 we must first make the parallel translation along @xmath59 identifying the normal spaces to @xmath59 at @xmath58 and at @xmath67 . the orientation of a normal space is defined by projecting it along oriented @xmath59 to its initial end into the tangent space to @xmath65 , which is canonically oriented by the complex structure . in this case we consider that the hyperbolic manifold @xmath1 is the hyperbolic space @xmath45 with riemann - sphere as its boundary at infinity . for this case we have [ p1.5 ] for @xmath68 and @xmath69 in @xmath35 , denote by @xmath70 a meromorphic function on @xmath35 with the divisor @xmath71 . then we have @xmath72 and @xmath73 mobius transformations preserve hyperbolic distance and angel . then both sides of ( [ f1.3 ] ) and ( [ f1.4 ] ) are invariant under these transformations . hence it suffices to consider the case when @xmath74 in @xmath35 . then the geodesic @xmath75 is the @xmath76 semi - axis in @xmath77 coordinate for @xmath45 and the geodesics joining the points @xmath78 and @xmath79 normally to this semi - axis are half circles passing from these points with center in @xmath80 and normal to @xmath50 ( figure [ fig1.11 ] ) . then we have @xmath81 and @xmath82 also from the properties of cross - ratio we have @xmath83 this gives ( [ f1.3 ] ) directly . to see ( [ f1.4 ] ) , we know that angles in @xmath45 can be calculated using the euclidean metric @xmath84 and also the parallel transport along @xmath76 semi - axis coincides with the euclidean one . as it s shown in figure [ fig1.11 ] the vectors @xmath85 and @xmath86 tangent to the geodesics from the points @xmath15 and @xmath87 , are normal to @xmath88 . hence if we transport ( parallel ) the vector @xmath86 in the point @xmath89 to the point @xmath90 then both vectors are in an euclidean plane normal to @xmath76 semi - axis . then the angel from the vector @xmath85 to the vector @xmath86 can be calculated via there angels from @xmath91 axis . then we have @xmath92 this proves ( [ f1.4 ] ) . the part @xmath93 in the proposition above is the classical cross - ratio of four points on the riemann sphere @xmath35 , for which it is convenient to have a special notation @xmath94 let @xmath68 and @xmath69 be in @xmath35 . then we have + @xmath95 + from the formula ( [ f0.11 ] ) for the green s function in the case that the divisors are principal , we have + @xmath96 + then from proposition ( [ p1.5 ] ) and notation ( [ f1.5 ] ) we have ( [ f1.6 ] ) . in this case the group that uniformize the hyperbolic 3 - manifold with its boundary at infinity is a cyclic group , and there is a nice explicit formula for the basic green s function on the boundary at infinity of 3-manifold ; but because we do not use it here we do not want to point to it here . but one can find it in @xcite ( or for a physical point of view in @xcite ) . of course the process in the next part can be used for this case too . we have pointed some notes about this case in section [ s1.5 ] . consider the complex projective linear transformations group @xmath98 . _ ( i ) kleinian groups . _ a subgroup @xmath99 of @xmath98 is called a kleinian group if @xmath99 acts on @xmath45 properly discontinuously . for a kleinian group @xmath99 and a point @xmath100 in @xmath45 lets denote by @xmath101 , the orbit of the point @xmath102 under the action of @xmath99 . since @xmath99 acts on @xmath45 properly discontinuously , @xmath101 has accumulation points only on @xmath35 . they are independent of the choice of the reference point @xmath100 and are called the _ limit set _ of @xmath99 , which is denoted by @xmath103 . equivalently , it is the closure of the set of all fixed points of the elements of @xmath99 other than the identity . @xmath103 is the minimal non - empty closed @xmath99 - invariant set . the complement of the limit set @xmath104 is denoted by @xmath105 and is called the _ region of discontinuity _ of @xmath99 . + the kleinian group @xmath99 is called of the _ first kind _ if @xmath106 and of the _ second kind _ if @xmath107 . in the case that @xmath99 is of the second kind @xmath99 acts on @xmath105 properly discontinuously . therefor in this case @xmath105 is the maximal open @xmath99 - invariant subset of @xmath35 where @xmath99 acts properly discontinuously . the quotient space @xmath108 has the complex structure induced from that of @xmath105 . thus @xmath108 is a countable union of riemann surfaces lying at infinity of the complete hyperbolic 3 - manifold @xmath109 . for a torsion free kleinian group @xmath99 , a manifold @xmath110 possibly with boundary is denoted by @xmath111 and is called a _ kleinian manifold_. the interior @xmath109 of @xmath111 which admits the hyperbolic structure is denoted by @xmath112 . + _ ( ii ) loxodromic , parabolic and elliptic elements . _ an element @xmath113 in @xmath98 is called loxodromic if it has exactly two different fixed points in @xmath35 . these points are denoted by @xmath114 and @xmath115 and are called attracting one and repelling one . for any @xmath116 and @xmath117 , we have @xmath118 . if we denote by @xmath119 the eigenvalue of @xmath113 on the complex tangent space to @xmath114 then there is a local coordinate for @xmath35 that in this coordinate @xmath113 is represented by @xmath120 . for this reason @xmath119 is called the multiplier of @xmath113 , and we have @xmath121 , @xmath122 . also by definition @xmath113 is called a parabolic element if it has precisely one fixed point in @xmath35 , and elliptic if it has fixed points in @xmath45 . in fact an elliptic element fixes all points of a geodesic in @xmath45 joining two fixed points in @xmath35 . for a loxodromic or elliptic element @xmath113 the geodesic joining two fixed points in @xmath35 is called axis of @xmath113 . an element is elliptic if and only if it has finite order , then elliptic elements cause a singularity in @xmath112 . for this reason we consider kleinian groups without elliptic elements . this means that the group is torsion free or equivalently acts freely on @xmath45 . + _ ( iii ) schottky groups_. a finitely generated , free and purely loxodromic kleinian group is called a schottky group . purely loxodromic means , all elements except the identity are loxodromic . for such a group @xmath123 the number of a minimal set of generators is called the genus of @xmath123 . a marking for a schottky group @xmath123 of genus @xmath100 by definition is a family of @xmath124 open connected domains @xmath125 in @xmath35 and a family of generators @xmath126 with the following properties . for @xmath127 + a ) the boundary @xmath128 of @xmath129 is a jordan curve homeomorphic to @xmath130 and closures of @xmath129 are pairwise disjoint . \b ) @xmath131 and @xmath132 . + we say that a marking is classical , if all @xmath128 are circles . it is known that every schottky group admits a marking . in fact each schottky group admit infinitely many marking but there are some schottky groups for which no classical marking exists . + _ ( iv ) @xmath123 - invariant sets of the schottky groups and there quotient spaces_. a schottky group @xmath123 is a kleinian group , then it acts on @xmath45 properly discontinuously . this action is free too . then the quotient space @xmath133 has a complete hyperbolic 3-manifold structure this means that it is a non - compact riemann space of constant curvature -1 . topologically , if @xmath123 is of genus @xmath100 then @xmath133 is the interior of a handlebody of genus @xmath100 . + now lets consider the marking @xmath134 for the schottky group @xmath123 of genus @xmath100 and put @xmath135 the set @xmath136 is the region of discontinuity of @xmath123 and @xmath137 is a fundamental domain for the action of @xmath123 on @xmath35 . then @xmath123 acts on @xmath136 freely and properly discontinuously . so the quotient space @xmath138 is a complex riemann surface of genus @xmath100 . all compact riemann surfaces can be obtained in this way ( see @xcite ) and every compact riemann surface admits infinitely many different schottky covers . + the cayley graph of a schottky group @xmath123 of genus @xmath100 is an infinite tree with multiplicity of @xmath124 at each vertex . @xmath123 is free , then this tree is without loops and each path between two points is unique . now , by the definition of a marking for @xmath123 , each generator @xmath139 takes @xmath140 to the inside of @xmath141 and again the generator @xmath142 takes this part to the inside of the second copy of @xmath143 inside @xmath141 . this means that for each element @xmath113 in @xmath123 we can associate a path in the cayley graph that the points of @xmath140 moves along that path on a tubular neighborhood . this express the action of @xmath123 on the set @xmath136 ( figure [ fig1.12 ] illustrating this for the case @xmath144 ) . we can consider the riemann surface @xmath145 as the boundary at infinity of the manifold @xmath146 by identifying the points of @xmath145 with the points of the set of equivalence classes of unbounded ends of oriented geodesics in @xmath146 modulo the relation `` distance=0 '' . + by the definition of the limit set for a kleinian group , the complement @xmath147 is the limit set of @xmath123 . for the cyclic schottky groups ( of genus 1 ) the limit set @xmath148 consists of two points which can be chosen as @xmath149 , but for genus@xmath150 this set is an infinite cantor set ( see @xcite ) . + lets consider the irreducible left - infinite words like @xmath151 . where @xmath152 , @xmath153 , and @xmath154 is a point in the fundamental domain @xmath137 . and put @xmath155 this is a well defined point of @xmath148 and is independent of the point @xmath154 . since @xmath123 is a free group , the map @xmath156 establishes a bijection as following irreducible left - infinite words in @xmath139 + @xmath157 + ends of the cayley graph of @xmath123 , @xmath158 + @xmath157 + points of @xmath159 + denote by @xmath160 the hausdorff dimension of the set @xmath148 . it can be characterized as the convergence abscissa of any poincare series @xmath161 where @xmath15 is any coordinate function on @xmath35 with a zero and a pole in @xmath136 . for @xmath162 that re@xmath163 , this series converges uniformly on compact subsets of @xmath136 . generally we have @xmath164 ( see @xcite ) . in the next section we will consider @xmath165 to have the convergency of some infinite products defining some @xmath123 - automorphic functions . this class of schottky groups was characterized by bowen @xcite in following way : @xmath165 if and only if @xmath123 admits a rectifiable invariant quasi - circle ( which then contains @xmath148 ) . + choose marking for @xmath123 of genus @xmath100 and denote by @xmath166 the image of @xmath128 in @xmath167 with induced orientation . also for @xmath127 choose the points @xmath168 in @xmath128 and denote by @xmath169 the images in @xmath167 of oriented pathes from @xmath168 to @xmath170 lying in @xmath137 . these images are obviously closed paths and we can choose them in such a way that they do nt intersect . if we denote the classes of these pathes in 1- homology group @xmath171 by the same notations , then the set @xmath172 form a canonical basis of this group i.e. for all @xmath127 we have @xmath173 moreover the kernel of the map @xmath174 which is induced by the inclusion @xmath175 is generated by the classes @xmath166 . lets consider the cyclic schottky group @xmath176 , for @xmath177 , @xmath178 . in this case @xmath179 and consequently @xmath180 . then a differential of the first kind on @xmath180 can be written as @xmath181 where @xmath154 is any point @xmath182 . and for @xmath183 , this differential determines a differential of the first kind on @xmath167 . in general case for a schottky group @xmath123 of genus @xmath100 we can make a differential of the first kind @xmath184 for any @xmath185 on @xmath136 and @xmath167 by an appropriate averaging of this formula . lets consider a marking for @xmath123 and denote by @xmath186 a set of representatives of @xmath187 ; by @xmath188 a similar set for @xmath189 ; and by @xmath190 the conjugacy class of @xmath113 in @xmath123 . then for any @xmath191 we have \(a ) if @xmath192 , the following series converges absolutely for @xmath193 and determines ( the lift to @xmath145 of ) a differential of the first kind on @xmath145 : @xmath194 this differential does not depend on @xmath154 , and depends on @xmath113 additively . also if the class of @xmath113 is primitive ( i.e. non - divisible in h:@xmath195 $ ] ) , @xmath184 can be rewritten as following @xmath196 ( b ) if @xmath139 form a part of the marking of @xmath123 , and @xmath166 are the homology classes described before , we have @xmath197 it follows that the map @xmath113 mod @xmath198\mapsto \omega_g$ ] embeds @xmath199 as a sublattice in the space of all differentials of the first kind . + ( c ) denote by @xmath200 the complementary set of homology classes in @xmath201 as in before . then we have for @xmath202 , with an appropriate choice of logarithm branches : @xmath203 and @xmath204 where @xmath205 is @xmath206 without the identity class . for the proofs , see @xcite and @xcite . notice that our notation here slightly differs from @xcite ; in particular , @xmath207 here corresponds to @xmath208 of @xcite . lets consider the points @xmath54 and @xmath55 in @xmath140 the fundamental domain of @xmath123 and put @xmath209 . then assuming @xmath165 , we see that this series absolutely converges and because it s @xmath123 - automorphic it gives us a differential of the third kind on @xmath145 with residues @xmath210 at the images of @xmath64 in @xmath145 . moreover , since both points @xmath64 are out of the circles @xmath128 , its @xmath166 - periods vanish . now , if we consider the linear combination @xmath211 with real coefficients @xmath212 , then it will have pure imaginary @xmath166 - periods and if we find real coefficients @xmath212 so that the real part of @xmath169 - periods of the form @xmath213 vanish , we will be able to use this differential in order to calculate conformally invariant green s functions . the set of the equations for calculating the coefficients @xmath214 are as following @xmath215 for @xmath127 . the parts @xmath216 are calculated by means of formula ( [ f1.10 ] ) and ( [ f1.11 ] ) , and @xmath169 - periods of @xmath217 are given in @xcite . + now if we denote the points @xmath218 , and @xmath69 in @xmath145 and their images in the fundamental domain @xmath137 by the same notations , then we have @xmath219 and finally we have the green s function on @xmath167 as following @xmath220 all hyperbolic riemann surfaces are uniformized by fuchsian groups . also fuchsian groups act on hyperbolic space by poincare extension and uniformize some hyperbolic 3-manifolds with two boundaries at infinity with the same genera . but they do not uniformize all such manifolds ; in fact they are uniformized by an extension of fuchsian groups that are called quasi - fuchsian groups . in this chapter first we intend to extend the method in previous chapter to fuchsian groups and then for quasi - fuchsian groups . + consider hyperbolic plane @xmath221 with the upper half plane model and coordinate @xmath222 with @xmath223 , equipped by the hyperbolic distance function corresponding to the metric @xmath224 of the constant curvature -1 . the geodesics in this model are vertical half - lines @xmath225 and vertical half - circles orthogonal to the line at infinity @xmath226 ( i.e , for @xmath51 ) . + if we consider the end points of geodesics ( including @xmath43 for the end of vertical half - lines other than the end in @xmath226 ) then we can consider circle @xmath227 ( or @xmath228 in unit disc model ) as boundary at infinity of @xmath221 . each subgroup @xmath229 of @xmath230 , the general projective linear group over @xmath226 , that acts freely and properly discontinuously on @xmath221 is called a fuchsian group . similar to the kleinian groups the limit set @xmath231 and the region of discontinuity @xmath232 are defined but in @xmath233 , the boundary at infinity of @xmath221 . and the properties are similar to them . also similar to the loxodromic elements , a hyperbolic element is an element @xmath234 with two fixed points @xmath235 in @xmath233 . for a fuchsian group @xmath229 the quotient space @xmath236 is a hyperbolic riemann surface possibly with boundary @xmath237 ( if @xmath238 ) . + let @xmath229 be a fuchsian group such that @xmath239 is a compact riemann surface with the genus @xmath240 . then @xmath229 is a purely hyperbolic group of finite order @xmath124 ( for example see @xcite ) . lets denote by @xmath241 the generators of @xmath229 and by @xmath242 the fixed point set of @xmath241 . also consider that @xmath102 is the fundamental polygon of @xmath229 with @xmath243 as it s sides , such that @xmath244 and @xmath245 for @xmath246 . we represent @xmath102 as following : @xmath247 where @xmath248 and @xmath249 are the intersection points of @xmath166 and @xmath169 , @xmath250 and @xmath251 and so on . now we have @xmath252=i \rangle\ ] ] and also ( see for example @xcite ) . @xmath253 if we denote by @xmath254 and @xmath255 the images of @xmath166 and @xmath169 in @xmath0 , then @xmath256 generate @xmath257 . @xmath229 acts similarly , freely and properly discontinuously on lower half plane @xmath258 of @xmath50 too , and @xmath258 is @xmath229-invariant . lets denote by @xmath229 the poincare extension of @xmath229 on @xmath45 too ( see @xcite or @xcite ) . by this extension @xmath229 can be considered as a kleinian group . and we have @xmath259 ( see @xcite ) . and @xmath260 has a hyperbolic structure , @xcite . we know that @xmath231 the limit point set of @xmath229 is @xmath261 ( @xmath130 in unit disc model ) , @xcite . then as a kleinian group @xmath262 and the region of discontinuity of @xmath229 considering as a kleinian group is @xmath263 and also the kleinian manifold is @xmath264.\end{aligned}\ ] ] now because @xmath0 is compact then @xmath265 . consequently @xmath260 is a hyperbolic 3 - manifold with two compact boundary component at infinity @xmath266 and @xmath267 with the same genus @xmath100 . for convenience and having a simple intuition lets consider the unit disc model for @xmath221 in this section . we can code points of @xmath268 and geodesics with beginning and end points on @xmath268 as following : + lets mark semicircles including sides of p by @xmath269 in the counter clockwise direction around @xmath130 and put @xmath270 , and so on . in general for @xmath271 , @xmath272 . and label end points of @xmath273 on @xmath130 by @xmath274 and @xmath275 ( with @xmath276 ) with @xmath274 occurring before @xmath275 in the counter clockwise direction , ( see figure [ fig2.13 ] ) . and define @xmath277 @xmath278 then @xmath279 is a well defined map and is called a markov map related to the fuchsian group @xmath229 . we have the following lemma the map @xmath279 and the group @xmath229 are orbit equivalent on @xmath130 , namely except for the pairs @xmath280 , for @xmath281 ; for each x , y in @xmath130 , @xmath282 for some @xmath283 in @xmath229 if and only if there exists nonnegative integers @xmath284 such that @xmath285 . ( see @xcite ) . now label each arc @xmath286 by @xmath139 , and for each element @xmath16 in @xmath130 put @xmath287 . where the component @xmath288 is the label of the segment to which @xmath289 belongs . we know that for each generator @xmath241 of @xmath229 , the fixed point @xmath290 is in the circle @xmath291 and @xmath292 is in @xmath293 then from the definition of markov map we have @xmath294 next , if @xmath295 be a geodesic with the beginning point @xmath296 and end point @xmath297 then put @xmath298 now for each @xmath299 let @xmath300 show the geodesic arc between @xmath301 and @xmath302 ( the axis of @xmath303 ) and @xmath304 the image of @xmath305 in @xmath260 . then we have the following geodesics in @xmath260 * closed geodesics : a geodesic in @xmath260 is closed if and only if it s the projection of the axis of a hyperbolic element in @xmath229 , * images of the geodesics with beginning and end points in @xmath231 , * geodesics with beginning points in @xmath306 that are limit cycle to @xmath307 , for a generator @xmath241 of @xmath229 , * geodesics with beginning and end points in @xmath308 or @xmath309 . , * geodesics with beginning point in @xmath308 and end point in @xmath309 and vis versa . for the riemann surface @xmath0 and associated fuchsian group @xmath229 as above let @xmath310 be the smallest normal subgroup of @xmath229 including @xmath241 for @xmath311 . then it s obvious that the factor group @xmath312 is a free group generated by @xmath100 generators . if we consider the covering @xmath313 associated to @xmath310 , then from normality of @xmath310 the group of deck transformations of this cover is @xmath312 and according to the classical koebe uniformization theorem @xcite ( see also @xcite)there is a planar region @xmath314 that is a region of discontinuity of a schottky group @xmath315 and @xmath316 . @xmath317 is covering isomorphic to @xmath318 and for coverings @xmath319 , @xmath320 and @xmath321 we have @xmath322 i.e. the following diagram is commutative @xmath323 + @xmath324 + @xmath0 also we have @xmath325 for @xmath127 and @xmath326 for each @xmath283 in @xmath310 and each mappings is complex - analytic covering and @xmath123 is uniquely determined to within conjugation in @xmath327 @xcite . + for an extension of @xmath328 to a set some more than @xmath221 that we need it in the later computations , lets denote by @xmath329 the free subgroup of @xmath229 generated by @xmath330 . we know that @xmath331 is a subset of @xmath231 . @xmath331 is @xmath329 invariant and is out of the closer of the fundamental domain of @xmath329 , then all points of @xmath331 are in only the circles @xmath128 that are related to the generators of @xmath329 and their inverses . this shows that the coding of each point in @xmath331 includes only the generators of @xmath329 and their inverses . also since each generator of @xmath229 like @xmath283 is an isometry of @xmath221 , it s a composition of two reflections , then @xmath332 is not in the isometry circle of @xmath333 for each point @xmath16 in @xmath130 . then by the definition of the markov map @xmath279 all the codings of the points of @xmath130 are irreducible . this means that each point @xmath16 in @xmath331 can be coded by the irreducible formal combination @xmath334 for @xmath335 . this fuchsian coding is suitable for expressing geodesics in @xmath260 or @xmath336 when the beginning and end points of the geodesics are in @xmath331 and also can be used for extending the map @xmath328 on @xmath331 and maybe on some more . + now we want to extend @xmath328 to @xmath337 ( onto @xmath338 ) . for this , we know that each element of @xmath148 can be coded by the infinite words @xmath339 for a constant @xmath154 in @xmath136 and @xmath340 . similarly , since @xmath329 is free and purely loxodromic group then it s a schottky group too and we can use schottky coding for it too . for conveniences in some proofs we will use schottky codings for the points of @xmath331 . then each point @xmath16 in @xmath331 can be coded by the infinite words @xmath341 for @xmath335 and a constant point @xmath342 in @xmath343 ( also @xmath344 ) . then we can consider @xmath342 in @xmath345 such that @xmath346 . also from ( [ f2.10 ] ) and using the properties of a loxodromic element and invariance of the limit set under @xmath229 ( or @xmath329 ) it is not hard to show that on @xmath331 we have the following relation between fuchsian and schottky coding @xmath347 and also this is true for repelling points too because of the relation @xmath348 for each @xmath283 in @xmath229 . now , lets put @xmath349 in other words , by the definition of infinite words , the continuity of @xmath328 and the relation @xmath350 on @xmath221 @xmath351 then the map @xmath352 is well defined and onto . also from ( [ f2.9 ] ) we see that we can explain @xmath328 and consequently the functions ( spatially green s function ) on @xmath308 by the fuchsian coding , when these functions are expressed via this map . + for each point @xmath16 in @xmath331 and element @xmath283 in @xmath329 equivalent to @xmath59 ( i.e. @xmath353 , where @xmath335 and @xmath241 is replaced by @xmath354 and vice versa ) we have @xmath355 then @xmath356 and @xmath357 . let @xmath329 be the free group generated by @xmath330 and @xmath358 be a divisor with support @xmath359 in @xmath221 . put @xmath360 and again lets denote by @xmath361 a meromorphic function on @xmath35 with the divisor @xmath6 , and for an element @xmath342 in @xmath362 define : @xmath363 for the schottky group @xmath123 associated to the fuchsian group @xmath229 , if @xmath165 , then the product ( [ f2.91 ] ) converges absolutely and uniformly on any compact subset of @xmath221 , after deleting a finite number of factors that may have a pole or zero on this subset . let @xmath364 be a compact subset of @xmath221 . since @xmath229 acts on @xmath221 properly discontinuously the set @xmath365 is finite . if we delete the set @xmath366 from the index set of the product ( [ f2.91 ] ) then for each @xmath367 when @xmath332 and @xmath368 lie outside a fixed compact neighborhood of @xmath359 we have @xmath369 where @xmath370 is a constant , @xmath371 , @xmath346 and @xmath372 ( and @xmath373 ) . the last inequality comes from the equality @xmath374 . now since @xmath165 the series @xmath375 converges uniformly on the compact subsets of @xmath136 . then the product ( [ f2.91 ] ) is convergent . in @xmath376 , if we change the point @xmath342 to @xmath377 , then we have @xmath378 . where @xmath379 is a nonzero complex number that depends on the points @xmath342 and @xmath377 and @xmath380 also , for @xmath381 and @xmath382 we have @xmath383 @xmath384 is a nonzero complex number multiplicative on @xmath385 and @xmath283 and also independent of @xmath342 . we can see this as following @xmath386 then @xmath387 . when @xmath123 is a cyclic group @xmath388 . this shows that @xmath389 is not @xmath229 automorphic function on @xmath221 in general . \a ) lets denote by @xmath390 a set of the representatives of @xmath391 . then for @xmath392 $ ] in @xmath329 we have @xmath393 and by defining @xmath394 we have @xmath395 where @xmath396 is the set @xmath397 without the identity class . + b ) lets denote by @xmath398 a set of representatives of @xmath399 and by @xmath400 the conjugacy class of @xmath283 in @xmath329 . then for some @xmath377 in @xmath221 such that @xmath401 stays in @xmath402 , we have @xmath403 and this is independent of @xmath377 . lets put @xmath404 , @xmath405 and @xmath406 then we have @xmath407 the part ( [ f2.8 ] ) comes from the equation @xmath408 and the invariance of the cross - ratio on the action of mobius transformations . the second part of a ) and b ) can be proved similarly . for b ) we should consider that @xmath409 . in above theorem in the case that the divisor is @xmath71 we have @xmath410 by previous theorem @xmath411 is a meromorphic function without any poles and zeroes in @xmath221 then it is a holomorphic function on @xmath221 . also as we see in the expression , it is independent of the point @xmath377 . for each @xmath283 in @xmath329 lets put @xmath412 however @xmath389 is not @xmath229- automorphic function on @xmath221 in general but we have @xmath413 and this shows that the differential @xmath414 is @xmath229 automorphic . then @xmath415 is a differential of the first kind on @xmath308 . + according to the classical theorem of cuts we can choose @xmath136 the region of discontinuity of the schottky group @xmath123 with the marking @xmath416 with @xmath417 such that @xmath418 the image of @xmath166 for @xmath127 in @xmath308 be coincident with the image of @xmath141 . and these together with @xmath419 the image of @xmath169 for @xmath127 in @xmath308 make a canonical base for @xmath420 i.e. @xmath421 \a ) @xmath422 is a riemann s basis for the space of differentials of the first kind on @xmath308 by choosing the previous base for @xmath423 i.e. @xmath424 b ) if we denote by @xmath390 a set of the representatives of @xmath391 and by @xmath425 , the set @xmath390 without the identity class , then for @xmath426 we have @xmath427 and for @xmath428 by defining @xmath429 we have @xmath430 we have @xmath431 where @xmath432 , @xmath401 and @xmath433 . if we show this equality for @xmath434 by @xmath435 then for a ) we have @xmath436 the last equality comes from this fact that if @xmath428 since @xmath437 and @xmath438 , only for @xmath439 the first alternative and for all @xmath440 the third one is valid . if @xmath426 only the third alternative is right for all @xmath441 . we can see these from the figure 1 . + for the first part of b ) . we have shown by the point @xmath442 the intersection of @xmath443 and @xmath444 in the representation of @xmath102 , the fundamental domain of @xmath229 . then @xmath445 is the image of the part of @xmath169 that is between the points @xmath442 and @xmath446 in the fundamental domain . now , from ( [ f2.11 ] ) we have @xmath447 where the last equality comes from the definition in chapter one . similarly we can reach to the second part of b ) using ( [ f2.12 ] ) . for @xmath448 lets denote by the same words the corresponding points in @xmath449 the fundamental domain of @xmath229 ( then @xmath450 and @xmath451 are in @xmath452 the fundamental domain of @xmath123 ) and put @xmath453 . then @xmath217 is a differential of third kind on @xmath308 with the residues 1 and -1 at the images of @xmath54 and @xmath55 . also , because the points @xmath54 and @xmath55 are in the fundamental domain then they are out of the circles @xmath128 . then @xmath418 - periods of @xmath217 are zero , and from ( [ f2.13 ] ) the @xmath454 - periods are @xmath455 then we have @xmath456 now , we can reach to a differential of the third kind with pure imaginary periods by defining @xmath457 where the real coefficients @xmath458 are such that the set of the equations @xmath459 for @xmath460 are satisfied . in fact the coefficients @xmath458 kill the real part of the @xmath419 - periods of @xmath217 . notice that the new differential form @xmath461 probably have nonzero @xmath418 - periods , but since the coefficients @xmath458 are real and @xmath462 are pure imaginary then they are pure imaginary too . + finally , if we denote the points @xmath218 , and @xmath69 in @xmath463 and the images of these points in the fundamental domain of @xmath229 by the same notations , then we have @xmath464 and by ( [ t2.11 ] ) @xmath465 then the green s function on @xmath308 can be computed as following @xmath466 because of the commutativity of the diagram @xmath467 @xmath468 @xmath469 the image of the point @xmath382 and @xmath470 are the same in @xmath308 . then the above formula for the green s function on @xmath308 gives an expression via the points on @xmath308 . in fact for the points @xmath64 in @xmath232 the images of the geodesics @xmath471 and @xmath472 in the kleinian manifold @xmath473 are the same . we can see this by using the following covering spaces @xmath474 for all parts @xmath475 , and again we have @xmath476 . also since @xmath328 is extended on @xmath331 in a natural way then for each @xmath477 the image of the geodesics @xmath478 and @xmath479 in @xmath260 are the same . + one can reach to a formula for the green s function on @xmath309 similarly by replacing @xmath258 and @xmath480 instead of @xmath221 and @xmath102 . a finitely generated , torsion free kleinian group @xmath481 is called quasi - fuchsian if the limit set @xmath482 be a jordan curve and each of the two simply connected components of @xmath483 be @xmath481-invariant . given a quasi - fuchsian group @xmath481 , there exist a fuchsian group @xmath229 and a quasiconformal diffeomorphism between kleinian manifolds @xmath484 and @xmath473 . ( see @xcite ) . lets denote by @xmath485 and @xmath486 the simply connected components of @xmath483 for quasi - fuchsian group @xmath481 . then there are two fuchsian groups @xmath487 and @xmath488 related to @xmath481 such that @xmath489 is homeomorphic to @xmath490 . then each @xmath491 is isomorphic to @xmath481 . also since @xmath492 is topologically conjugate to @xmath493 ( or @xmath494 in unit disk model ) then naturally there is a markov map @xmath495 like fuchsian case . then all the statements for fuchsian groups can be extended to the quasi - fuchsian groups . in this case because the jordan curve @xmath482 is generally not smooth nor even rectifiable ( @xcite p.263 ) then the hausdorff dimension of the subgroup @xmath496 of @xmath481 ( like @xmath329 for fuchsian case @xmath229 ) may be not less than 1 in general . we have the following theorem for a finitely generated kleinian group @xmath99 the following conditions are equivalent + ( 1 ) @xmath497 is diffeomorphic to @xmath498 $ ] where @xmath0 is a component of @xmath499 ; + ( 2 ) @xmath99 is quasi - fuchsian ; + ( 3 ) @xmath105 has an invariant component that is a jordan domain ; + ( 4 ) @xmath105 has two invariant components ; ( see @xcite , page 125 ) . also two homeomorphic riemann surfaces can be made uniform simultaneously by a single quasi - fuchsian group ( theorem of bers on simultaneous uniformization ) . then we have the computations for all hyperbolic 3 - manifolds with two compact riemann surfaces with the same genus as it s boundary components . already , we have computed the green s function for the boundary components of a hyperbolic 3-manifolds with two boundary at infinity with the same genera . in this chapter we want to do the problem for the most general case that the green s function can be defined i.e. for manifolds with @xmath41 boundary components that are compact riemann surfaces probably with different genera . the case with two boundary is different genera are included in this case too . such manifolds are uniformized by some kleinian groups and for the first , we will try to find a decomposition for such kleinian groups . this , help us to find invariants of these groups that are necessary for computing the automorphic functions that will be used for computing the differentials and the green s functions . in all of this section we consider @xmath99 to be a finitely generated and torsion free kleinian group of the second kind ( i.e. @xmath501 ) . * ( the ahlfors finiteness theorem ) * let @xmath99 be a finitely generated and torsion free kleinian group . then @xmath502 is a finite union of analytically finite riemann surfaces ( closed riemann surface from which a finite number of points are removed ) . for the rest of the section we consider @xmath503 such that for each @xmath504 , @xmath3 is a compact riemann surface of genus @xmath505 and without cusped point . then we should consider that @xmath99 does nt have any parabolic elements . also consider that the component @xmath506 is the one that for @xmath507 , @xmath508 . in this case @xmath99 is a function group ( a finitely generated non - elementary kleinian group which has an invariant component in its region of discontinuity ) and there is an invariant component @xmath509 in @xmath105 . we consider two cases for @xmath509 in the proposition above : + * case1 : * the component @xmath509 is simply connected . then in this case @xmath99 is a b - group ( a function group with simply connected invariant component ) . and we know that a b - group with a compact boundary component say @xmath511 is a quasi - fuchsian group ( see @xcite page 411 ) . + * case2 : * the component @xmath509 is not simply connected . in this case we have the following theorem . let @xmath99 be a function group with invariant component @xmath509 that is not a simply connected component . then @xmath99 has a decomposition into a free product of subgroups as following @xmath512 where each @xmath513 is a b - group , @xmath514 is a free abelian group of rank two with two parabolic generators , @xmath515 is the cyclic group generated by the parabolic transformation @xmath505 and @xmath516 is the cyclic group generated by the loxodromic transformation @xmath241 . furthermore , each parabolic transformation in @xmath99 is conjugate in @xmath99 to one a listed subgroup . @xmath99 has finite - sided fundamental polyhedron if and only if all the groups @xmath513 do . this theorem shows that in our special case we have @xmath517 for b - groups if one boundary component be compact then it s a quasi - fuchsian group . and @xmath518 is a free and purely loxodromic group and consequently it s a schottky group @xcite . hence we have @xmath519 where @xmath520 is a quasi - fuchsian group and @xmath123 is a schottky group . then by van kampen theorem we have @xmath521 and if @xmath522 then @xmath523 we will denote by @xmath524 the genus of schottky group @xmath123 in the decomposition of the group @xmath99 . we intend to analyze the structure of the region of discontinuity of @xmath99 . for the first , we consider the case that we do nt have the schottky group @xmath123 in the decomposition of @xmath99 . lets denote by @xmath129 and @xmath525 the simply connected components of the quasi - fuchsian group @xmath520 . from proposition ( [ p3.11 ] ) all other components than @xmath509 of @xmath105 are simply connected . now , since @xmath531 and from lemmas ( [ l3.11 ] ) , ( [ l3.12 ] ) and the decomposition of @xmath499 we see that @xmath532 is a complete system of representatives of the equivalent classes ( @xmath533 and @xmath534 are equivalent or conjugate if and only if @xmath535 is conjugate in @xmath99 to @xmath536 ; and since @xmath537 then @xmath533 and @xmath534 are equivalent or conjugate if and only if there is a @xmath113 in @xmath99 such that @xmath538 ) of the components of @xmath105 . since @xmath509 is @xmath99 - invariant the class of @xmath509 has one member . then we have in this case we have @xmath549 + now , lets consider the general case that there is a schottky group @xmath550 ( of order @xmath524 ) in the decomposition of the group @xmath99 . again we know that all other components of @xmath105 than @xmath509 is simply connected . then @xmath551 can not be glued to the components other than @xmath506 in @xmath499 . also this means that in this case we have @xmath552 and @xmath553 in this case if @xmath554 be a marking for the schottky group @xmath555 , then the fundamental domain is @xmath556 and also for the genus of the component @xmath506 we have @xmath557 . lets denote by @xmath310 the smallest normal subgroup of the group @xmath560 including the generators @xmath561 for @xmath507 and @xmath562 and @xmath563 be the smallest normal subgroup of the group @xmath520 including the generators @xmath561 for @xmath562 . then we have @xmath564 now if @xmath310 be the smallest normal subgroup of the group @xmath99 including the generators @xmath561 for @xmath507 and @xmath562 , then by the previous lemma one can see that @xmath568 is isomorphic to the group @xmath569 then @xmath568 is a free group of order @xmath570 . + now lets consider the covering space map @xmath571 . since @xmath310 is a normal subgroup of @xmath99 then the covering space @xmath572 is regular and is between @xmath573 with the covering transformations group @xmath574 that is a free group of order @xmath570 . then similar to the previous section we have a schottky group @xmath575 ( we can arrange the generators of the group @xmath123 in this way because of the isometry ) and a complex - analytic covering mapping @xmath576 such that the following diagram is commutative and for @xmath507 and @xmath579 , @xmath580 and for @xmath581 , @xmath582 and @xmath583 ( we have considered the same notations for the generators of @xmath584 and @xmath123 ) . as a matter of fact there is a fuchsian group @xmath229 and there are normal subgroups @xmath585 and @xmath586 of @xmath229 such that the sequence @xmath587 is the composition of analytic covering maps . then we have the composition of covering maps @xmath588 and since @xmath589 is a free group , like the previous section we have @xmath590 for some schottky group @xmath123 . in fact , for the first , because of the isometries @xmath591 we can choose the fuchsian group @xmath592=i\quad \rangle\end{aligned}\ ] ] such that the equations @xmath593 for @xmath507 and @xmath594 are satisfied . where @xmath595 are the generators of @xmath99 . then like the previous section we can choose uniquely up to conjugation in @xmath596 the schottky group @xmath123 that satisfies the equations @xmath597 then since @xmath598 is onto , the relations @xmath599 and @xmath600 are satisfied automatically for @xmath507 and @xmath594 . this means that the following diagram is commutative in all loops @xmath601 @xmath602 @xmath601 this gives a proof for the existing the of covering space @xmath136 and schottky group @xmath123 satisfying the properties that we need and also shows the relations between the fuchsian group @xmath229 , kleinian group @xmath99 and the schottky group @xmath123 associated to @xmath506 . + now lets put @xmath603 . where @xmath604 is the free subgroup of @xmath520 generated by the elements @xmath605 . for the extension of @xmath328 to @xmath606 i.e. defining the map @xmath607 , we know that each @xmath604 is free and purely loxodromic then it is a schottky group . like the previous section we can code an element of @xmath608 by schottky coding @xmath609 for @xmath610 and a point @xmath342 in @xmath611 . for each element @xmath113 in @xmath612 lets denote by @xmath59 the element in @xmath123 corresponding to @xmath113 , such that @xmath613 . now put @xmath614 @xmath615 where @xmath346 . and for @xmath507 put @xmath616 @xmath617 and finally lets define @xmath618 @xmath619 + then like the previous section we can show that for each element @xmath113 in @xmath612 and @xmath59 the element of @xmath123 corresponding to @xmath113 and each @xmath16 in @xmath606 we have @xmath620 and @xmath621 . also like the previous chapter we can consider the fuchsian coding for the points of @xmath606 and from ( [ f2.9 ] ) we can explain @xmath328 and consequently the green s function on @xmath506 via the fuchsian coding . if we consider the condition @xmath165 and use the subgroup @xmath622 instead of @xmath329 in the previous section then one can bring all of the definitions like before with some changes , and compute the green s function on @xmath506 and the other parts in the same way . in this case we should notice that if @xmath623 be a set that makes a base for @xmath624 then each member of it has a class of images in @xmath509 . but we can consider that the representatives that are used in the computations are in the fundamental domain . as its shown in figure [ fig3.14 ] . in this case these images are in @xmath625 for @xmath507 and also in @xmath626 for @xmath627 . then the representatives for @xmath3 and @xmath506 are coincide if we uniformize @xmath3 by @xmath628 , i.e. identifying both components of @xmath629 . final formula for the green s function on @xmath506 is as following @xmath630 when we have infinitely many boundary components in the boundary of @xmath112 that are riemann surfaces without puncture , according to the ahlfors finiteness theorem , @xmath99 is an infinitely generated kleinian group . and hence one of the boundary components of @xmath111 is with infinity genus and is not a compact riemann surface then for this component the green s function is not defined . but for the other components we can bring the computations similar to the previous condition using some subgroups of @xmath99 that are isomorphic to the kleinian groups in the previous sections . 30 l. v. ahlfors . `` finitely generated kleinian groups '' . amer.j . math . 86 ( 1964),413 - 429 and 87(1965),759 . `` uniformization , moduli and kleinian groups '' . bull . london math . , 4(1972),257.300 . + r. bowen . `` hausdorff dimension of quasi - circles '' publ . math . ihes 50,11 - 25(1979 ) . c. series . `` markov maps associated with fuchsian groups '' . publications mathmatiques de li.h.e.s.,tome 50(1979),p.153 - 170 . g. cornel , j.h . `` arithmetic geometry '' . springer verlag ( 1986 ) . g. faltings . `` calculating on arithmetic surfaces '' . 119 , 387 - 424 ( 1982 ) . l. r. ford . `` automorphic functions '' . mcgraw - hill 1929 . d. hejhal . `` on schottky and teichmuller spaces '' . 15 , 133 - 156 ( 1975 ) . `` compact riemann surfaces '' . `` uber die uniformisierung der algebraischen kurven '' . iv , math . ann . 75 ( 1914 ) `` introduction to arakelov theory '' . new york berlin heidelberg : springer 1988 . `` three - dimensional hyperbolic geometry as @xmath43 - adic arakelov geometry '' . invent . 104 ( 1991 ) , no . 2 , 223243 . yu . i. manin , v. drinfeld , `` periods of p - adic schottky groups '' . j. reine angew . math . 262/263 ( 1973 ) , 239247 . i. manin , matilde marcolli . `` holographi principle and arithmetic curves '' . k. matsuzaki , m. taniguchi . `` hyperbolic manifolds and kleinian groups '' . oxford university press , 1998 . c. mcmullen . `` riemann surfaces and the geometrization of 3 - manifolds '' . bull . 27 ( 1992 ) , no . 2 , 207216 . a. marden . `` the geometry of finitely generated kleinian groups '' . the annals of mathematics , 2nd ser . 99 , no . 3 ( may , 1974 ) , pp . 383 - 462 . j. g. ratcliffe . `` foundations of hyperbolic manifolds '' springer verlag ( 1991 ) . p. zograf , l. takhtajan . `` on the uniformization of riemann surfaces and on the weil - petersson metric on the teichmuller and schottky spaces '' . math . ussr - sb . 60 ( 1988 ) , no.2 , 297 - 313 .

we study the distribution of multivalent counterions next to a dielectric slab , bearing a quenched , random distribution of charges on one of its solution interfaces , with a given mean and variance , both in the absence and in the presence of a bathing monovalent salt solution . we use the previously derived approach based on the _ dressed multivalent - ion theory _ that combines aspects of the strong and weak coupling of multivalent and monovalent ions in a single framework . the presence of quenched charge disorder on the charged surface of the dielectric slab is shown to substantially increase the density of multivalent counterions in its vicinity . in the counterion - only model ( with no monovalent salt ions ) , the surface disorder generates an additional logarithmic attraction potential and thus an algebraically singular counterion density profile at the surface . this behavior persists also in the presence of a monovalent salt bath and results in significant violation of the contact - value theorem , reflecting the _ anti - fragility _ effects of the disorder that drive the system towards a more ` ordered ' state . in the presence of an interfacial dielectric discontinuity , depleting the counterion layer at the surface , the charge disorder still generates a much enhanced counterion density further away from the surface . likewise , the charge inversion and/or overcharging of the surface occur more strongly and at smaller bulk concentrations of multivalent counterions when the surface carries quenched charge disorder . overall , the presence of quenched surface charge disorder leads to sizable effects in the distribution of multivalent counterions in a wide range of realistic parameters and typically within a distance of a few nanometers from the charged surface .

1409.2609

nature is not perfect and various types of disorder are ubiquitous . disorder often causes large changes in the properties of condensed matter systems as predicted on the basis of naive idealized models that assume perfect regularity . electron properties in two - dimensional @xcite and three - dimensional disordered media @xcite , crystalline lattices with structural defects @xcite , spin glasses with random interactions @xcite and systems exhibiting criticality modified by the presence of disorder @xcite , are all instances of pronounced disorder effects in the bulk of the materials that can fundamentally change the behavior of idealized model systems . apart from their fundamental importance in modifying the bulk properties , disorder effects at surfaces and interfaces are particularly important in the context of the solid - electrolyte interphases @xcite relevant also for energy generation and storage technologies @xcite . the structural disorder in the charge distribution and/or dielectric response spatial profile in the vicinity of the material interfaces couple to long - range electrostatic interactions , leading effectively to long - range disorder effects as well @xcite , that can not be understood in terms of the usual assumptions of piecewise homogeneous charge distribution and/or dielectric properties , underpinning so much of colloid science and electrochemistry @xcite . the coupling between electrostatic interactions and disorder has been already noted and discussed in other important cases @xcite , including surfactant - coated surfaces @xcite , random polyelectrolytes and polyampholytes @xcite , and contaminant adsorption onto macroscopic surfaces or in amorphous films showing grain structure after being deposited on crystalline substrates @xcite . in all these cases the charge distribution often shows a fundamentally disordered component that often remains unaltered after the assembly or fabrication of the materials , thus exhibiting a frozen , or _ quenched _ , type of disorder ( see , e.g. , refs . @xcite for examples of surfaces with _ annealed _ charge distributions that will not be considered in this paper ) . this charge disorder coupled to the long - range electrostatic interactions can then leave its fingerprint also on the interactions between macromolecular surfaces that in their turn can play a fundamental role in the stability of colloidal systems @xcite . in fact , this coupling between disorder and coulomb interactions has been suggested to underly the anomalously long - ranged interactions observed in ultrahigh sensitivity experiments on casimir - van der waals interactions between surfaces _ in vacuo _ @xcite . the intricate experimental details of accurate measurements of these interactions can be properly accounted for only if one considers also the disordered nature of charges on and within the interacting surfaces by invoking the so - called _ patch effect _ @xcite , where the disorder stems , for instance , from the adsorption of charged contaminants and/or impurities that can give rise to monopolar random surface charges , and/or the variation of the local crystallographic axes of the exposed surface of a clean polycrystalline sample and the corresponding electron work function that can cause a variation of the local surface potential . such random distributions of surface charges can be measured directly by kelvin force microscopy measurements @xcite . the salient features of electrostatic interactions themselves , even for homogeneous charge distributions in the absence of any disorder , are however quite involved ( see , e.g. , refs . @xcite and references therein ) . it has been recognized some time ago that electrostatic interactions in fact come in several varieties , depending on the strength of electrostatic coupling in the system @xcite . in the presence of mobile monovalent counterions , they are standardly described by the poisson - boltzmann ( pb ) theory stemming from the mean - field , collective description of coulomb fluids @xcite that gives rise to pronounced repulsive interactions between like - charged macromolecules ( such as polymers , colloids and nano - particles ) . on the contrary , in the presence of multivalent counterions , electrostatic interactions exhibit basically a single - particle character and mediate strong attractive interactions between like - charged macromolecules this attraction led to a new understanding of the theory of electrostatic interactions in colloidal domain based on the _ strong - coupling ( sc ) limit _ @xcite , devised to describe the equilibrium properties of coulomb fluids when charges involved become large . in the simple case of a counterion - only system , the transition from the mean - field pb description , dubbed also the _ weak - coupling ( wc ) limit _ , to the sc limit is governed by a single dimensionless _ electrostatic coupling parameter _ @xcite , being a ratio of the bjerrum length , which identifies coulomb interaction between counterions themselves , and the gouy - chapman length , which describes electrostatic interaction between the counterions and the charged ( macromolecular ) surfaces . the emerging picture of equilibrium properties of coulomb fluids has thus become much richer than conveyed for many years by the standard dlvo paradigm of colloid science @xcite . however , this is still not the complete story . the most relevant case of a coulomb fluid is in fact not a counterion - only system , but an asymmetric mixture of multivalent ions in a bathing solution of monovalent ions , a particularly relevant situation specifically in the context of bio - macromolecules , where multivalent ions together with the screening properties of the monovalent salt are believed to play a key role in the stability of macromolecular aggregates such as liquid crystalline mesophases of semiflexible biopolymers @xcite , or dna condensates that form in the bulk @xcite or within viruses or virus - like nano - capsids @xcite . in an asymmetric mixture , multivalent counterions and monovalent ions are coupled differently to the macromolecular charges : multivalent ions strongly , while monovalent ions only weakly , as evidenced from their respective electrostatic coupling parameters . since usually multivalent ions are present at very low concentrations , e.g. , around just a few mm , their behavior is expected to be properly described within the virial expansion in powers of their fugacity ( or bulk concentration ) @xcite . a _ dressed multivalent - ion theory _ then emerges naturally within this context @xcite since the degrees of freedom due to weakly coupled monovalent ions can be traced out from the partition function , leading to an effective formalism based on screened interactions between the remaining dressed multivalent ions and fixed macromolecular charges . the dressed multivalent - ion theory for complicated asymmetric mixtures of multivalent ions in a bathing solution of monovalent ions can then seamlessly bridge between the standard wc and sc limits @xcite . these baroque features of electrostatic interactions furthermore give their imprint also on the effects of disordered charge distribution along macromolecular interfaces @xcite . while on the wc level and for homogeneous planar systems the quenched disorder effects are nonexistent @xcite , they can lead to qualitative changes in the stability properties of the system once the dielectric contrast between the solution and macromolecular interfaces ( or the inhomogeneous distribution of salt ions ) is taken fully into account @xcite . nevertheless , it is in the sc limit that the coupling between electrostatic interactions and the quenched disorder in the external interfacial charge distributions gives rise to fundamentally novel and unexpected phenomena @xcite . while studying the interaction between two disordered charged surfaces it was noticed @xcite that disorder can in fact lead to a lowering of the effective temperature of the system , engendering a distribution of the multivalent counterions between the interacting surfaces that is characterized by less effective entropy . this is intuitively difficult to foresee , as one would perhaps naively assume that thermal and externally imposed charge disorder would somehow enhance one another . in order to properly understand and identify all salient features of the coupling between quenched charge disorder and long - range electrostatic interactions , we now proceed to characterize more closely the consequences of coupling between charge disorder and _ electrostatically strongly coupled _ multivalent counterions immersed in a monovalent salt solution bath . in particular , we will identify the defining feature of this strongly coupled , disordered system as belonging to the _ anti - fragility _ @xcite exhibited by this system . in the present context , anti - fragility simply refers to the fact that an externally imposed , quenched charge disorder , effectively diminishes the intrinsic thermal disorder in the system , forcing its behavior to be more ` ordered ' . we will show that this behavior stems from the interplay between the _ translational entropy _ of the multivalent counterions and the _ configurational entropy _ due to the averaging over different realizations of the quenched disorder . in the particular example of the counterion - only model ( with no monovalent salt ions and no interfacial dielectric discontinuity ) , we show that multivalent counterions experience an additional logarithmic attraction towards the surface due to the presence of the surface charge disorder in a way that their density profile exhibits an algebraically singular behavior at the surface with an exponent that depends on the disorder strength ( variance ) . this behavior persists also in the presence of a monovalent salt bath and results in significant violation of the contact - value theorem @xcite . in the presence of an interfacial dielectric discontinuity , depleting the counterion layer at the surface , the charge disorder still generates a much enhanced counterion density further away from the surface . likewise , the charge inversion and/or overcharging of the surface are predicted to occur more strongly and at smaller bulk concentrations of multivalent counterions when the surface carries quenched charge disorder . the organization of the paper is as follows : in section [ sec : model ] , we introduce our model and then in section [ sec : formal ] present the theoretical formalism that will be used to study the distribution of multivalent counterions next to a randomly charged dielectric interface . we then proceed to present our results in section [ sec : results ] , where we discuss the case of counterion - only systems and the effects due to the presence of a bathing salt solution , an interfacial dielectric discontinuity , and also the overcharging and charge - inversion phenomena in the system . we conclude our discussion in section [ sec : discussion ] . our model is comprised of an infinite , planar dielectric slab of thickness @xmath0 and dielectric constant @xmath1 , immersed in a bathing ionic solution of dielectric constant @xmath2 ( see fig . [ f : schematic ] ) . the ionic solution is assumed to consist of a mixture of a monovalent 1:1 salt of bulk concentration @xmath3 as well as of a multivalent @xmath4:1 salt of bulk concentration @xmath5 . the dielectric slab is assumed to be impermeable to mobile ions and occupy the region @xmath6 with its surface normal oriented in the direction of the @xmath7-axis . the bounding surface of the slab at @xmath8 is assumed to be charged , carrying a _ quenched _ spatial distribution of random monopolar charges , @xmath9 , while the other surface of the slab is uncharged . the multivalent counterions are assumed to be in contact only with the charged surface ( see below ) . the disordered ( random ) charge distribution is described by the gaussian probability distribution function @xmath10 = c\ , \exp\bigg(\!\ ! - \frac{1}{2 } \int { \mathrm{d } } { \mathbf r}\ , g^{-1}({\mathbf r})\ , [ \rho({\mathbf r } ) - \rho_0({\mathbf r } ) ] ^2 \bigg ) , \label{eq : pdf}\ ] ] where @xmath11 is a normalization factor , @xmath12 is the mean and @xmath13 the variance of the spatial distribution of random surface charges . it is obvious that the above probability distribution function entails an uncorrelated disorder , i.e. , @xmath14 , where double - brackets denote the configurational ( quenched ) average @xmath15\ , \big(\cdots\big)$ ] . and dielectric constant @xmath1 immersed in a solution of dielectric constant @xmath2 , containing a mixture of mono- and multivalent salts . the slab boundary at @xmath8 is randomly charged . the multivalent ions ( counterions ) with charge valency @xmath4 are shown by large blue spheres and monovalent salt anions and cations are shown by small orange and blue spheres . the multivalent counterions are in contact only with the charged surface ( right).,width=302 ] in what follows , the general formalism is valid for an arbitrary shape of the charged boundaries , but for the later developments in this paper we shall delimit ourselves to the specific example of a planar slab as noted above . in this case , the mean charge distribution and its variance are given by @xmath16 where @xmath17 , and , without loss of generality , we assume @xmath18 and , for the multivalent counterions , @xmath19 . the monovalent salt is assumed to be present on both sides of the slab while the multivalent counterions are restricted to the right half - space @xmath20 . this spatial constraint can be taken into account formally by introducing the indicator functions @xmath21 for the multivalent counterions , and @xmath22 for the monovalent ions , where @xmath23 is the heaviside s step function . this constraint can be realized by enclosing the region on the left side of the slab in a membrane impermeable to multivalent ions ; such a constraint will be relevant in our analysis only in the situations where the slab thickness is small relative to the salt screening length and will otherwise have vanishing impact on the distribution of multivalent counterions next to the charged surface in the regime of parameters that will be of interest in this paper . for the most part , however , we shall focus on the case of semi - infinite slabs . another point to be noted here is that , for the sake of simplicity , we model the monovalent and multivalent ions as point - like particles . since the monovalent ions will be treated implicitly ( section [ sec : formal ] ) , the generalization of our formalism to include a finite size for the multivalent counterions ( which can have a relatively large size as compared with the monovalent ions ) is quite straightforward and we shall return to it in section [ subsec : ion - size ] , where we also analyze its consequences . for a given realization of the fixed charge distribution @xmath9 , the grand - canonical partition function of the above model can be written exactly in a functional - integral form as @xcite @xmath24 } , \label{eq : z}\ ] ] where @xmath25 and @xmath26 is the fluctuating ( electrostatic ) potential and the effective field - action " reads @xmath27 = \frac{1}{2 } \int \rmd { \mathbf r}\ , \rmd { \mathbf{r } ' } \phi(\mathbf{r } ) g^{-1}({\mathbf r } , { \mathbf r } ' ) \phi(\mathbf{r } ' ) + \nonumber\\ & & \qquad\quad + \,\rmi \int \rmd { \mathbf r}\,\rho({\mathbf r})\phi({\mathbf r } ) - k_{\mathrm{b}}t \int \rmd { \mathbf r}\,{\mathcal v}(\phi(\mathbf{r})),\end{aligned}\ ] ] where @xmath28 is the operator inverse of the coulomb interaction ( or the bare green s function ) , @xmath29 , in the presence of dielectric boundaries satisfying the equation @xmath30 . the field self - interaction " term is given by @xcite @xmath31 where @xmath32 and @xmath33 are the ionic fugacities of multivalent counterions ( @xmath34 ) and monovalent anions ( @xmath35 ) and cations ( @xmath36 ) , respectively . note that , since the multivalent counterions result from a @xmath4:1 salt of bulk concentration @xmath5 mixed with a 1:1 salt of bulk concentration @xmath3 , we have @xmath37 , @xmath38 and @xmath39 . the above field action obviously represents a highly asymmetric coulomb fluid especially when counterions are multivalent , @xmath40 . these counterions couple strongly to the fixed ( mean ) surface charge , whereas the monovalent salt species couple weakly . this leads to a complex situation where different components of the coulomb fluid couple differently to the same surface charges , thus making the analytical progress and , in particular , obtaining exact solutions @xcite , very difficult . nevertheless , progress is possible and systems of this type can be treated using a combined weak - strong coupling approximation , which has been discussed in a series of recent works @xcite . it was shown , by employing both analytical approaches as well as implicit- and explicit - ion simulations that , in a wide range of realistic system parameters , the monovalent ions can be treated safely within the debye - hckel ( dh ) framework , while the multivalent ions can be handled by means of a standard virial expansion scheme @xcite of the strong coupling approximation . the dh - type terms in this context follow by expanding the last two terms in eq . ( [ eq : v_s ] ) up to the second order in @xmath26 ( which can be justified on a systematic basis in highly asymmetric systems with @xmath41 @xcite ) , i.e. , @xmath42 where @xmath43 is the bulk concentration due to all monovalent ions @xmath44 . on the analytical level , the above procedure allows one to trace the partition function over the degrees of freedom associated with monovalent ions and thus one remains with only dressed " multivalent ions and surface charges that then interact through a dh - type interaction kernel ( or the screened green s function ) defined via @xmath45 where the debye ( or salt ) screening parameter @xmath46 is non - zero only outside the dielectric slab and is given by @xmath47 with @xmath48 being the bjerrum length . this type of methodology leads to the so - called _ dressed multivalent - ion theory _ @xcite , which is a direct generalization of the standard counterion - only sc theory @xcite . in fact , the dressed multivalent - ion theory has a hybrid character in that it reproduces both the counterion - only sc theory and the dh theory as two asymptotic limits at small and large salt screening parameters , respectively . in what follows , we shall use this framework to study the effects of surface charge disorder on the distribution of multivalent counterions . the density profile of multivalent counterions follows standardly from the general formalism defined by eqs . ( [ eq : z ] ) and ( [ fieldaction ] ) as @xcite @xmath49 ) = \lambda_c \omega_c({\mathbf r})\langle \rme^{-\rmi\beta q e_0 \phi}\rangle,\ ] ] where @xmath50 denotes the thermal ( ensemble ) average over the fluctuating field @xmath26 . for a given ( quenched ) realization of @xmath9 , this average can be calculated analytically in both limits of weak and strong coupling in the counterion - only case @xcite and also in the more general context of dressed multivalent ions in asymmetric coulomb fluids @xcite . as noted before , the effects due to multivalent ions can be investigated by virial expanding the partition function in terms of their fugacity and keeping only the leading order contributions ( see refs . @xcite for further details ) . this procedure leads to a single - particle form for the density profile of multivalent counterions as @xmath51 ) = \lambda_c \omega_c({\mathbf r } ) \ , \rme^{-\beta u({\mathbf r } ; [ \rho])},\ ] ] where @xmath52)$ ] is the single - particle interaction energy @xmath53 ) = qe_0 \int { \mathrm{d}}{\mathbf r } ' \,g({\mathbf r } , { \mathbf r}')\rho({\mathbf r } ' ) + \frac{q^2e_0 ^ 2}{2 } g_{\mathrm{im}}({\mathbf r } , { \mathbf r } ) . \label{eq : u_rho}\ ] ] here , @xmath54 is the generalized born energy contribution that stems purely from the dielectric and/or salt polarization effects ( or the so - called image charges " ) . in other words , @xmath55 , where @xmath56 is the formation ( self-)energy of individual counterions in a homogeneous background , which is obtained from the free - space screened green s function defined via @xmath57 . in the present model , the fixed surface charge distribution , @xmath9 , has a disorder component and thus , in order to obtain the measurable counterion density profile , one must average eq . ( [ eq : sc_density ] ) over different realizations of this disorder field using the gaussian weight ( [ eq : pdf ] ) . this can be done straightforwardly by computing @xmath58 ) \rangle\ ! \rangle = \lambda_c \omega_c({\mathbf r } ) \langle \ ! \langle \rme^{-\beta u({\mathbf r } ; [ \rho ] ) } \rangle\ ! \rangle,\ ] ] which then gives @xmath59 where the _ effective _ single - particle interaction energy now reads @xmath60 with @xmath61 ^ 2 . \label{eq : u2}\end{aligned}\ ] ] the three terms , respectively , originate from the contribution of the interaction of multivalent ions with the mean surface charge density ( first term ) , the contribution of self - interactions ( interactions with image charges , second term ) and the contribution of the surface charge disorder ( third term ) . the latter can be viewed as an effective surface - counterion interaction which is induced by the quenched randomness in the surface charge . it is proportional to the disorder variance but also shows an explicit _ temperature dependence_. another interesting point is that the disorder interaction term exhibits a quadratic dependence on the green s function and likewise also on the multivalent ion charge valency @xmath4 . these features can be understood by noting that the disorder term in fact represents the _ sample - to - sample fluctuations ( or variance ) _ of the single - particle interaction energy , @xmath52)$ ] , eq . ( [ eq : u_rho ] ) , which is a linear functional of the gaussian field @xmath9 , and hence one can show that @xmath62)\rangle\ ! \rangle - \beta \langle \ ! \langle u^2({\mathbf r } ; [ \rho])\rangle\ ! \rangle_c/2 , \label{eq : u_v2}\ ] ] where @xmath63)\rangle\ ! \rangle_c\equiv \langle \ ! \langle u^2({\mathbf r } ; [ \rho])\rangle\ ! \rangle - \langle \ ! \langle u({\mathbf r } ; [ \rho])\rangle\ ! \rangle^2 $ ] , and @xmath64)\rangle\ ! \rangle = u_0({\mathbf r})+ u_{\mathrm{im}}({\mathbf r } ) , \\ & & \langle \ ! \langle u^2({\mathbf r } ; [ \rho])\rangle\ ! \rangle_c= -2k_{\mathrm{b}}t\,u_{\mathrm{dis}}({\mathbf r}).\end{aligned}\ ] ] in order to proceed , we introduce the dimensionless ( rescaled ) quantities @xmath65 where @xmath66 is the gouy - chapman length and @xmath67 is the electrostatic coupling parameter associated with the mean surface charge @xcite . analogously , one can define the dimensionless _ disorder coupling ( or disorder strength ) parameter _ @xmath68 which is proportional to the disorder variance ; its dependence on the counterion valency , @xmath4 , is different from that of the mean electrostatic coupling parameter as first noted in ref . let us first consider the special case of multivalent counterions next to a charged surface in the absence of any salt screening ( @xmath69 ) and dielectric image charge effects ( i.e. , in a dielectrically homogeneous system with @xmath70 ) . this case has been considered in a previous work on the effective interaction between two randomly charged surfaces @xcite , which however did not investigate the distribution of counterions . in this case , the number of counterions , @xmath71 , is fixed by the mean charge on the surface through the electroneutrality condition @xmath72 , where @xmath73 is the surface area . the fugacity of counterions is thus given by @xcite @xmath74 the effective single - particle interaction is obtained by using eqs . ( [ eq : u])-([eq : u2 ] ) and noting that , in this case , @xmath75 . hence , in rescaled units , we have @xmath76 , and @xmath77 ( up to irrelevant additive constants ) . therefore , @xmath78 the rescaled density profile of counterions is then obtained using eq . ( [ eq : sc_density_av ] ) as @xmath79 where @xmath80 is the gamma function . this expression shows a standard sc exponential decay of the multivalent counterion density , which dominates at large separations from the surface and is a well - established result within the sc context @xcite . but it also exhibits an algebraic dependence on @xmath81 , which dominates at small separations from the surface and thus shows that , in the presence of surface charge disorder , the counterion density _ diverges _ in the immediate vicinity of the surface . the presence of disorder thus clearly violates the contact - value theorem which was derived for uniformly charged surfaces @xcite ; this theorem entails a contact value of @xmath82 in the whole range of coupling parameters . note that , nevertheless , the electroneutrality is exactly satisfied as @xmath83 . the behavior of the density profile , eq . ( [ eq : dens_0 ] ) , is shown in fig . [ f : densityp_fin ] for a few different values of the disorder coupling parameter . as seen , due to the singular behavior at the surface , the presence of charge disorder enhances ( suppresses ) the density of counterions at small ( large ) separations , in general agreement with the previous findings in the case of non - disordered but heterogeneously charged surfaces such as surfaces carrying discrete charge patterns ( see , e.g. , @xcite and references therein ) . the above results can be illuminated further by analyzing the averaged cumulative charge defined as @xmath84,\ ] ] or , in rescaled units , @xmath85 where @xmath86 is the incomplete gamma function . note that @xmath87 is normalized such that @xmath88 , which represents the mean surface charge density at @xmath89 , and @xmath90 for @xmath91 , which reflects the global electroneutrality of the system . as seen in fig . [ f : q_densityp_fin ] , nearly all of the counterions become strongly localized in the vicinity of the surface ( i.e. , @xmath92 for finite @xmath93 ) as @xmath94 is increased . in fact , the counterion density profile tends to zero at any finite separation from the surface @xmath95 when @xmath94 tends ( from below ) to the threshold value @xmath96 because the gamma function in the denominator of the normalization factor in eqs . ( [ eq : dens_0 ] ) and ( [ eq : q0 ] ) goes to infinity . it should be emphasized that the strong accumulation of multivalent counterions in the immediate vicinity of the surface does not give rise to a renormalized mean surface charge density ( see also appendix c in ref . @xcite ) and that the singular behavior of counterions at the surface in the present context should be distinguished from the surface adsorption or counterion condensation phenomena @xcite . the foregoing results clearly show that the excess accumulation of counterions near the surface is driven by the disorder - induced , single - particle interaction energy ( [ eq : u2_ci_only ] ) , which is attractive and depends logarithmically on the distance from the surface and thus generates the singular behavior of the counterion density profile at the surface . this suggests that the presence of quenched surface charge disorder drives the system towards a state of lower thermal ` disorder ' . this point can be established systematically by calculating the difference in the entropy of counterions in the presence and in the absence of disorder , @xmath97 . we find @xmath98 where @xmath99 is the digamma function . it thus follows that @xmath100 . the entropy reduction is larger for larger disorder strength , @xmath94 , and diverges as @xmath101 . in other words , the reduction in the _ translational entropy _ of multivalent counterions in the solution is driven by introducing a finite degree of _ configurational entropy _ due to the presence of quenched randomness in the surface charge distribution . formally , this latter type of entropy is generated by the non - thermal ( quenched ) average taken over different realizations of the surface charge disorder . this subtle interplay between the different kinds of entropy is therefore essential in generating the singular behavior of counterions near the disordered surface . it thus seems appropriate to refer to this type of behavior of the multivalent counterions as _ anti - fragile _ @xcite , since introducing an external ( quenched ) disorder source effectively diminishes the intrinsic thermal disorder in the system and drives it towards a more ` ordered ' state . another interesting point to be noted here is that the internal energy of the system also decreases due to the presence of disorder but in such a way that leads to a decrease in the free energy of the system as can be seen from the free energy difference , @xmath102 . therefore , the system also attains a thermodynamically more stable state , which is again a direct consequence of the singular behavior of counterions near the disordered surface . by contrast , one can show that in the case of counterions next to a uniformly charged surface , which exhibits a regular potential , the attraction of counterions towards the surface leads to a larger free energy as compared with the ideal case where the system is uncharged . we now turn to the effects due to a monovalent salt bath by assuming that in addition to the multivalent counterions ( of bulk concentration @xmath5 ) , we also have a finite amount of monovalent salt in the system , giving a total bulk monovalent ion concentration of @xmath103 ( sections [ sec : model ] and [ sec : formal ] ) . we take a semi - infinite slab ( @xmath104 ) impermeable to all ions and assume that the system is again dielectrically homogeneous , i.e. @xmath70 . this helps to disentangle the polarization effects due to the inhomogeneous distribution of salt ions ( salt image effects " ) from those resulting from the inhomogeneous distribution of the dielectric constant ( dielectric image effects " ) . for a semi - infinite slab , the fourier - bessel transform of the green s function can be obtained by standard methods as @xmath105 , \label{eq : inf_green}\ ] ] where @xmath106 note that because of the translational and rotational symmetry with respect to the transverse ( in - plane ) coordinates @xmath107 and @xmath108 , the green s function depends on these coordinates only through @xmath109 , i.e. , @xmath110 , and thus its fourier - bessel transform is defined via @xmath111 . , @xmath112 and in the absence of dielectric image charge effects ( @xmath70 ) . different curves correspond to different values of the disorder coupling parameter as shown on the graph . inset shows the diverging behavior of the density profile at small distances from the surface in the log - log scale . ] the contributions to the effective single - particle interaction energy , @xmath113 , eqs . ( [ eq : u])-([eq : u2 ] ) , are now obtained , in rescaled units , as @xmath114 where @xmath115 . these equations can be used along with eq . ( [ eq : sc_density_av ] ) in order to compute the density profile of counterions . as seen in fig . [ f : contact_inf ] , the rescaled density profile of counterions shows a clear depletion effect in the absence of disorder ( note that here we have rescaled the density of counterions with their bulk concentration ) . this behavior is caused by the salt image effects that are produced by the second term in eq . ( [ eq : inf_green ] ) . the depletion effect becomes weaker when the surface is randomly charged as the counterions are again attracted more strongly to the surface in the presence of charge disorder . the interplay between salt image depletion and disorder attraction leads to a non - monotonic behavior in the counterion density profile as the disorder effects dominate at small separations while the salt image effects dominate at intermediate separations . at large distances from the surface ( @xmath116 ) , the behavior of the density profile is dominated by the salt screening effects and we get @xmath117 for @xmath118 . the accumulation of counterions in the vicinity of the surface is suppressed when the bulk salt concentration is increased or when a finite dielectric discontinuity is introduced at the surface ( see section [ subsec : delta ] ) . note that the counterion density profiles show a singular bahavior at the surface even in the presence of additional salt . this behavior is in fact present at _ any _ finite values of @xmath94 ( as may be discerned more clearly from the log - log plot in the inset of fig . [ f : contact_inf ] ) and coincides with the same algebraic divergence @xmath119 on approach to the surface as we found in the counterion - only case in the previous sections . this is intuitively expected because the salt bath effects diminish at separations much smaller than the screening length @xmath120 . the effects of a finite slab thickness , @xmath0 , can be examined by using the appropriate form of the green s function in this case that can be obtained by means of standard methods as @xmath121.\end{aligned}\ ] ] hence , the three different terms in the effective single - particle interaction energy follow in rescaled units as @xmath122 the finiteness of the slab thickness is expected to be relevant mainly in the regime where the thickness is comparable with or smaller than the screening length , i.e. @xmath123 . as it can be seen directly from eqs . ( [ eq : b_u0])-([eq : b_u2 ] ) , both the attraction experienced by multivalent counterions due to the mean surface charge and its disorder variance and the repulsion due to the salt image effects become stronger as the slab becomes thicker or , in other words , as the system becomes more strongly inhomogeneous in terms of the salt distribution . the overall effect is such that the counterion density close to the charged surface becomes smaller for smaller @xmath124 as shown in fig . [ f : contact_inf_b ] , inset ( compare also figs . [ f : contact_inf ] and [ f : contact_inf_b ] ) . this behavior can be understood also by noting that for thinner slabs the salt ions on the left side of the slab also contribute to the screening effects and , hence , further suppress the multivalent counterion density on the right side ( see fig . [ f : schematic ] ) . this however leaves the effects resulting from the surface charge disorder qualitatively unchanged , especially at small distances from the surface , where the singular behavior persists . but for fixed rescaled slab thickness @xmath125 and different values of the disorder coupling parameter as shown on the graph ( main set ) and also for fixed disorder coupling parameter @xmath126 and different values of the rescaled slab thickness ( inset).,width=302 ] it is to be noted that the expressions for the effective single - particle interaction energy , eqs . ( [ eq : b_u0])-([eq : b_u2 ] ) , can correctly reproduce the counterion - only result , eq . ( [ eq : u_z ] ) , when the limit @xmath127 , which also gives @xmath128 , is taken ( in this case , the thickness @xmath124 will be irrelevant ) . the counterion - only limit can not be recovered if we start with the infinite - thickness expressions ( [ eq : inf_u0])-([eq : inf_u2 ] ) ( where the thickness of the slab is strictly set equal to infinity ) and then take the limit of zero salt . the difference would be in a factor 2 in the expression for @xmath129 , indicating that the two limits @xmath130 and @xmath131 do not commute . alternatively , one can recover eq . ( [ eq : u_z ] ) from eqs . ( [ eq : b_u0])-([eq : b_u2 ] ) by first taking the limit of a thin slab @xmath132 and then @xmath127 . , @xmath112 , @xmath133 and different values of the disorder coupling parameter as shown on the graph . ( b ) same as panel ( a ) but for @xmath126 and different values of the rescaled salt screening parameter . , title="fig : " ] ( a ) .3 cm , @xmath112 , @xmath133 and different values of the disorder coupling parameter as shown on the graph . ( b ) same as panel ( a ) but for @xmath126 and different values of the rescaled salt screening parameter . , title="fig : " ] ( b ) so far we focused on the cases with no dielectric discontinuity at the boundaries of the slab . the case of a dielectrically inhomogeneous system with @xmath134 can be studied by simply replacing the definition of @xmath135 in the expressions in the previous section ( e.g. , eqs . ( [ eq : inf_u0])-([eq : inf_u2 ] ) ) with a more general one , i.e. @xmath136 in the absence of a salt bath ( @xmath137 ) , the dielectric image charges lead to very strong repulsions from the surface when @xmath138 ( as is often the case for aqueous solvents and macromolecular surfaces ) . this effect enters through the image interaction term @xmath139 , where the dielectric discontinuity parameter is defined as @xmath140 the image interaction term diverges at the surface and thus implies a vanishing contact density @xmath141 for @xmath142 , a behavior that is very distinct from that generated merely by salt images ( figs . [ f : contact_inf ] and [ f : contact_inf_b ] ) , as the latter can not be described generally in terms of point - like image charges " and generate much weaker repulsive forces on multivalent ions than the dielectric images . in the most general case with both salt and dielectric image effects , the density profile of counterions can be calculated via eqs . ( [ eq : sc_density_av ] ) and ( [ eq : inf_u0])-([eq : inf_u2 ] ) with the definition in eq . ( [ eq : delta_s_general ] ) . the results are shown in fig . [ fig : density_inf_delta ] for @xmath112 and @xmath133 , corresponding to the dielectric discontinuity at the water / hydrocarbon interface ( with @xmath143 and @xmath144 ) . clearly , both added salt and charge disorder effects become irrelevant in the small - distance regime , where the dielectric - image repulsions dominate and generate a wide depletion zone near the surface . the counterion density thus again vanishes at the surface and tends to the bulk value at large separations . although the qualitative form of the density profile remains the same in the absence and in the presence of charge disorder ( which is in contrast with what we found in section [ subsec : salt ] ) , the peak of the density profile becomes more pronounced and shifts to smaller values of the distance from the surface as the disorder strength is increased , see fig . [ fig : density_inf_delta]a . a similar effect is seen in fig . [ fig : density_inf_delta]b , where the salt screening parameter is decreased ; in this case the location of the peak of the density profile remains nearly unchanged while its height increases by almost an order of magnitude when the salt screening parameter is decreased by only a factor of 2 . the preceding results suggest that the charge inversion and/or overcharging of the surface , which are known to occur with asymmetric coulomb fluids ( see , e.g. , refs . @xcite and references therein ) , may be enhanced when the surface carries a random charge component . this can be inferred from the averaged cumulative charge , @xmath87 , being the sum of the average charges due to fixed and mobile charges ( including both monovalent and multivalent ions ) within a finite distance @xmath145 from the surface , i.e. , @xmath146 , \label{eq : q_av}\ ] ] where @xmath147 represent the averaged ( dh ) density of monovalent ions and @xmath148 the density of dressed multivalent counterions . as shown in ref . @xcite , the cumulative charge can be written only in terms of the counterion density by using the fact for monovalent ions we have @xmath149 , where @xmath150 is the mean electrostatic potential generated by the explicit charge densities , i.e. , @xmath151 . \label{eq : psi_av}\ ] ] hence , using the green s function expressions ( [ eq : inf_green ] ) for an infinite slab , we have @xcite @xmath152\hat c(\tilde z ' ) , \nonumber\end{aligned}\ ] ] where @xmath153 and @xmath154 with the definition @xmath155 , @xmath156 , @xmath157 , @xmath133 and different values of the disorder coupling parameter as indicated on the graph . ( b ) phase diagram " showing the minimal amount of multivalent counterion concentration , @xmath158 ( in rescaled units ) , required to find the charge - inversion ( main set ) or overcharging effect ( inset ) as a function of the rescaled salt screening parameter in the absence ( @xmath159 ) and in the presence of surface charge disorder ( @xmath160 ) . the region above the curves shows where charge inversion ( main set ) or overcharging ( inset ) is predicted to occur . , title="fig : " ] ( a ) .3 cm , @xmath156 , @xmath157 , @xmath133 and different values of the disorder coupling parameter as indicated on the graph . ( b ) phase diagram " showing the minimal amount of multivalent counterion concentration , @xmath158 ( in rescaled units ) , required to find the charge - inversion ( main set ) or overcharging effect ( inset ) as a function of the rescaled salt screening parameter in the absence ( @xmath159 ) and in the presence of surface charge disorder ( @xmath160 ) . the region above the curves shows where charge inversion ( main set ) or overcharging ( inset ) is predicted to occur . , title="fig : " ] ( b ) as seen in fig . [ f : phasediag]a , the cumulative charge shows a _ positive _ hump at intermediate separations from the surface , also known as the _ charge - inversion effect_. the charge - inversion degree is usually found to amount to a fraction of the total charge ( i.e. , the maximum value of @xmath161 is smaller than unity ) @xcite . in the presence of charge disorder , a substantially larger amount of multivalent counterions are attracted towards the surface and also a much larger charge - inversion degree ( corresponding to the height of the hump ) is predicted to occur . there is a narrow region at small separations from the surface where one can see a decrease in @xmath161 as the disorder strength , @xmath94 , is increased . for sufficiently large @xmath94 , it exhibits a short - distance dip with @xmath162 . in this region , the cumulative charge has the same sign as the _ mean _ surface charge but with a larger magnitude and , therefore , represents the so - called _ overcharging effect_. note that the overcharging effect can be present even in the absence of disorder and depends on the bulk concentration of multivalent counterions , which enters in eq . ( [ eq : q ] ) through the rescaled parameter @xmath158 . in fig . [ f : phasediag]b , we show the minimal amount of bulk multivalent counterion concentration ( represented by @xmath158 ) that is required to achieve the charge - inversion ( main set ) and overcharging effects ( inset ) for a wide range of rescaled salt screening parameters . the region above the curves pertains to the parameter values where we find charge inversion or overcharging of the mean surface charge . as seen , for larger salt screening parameters , a larger bulk concentration of multivalent counterions are required to achieve these effects , and for a given salt screening parameter , a larger concentration of multivalent counterions are required to cause overcharging of the surface than its charge inversion . the presence of surface charge disorder facilitates both these effects as they can occur for smaller threshold values of @xmath158 , especially at intermediate to large values of the salt screening parameter . the aforementioned features of the averaged cumulative charge depend strongly on the dielectric discontinuity at the surface , which generates the image - charge repulsion that competes with the disorder - induced attraction of ions towards the surface . note that these two latter mechanisms affect the spatial distribution of multivalent counterions ( figs . [ fig : density_inf_delta]a and b ) , and to a lesser extent , also the spatial distribution of monovalent ions ( not shown ) that are treated implicitly in this work ; this latter quantity can be calculated from the mean potential , eq . ( [ eq : psi_av ] ) , which depends on various system parameters through the green s function and the distributions of explicit charges ( see refs . @xcite for explicit - ion simulations of charge inversion and overcharging phenomena at uniformly charged surfaces that incorporate the dielectric image charges as well ) . the exact form of @xmath161 , and the ensuing charge - inversion and/or overcharging effects , thus follow from the interplay between the contributions from the explicit multivalent counterions and the implicit monovalent ions to the averaged cumulative charge at any given set of values for the system parameters . finally , it should be noted that , while the predicted boundaries of the parameter space pertaining to the onset of the charge inversion and/or overcharging are expected to be relatively accurate @xcite , the single - particle approximation that lies at the heart of the dressed multivalent - ion description is not expected to be adequate in the regime of parameters deeply within the regime of charge inversion and/or overcharging due to non - negligible many - body contributions @xcite . these considerations and the role of other possible factors such as ion - ion excluded - volume repulsions @xcite remain to be assessed further in future simulations . , @xmath156 , @xmath157 , @xmath133 , @xmath160 and different values of the rescaled radius of the multivalent counterions , @xmath163 , as indicated on the graph . ] we have so far assumed that the multivalent counterions are modeled as point - like particles and can therefore access the whole volume in the region @xmath164 . on the microscopic level and within the primitive model considered here , the counterion size effects enter through the excluded - volume repulsions between the counterions themselves as well as through the excluded - volume repulsion between individual counterions and the randomly charged surface , which leads to the steric depletion of counterions from the immediate vicinity of the surface . within the strong - coupling approach or its generalization , the dressed multivalent - ion theory on which our approach is based , the partition function is virial expanded systematically to the leading order ( section [ sec : formal ] ) and , as such , involves only the single - particle contributions . this means that only the counterion - surface excluded - volume repulsion enters in the leading - order theory and the excluded - volume repulsions between counterions enter in the subleading terms that become important at lower electrostatic couplings , falling outside the regime of interest in the present analysis @xcite . ( the dressed multivalent - ion theory of course has a range of validity @xcite that can be ascertained in detail once simulations or , indeed , alternative theoretical approaches for the present problem become available ; see the discussion ) . the incorporation of the counterion size effects on the leading virial level is thus straightforward . for the sake of simplicity , we model the multivalent counterions as hard - sphere particles with radius @xmath165 . therefore , one needs to restrict the volume accessible to multivalent counterions to the region @xmath166 , which can be done by re - defining the indicator function as @xmath167 . it is thus evident that the form of the single - particle interaction energy and , therefore , the @xmath7-dependent form of the density profile remain unchanged , with the proviso that now one needs to impose the constraint @xmath168 . the prefactor @xmath32 in eq . ( [ eq : sc_density_av ] ) is still given by the bulk concentration of multivalent counterions , @xmath5 , in the presence of a salt bath ( sections [ subsec : salt ] and [ subsec : delta ] ) and is determined from eq . ( [ eq : lambda_c ] ) within the counterion - only model ( section [ subsec : ci - only ] ) . in this latter case , we find the rescaled density profile of hard - sphere counterions of rescaled radius @xmath169 as @xmath170 in other words , the singular behavior of the single - particle interaction energy remains intact and , therefore , the @xmath7-dependent form of the density profile still involves an algebraic factor and thus differs qualitatively from that in the case of a uniformly charged surface , only that the multivalent counterions can not probe the singular point at @xmath8 and do not exhibit the diverging surface density obtained with point - like counterions . the exclusion of multivalent counterions from the vicinity of the surface for large counterion radii can suppress the charge - inversion and/or overcharging effects discussed in the previous section . in fig . [ f : qc_a ] , we show the averaged cumulative charge , @xmath161 , as a function of @xmath81 for a few different values of the rescaled counterion radius , ranging from @xmath171 up to @xmath172 ( in actual units , and assuming @xmath173 nm , see table [ table ] , these values correspond to counterion radii of up to around 5.8 ) . as seen , upon increasing the counterion radius , the height of the positive hump ( the charge - inversion degree ) is decreased and its location shifts to larger distances from the surface . the location of the negative ( overcharging ) dip also shifts to larger separations ; however , the overcharging effect is enhanced at first when the counterion radius is increased and diminishes only when the latter is increased beyond @xmath174 . it is also important to note that for sufficiently small counterion radii , the cumulative charge nearly coincides with that of point - like counterions ( black solid curve ) . in fact , the effects of finite counterion size show up only when the counterion radius becomes larger than the size of the depletion zone generated by dielectric images in the vicinity of the surface , which , in the case fig . [ f : qc_a ] , corresponds to counterion radii @xmath175 ( see also figs . [ fig : density_inf_delta]a and b ) . in other words , the charge inversion and/or overcharging of the surface are affected also by the interplay between the image - charge depletion and the steric depletion of multivalent counterions from the dielectric interface . finally , we note that multivalents counterions may have an internal structure that can introduce higher - order multipolar effects ; these effects can be relevant especially for multivalent counterions with an extended structure ( such as rod - like polyamines including the trivalent spermidine and tetravalent spermine @xcite ) , in which case the present hard - sphere model may not be appropriate in order to account for the counterion size effects ( see , e.g. , refs . @xcite and references therein ) . rrrrrrrr|rrrrrrr & + @xmath176 ' '' '' & @xmath177 & @xmath178 & @xmath179 & & & & & & @xmath94&&&@xmath180 ) & & + @xmath181 mm & 109 mm & 158 mm & 283 mm & @xmath1821.1 mm & @xmath183= 0.1&&&&0.5 & & & 0.01 & & + 57 mm & 98 mm & 147 mm & 272 mm & 7 mm & 0.25 & & & & 1.0&&&0.02 & & + 51 mm & 91 mm & 140 mm & 270 mm & 10 mm & 0.3 & & & & 2.0&&&0.04 & & + 36 mm & 75 mm & 125 mm & 250 mm & 18mm&0.4 & & & & 4.0&&&0.08 & & + we have investigated the distribution of multivalent counterions close to a dielectric slab bearing a quenched , random distribution of monopolar surface charges on one of its solution interfaces with set mean value and variance , both in the absence and in the presence of an asymmetric coulomb fluid , comprised of a mixture of multivalent counterions in a bathing solution of monovalent ions . such asymmetric coulomb fluids are commonplace in many experimental examples such as in the condensation of dna by multivalent cations in the bulk @xcite or in viruses and virus - like nano - capsids @xcite . our analysis is done within the framework of the _ dressed multivalent - ion theory _ , which reproduces the strong - coupling theory of multivalent counterions @xcite in the zero salt limit and takes into account the surface - counterion as well as counterion - image correlations on the leading order and in the presence of a bathing salt solution as discussed in detail elsewhere @xcite . ( note that in the opposite regime of weak coupling , where , e.g. , all ions are monovalent , the quenched charge disorder effects turn out to be small and do not lead to any qualitatively new features in the behavior of the system @xcite . ) in the case of counterions only , we show that a randomly charged surface generates a singular density profile for multivalent counterions with an algebraically diverging behavior at the surface ; the latter is characterized by an exponent which is determined by the disorder strength ( variance ) . thus , multivalent counterions are predicted to accumulate strongly in the immediate vicinity of the randomly charged surface in a way that violates the contact - value theorem , which describes the behavior of counterions at uniformly charged surfaces and predicts a finite contact density @xcite . this behavior stems from the interplay between the translational entropy of the solution ions and the ( non - thermal ) configurational entropy due to the averaging over different realizations of the quenched disorder . therefore , by introducing an external ( quenched ) disorder component , we find that the system is driven towards a more ` ordered ' state characterized by a diminished intrinsic thermal ` disorder ' in the system . it thus seems appropriate to characterize this response of the system to an externally imposed quenched disorder as the _ anti - fragile behavior _ @xcite of multivalent counterions in the presence of quenched charge disorder . it is to be noted that , in the presence of disorder , the system also attains a thermodynamically more stable state because the internal energy of the system drops in a way that leads to a lowered free energy . the singular behavior of multivalent counterions persists also when counterions are immersed into a bath of a monovalent salt solution and there are no dielectric inhomogeneities in the system . in this case , the slab defines an ion - excluded region , creating salt image effects . the interplay between the disorder - induced attraction and the salt - image depletion leads to a non - monotonic density profile for counterions close to the surface . the amount of multivalent counterions accumulated near the surface is again enhanced strongly when the surface is randomly charged . this holds also in the case of a finite discontinuity in the dielectric constant ( even though dielectric image charges , unlike salt images , eliminate the singularity and create a counterion - depleted zone in the immediate vicinity of the charged surface ) and/or when the multivalent counterions have a finite size ( that prevents them from probing the singular point of the single - particle interaction energy on the charged surface ) . the charge disorder can thus make the overcharging and/or charge inversion of the mean surface charge highly pronounced . our results are presented in terms of rescaled ( dimensionless ) parameters such as the rescaled screening parameter and the electrostatic and disorder coupling parameters , which can be mapped to a wide range of values for counterion and salt bulk concentration , mean surface charge density , counterion valency , etc . a few examples of the actual values for these latter quantities ( corresponding to the typical values of the rescaled parameters that were used to plot the figures in the previous sections ) are shown in table [ table ] . note that other sets of actual parameter values than those given in the table ( e.g. , using divalent and trivalent counterions ) are just as conceivable , as long as they correspond to the same set of dimensionless parameters . the typical values of the disorder coupling parameter that we used in our study , e.g. , @xmath184 , correspond to a relatively small degree of charge disorder on the surface @xmath185nm@xmath186 . assuming that the disorder originates from a quenched , random distribution of positive and negative impurity charges , @xmath187 , residing on the surface with a surface density of @xmath188 , we find @xmath189 @xcite . therefore , the above - mentioned values of @xmath190 can be obtained by relatively small densities @xmath191nm@xmath186 of impurity charges as compared with the mean number of surface charges ( typically @xmath192nm@xmath186 ) and can be thus easily realized in actual systems . hence , we conclude that the effects due to charge randomness , even at such small amounts , can be quite significant ! we should emphasize that our results are valid strictly in the case of highly asymmetric coulomb fluids , where the dressed multivalent - ion approach can be justified @xcite . the dressed multivalent - ion theory , that was implemented here , follows as a limiting single - particle theory from the virial expansion of the partition function up to the leading order in the fugacity of multivalent counterions and , as such , is expected to be applicable in two distinct regimes @xcite : ( i ) when the electrostatic interactions are strong enough giving rise , on the leading order , to a strong - coupling , single - particle behavior for multivalent counterions next to an oppositely charged boundary ( typically at low salt concentrations or in counterion - only systems ) @xcite , and ( ii ) when multi - particle interactions between counterions are sufficiently weak due , e.g. , to high salt screening effects , allowing again for a single - particle description ( typically at moderate to high salt concentrations ) @xcite . this analytical approach is thus expected to be valid only at relatively small bulk concentrations of multivalent counterions around , for instance , just a few mm , which is in fact often the case in experiments ( see , e.g. , refs . @xcite ) . the dressed multivalent - ion theory has been tested extensively against implicit- and explicit - ion simulations @xcite and turns out to have a wide range of validity in the parameter space when the surfaces bear uniform charge distributions . similar simulations are still missing in the case of randomly charged surfaces with multivalent ions mostly because of a significantly large increase in the computational time , which would be required in oder to produce reliable quenched disorder averages . our results , however , produce concrete predictions that can be tested against simulations . the fingerprints of charge disorder are expected to show up in appropriately designed experiments as well @xcite , although one should note that experiments on systems containing solutions of multivalent ions face certain difficulties as is , for instance , the case @xcite in electrophoresis measurements conducted to show the charge inversion effect ( see refs . @xcite for other recent methods such as streaming currents or atomic force microscopy measurements ) . in general , we expect that the previously determined regimes of validity of the dressed multivalent - ion theory @xcite roughly hold also for the present case with disordered surfaces . one particular case that should be treated with caution in systems containing added monovalent salt is the situation where the mean electrostatic potential near the randomly charged surface becomes large , e.g. , when the disorder strength is very large and/or the dielectric discontinuity parameter is small , in which case the validity of the underlying dh approximation used for the monovalent ions can break down @xcite . another case that goes beyond the present approach is the situation where nonlinear charge renormalization and/or bjerrum pairing effects become relevant ( see , e.g. , refs . @xcite ) ; however , these effects turn out to be negligible in the regime of parameters that is of concern here @xcite . also , while we expect that the predicted boundaries of the parameter space pertaining to the onset of the charge inversion and/or overcharging would be relatively accurate , the single - particle approximation that lies at the heart of the dressed multivalent - ion description is not expected to be adequate in the regime of parameters deeply within the regime of charge inversion and/or overcharging due to non - negligible many - body contributions @xcite . these considerations and the role of other possible factors such as higher - order virial corrections @xcite , the discrete nature and the finite size of monovalent salt ions @xcite and the ion - ion excluded - volume repulsions @xcite , etc , that are expected to become relevant especially at intermediate electrostatic couplings and/or within the regime of charge inversion / overcharging , remain to be assessed further in future simulations . our model is based on a few simplifying assumptions and , as such , neglects several other factors including solvent structure ( see , e.g. , refs . @xcite and references therein ) , the polarizability of mobile ions ( see , e.g. , refs . @xcite and references therein ) , specific surface ion - adsorption effects @xcite , etc . we have also neglected the internal structure of counterions that can introduce higher - order multipolar effects ( see , e.g. , refs . @xcite and references therein ) ; these effects can be relevant especially for multivalent counterions that have an extended structure such as rod - like polyamines like the trivalent spermidine and tetravalent spermine with chain lengths of up to 1 - 1.5 nm @xcite . on the other hand , we have assumed that the charge disorder is distributed according to a gaussian weight and that it is uncorrelated in space . spatial correlations can be included in our formalism in a straightforward manner @xcite and will be considered in future works . it is important to note that the precise statistical characteristics of charge disorder in real systems can be highly sample and material dependent , involving also the method of preparation , features that should all be considered if the theoretical findings are to be compared with experiments . furthermore , annealed as opposed to quenched disordered surfaces , containing mobile surface charges that are in thermal equilibrium with the rest of the system @xcite , as well as surfaces containing partially quenched or partially annealed charge distributions @xcite , and also charge regulating surfaces @xcite constitute other interesting examples that can be studied in the present context . all of these additional features we plan to address in the future . another interesting problem , which is closely related to the present work and can be studied using similar methods , is the strong - coupling interaction between randomly charged surfaces immersed in an asymmetric coulomb fluid @xcite . it is also worth mentioning that some of the key findings in the present study , such as the singular behavior of the density profile of multivalent ions , remain valid even in the case of _ net - neutral _ surfaces that carry no mean charge density but only a finite charge disorder variance . the case of net - neutral surfaces has been studied recently in a series of works in the context of casimir interactions @xcite and the role of an asymmetric coulomb fluid in this case will be discussed elsewhere @xcite . a.n . acknowledges partial support from the royal society , the royal academy of engineering , and the british academy ( uk ) . acknowledges support from the school of physics , institute for research in fundamental sciences ( ipm ) , where she stayed as a short - term visiting researcher during the completion of this work . r.p . acknowledges support from the arrs through grants no . p1 - 0055 and j1 - 4297 . we thank m. kandu and e. mahgerefteh for useful comments and discussions . kim , m. brown - hayes , d.a.r . dalvit , j.h . brownell and r. onofrio , phys . a * 78 * , 020101(r ) ( 2008 ) ; * 79 * , 026102 ( 2009 ) ; r.s . decca , e. fischbach , g.l . klimchitskaya , d.e . krause , d. lpez , u. mohideen and v.m . mostepanenko , phys . a * 79 * , 026101 ( 2009 ) ; s. de man , k. heeck and d. iannuzzi , phys . a * 79 * , 024102 ( 2009 ) . lukatsky , s.a . safran , a.w.c . lau and p.a . pincus , europhys . lett . * 58 * , 785 ( 2002 ) . henle , c.d . santangelo , d.m . patel and p.a . pincus , europhys . lett . * 66 * , 284 ( 2004 ) . w. pezeshkian , n. nikoofard , d. norouzi , f. mohammad - rafiee and h. fazli , phys . e * 85 * , 061925 ( 2012 ) .

experimental tests are in progress to evaluate the accuracy of the modeled iron opacity at solar interior conditions , in particular to better constrain the solar abundance problem [ s. basu and h.m . antia , physics reports 457 , 217 ( 2008 ) ] . here we describe measurements addressing three of the key requirements for reliable opacity experiments : control of sample conditions , independent sample condition diagnostics , and verification of sample condition uniformity . the opacity samples consist of iron / magnesium layers tamped by plastic . by changing the plastic thicknesses , we have controlled the iron plasma conditions to reach i )=167 ev and , ii )=170 ev and , and iii )=196 ev and , which were measured by magnesium tracer k - shell spectroscopy . the opacity sample non - uniformity was directly measured by a separate experiment where al is mixed into the side of the sample facing the radiation source and mg into the other side . the iron condition was confirmed to be uniform within their measurement uncertainties by al and mg k - shell spectroscopy . the conditions are suitable for testing opacity calculations needed for modeling the solar interior , other stars , and high energy density plasmas .

1405.4012

opacity quantifies photon absorption in matter and plays a crucial role in many high energy density ( he d ) plasmas , including inertial fusion plasmas and stellar interiors . modeling opacity is especially challenging when the plasma is composed of partially - ionized atoms with many bound electrons . first , one must accurately compute atomic data ( e.g. , energy level structures , photoionization cross - sections , and oscillator strengths ) ; then , atomic level populations have to be computed for the given conditions including a sufficiently - complete set of atomic levels ; and finally , line shapes and spectral formation must be taken into account in order to compute the frequency - dependent opacity . the complexity of such models makes experimental tests critical . however , the range of plasma conditions where opacity models are experimentally validated is extremely limited because opacity measurement requirements are also challenging . opacity measurements at he d conditions , such as those found in stellar interiors and inertial fusion ablators , are particularly rare@xcite . without accurate opacity measurements , the systematic uncertainties of modeled opacities are not known . this limits our understanding of plasmas and sometimes complicates the interpretation of plasma hydrostatic / hydrodynamic simulations when they disagree with measurements or observations . around 1980 , it was reported that simulated cepheid variable stars could not reproduce the observed pulsations in certain regimes@xcite . in an attempt to resolve the discrepancies , various modifications to the cepheid models were tested , such as reducing mass , increasing helium abundance , and introducing magnetic fields . however , it turned out that the discrepancies were not caused by the cepheid models but mostly by inaccuracy of the calculated opacities used in the models . opacities re - computed using improved models were significantly larger at cepheid envelope conditions and resolved the discrepancies@xcite . calculating opacities is difficult , and this example illustrates the possible consequences of their inaccuracy . more recently , calculated solar interior opacities , especially at the base of the solar convection zone ( cz ) , are in question@xcite . standard solar models ( i.e. , hydrostatic models ) were in excellent agreement with helioseismic measurements until the solar metal abundances were reduced in early 2000s @xcite . this lowered the abundances of c , n , o , ne , and ar by 35 - 45% , which lowered the calculated solar mixture opacities accordingly . the downward shift in the opacity altered the radiative heat transfer in the solar models , affecting their agreement with helioseismic measurements , especially at the cz base ( i.e. , @xmath5)@xcite . as with the cepheid variable problem , the inaccuracy of calculated opacity has been suggested as the origin of this cz problem ; a 10 - 30% increase in opacity would bring solar models and helioseismology into agreement@xcite . until high - accuracy opacity measurements are conducted , opacity uncertainty will continue to be a potential explanation for many he d and astrophysical model - observation discrepancies . while a discrepancy often suggests how much the mean opacity should be increased ( or decreased ) , it does not provide anything conclusive on calculated opacity accuracy . are the discrepancies really caused by opacity inaccuracy ? exactly how are calculated opacities different from true opacities ? performing benchmark experiments measuring opacity is crucial to answer these questions and to guide potential opacity model refinements . such refinements would help to improve our understanding of the sun , many other stars , and a variety of other he d plasmas . opacity measurement techniques have been developed and refined over the past two decades@xcite . nevertheless , performing reliable experiments is demanding , and few high quality measurements exist . there are many sources of difficulty@xcite . first , one has to tailor the sample to the target conditions while also achieving sample uniformity , steady state , and local thermodynamic equilibrium ( lte ) . when transmission is measured through significant gradients , it is difficult to derive definitive conclusions on opacity . steady state and lte are common assumptions in opacity models , and thus have to be realized in the sample . to achieve lte , the experiment must either reach high densities or include a strong external radiation field that is reasonably close to equilibrium with the sample plasma , and these conditions must be maintained over a duration sufficient to establish steady - state ( i.e. , @xmath6 for each level population @xmath7 ) . second , once the sample is tailored to the target conditions , the sample transmission has to be accurately measured . to benchmark the detailed atomic and plasma physics in modern opacity models , it is necessary to measure the frequency - resolved transmission . this requires a backlighter that is bright , to mitigate the sample self - emission effects on the emergent transmission spectra@xcite , and spectrally smooth , to avoid unnecessary complications due to instrumental broadening@xcite . reliable frequency - resolved opacity measurement also requires spectrometers with high resolving power to observe detailed bound - bound line features . these spectrometers must be free of any artifacts such as crystal defects and film scanning flaws that could result in disturbing the data . finally , the fe plasma conditions have to be independently measured to model the fe opacity and compare it with the measurement . in 2007 , fe opacity experiments were performed to constrain the cz problem discussed earlier , and fe transmission spectra were successfully measured at charge state distributions similar to fe at the cz base for the first time@xcite . fe was studied for two reasons . first , fe is one of the biggest opacity contributors at the cz base . second , fe at the cz base has many bound electrons , making it more difficult to model than many other elements in the solar matter . uncertainty of modeled fe opacity may therefore be larger than the other elements . the fe samples were heated by the z - pinch dynamic hohlraum ( zpdh ) radiation source at the sandia national laboratories z - machine . the backlighter was provided by the stagnation of the zpdh implosion and was characterized as a smooth 314 ev blackbody radiator . this backlighter was bright enough to mitigate sample self - emission at the measured fe conditions . the fe reached an electron temperature , @xmath0 , of 156 ev and an electron density , @xmath8 , of 6.9@xmath910@xmath10 . the conditions were independently measured by k - shell transmission spectroscopy of mg mixed with the sample . fe l - shell transmission spectra computed from detailed opacity models were shown to match reasonably well with the measurements . in order to benchmark fe opacity models at the cz base , more experiments are needed . first , while the fe transmission spectra were measured at the charge state distribution similar to that of fe at the cz base , the inferred @xmath0 and @xmath8 were significantly lower than the conditions at the cz base ( i.e. , 185 ev and 9@xmath910@xmath11 ) . thus , fe opacity models need to be tested at higher @xmath0 and @xmath8 . this is important because additional effects become important as @xmath0 and @xmath8 increase . in lte , the same average ionization can be achieved by increasing both @xmath0 and @xmath8 . however , since population within each charge state is distributed following the boltzmann relation , @xmath12 where @xmath13 is the energy difference between the excited and ground states , ions produced by higher @xmath0 and @xmath8 would produce more population in the excited states . this makes the accuracy of opacity calculations more sensitive to the accuracy of the atomic data of excited levels and to the treatment of highly - excited levels ( doubly , triply , .. , multiply excited levels ) . as density increases , high density effects such as ionization potential depression and stark line broadening become more important . in particular , stark line broadening in l - shell line transitions has not been explored as extensively as that of k - shell lines . furthermore , even detailed modern opacity models employ rather simple approximations similar to the formalism discussed in griem _ _ et al.__@xcite . in order to disentangle these physical details and benchmark opacity models , it is crucial to control the plasma conditions and repeat fe opacity experiments at several different @xmath0 and @xmath8 while maintaining similar charge state distributions . such a collection of measurements would promote investigations of how the effects of higher @xmath0 and @xmath8 gradually change the opacity and how well opacity models can predict them . it is also important to make direct measurements of the sample uniformity . in bailey _ _ et al.__@xcite , the sample uniformity assertion was supported by the use of volumetric heating provided by zpdh radiation and by the fact that the mixed fe / mg transmission spectra were successfully modeled by a single @xmath0 and @xmath8 . however , it is preferable to experimentally quantify the level of the sample non - uniformity . in this article , we describe measurements addressing three of the key requirements for opacity experiments : i ) control of the fe conditions , ii ) independent plasma diagnostics , and iii ) verification of the sample condition uniformity . the fe sample conditions can be controlled by the thickness of the tamping plastic ( @xmath14 ) as suggested by hydrodynamic simulations @xcite . thus , we performed fe opacity experiments with three different @xmath14 tamper thicknesses : i ) 10 @xmath15 m , ii ) 35 @xmath15 m , and iii ) 68 @xmath15 m . the fe conditions are independently measured by mg k - shell spectroscopy and found to be i ) @xmath0=167@xmath16 ev and @xmath17 , ii ) @xmath0=170@xmath18 ev and @xmath19 , and iii ) @xmath0=196@xmath20 ev and @xmath21 , respectively . these conditions nearly reproduced the charge state distribution of fe at the cz base , but with different levels of high temperature and density effects . also , one experiment was designed and performed to directly investigate the sample non - uniformity . for this particular experiment , al was mixed into the side of the fe sample facing the radiation source and mg in the opposite side . fe conditions in both sides are inferred by al and mg k - shell spectroscopy . we confirmed that there is no measurable axial non - uniformity in the fe sample . the discussion of other requirements , such as steady - state and lte , as well as details of the transmission determination , the comparison with fe opacity models , and the implications to astrophysics , atomic physics , and he d physics are beyond the scope of this article and will be discussed elsewhere . section [ sec : z - pinch - dynamic - hohlraum ] discusses the experiments and data reduction . section [ sec : k - shell - spectroscopy - for ] explains the mg k - shell spectroscopy , how to model the spectra , and how to find the optimal @xmath0 and @xmath8 from the measured mg spectra . section [ sec : fe - condition - analysis ] summarizes fe conditions analyzed for experiments using different @xmath14 tamper thicknesses . section [ sec : uniformity - measurement ] summarizes the sample non - uniformity investigation . fe opacity experiments were performed at the sandia national laboratories z - machine . the fe sample is volumetrically heated by a z - pinch dynamic hohlraum ( zpdh ) and backlit at its stagnation . a detailed description of the z - pinch dynamic hohlraum is discussed elsewhere@xcite . figure [ fig : how_zpdh_works](a ) illustrates a cross - sectional schematic of the cylindrical zpdh initial setup . an azimuthal magnetic field is generated by the electrical current , @xmath22 , running through tungsten wires strung in a cylindrical array . the current and self - generated magnetic field produce a @xmath23 force that pushes the tungsten plasma towards the axis where a 14.5 @xmath24 cylindrical plastic ( @xmath25 ) foam is located . when the tungsten plasma collides with the foam , it generates a radiative shock in the @xmath25 [ fig . [ fig : how_zpdh_works](b ) ] . the shock radiation is trapped inside the thick tungsten plasma wall due to the high opacity of the tungsten plasma . the fe / mg sample placed on the top exit hole is heated by the hohlraum radiation , and it is backlit when the radiating shock stagnates at the z - pinch axis [ fig . [ fig : how_zpdh_works](c ) ] . cross - sectional schematics of the cylindrical z - pinch dynamic hohlraum ( zpdh ) . ( a ) initial setup of the zpdh , which consists of 360 tungsten ( w ) wires concentrically arranged at two radii with a cylindrical low - density plastic ( @xmath25 ) foam located at the z - pinch center . ( b ) when the imploding w plasma collides with the @xmath25 foam , shock - generated radiation is trapped inside the high - opacity w plasma wall and heats the sample . ( c ) at stagnation , it provides an intense backlighter along the cylindrical axis . , width=321 ] figure [ fig : zpdh_radiation ] illustrates the difference in the zpdh radiation seen by the sample and our spectrometers . while any point in the fe / mg sample sees the entire zpdh emitting surface , the detector located @xmath264100 mm away from the radiation source only sees the central region of its surface because the detector view is limited by an aperture and @xmath2650 @xmath15 m slits , which are located @xmath2617 mm and @xmath262000 mm away from the source , respectively . during a zpdh implosion , the radiation intensity increases while the area of the bright region decreases . the radiation power from the zpdh exit hole is proportional to intensity times emitting surface area ( @xmath27 ) and thus changes slowly compared to the change in the implosion itself . the heating radiation is energetic : its energy distribution maximum exceeds 700 ev . the fe sample is almost transparent to this radiation , and is therefore volumetrically heated@xcite . as the implosion continues , the heating radiation power begins to drop as the decrease in emitting area exceeds the increase in intensity . a few nanoseconds after the heating radiation peak , the radiative shock reaches the z - pinch axis , generating extremely bright radiation that provides the backlighter for the transmission measurements [ fig . [ fig : zpdh_radiation](b ) ] . while the data are recorded by a time - integrated spectrometer , time resolution is provided by the few - nanosecond duration of the backlighter . schematics showing the zpdh radiation emitting surfaces and the view factors from the samples and the detectors at ( a ) the time of the heating radiation peak and ( b ) the time of the backlighter radiation peak . since each point on the sample sees the entire emitting surface area , the sample experiences maximum radiation ( i.e. , @xmath28 ) before stagnation . the detector records most of the radiation during stagnation because its view is limited to the central region by an aperture and 50 @xmath15 m slits . the distances specified for the various components are measured from the x - ray source . , width=321 ] the zpdh backlighter is attenuated by the fe / mg sample , and the spectrum is recorded with time - integrated potassium acid phthalate convex curved crystal spectrometers equipped with kodak rar2492 x - ray film . the spectrometers have between four and six slits @xmath262000 mm away from the sample and the backlighter , making the magnification @xmath261 . the typical slit width is @xmath2650 @xmath15 m , which provides a spatial resolution of @xmath26100 @xmath15 m . the crystal curvature radius is 4 or 6 inches , and an x - ray film is located at 8 cm from the crystal . the spectral dispersion axis is determined by ray tracing using prominent lines as wavelength references . film optical densities are converted to exposure based on henke _ et al . filter transmissions and crystal reflectivity are corrected based on the data available from the center for x - ray optics with the instrument geometry effects taken into account . recording multiple slit images has several advantages . first , one can increase the signal - to - noise ratio of the data without giving up the instrumental spatial resolution . second , a single image could be flawed by systematic issues such as crystal defects , slit width variations along its length , or reflectivity variations over the crystal surface . however , these systematic issues are random across the different slit images and can be mitigated by averaging over multiple slit images . for example , fig . [ fig : data_reduction_average_image ] ( a ) and ( b ) show two of the six individual slit images of fe / mg absorption measurements from a single experiment ( z2364 ) and contain crystal defects at random locations . when averaged over the six slit images [ fig . [ fig : data_reduction_average_image ] ( c ) ] , those artifacts have been greatly reduced . fe / mg absorption images over the mg k - shell spectral range . strong line features are from mg k - shell bound - bound line absorptions such as mg he-@xmath29 , ly-@xmath29 , he-@xmath30 , he-@xmath31 , and ly-@xmath30 . ( a ) and ( b ) show two single slit images from z2364 , which contain many crystal defects . ( c ) shows the image averaged over six slit images of the same experiment . averaging improves the signal - to - noise ratio and helps to mitigate any bias towards individual slit images due to crystal defects or other instrumental problems . the spatial range of interest is from 100 @xmath32 to 400 @xmath32 where the backlighter is bright and the fe / mg exists in the sample . in this experiment , there is no fe / mg in the bottom half of the sample so that we have the option to extract transmission in a single experiment by using the fe / mg attenuated spectrum from the top of the image and the unattenuated spectrum from the bottom@xcite . , width=321 ] the averaged spectra can be extracted either from the averaged slit image or by averaging the spectra extracted from individual slit images . these two methods are identical as long as the crystal defects are treated in the same way . the uncertainties in the spectra are estimated by taking the standard deviation of the spectra extracted from n individual slit images divided by @xmath33 . an example is shown in fig . [ fig : six - spectra - extracted ] . the black and the red curves are , respectively , the averaged spectrum and its uncertainty , which for this data set is approximately @xmath34 . fe / mg absorption spectra from a 300 @xmath32 wide region in fig . [ fig : data_reduction_average_image ] . prominent line features are mg bound - bound line absorptions such as mg he-@xmath29 , ly-@xmath29 , he-@xmath30 , he-@xmath31 , and ly-@xmath30 . six spectra in different colors are extracted from individual slit images . the averaged spectrum is overlaid in black , and its uncertainty is shown at the bottom in red . the average percent uncertainty is @xmath12.5% . , width=321 ] fe conditions have to be measured independently in order to compare the measured fe transmission with the modeled transmission . fe conditions predicted by hydrodynamic simulations are not appropriate for this purpose for two reasons . first , we are investigating fe opacity because the opacity model accuracy has been called into question . since the sample is heated by radiation in our experiments , the fe opacity plays a key role in the plasma evolution . if the fe opacity is in question , we should not expect the hyrdosimulation to predict plasma conditions accurately . second , hydrodynamic simulations require many input parameters . providing all of the required information for every experiment is itself a challenge . propagating all of the input uncertainties to the fe conditions uncertainties is even more challenging . thus , we measure fe conditions directly through k - shell spectroscopy of mg mixed into the fe sample . while l - shell opacity and spectra are not completely understood , particularly at high temperatures and densities , k - shell spectra from h- , he- , and li - like ions have been extensively researched and used to diagnose plasma conditions@xcite . due to the small number of bound electrons , the required atomic data for singly and multiply excited states can be calculated with high accuracy and fine structure detail . additionally , many of the atomic data have been experimentally validated@xcite . k - shell line shapes for h- , he- , and li - like ions have also been extensively investigated over the last 50 years@xcite and are understood much better than l - shell line shapes . different k - shell spectral models would infer slightly different @xmath0 and @xmath8 due to differences in details such as stark line shape calculation , continuum lowering , atomic data , and numerical approach . the fe condition uncertainties due to the spectral model details are under investigation and will be discussed elsewhere . line emission from k - shell ions is sensitive to temperature and density . the bound - bound ( bb ) line transmission is defined as follows : @xmath35 @xmath36 where @xmath37 is the line optical depth at frequency @xmath38 due to photoexcitation from a lower level @xmath39 to an upper level @xmath40 , @xmath41 is the mg areal number density measured by rutherford backscattering , @xmath42 is the oscillator strength of the transition , @xmath43 is the line shape associated with the transition , and @xmath44 is the fractional population in the lower state of the transition . when the oscillator strengths do not depend on plasma conditions , the only unknowns are @xmath43 and @xmath44 . the line shape , @xmath43 , is determined by doppler broadening , natural broadening , and stark broadening . we also estimated the possible zeeman effect on the line shape based on an upper limit magnetic field in our experiments . the effect was confirmed to be much smaller than the instrumental spectral resolution and thus neglected in our line shape calculations . at the density of interest here , the dominant line broadening mechanism is stark broadening , which is sensitive to electron density . figure [ fig : density_from_stark_broadening ] shows area normalized line shapes for mg he-@xmath31 ( @xmath45 ) and mg ly-@xmath30 ( @xmath46 ) calculated at three different conditions using a detailed stark line shape calculation code , merl@xcite . one can clearly see that , as electron density increases , the line shapes become broader . this is more significant for the transitions involving high principal quantum numbers , and thus , in this article , the transitions with upper quantum number greater than 2 ( e.g. , @xmath30 , @xmath31 , @xmath47 ) are used to extract @xmath8 . area - normalized detailed stark line shapes calculated at @xmath8=@xmath48 , @xmath49 , @xmath50 for ( a ) mg he-@xmath31 and ( b ) mg ly-@xmath30 . as density increases , the line shapes become broader . , width=321 ] to diagnose temperature , we use line ratios that reflect the @xmath0-dependent ionization balance . the ratio of optical depths integrated over line shapes with @xmath51 is : @xmath52 where @xmath53 and @xmath54 are the oscillator strengths of ly-@xmath30 and he-@xmath31 line transitions , respectively . since the h - like ground state population and he - like ground state population are directly affected by charge state distributions , their ratio is a strong temperature diagnostic at a given density . figure [ fig : te_from_line_ratios ] shows mg he-@xmath31 and ly-@xmath30 line transmissions computed at @xmath0=160 , 165 , and 170 ev with a fixed electron density @xmath8=@xmath55 using a collisional radiative model , prismspect in lte mode@xcite . as the temperature increases from 160 to 170 ev , the he - like ground state population , @xmath56 , decreases by 4% ( [email protected] ) while the h - like ground state population , @xmath58 , increases by about a factor of two ( [email protected] ) , monotonically increasing the ratio of these initial states , @xmath59 , from @xmath60 to @xmath61 . thus , from eq . ( [ eq : line_ratio ] ) , one can see that relative line strengths from adjacent charge states are strongly dependent on the electron temperature . due to the monotonic relationships between line shapes and @xmath8 and between line ratios and @xmath0 , the measured mg lines constrain the @xmath0 and @xmath8 of the mg embedded region . since mg is mixed throughout the fe sample , the fe plasma conditions can be inferred from the mg k - shell lines . mg he-@xmath31 and ly-@xmath30 transmissions at @xmath0=160 , 165 , and 170 ev with a fixed electron density @xmath8=@xmath62 . as temperature increases , the line ratio changes due to the change in ionization balance . , width=321 ] in the present analysis , mg spectra are computed by radiator , which is a framework to couple a tabulated emissivity and opacity database with a radiation transport solver and compute emergent spectra by taking into account a given instrumental broadening@xcite . this framework is combined with a multi - objective global optimization program , galm@xcite , which is based on a genetic algorithm ( ga ) followed by levenberg - marquardt non - linear least squares minimization ( lm)@xcite . the combination of the ga and lm is an efficient optimization algorithm that finds the solution quickly by exploring small fractions of the parameter space . to use radiator+galm , two things have to be done : i ) create a tabulated emissivity and opacity for mg and ii ) add a radiation transport solver for slab geometry to radiator . the local emissivity and opacity database for mg is computed by prismspect in lte mode together with the detailed stark line shape database computed by merl@xcite . the details on how to extract local emissivity and opacity out of the prismspect calculation are discussed in appendix a. a general transport radiation solver is developed for slab geometry that can take into account mixtures , linear gradients , background , and self - emission options as needed . while the simple pure transmission approximation is sufficient to extract the sample conditions , the general formalism is useful to investigate various effects on the emergent spectra for many applications and thus is discussed in detail in appendix b. the formalism is not exclusive to prismspect and can be used with other collisional radiative models@xcite . the pure transmission approximation assumes that the backlighter is much brighter than the plasma self - emission and is a slowly changing function of frequency over the instrumental broadening width ( appendix b ) : @xmath63 @xmath64 where the first equation convolves the computed transmission with the instrumental broadening , @xmath65 , and the second equation is a transmission calculation using a tabulated mg opacity database computed by prismspect ( psdb ) . @xmath41 is the mg areal number density in @xmath66 , and @xmath67 is the tabulated frequency - dependent mg fractional absorption coefficient in @xmath68 . @xmath69 contains bound - bound ( bb ) , bound - free ( bf ) , and free - free ( ff ) contributions : @xmath70 which includes , @xmath71 the merl database used in prismspect contains detailed line shapes for mg he-@xmath29 , @xmath30 , @xmath31 , @xmath47 , and @xmath72 , and mg ly-@xmath29 and @xmath30 . the ion microfields are computed by apex@xcite assuming an fe : mg mixture of 1:1 . in the temperature range of interest for this work , ion dynamic effects are assumed to be negligible and omitted from the calculation . both temperature and density sensitivities are encoded in @xmath69 via the detailed line shape , @xmath73 , and the lower population , @xmath44 , in the @xmath74 contribution . the areal number density @xmath41 controls the line depths . radiator+galm optimizes @xmath0 , @xmath8 , and @xmath41 that best reproduce the measured mg spectra . while the condition sensitivities exist in @xmath44 and @xmath75 of the optical depths @xmath76 , what we measure is transmission @xmath77 convolved with instrumental broadening . one can extract @xmath76 out of @xmath77 perfectly if and only if the instrumental spectral resolving power is infinite , which is never the case for experiments . the spectral resolving power of our instrument was measured to be @xmath78@xmath79@xcite . this finite spectral resolution obscures @xmath0 and @xmath8 sensitivities embedded in the transmission@xcite , and lines with higher optical depths lose the sensitivities more significantly by this effect due to the non - linear relationship between @xmath76 and @xmath77 [ eq . ( [ eq : transmission ] ) ] . the rest of this section synthetically illustrates how our instrumental spectral resolution affects the measured lines , and determines what mg lines are best used for the fe plasma @xmath0 and @xmath8 analysis . the blue curves in fig . [ fig : weak_lines_should_be_fitted ] ( a ) and ( b ) are the calculated optical depths , @xmath76 , and transmissions , @xmath77 , respectively , for mg he-@xmath29 ( @xmath80 ) , he-@xmath30 ( @xmath81 ) , and he-@xmath31 ( @xmath82 ) at @xmath0=163 ev , @xmath8=@xmath83 , and @xmath84 . these are typical conditions achieved in our experiments . while these lines share the same initial state population @xmath56 in eq . ( [ eq : optical_depth ] ) , their optical depths are very different due to the differences in their oscillator strengths and line shapes . when the modeled transmissions are convolved with the instrumental spectral shape , transmissions at the line centers are overestimated , and all the detailed structures are smoothed out as shown in the red curves in fig . [ fig : weak_lines_should_be_fitted](b ) . one then can not successfully recover the calculated optical depths when converting from these convolved transmissions , reducing our ability to accurately diagnose @xmath0 and @xmath8 . red lines in fig . [ fig : weak_lines_should_be_fitted](a ) show optical depths converted from transmission with the instrumental spectral resolution effect . one can clearly see that the saturation due to the instrumental broadening affects the stronger lines more . while the he-@xmath31 line preserves both the shape and the strength , the he-@xmath29 and @xmath30 lines are heavily altered . the saturation in the he-@xmath29 is particularly severe and the red line is barely visible in fig . when optical depths are larger than 1 , the apparent line shape and strength are strongly affected by instrumental broadening . thus , in this article , we decided to analyze weaker mg lines ( @xmath851 ) for the purpose of the iron plasma @xmath0 and @xmath8 diagnostics . ( a ) optical depths , @xmath76 and ( b ) transmission , @xmath77 , of he-@xmath29 , @xmath30 , @xmath31 calculated at @xmath0=163 ev and @xmath8=@xmath83 with @xmath86 . blue and red curves are before and after the instrumental broadening effects , respectively . , width=321 ] transmission spectra can be extracted by dividing fe / mg absorption spectra by unattenuated backlighter spectra . however , fitting the modeled mg transmission spectra directly to the measured transmission spectra is difficult because the measured transmission has additional fe attenuation in the mg b - b line spectral region . figure [ fig : b - b transmission](a ) shows fe / mg mixture ( red ) and pure mg ( blue ) synthetic transmission spectra computed at the same conditions . because of the fe attenuation , fe / mg mixture transmission is lower than pure mg transmission . also , while fe attenuation in this spectral range is smooth and slowly varying with photon energy , there are smooth b - f contributions from mg above 1700 ev . it is difficult to objectively separate mg b - f from the fe b - f in the measured data . fitting the modeled fe / mg transmission spectra to the measured transmission spectra is not ideal because we do not want to rely on calculated fe opacity to diagnose the fe conditions . since plasma condition sensitivity mostly comes from mg b - b lines as in eq . [ eq : transmission ] , one strategy is to extract mg b - b line transmission spectra both from the measured fe / mg spectra and the modeled mg transmission spectra , and then compare them . to this end , we first remove mg b - b lines and then replace them with straight lines to define baselines . by dividing the spectra by the baselines , one can extract mg b - b line transmission spectra . this removes not only fe b - f but also mg b - f from the spectra . for the modeled mg spectra , we could compute the b - b line transmission for specific lines of interest as in eq . [ eq : transmission ] . however , the details of composite spectral formation such as overlapping lines , satellite line contributions , and continuum lowering could affect the instrumental broadening , the baseline determination , and the emergent spectral line shapes . thus , complete transmission spectra are computed first , and then the baselines are defined to extract b - b line transmission spectra in the same way as for the data . figures [ fig : b - b transmission](b ) and ( c ) illustrate how baselines are defined for the synthetic data in fig . [ fig : b - b transmission](a ) and how the resultant mg b - b line transmission spectra agree between fe / mg and mg spectra . the red and blue dashed lines in fig . [ fig : b - b transmission](b ) are the baselines defined for fe / mg and pure mg spectra , respectively . by dividing the spectra by the baselines , mg b - b line transmission spectra are extracted [ fig . [ fig : b - b transmission](c ) ] . this study shows that mg b - b spectra extracted from fe / mg spectra are identical to those from pure mg spectra . thus , fe conditions can be inferred by fitting modeled spectra to the measured spectra in mg b - b line transmissions . there are two comments on mg b - b line transmission analysis . one is on the fe / mg mixture effects on the emergent spectra . in this article , we assume lte . in lte plasmas , level populations are fully determined by @xmath0 and @xmath8 regardless of whether the plasma is a mixture or not , and mixture effects will only appear in the line shapes via ion microfields contributed from both fe and mg ions . as long as mixture effects are included in the mg stark line shapes calculation , mg b - b lines extracted from fe / mg spectra and from pure mg spectra should be identical [ fig . [ fig : b - b transmission](c ) ] . another comment is on condition uncertainty due to the baseline determination from the measured spectra . this is a concern because there is noise in the measured spectra . however , baselines are determined based on many data points on both sides of b - b lines , and thus the baseline uncertainty is smaller than the uncertainty of any of those data points . the uncertainty due to the baseline determination is included in the uncertainty due to experiment - to - experiment variation in the following sections . this figure compares calculated transmission spectra from an fe / mg mixture ( red ) and from pure mg ( blue ) . ( a ) fe / mg transmission is lower than mg transmission due to additional attenuation by fe b - f but does not have much fe b - b contribution except @xmath871450 ev . ( b ) blow - up of the dashed region of ( a ) . the red and blue dashed lines are baselines defined by replacing the apparent mg b - b lines by straight lines . ( c ) mg b - b line transmission spectra defined by dividing transmission spectra [ solid curves in ( b ) ] by the baselines [ dashed lines in ( b ) ] . , width=321 ] in order to disentangle the complex physical processes included in opacity models and benchmark those models , it is important to measure fe opacity at different conditions , but with similar charge state distributions . used lasnex 2d@xcite hydrodynamic simulations to explore a way to control fe sample conditions by changing the target configuration@xcite . they predicted that the fe sample @xmath8 could be controlled by changing the rear @xmath14 tamper thickness . adding more tamping mass on the back slows the expansion speed and maintains higher density at the time of the backlight . the slower expansion would also produce modestly higher temperatures . based on their suggestions , fe opacity experiments were performed with three different rear @xmath14 tamper thicknesses as shown in fig . [ fig : target_designs ] : ( a ) two experiments with 10 @xmath15 m , ( b ) one experiment with 35 @xmath15 m , and ( c ) six experiments with 68 @xmath15 m . by increasing this rear @xmath14 tamper thickness , the backlighter radiation is attenuated more . in order to minimize this extra attenuation , we reduced the front @xmath14 thickness from 10 @xmath15 m to 2 @xmath15 m for ( b ) and ( c ) assuming that fe / mg samples do not expand downward due to the pressure provided by the zpdh @xmath25 foam plasma . in this section , we summarize the fe sample conditions for each target configuration , which are analyzed by mg k - shell line transmission spectroscopy . figure [ fig : whole_fe_mg_spectra ] ( a ) , ( b ) , and ( c ) show the fe / mg absorption spectra recorded from the different target configurations shown in fig . [ fig : target_designs ] ( a ) , ( b ) , and ( c ) , respectively . one can qualitatively observe that mg k - shell lines become broader as the rear @xmath14 tamper thickness increases , which indicates that @xmath8 becomes higher with increasing tamper thickness . for example , he-@xmath47 becomes broader from fig . [ fig : whole_fe_mg_spectra](a ) to ( b ) , and becomes even broader and almost merged into the continuum in ( c ) . also , he - like satellite lines of ly-@xmath29 are not visible in ( a ) , become visible in ( b ) , and even more prominent in ( c ) . these transitions are @xmath88 or @xmath89 and start from excited levels . this is proof that , as going from ( a ) to ( c ) , there is a larger fraction of the population in excited levels and a sign that both @xmath0 and @xmath8 are higher . this means that the plasma must be hot and dense enough for the collisional excitation rates to be comparable to the spontaneous radiative decay rates . the actual condition of each sample is quantitatively analyzed based on the method discussed in sec . [ sec : k - shell - spectroscopy - for ] . fe / mg samples are tamped by three different thicknesses of plastic ( @xmath14 ) . ( a ) rear : 10 @xmath15 m , front : 10 @xmath15 m ( b ) rear : 35 @xmath15 m , front : 2 @xmath15 m , ( c ) rear : 68 @xmath15 m , front : 2 @xmath15 m . based on the hydrosimulation , thicker rear @xmath14 tamper would reach higher @xmath0 and @xmath8 in fe / mg plasmas . the pictures are not to scale , but exaggerated to illustrate the different tamper thicknesses . , width=321 ] fe / mg spectra recorded through the samples with rear @xmath14 tamper thicknesses of ( a ) 10 @xmath15 m , ( b ) 35 @xmath15 m , ( c ) 68 @xmath15m.,width=321 ] first , we analyzed fe conditions using the thin @xmath14 tamper target [ fig . [ fig : target_designs ] ( a ) ] , where the fe / mg sample is sandwiched by 10 @xmath90 thick plastic . figure [ fig : whole_fe_mg_spectra ] ( a ) shows the fe / mg absorption spectrum from z2301 ( atomic ratio of fe : mg=0.5:1 ) . the mg lines with @xmath91 are he-@xmath31 , he-@xmath47 , and ly-@xmath30 . optical depths for ly-@xmath29 and he-@xmath30 are approximately 4 and 6 , respectively , and excluded from the analysis . thus , the analysis focuses on he-@xmath31 , he-@xmath47 , and ly-@xmath30 , which are simultaneously analyzed with radiator+galm . the inferred conditions are @xmath92 ev and @xmath93 . the individual fits are shown in fig . [ fig : thin_ch_results ] . another experiment , z2221 , used the same @xmath14 configuration and fe : mg=0.9:1 . the simultaneous analysis of its mg he-@xmath31 , he-@xmath47 , and ly-@xmath30 infers @xmath94 ev and @xmath95 , in agreement with z2301 within uncertainties . the means and the standard deviations from these two experiments are @xmath96 ev and @xmath97 where these standard deviations indicate the experiment - to - experiment variation in the fe conditions . the mean measurement uncertainties are @xmath98 ev and @xmath99 . total uncertainties are computed by adding the mean measurement uncertainties to the experimental variation in quadrature : @xmath100 ev and @xmath101 . thus , the fe conditions inferred from the thin rear @xmath14 tamper target are @xmath102 ev and @xmath17 . this density result is consistent with the density reported in 2007 ; however , this temperature is significantly higher than the corresponding temperature , @xmath103 ev @xcite . one reason is that the data reported in 2007 were recorded before the z - machine refurbishment of that same year , which increased the electrical power delivered to the load . this difference is evidence that the refurbished z - machine produces higher radiation power , therefore reaching higher temperatures in the fe sample . model fits ( blue ) to the mg he-@xmath31 , he-@xmath47 , and ly-@xmath30 lines recorded from experiment 2301 ( red ) . the inferred conditions are @xmath92 ev and @xmath93.,width=321 ] we performed only one experiment , z2363 , using the 35 @xmath15 m rear @xmath14 tamper target , whose spectra are shown in fig . [ fig : whole_fe_mg_spectra](b ) . the fe : mg atomic ratio is 0.9:1 . the lines with @xmath91 are mg he-@xmath31 and ly-@xmath30 . these lines are simultaneously analyzed , and the fits are given in fig . [ fig : fits_intermed_ch](a ) and ( b ) , respectively . the inferred conditions are @xmath104 ev and @xmath105 . we confirmed a slight increase in @xmath0 and more than a factor of two increase in @xmath8 compared to the conditions achieved by the 10 @xmath15 m rear @xmath14 tamper target . since there is only one experiment from this target configuration , the uncertainties do not include experiment - to - experiment sample conditions reproducibility . mg he-@xmath31 and ly-@xmath30 are simultaneously fit for experiment z2363 . the best fit is provided by @xmath106 ev and @xmath107 . , width=321 ] we performed six experiments using the 68 @xmath15 m rear @xmath14 tamped target illustrated in fig . [ fig : target_designs](c ) . mg lines with @xmath1081 are he-@xmath31 , ly-@xmath30 , and the he - like satellite lines of ly-@xmath29 . thus , these lines from z2364 [ i.e. , fig . [ fig : whole_fe_mg_spectra](c ) ] are simultaneously analyzed using radiator+galm . the fits to these lines are shown in the fig . [ fig : thick_ch_results ] , and the inferred conditions are @xmath109 ev and @xmath110 . while the @xmath14 configurations are the same for the six experiments , fe areal number densities are different to test the reliability of the measured fe opacity by applying beer s law@xcite . figure [ fig : thick_ch_all_results ] shows the conditions inferred for each of the six experiments in ( a ) @xmath0 and ( b ) @xmath8 . experiments z2364 and z2366 were performed with the thinnest fe samples , whose areal densities were @xmath111 ( atomic ratio of fe : mg=0.5:1 ) . experiments z2242 and z2270 were performed with intermediate fe thicknesses with areal number densities of @xmath112 ( fe : mg=1:1 ) . experiments z2244 and z2267 were performed with the thickest fe samples with areal number densities of @xmath113 ( fe : mg=2.3:1 ) . we did not observe any correlation between fe thickness and the inferred fe conditions . the average and standard deviation are indicated by blue solid and dashed lines , respectively : @xmath114 ev and @xmath115 . based on the standard deviations , the fe condition reproducibility is @xmath13% in @xmath0 and @xmath113% in @xmath8 . the mean individual measurement uncertainties are 3 ev and 0.6@xmath116 . the total @xmath0 and @xmath8 uncertainties are computed by adding the experiment - to - experiment variations and the mean individual measurement uncertainties in quadrature , which are 6 ev ( @xmath13% ) and @xmath117 ( @xmath121% ) , respectively . fits to mg he - like satellites , he-@xmath31 , and ly-@xmath30 recorded on experiment z2364 . the inferred conditions are @xmath109 ev and @xmath110.,width=321 ] fe condition analysis result summary for thick rear @xmath14 tamper targets . the average and standard deviation are indicated by blue solid and dashed lines , respectively : @xmath114 ev and @xmath115 . the experiment - to - experiment variations of @xmath0 and @xmath8 over the six experiments are @xmath13% and @xmath113% , respectively . , width=321 ] uniformity of the fe sample is asserted based on volumetric heating provided by the powerful zpdh radiation . in sec . [ sec : fe - condition - analysis ] , similar fe conditions are inferred from different fe thicknesses , which also supports the sample axial uniformity . however , since volumetric heating is never perfect and heating radiation is supplied from one side , it is worthwhile to examine the axial uniformity assumption with more explicit experimental evidence . to this end , an experiment was designed to investigate the sample axial non - uniformity . instead of mixing mg throughout the fe sample , mg is mixed in the observer ( rear ) side of fe and another dopant , al ( z=13 ) , is mixed in the radiation source ( front ) side of the fe ( fig . [ fig : uniformity_measurement_design ] ) , which are separated by a pure fe region . one can infer fe conditions in the radiation source side and in the observer side by al and mg k - shell spectroscopy , respectively . spectrometers on axis and at 9@xmath118 sample transmission spectra from slightly different spatial centers . the analyses of the measured spectra would reveal not only axial non - uniformity but also a lateral gradient if it were significant . , width=321 ] comparison of line transmission spectra recorded along the vertical axis ( at 0@xmath118 ) and at 9@xmath118 . the good agreement indicates lateral gradients are insignificant.,width=321 ] the resultant fe / al / mg ( atomic ratio of 2:1:1 ) absorption spectra were recorded along two different lines of sight ; one along the vertical axis ( at 0@xmath118 ) and the other at 9@xmath118 from the vertical axis as in fig . [ fig : uniformity_measurement_design ] . the distance between the sample and the backlighter is measured on similar experiments , based on parallax , and found to be [email protected] mm . due to this source - to - sample distance , when the spectrometers observe the backlighter through the fe / al / mg sample , they see through slightly different spatial regions in the sample , which are separated by 232@xmath136 @xmath15 m as indicated in fig . [ fig : uniformity_measurement_design ] . the spectra are extracted by integrating the backlighter images over the central 300 @xmath15 m . since their backlighter centers appear at different points in the sample , which are separated by 236 @xmath15 m , 300 @xmath15 m integrations span different lateral regions in the sample with a small amount of overlap ( @xmath2664 @xmath15 m ) . thus , in addition to measuring an axial gradient , these data will reveal a lateral gradient if it is significant . figure [ fig : al_mg_spectra ] shows the extracted line transmission spectra . in addition to mg , lines from al are apparent . the lines with @xmath91 are he-@xmath31 and ly-@xmath30 from mg and ly-@xmath29 and he-@xmath31 from al . these spectra are analyzed using radiator+galm and the inferred @xmath0 and @xmath8 are summarized in table [ tab : sample - uniformity - analysis ] . figure [ fig : uniformity_results ] shows the resultant @xmath0 ( left ) and @xmath8 ( right ) . red and blue data points are the plasma conditions inferred for the radiation source side ( al mixed side ) and the observer side ( mg mixed side ) of the fe sample , respectively . the red and the blue points agree within their measurement uncertainties . thus , it is confirmed that there is no measurable axial gradient in the sample . also , the conditions inferred from different spectrometers ( i.e. , 0@xmath118 and @xmath119 ) agree within their uncertainties . these results indicate that there is no gradient over @xmath120 m from the heating center ( i.e. , the point in the sample where the backlighter center appears from the spectrometer at @xmath121 ) . .[tab : sample - uniformity - analysis]sample uniformity analysis results . no measurable axial nor lateral gradients are observed . [ cols="^,^,^",options="header " , ] measured @xmath0 ( left ) and @xmath8 ( right ) from uniformity analysis . the agreement between the conditions measured in the radiation source side ( red ) and observer side ( blue ) of the sample confirm that there is no axial gradient within uncertainties . similarly , the agreement between the 0@xmath118 and @xmath119 line of sight conditions show that there is no measurable lateral gradient.,width=321 ] we addressed three of the key opacity measurement requirements : sample conditions control , independent sample condition measurements , and sample uniformity verification . the opacity experiments were performed at the sandia national laboratories z - machine to benchmark opacity models for fe at the base of solar convection zone where @xmath0 and @xmath8 are 185 ev and @xmath122 , respectively . the fe sample conditions were controlled by the tamping @xmath14 thicknesses . increasing the rear @xmath14 thickness slows down the expansion , and the fe samples remain dense at the time of backlighting . three different rear @xmath14 thicknesses are tested : i ) 10 @xmath15 m , ii ) 35 @xmath15 m , and iii ) 68 @xmath15 m . the resultant fe conditions are independently measured by mg k - shell spectroscopy and found to be i ) @xmath0=167@xmath16 ev and @xmath17 , ii ) @xmath0=170@xmath18 ev and @xmath19 , and iii ) @xmath0=196@xmath20 ev and @xmath21 , respectively . one experiment was designed and performed to specifically investigate the sample uniformity , with al mixed into the side of the sample facing the radiation source ( front ) , and mg mixed in the opposite side of the sample ( rear ) . the conditions at the bottom and the top of the fe sample were inferred by al and mg k - shell spectroscopy , respectively , and confirmed that there are no measurable gradients in the sample . the inferred condition uncertainties due to k - shell spectral model details are under investigation , as well as the effects of non - lte , sample self - emission , and temporal gradients . while different k - shell spectral models might infer slightly different @xmath0 and @xmath8 , those differences should be systematic and would not affect the conclusions on the condition reproducibility and the sample uniformity . the sample condition uncertainties due to k - shell spectral model details are under investigation . fe opacity experiments at three different conditions with similar charge state distributions provide crucial information to disentangle complex physical processes in opacity models . opacity models have to accurately compute relevant atomic data , atomic level populations , and then spectral formation , taking into account correct line shapes , and the complexity is enhanced as temperature and density increase . in lte , the same average ionizations achieved by different @xmath0 and @xmath8 have different population distribution within each charge state because of the temperature dependence on the boltzmann relationship [ @xmath12 ] . as temperature increases , more population is in excited states and absorption starting from those excited states becomes more important . thus , at higher temperature , the accuracy and the treatment of excited states ( i.e. , singly- , doubly- , ... , multiply - excited states ) become more crucial to accurately solve for population and model opacity . as density increases , the effects of continuum lowering and stark broadening become more important . since our data are measured at different @xmath0 and @xmath8 with similar charge states , we can study how these effects gradually change the true opacity , and how well state - of - the - art opacity models can calculate them . we are grateful to r. falcon and t. lockard for their help in refining the manuscript . sandia is a multiprogram laboratory operated by sandia corporation , a lockheed martin company , for the united states department of energy under contract de - ac04 - 94al85000 . prismspect is a collisional radiative model that computes atomic level populations and then computes opacity ( i.e. , mass absorption coefficient ) , @xmath124 , and emergent spectral radiance , @xmath125@xcite . in slab geometry , these two outputs are related by the following equations : @xmath126\label{eq : radiation_transport_slab}\ ] ] @xmath127 where @xmath128 is the thickness of the slab , and @xmath129 and @xmath130 are emission coefficient ( or @xmath131 ) and absorption coefficient ( or @xmath132 ) , respectively@xcite . for the rest of appendix a and b , opacity refers to absorption coefficient , @xmath130 ( not mass absorption coefficient , @xmath124 ) . since both emissivity and opacity are proportional to the ion number density , @xmath133 , it is convenient to invert the prismspect outputs , @xmath125 and @xmath124 , into what we call fractional emissivity and opacity as follows : @xmath134 @xmath135 the quantities @xmath136 and @xmath137 are proportional to the initial state fractional population of relevant radiative processes and depend only on @xmath0 and @xmath8 in the lte assumption . the database of @xmath138 and @xmath139 can be very useful for quick spectra calculations . one can build a prismspect fractional emissivity and opacity database of an element @xmath140 , @xmath141 and @xmath142 , by performing a single element , lte prismspect calculation for ranges of temperature and density and extracting @xmath136 and @xmath137 from each condition using eqs . ( [ eq : fractional_opacity ] ) and ( [ eq : fractional_emissivity ] ) . we note that prismspect requires ion number density , @xmath133 , as an input ( not @xmath8 ) . thus , @xmath8 has to be extracted from the prismspect output using @xmath143 , where @xmath144 is the mean charge of the plasma . interpolations are required both to build the database and to use the database . the required database grid spacing in @xmath0 , @xmath8 , and @xmath145 , and prismspect calculation details depend on the required accuracy , and have to be carefully investigated for each application . the database used in this article is optimized for the emergent spectra accuracy , and temperature and density diagnostics . the spectra based on the database with linear interpolation agree with the spectra directly computed by prismspect within 1% . the uncertainties in inferred @xmath0 and @xmath8 due to the use of the database and the linear interpolation are within 0.5 % and 1% , respectively . for mg and al database calculations , prismspect employed tabulated stark line shapes for their k - shell lines , which were computed in detail by merl@xcite . the advantage of the database is speed and flexibility . fractional emissivity and opacity of an element are dependent only on @xmath0 and @xmath8 and independent of surrounding species , its own ion number density , and geometry . thus , once the database is extracted for each element , one can compute spectra for either single or multiple elements , uniform or non - uniform , emission or absorption by solving radiation transport for a given geometry , without re - running collisional radiative models . the database extraction is not exclusive to prismspect ; databases can be extracted from other collisional models . the accuracy of the database depends on the details of the collisional model used and the numerical details of the extraction scheme . the idea of the database can be extended for non - lte as long as the non - lte effects are independent of geometry such as for the optically - thin approximation or for non - lte effects dominated by strong external radiation . once fractional emissivity and opacity databases are computed for each element @xmath140 [ i.e. , @xmath141 and @xmath142 in appendix a ] , one can solve the slab radiation transport equation to compute emergent spectra for any mixture and any gradient . geometry can even be extended to arbitrary shapes@xcite . assume there is an m species plasma , which is discretized into n zones along the observation line of sight where zone n is closest to the observer . by assuming that an event is observed very far from the plasma , emergent spectra seen from the observer can be computed by the following equation : @xmath146 where @xmath147 is the incident spectral irradiance ( or backlighter ) , @xmath76 is the net optical depth of the plasma , @xmath148 is the plasma self - emission taking into account its self - absorption effect , and @xmath149 is a potential additional background . @xmath148 can be computed recursively as follows : where @xmath155 is the emergent spectral irradiance at the end of zone @xmath156 of the discretized plasma , and @xmath157 and @xmath158 are local emissivity and opacity ( i.e. , absorption coefficient ) in zone @xmath156 , respectively . the local emissivity and opacity at zone @xmath156 ( i.e. , @xmath157 and @xmath158 ) can be computed by summing element emissivity , @xmath159 , and opacity , @xmath160 , over all of the species in zone @xmath156 , where @xmath161 are the ion number density of element @xmath162 at zone @xmath156 , and @xmath163 and @xmath164 are electron temperature and density at zone @xmath156 , respectively . by multiplying @xmath157 and @xmath158 by the zone length , @xmath165 , one can compute self - emission in the optically - thin approximation , @xmath166 , and the optical depth , @xmath167 , of zone @xmath156 , respectively . furthermore , @xmath168 and @xmath169 gives the net self - emission in optically - thin approximation and the net optical depth of the plasma , respectively . this formula is general and works for either emission or absorption spectra . if the incident flux ( or backlighter ) is weaker than the sample self emission ( i.e. , @xmath170 ) , eq . ( [ eq : radiation_transport_mixture3 - 1 ] ) gives emission spectra . if the backlighter is brighter ( i.e. , @xmath171 ) , this formula naturally produces absorption spectra . the measured spectra can be simulated by convolving @xmath125 with instrument spectral shape , @xmath65 , as follows : @xmath172 where @xmath173 . the measured transmission spectra , @xmath174 , can be computed by dividing @xmath175 by the convolved backlighter , @xmath176 . there are two limiting cases of eq . ( [ eq : radiation_transport_mixture3 - 1 ] ) . for these limiting cases , @xmath149 is neglected for communication purposes . one limiting case is the optically - thin approximation for emission spectra ( i.e. , @xmath177 and @xmath178 ) . this approximation simplifies eq . ( [ eq : radiation_transport_mixture3 - 1 ] ) to @xmath179 . the other limiting case is pure transmission or the bright - backlighter approximation ( i.e. , @xmath180 ) , which simplifies eq . ( [ eq : radiation_transport_mixture3 - 1 ] ) to the following : @xmath181 when the backlighter is not only bright but also spectrally smooth , one can approximate the instrumental broadening effect as follows : @xmath182 which is eq . ( [ eq : approximated_transmission ] ) . this approximation is valid when @xmath147 changes slowly over the instrumental spectral shape , @xmath65 . in our application , areal number densities of each element ( i.e. , fe , mg , and al ) are measured prior to the experiments using rutherford backscattering ( rbs ) . they can also be inferred by radiator+galm because of their sensitivities to the line depths . thus , rewriting the @xmath157/@xmath158 , @xmath166 , and @xmath167 in terms of element areal number density , @xmath183 , and fraction of the element areal number density in each zone @xmath156 , @xmath184 , is also useful : @xmath185 @xmath186 @xmath187 where @xmath188 . by assuming the fraction @xmath189 is determined by the initial target design@xcite , one can compute emergent spectra of multi - species plasmas just by providing @xmath163 and @xmath164 at each zone @xmath156 . the areal number density of each element , @xmath183 , can either be provided by rbs measurements or extracted from the analysis . 55ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ , ( ) @noop * * , ( ) _ _ , @noop ph.d . thesis , , ( ) @noop _ _ , pure and applied physics series ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ ( ) p. @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , , ) @noop _ _ , ed . ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , a series of books in astronomy and astrophysics ( , ) @noop * * , ( )

the theory of dissipativity has been primarily developed for controllable systems / behaviors . for various reasons , in the context of uncontrollable systems / behaviors , a more appropriate definition of dissipativity is in terms of the dissipation inequality , namely the _ existence _ of a storage function . a storage function is a function such that along every system trajectory , the rate of increase of the storage function does not exceed the power supplied . while the power supplied is always expressed in terms of only the external variables , whether or not the storage function should be allowed to depend on only the external variables and their derivatives or also unobservable / hidden variables has various consequences on the notion of dissipativity : this paper thoroughly investigates the key aspects of both cases , and also proposes another intuitive definition of dissipativity . we first assume that the storage function can be expressed in terms of the external variables and their derivatives only and prove our first main result that , assuming the uncontrollable poles are unmixed , i.e. no pair of uncontrollable poles add to zero , and assuming a strictness of dissipativity at the infinity frequency , the dissipativities of a system and its controllable part are equivalent ; in other words once the autonomous subsystem satisfies a lyapunov equation solvability - like condition , it does not interfere with the dissipativity of the system . we also show that the storage function in this case is a static state function . this main result proof involves new results about solvability of the algebraic riccati equation , and uses techniques from indefinite linear algebra and hamiltonian matrix properties . we then investigate the utility of unobservable / hidden variables in the definition of storage function : we prove that lossless uncontrollable behaviors are ones which require storage function to be expressed in terms of variables that are unobservable from the external variables . we next propose another intuitive definition : a behavior is called dissipative if it can be embedded in a controllable dissipative _ super - behavior_. we show that this definition imposes a constraint on the number of inputs and thus explains unintuitive examples from the literature in the context of lossless / orthogonal behaviors . these results are finally related to rlc realizability of passive networks , specifically to the nonrealizability of the nullator one - port circuit using rlc elements .

1110.2102

the theory of dissipativity for linear dynamical systems helps in the analysis and design of control systems for several control problems , for example , lqr / lqg control , @xmath0 , synthesis of passive systems , and optimal estimation problems . when dealing with lti systems , it is straightforward to define dissipativity for controllable systems due to a certain property of such systems that their compactly supported system trajectories are , loosely speaking , ` dense ' in the set of all allowed trajectories . however , this is not the case for uncontrollable systems , and this situation is the central focus of this paper . we elaborate more on this point when we define dissipativity and review equivalent conditions for controllable systems in section [ sec2 ] . in this paper , we use a less - often - used definition of dissipativity for systems , possibly uncontrollable , and generalize key results using some techniques from indefinite linear algebra ( see @xcite ) for solving algebraic riccati inequalities in the context of an uncontrollable state space system . like in @xcite , we define a system as dissipative if there exists a storage function that satisfies the dissipation inequality for all system trajectories . the existential aspect of this definition raises key issues that this paper deals with . the main result we show is that if the uncontrollable poles of an lti system are such that no two of them add to zero , and if the controllable subsystem strictly dissipates energy at frequency equal to infinity , then the dissipativity of the controllable subsystem is equivalent to the system s dissipativity . we also show that , using the concatenability axiom of the state , the energy stored in a system is a static function of the state variables . further , we also show that this state is ` observable ' from the external variables , i.e. the state is a linear combination of the external variables and possibly their derivatives . this is intuitively expected in view of the fact that energy exchange between the system and its ambience takes place through the external variables . however , it appears that this may not be the case for lossless systems , i.e. systems that do nt dissipate any energy , nor contain a source within . in this section we include various definitions about the behavioral framework for studying dynamical systems ( subsection [ subsec2.1 ] ) and then introduce background results about dissipative systems ( subsection [ subsec2.2 ] ) . subsection [ subsec2.3 ] contains brief notation about indefinite linear algebra from @xcite . for this paper , @xmath1 denotes the set of all real numbers and @xmath2 $ ] the set of polynomials in the indeterminate @xmath3 and real coefficients ; matrices and polynomial matrices are denoted the obvious way . we use @xmath4 to leave a row dimension unspecified , for example , @xmath5}$ ] . the space of infinitely often differentiable functions from @xmath1 to say @xmath6 is denoted by @xmath7 and @xmath8 denotes the set of compactly supported functions within this space . when dealing with linear differential systems , it is convenient to use polynomial matrices for describing a differential equation . suppose @xmath9 , @xmath10 @xmath11 are constant matrices of the same size such that @xmath12 is a linear constant coefficient ordinary differential equation in the variable @xmath13 . we define the polynomial matrix @xmath14 , and represent the above differential equation as @xmath15 . a linear differential behavior , denoted by @xmath16 , is defined as the set of all infinitely often differentiable trajectories that satisfy a system of ordinary linear differential equations with constant coefficients , i.e. , @xmath17 where @xmath18 is a polynomial matrix having @xmath19 number of columns , i.e. , @xmath20}$ ] . we denote the set of all such linear differential behaviors with @xmath19 number of variables by @xmath21 . the linear differential behavior @xmath22 given by the above definition can also be represented by @xmath23 . one views the rows of @xmath24 as differential equations that the variable @xmath13 has to satisfy in order for a trajectory @xmath25 to be in the behavior @xmath26 . the matrix @xmath24 is not unique and one can use elementary row operations to modify @xmath24 and this does not change the set of solutions @xmath26 : this thus allows assuming without loss of generality that @xmath24 has full row rank ( see @xcite ) . the number of inputs of @xmath26 is defined as @xmath27 and is called the input cardinality . this integer depends only on @xmath26 and not on the @xmath24 used to define it ; @xmath28 denotes the number of inputs . an important fundamental concept is controllability of a system . a behavior @xmath29 is said to be _ controllable _ , if for every @xmath30 and @xmath31 , there exist @xmath32 and @xmath33 such that @xmath34 the set of all _ controllable _ behaviors with @xmath19 variables is denoted as @xmath35 . this patchability definition of controllability is known to have the traditional kalman state - space definition of controllability as a special case in @xcite . further , it is shown that @xmath36 is controllable if and only if @xmath37 has constant rank for all @xmath38 . it is also shown in @xcite that @xmath16 is controllable if and only if it can be defined as @xmath39 where @xmath40 $ ] . this representation of @xmath41 is known as an _ image representation_. the variable @xmath42 is called a _ latent _ variable : these are auxiliary variables used to describe the behavior ; we distinguish the variable @xmath13 as the manifest variable , the variable of interest . it is known that @xmath43 always allows an image representation with @xmath44 such that @xmath45 has full column rank for every @xmath46 . this kind of image representation is known as an _ observable _ image representation . in this paper , unless otherwise stated explicitly , we assume the image representations are observable . the use of the term ` observable ' is motivated by the fact that the variable @xmath42 is _ observable _ from the variable @xmath13 . this notion is defined as follows . for a behavior @xmath26 with variables @xmath13 and @xmath42 , we say @xmath42 is observable from @xmath13 if whenever @xmath47 and @xmath48 both are in @xmath26 , we have @xmath49 . observability of @xmath42 from @xmath13 in a behavior @xmath26 implies that there exists a polynomial matrix @xmath50 such that @xmath51 for all @xmath13 and @xmath42 in the behavior . we now define relevant notions in the context of uncontrollable behaviors . for a behavior @xmath16 , possibly uncontrollable , the largest controllable behavior contained in @xmath16 is called the _ controllable _ part of @xmath16 , and denoted by @xmath52 . an important fact about the controllable part of @xmath16 is that @xmath53 . the set of complex numbers @xmath54 for which @xmath37 loses rank is called the set of _ uncontrollable modes _ and is denoted by @xmath55 . for a detailed exposition on behaviors , controllability and observability we refer the reader to @xcite . the concept of quadratic differential forms ( qdf ) ( see @xcite ) is central to this paper . consider a two variable polynomial matrix with real coefficients , @xmath56 $ ] , where @xmath57 , the qdf @xmath58 induced by @xmath59 is a map @xmath60 defined as @xmath61 when dealing with quadratic forms in @xmath13 and its derivatives , we can assume without loss of generality that @xmath62 : such a @xmath59 is called a symmetric two variable polynomial matrix . a quadratic form induced by a real symmetric constant matrix @xmath63 is a special qdf . we frequently need the number of positive and negative eigenvalues of a nonsingular , symmetric matrix @xmath64 : they are denoted by @xmath65 and @xmath66 respectively . for a two variable polynomial matrix @xmath67 $ ] , we define the single variable polynomial matrix @xmath68 by @xmath69 . consider @xmath70 , a symmetric nonsingular matrix . a behavior @xmath71 is said to be dissipative with respect to the supply rate @xmath72 ( or @xmath72-dissipative ) if there exists a qdf @xmath73 such that @xmath74 the qdf @xmath73 is called a storage function : it signifies the energy stored in the system at any time instant . the above inequality is called the dissipation inequality . a behavior @xmath26 is called @xmath72-lossless if the above inequality is satisfied with an equality for some qdf @xmath73 . notice that the storage function plays the same role as that of lyapunov functions in the context of autonomous systems ; the notion of storage functions is a generalization to non - autonomous systems of lyapunov functions , as pointed in @xcite . the following theorem from @xcite applies to controllable behaviors . consider @xmath75 and let @xmath76 be an observable image representation . suppose @xmath77 is symmetric and nonsingular . then , the following are equivalent . 1 . there exists a qdf @xmath73 such that inequality is satisfied for all @xmath78 . @xmath79 for all @xmath80 , the compactly supported trajectories in @xmath26 . @xmath81 for all @xmath82 . the significance of the above theorem is that , for controllable systems , it is possible to verify dissipativity by checking non - negativity of the above integral over all compactly supported trajectories : the compact support signifying that we calculate the ` net power ' transferred when the system starts ` from rest ' and ` ends at rest ' . the starting and ending ` at rest ' ensures that for linear systems there is no internal energy at this time . the absence of internal energy allows ruling out the storage function from this condition : in fact , this is used as the definition of dissipativity for controllable systems . the same can not be done for uncontrollable systems due to the compactly supported trajectories not being dense in the behavior ( see @xcite ) . an extreme case is an autonomous behavior , i.e. a behavior which has @xmath83 : while the zero trajectory is the only compactly supported trajectory , the behavior consists of exponentials corresponding to the uncontrollable poles of @xmath26 . the issue of existence of storage functions is elaborated in ( * ? ? ? * remark 5.9 ) and in text following ( * ? ? ? * proposition 3.3 ) . the state variable is defined as a latent variable that satisfies the _ property of state _ , that is , if @xmath84 , @xmath85 and @xmath86 , then the new trajectory @xmath87 formed by concatenating @xmath84 and @xmath88 at @xmath89 , i.e. , @xmath90 also satisfies the system equation of @xmath91 in a distributional sense @xcite . it is intuitively expected that a variable @xmath92 has the state property if and only if @xmath13 and @xmath92 satisfy an equation that is at most first order in @xmath92 and zeroth order in @xmath13 : see @xcite for precise statement formulation and proof . when @xmath13 is partitioned into @xmath93 , with @xmath30 as the input and @xmath94 as the output , then @xmath91 admits the more familiar _ input / state / output _ ( i / s / o ) representation as @xmath95 one can ensure that @xmath92 is observable from @xmath13 ; this is equivalent to conventional observability of the pair @xmath96 . while such a state space representation is admitted by any @xmath26 , the pair @xmath97 is controllable ( in the state space sense ) if and only if @xmath16 is controllable ( in the behavioral sense defined above ) . in this paper , we use certain properties of matrices that are self - adjoint with respect to an _ indefinite _ inner product . we briefly review self - adjoint matrices and neutral subspaces ( see @xcite ) . let @xmath98 be an invertible hermitian matrix . this defines an indefinite inner product on @xmath99 by @xmath100 , where @xmath101 is the complex conjugate transpose of the vector @xmath102 . for a complex matrix @xmath103 , the complex conjugate transpose of @xmath103 is denoted by @xmath104 . consider matrices @xmath103 and @xmath105 with @xmath106 invertible and hermitian . the @xmath106-adjoint of the matrix @xmath103 , denoted by @xmath107}$ ] , is defined as @xmath107 } = p^{-1}a ^ * p$ ] . the matrix @xmath103 is said to be @xmath106-self - adjoint if @xmath108}$ ] , i.e. , @xmath109 . since @xmath106 is not sign - definite in general , the sign of @xmath110 is zero , positive or negative depending on the vector @xmath92 . a subspace @xmath111 is said to be @xmath106-neutral if the inner product @xmath112 for all @xmath113 . in this paper , we deal with systems which satisfy the dissipativity property , i.e. _ net _ energy is directed _ inwards _ along every system trajectory . as elaborated below , the ` total ' aspect of the energy involves an integral , thus bringing in the initial and final conditions of the trajectory being integrated . the convenience of starting - from - rest and ending - at - rest applies to only controllable systems as we will review soon . the notion of storage function helps in formulating the dissipation property as an inequality to be satisfied _ at each time - instant_. a central issue in this paper is what variables should the storage function be allowed to depend on . we explore dependencies on a latent variable @xmath42 , or on a state variable @xmath92 , or on the manifest variable @xmath13 : this is indicated in the storage function subscript ; the following definition has appeared in several works , see @xcite , for example . [ defn : dissip ] let @xmath70 be a nonsingular symmetric matrix , inducing the supply rate @xmath114 . consider a behavior @xmath115 with manifest variables @xmath13 and latent variable @xmath42 , with the corresponding full behavior @xmath116 . for the behavior @xmath26 , let @xmath92 be a state variable with the corresponding full behavior @xmath117 . with respect to the supply rate @xmath72 , the behavior @xmath26 is said to be dissipative if there exists a quadratic differential form @xmath73 such that @xmath118 1 . the function @xmath119 , a quadratic function of @xmath42 and its derivatives , is called a storage function a storage function @xmath119 is said to be an _ observable _ storage function if the latent variable @xmath42 is observable from the manifest variable @xmath13 . in this case , there exists a storage function @xmath120 such that @xmath121 for all @xmath122 . a storage function @xmath123 is said to be a _ state function _ if @xmath123 is equal to @xmath124 for some constant matrix @xmath125 . in this paper , we study dissipativity with respect to a constant , nonsingular , symmetric @xmath126 . it is known ( see ( * ? ? ? * remark 5.11 ) and ( * ? ? ? * proposition 2 ) ) that @xmath72-dissipativity of a behavior @xmath26 implies that the input cardinality of @xmath26 can not exceed @xmath127 , i.e. , @xmath128 . in this context it is helpful to perform a coordinate transformation in the variables @xmath13 so that @xmath72 is a diagonal matrix consisting of only @xmath129 s and @xmath130 s along the diagonal . moreover , there exists an input / output partition such that all the inputs correspond to @xmath129 s only and such that the transfer function from these inputs to all other variables is _ proper _ ( see ( * ? ? ? * remark 5.11 ) ) . in view of these facts and the inequality @xmath128 , we assume without loss of generality @xmath131\ ] ] where @xmath132 is the number of inputs in @xmath26 . the following theorem assumes an unmixing condition - no pair of uncontrollable poles are symmetric with respect to the imaginary axis . if this unmixing condition is satisfied for the uncontrollable poles , then the controllable part of a behavior being dissipative is equivalent to the dissipativity of the whole behavior . consider a behavior @xmath22 and a nonsingular , symmetric @xmath126 with the input cardinality of @xmath16 at - most the positive signature of @xmath72 , i.e. , @xmath128 . assume that the uncontrollable poles are such that @xmath133 . let @xmath16 have an observable image representation , @xmath134 , where @xmath44 is partitioned as @xmath135 , w_2 \in \mathbb{r}^{(\texttt{p+q } ) \times \texttt{m}}\left [ \xi \right].\ ] ] let @xmath136 and @xmath137 . assume @xmath52 is such that @xmath138 . then , @xmath16 is @xmath72-dissipative if and only if its controllable part @xmath52 is @xmath72-dissipative . [ main ] the above result gives conditions under which the autonomous part of a behavior plays no hindrance to dissipativity of the behavior after the controllable part is dissipative . one of the conditions for this is that the uncontrollable poles are not ` mixed ' , meaning no two of the uncontrollable poles add to zero . for autonomous lti systems , this condition is a necessary and sufficient condition for solvability of the lyapunov equation . of course , storage functions are just generalizations of lyapunov functions to non - autonomous systems . the other condition : @xmath138 on the controllable part is a kind of strictness of dissipativity ` at the infinity is finite , since the transfer function is proper ; see text before equation . ] frequency ' . while this condition allows the use of hamiltonian matrices in the proofs , this condition also rules out consideration of lossless systems from the above result . more significance of the assumption is noted in the remark below . while a behavior @xmath71 admits many i / o partitions , and for each i / o partition , admits many i / s / o representations , the matrix @xmath139 depends only on the behavior , and in particular , only on @xmath140 the controllable part . in other words , this matrix can be found from any image representation of @xmath140 . this matrix being positive semi - definite is a necessary condition for dissipativity of @xmath140 , and this denotes dissipativity at very high frequencies , i.e. as @xmath141 . positive definiteness of @xmath139 may hence be termed as ` strict dissipativity at the infinity frequency ' . this assumption helps in the existence of a hamiltonian matrix , our proofs use the hamiltonian matrix properties and techniques from indefinite linear algebra ( see @xcite ) . positive definiteness of this matrix is guaranteed , for example , by strict dissipativity of a behavior ; on the other hand , lossless controllable behaviors have this matrix as zero . another interpretation of the condition @xmath142 is that the ` memoryless ' part of the behavior is strictly dissipative ; the memoryless / static part of a behavior was defined in @xcite , and we do nt require this notion in this paper . we now review some existing results and formulate / prove new results about hamiltonian matrices in the context of dissipativity of controllable and uncontrollable systems . these are required for the proof of the main result . to prove that the behavior @xmath16 is @xmath72-dissipative , we use the proposition below and show that the existence of a symmetric solution to the dissipation lmi is sufficient for @xmath72-dissipativity . let the behavior have an input / output partition @xmath93 , @xmath143 as the input and @xmath144 as the output . let the input / state / output representation of the behavior @xmath26 be , @xmath145 with @xmath96 observable . the following result is well - known : see @xcite . [ lmiunctrb ] let a behavior @xmath146 have an i / s / o representation for an input / output partition @xmath93 with @xmath147 as state space matrices and @xmath97 possibly uncontrollable . the behavior @xmath16 is @xmath72-dissipative if there exists a symmetric @xmath148 which solves the following lmi @xmath149 \leqslant 0.\ ] ] further , under the condition that @xmath97 is controllable , existence of @xmath125 solving the above lmi is necessary and sufficient for @xmath150-disspativity of @xmath26 . as @xmath151 is invertible , the schur complement of @xmath151 in the above lmi gives the algebraic riccati inequality ( ari ) @xmath152 the corresponding equation is the algebraic riccati equation ( are ) and we use properties of this are in proving theorem [ main ] . we rewrite the above are as @xmath153 where @xmath154 and @xmath155 also define the hamiltonian matrix corresponding to the are : @xmath156.\ ] ] we define the following matrices ; they play a crucial role in the results we use from @xcite and in our proofs . @xmath157 \ : \ : \ : \text{and } \ : \ : \ : \widehat{p } : = { { j}}\ : \left [ \begin{array}{cc } 0 & i \\ -i & 0\\ \end{array } \right].\ ] ] it is well - known that a symmetric solution to are can be obtained from an @xmath158-dimensional , @xmath159-invariant , @xmath106-neutral subspace and we state this in the proposition below . let @xmath160 be a hermitian matrix . the graph subspace corresponding to matrix @xmath125 is defined as , @xmath161 . \ ] ] notice that @xmath162 is an n - dimensional subspace of @xmath163 . the following proposition from @xcite states solvability of the are in terms of @xmath106-neutrality with respect to the above @xmath106 . [ graphneutral](see @xcite ) consider the are with @xmath164 possibly uncontrollable . then @xmath160 is a hermitian solution of the are if and only if the graph subspace of @xmath125 is @xmath159-invariant and @xmath106-neutral . [ pro : invsln ] the proposition below allows use of simplified @xmath103 and @xmath165 for all later purposes when dealing with uncontrollable systems . ( see @xcite ) consider the behavior @xmath146 with an input / state / output representation @xmath166 , @xmath167 , where @xmath168 . then there exists a nonsingular matrix @xmath169 such that @xmath170 , \ : \ : \ : t^{-1}b = \left [ \begin{array}{c } b_c \\ 0 \\ \end{array } \right ] \ : \ : and \ : \ : ct = \left [ \begin{array}{cc } c_c & c_u \\ \end{array } \right].\ ] ] further , @xmath171 gives an i / s / o representation for the controllable part @xmath52 . let @xmath172 and @xmath173 denote the corresponding matrices for the controllable part @xmath52 as defined in with @xmath174 and @xmath175 defined accordingly in . [ partial ] suppose @xmath176 satisfies the assumption that @xmath138 . if @xmath52 is @xmath72-dissipative , then the partial multiplicities corresponding to the real eigenvalues of @xmath173 , if any , are all even . in order to prove lemma [ partial ] , we use a result from @xcite concerning the partial multiplicities of real eigenvalues of @xmath159 . let the generalized eigenspace of a matrix @xmath103 corresponding to an eigenvalue @xmath177 be denoted by @xmath178 and the controllable subspace of the pair @xmath164 be denoted by @xmath179 . [ partialm](see @xcite ) consider the behavior @xmath180 . assume @xmath181 and @xmath182 and there exists a hermitian solution @xmath183 to the are . suppose @xmath184 for every purely imaginary eigenvalue @xmath177 of @xmath185 . then , the partial multiplicities of corresponding real eigenvalues of @xmath173 are all even and are twice the partial multiplicities of those corresponding to the purely imaginary eigenvalues of @xmath185 . * proof of the lemma [ partial ] * : as the controllable part is @xmath72-dissipative , there exists a symmetric solution @xmath125 to the are by proposition [ lmiunctrb ] . since the behavior is controllable , the following is true : @xmath186 and hence @xmath187 for every purely imaginary eigenvalue @xmath177 of @xmath185 . thus , using proposition [ partialm ] , the partial multiplicities of all real eigenvalues of @xmath173 , if any , are all even . this completes the proof . we define a set called _ c - set _ as in @xcite . such a _ c - set _ , if exists , guarantees the existence of a unique @xmath106-neutral , @xmath159-invariant subspace @xmath188 under certain conditions . this is made precise in the proposition below . ( see @xcite ) let @xmath189 and let @xmath190 be a finite set of non - real complex numbers . @xmath190 is called a of @xmath159 if it satisfies the following properties 1 . @xmath191 2 . @xmath192 = @xmath193 , the set of all non - real eigenvalues of @xmath159 . let @xmath188 be an invariant subspace of @xmath159 and denote the restriction of @xmath103 to @xmath188 as @xmath194 . [ invariant](see @xcite ) let @xmath195 be a @xmath106-self - adjoint matrix such that the sizes of the jordan blocks of @xmath159 , say : @xmath196 , corresponding to real eigenvalues of @xmath159 are all even . then for every c - set @xmath197 there exists a unique @xmath106-neutral @xmath159-invariant subspace @xmath188 of dimension @xmath158 and @xmath198 , and the sizes of the jordan blocks of @xmath199 corresponding to the real eigenvalues are @xmath200 . it is easy to verify that @xmath159 and @xmath106 satisfy the following relation , @xmath201 . hence @xmath159 is @xmath106-self - adjoint and proposition [ invariant ] can be used . following proposition , which is a reformulation and combination of theorems a.6.1 , a.6.2 and a.6.3 in @xcite , states that the partial multiplicities of an eigenvalue are unaffected by pre - multiplying and/or post - multiplying by an unimodular matrix . this result is used in the proof of the theorem [ main ] . [ evenparmul ] consider @xmath202 and @xmath203 $ ] and let @xmath204 and @xmath205 $ ] for @xmath206 be the invariant polynomials of @xmath202 and @xmath207 respectively . suppose @xmath208 , for invertible @xmath209 . let @xmath38 and @xmath210 be the maximum integers , for @xmath211 , such that @xmath212 divides @xmath213 for @xmath214 . then @xmath215 for @xmath214 . in particular , if @xmath216\ ] ] and @xmath217\ ] ] and @xmath218 , then the partial multiplicities of @xmath202 and @xmath207 corresponding to @xmath54 are equal . before proving the main result theorem [ main ] , we state and prove another useful result . the following lemma relates the partial multiplicities of purely imaginary eigenvalues of the hamiltonian matrix corresponding to the controllable part to that of the uncontrollable behavior . [ evenpartial ] consider the behavior @xmath22 with the set of uncontrollable modes @xmath55 satisfying @xmath219 . let @xmath16 have an observable i / s / o representation as @xmath220 , @xmath221 induced by @xmath168 , such that @xmath138 . further , let @xmath222 . define @xmath223 and construct the hamiltonian matrix , @xmath224 as in . then 1 . @xmath225 . 2 . if the controllable part @xmath52 is @xmath72-dissipative , then the partial multiplicities corresponding to the purely imaginary eigenvalues of @xmath224 , if any , are all even . [ parmul ] * proof of lemma [ evenpartial ] * : * statement 1 * : see @xcite . * statement 2 * : we use the fact that if the controllable part is dissipative , then the partial multiplicities corresponding to purely imaginary eigenvalues of @xmath172 are even . without loss of generality the following i / s / o representation for @xmath16 is assumed @xmath226 = \left [ \begin{array}{cc } \hat{a}_c & \hat{a}_{cp } \\ 0 & \hat{a}_u \\ \end{array } \right ] \left [ \begin{array}{c } x_c \\ x_u \\ \end{array } \right ] + \left [ \begin{array}{c } \hat{b}_c \\ 0 \\ \end{array } \right ] w_1 \label{eq : kal1}\ ] ] @xmath227 \left [ \begin{array}{c } x_c \\ \end{array } \right ] + \hat{d } w_1 \label{eq : kal2}\ ] ] with @xmath228 controllable . then , the hamiltonian matrix gets the following form @xmath229 \label{eq : hamkal}\ ] ] where @xmath230 @xmath231 and @xmath232 @xmath233 and @xmath234 . the hamiltonian matrix for @xmath52 is @xmath235\ ] ] consider the polynomial matrix @xmath236 , @xmath237\ ] ] applying the following transformations to @xmath238 , we get @xmath239 @xmath240\ ] ] where @xmath241\ ] ] now , since @xmath242 , for @xmath243 , the matrix blocks @xmath244 and @xmath245 are invertible . pre - multiplying @xmath246 by @xmath247 and post - multiplying by @xmath248 , we get @xmath249 @xmath250 \ ] ] where @xmath251 , \ : \ : \ : \:\ ] ] and @xmath252\ ] ] and @xmath253 , @xmath254 , @xmath255 and @xmath256 . thus @xmath257\ ] ] where @xmath258\ ] ] thus by using proposition [ evenparmul ] , the partial multiplicities of purely imaginary eigenvalues of @xmath224 are even . this completes the proof of lemma [ parmul ] . * proof * : * if part * : assume that the controllable part @xmath52 is @xmath72-dissipative and the assumption in the theorem is satisfied . by propositions [ lmiunctrb ] and [ graphneutral ] , to prove that the behavior @xmath16 is dissipative , it suffices to show the existence of a @xmath259 such that the corresponding graph subspace is an @xmath158-dimensional , @xmath159-invariant , @xmath106-neutral subspace . to show the existence of such a @xmath125 , we use lemma [ invariant ] to construct a c - set such that the corresponding @xmath158-dimensional @xmath159-invariant , @xmath106-neutral subspace is also a graph subspace of @xmath125 . this @xmath125 is the solution to the are and storage function for the whole behavior would then be defined as @xmath260 , thus completing the proof . as the unmixing assumption on uncontrollable modes is assumed , @xmath261 means that @xmath262 and @xmath263 . as we have assumed that the controllable part is @xmath72-dissipative , from lemma [ evenpartial ] , the partial multiplicities of real eigenvalues of @xmath264 are all even . using this fact , lemma [ invariant ] can be used to infer that there exists a unique @xmath158-dimensional @xmath159-invariant , @xmath106-neutral subspace for every c - set . now , it remains to show the existence of a c - set such that the corresponding @xmath158-dimensional , @xmath159-invariant , @xmath106-neutral subspace is also a graph subspace . we choose a c - set @xmath190 such that @xmath265 and show that the corresponding @xmath158-dimensional , @xmath159-invariant , @xmath106-neutral subspace is a graph subspace . let @xmath266 be the @xmath158-dimensional , @xmath106-neutral , @xmath159-invariant subspace of @xmath163 corresponding to the c - set @xmath190 and suppose @xmath267 \label{eq : l}\ ] ] for matrices @xmath268 and @xmath269 @xmath270 . in order to prove that @xmath266 is a graph subspace it is enough to prove that @xmath268 is invertible . this is proved using contradiction : we assume @xmath268 is singular and show that we get a contradiction to the unmixing assumption on @xmath55 . this constitutes the rest of the proof of the ` if part ' . we prove this along the lines of @xcite . @xmath159-invariance of @xmath266 implies that @xmath271 \left [ \begin{array}{c } x_1 \\ \end{array } \right ] = \left [ \begin{array}{c } x_1 \\ x_2 \\ \end{array } \right ] t \nonumber\ ] ] for @xmath272 . in other words , @xmath273 then as @xmath266 is @xmath106-neutral @xmath274 \left [ \begin{array}{cc } -\tilde{c } & \tilde{a}^\ast \\ \tilde{a } & \tilde{d } \\ \end{array } \right ] \left[\begin{array}{c } x_1 \\ x_2 \\ \end{array } \right ] = 0 , \label{eq : neutral1}\ ] ] i.e. @xmath275 now suppose @xmath268 is singular . let @xmath276 . from , for every @xmath277 we have , @xmath278 which implies @xmath279 since @xmath280 , @xmath281 , i.e. , @xmath282 . now , for every @xmath277 , from equations we have , @xmath283 that is , @xmath284 this implies @xmath285 is @xmath286-invariant . hence there exists a non - zero @xmath287 which is an eigenvector of t such that @xmath288 corresponding to eigenvalue , say @xmath54 . we claim that @xmath54 can not be a real eigenvalue and prove this below . post - multiplying the second equation of by @xmath287 we get , @xmath289 @xmath290 this implies that @xmath291 is a left eigenvector of @xmath292 with eigenvalue @xmath293 and from equation we have @xmath294 . this means that @xmath293 is an uncontrollable eigenvalue of @xmath292 and @xmath295 , i.e. , @xmath296 . now , if @xmath54 were a real eigenvalue , then @xmath293 and @xmath297 belong to @xmath55 and contradicts the unmixing assumption @xmath298 . since @xmath299 ( by proposition [ invariant ] ) , @xmath300 . we now have @xmath301 but @xmath302 can not belong to @xmath190 . but we have @xmath296 and thus @xmath303 which is a contradiction by definition of c - set . this completes the proof for the if part . * only if part : * assume @xmath16 is @xmath72-dissipative . then there exists a storage function @xmath304 such that @xmath305 integrating both sides for every @xmath306 , @xmath307 which implies that @xmath52 is @xmath72-dissipative . this completes the proof of theorem [ main ] . the above proof is constructive in the sense that if a behavior @xmath71 satisfies the three conditions : * uncontrollable poles are unmixed , i.e. no two of them add to zero * the controllable part @xmath140 is dissipative , * the controllable part @xmath140 is ` strictly dissipative ' at infinity , i.e. @xmath138 where @xmath308 is the feed - through term of the transfer function , then we construct a storage function that satisfies the dissipation inequality for the whole behavior @xmath26 . further , the storage function we construct is equal to @xmath124 where @xmath125 is a solution to the corresponding algebraic riccati equation . further , we constructed the storage function by starting with a state representation in which the state @xmath92 is _ observable _ from the manifest variable @xmath13 . these facts lead to the following important corollary . let @xmath71 be an uncontrollable behavior whose uncontrollable poles @xmath55 are unmixed , i.e. @xmath309 . consider @xmath72 partitioned in accordance with the input cardinality of @xmath26 as @xmath310 suppose @xmath140 , the controllable part of @xmath26 , has an observable image representation , @xmath311 , where @xmath44 is partitioned as @xmath312 , w_2 \in \mathbb{r}^{(\texttt{p+q } ) \times \texttt{m}}\left [ \xi \right].\ ] ] let @xmath313 and @xmath314 . assume @xmath52 is such that @xmath138 . then , the following are equivalent . 1 . @xmath140 is dissipative . 2 . there exists a @xmath315}$ ] such that @xmath316 is a storage function , i.e. @xmath317 for all @xmath318 . 3 . there exists a matrix @xmath125 and an observable state variable @xmath92 such that @xmath319 for all @xmath318 . statement 2 tells that the storage function can be expressed as a quadratic function of the manifest variables @xmath13 and their derivatives . statement 3 says that the storage function is a ` state function ' , i.e. a static function of the states , and hence storage of energy requires no more memory of past evolution of trajectories than required for arbitrary concatenation of any two system trajectories . in this section we discuss two examples of uncontrollable systems that are dissipative . the riccati equations encountered in these cases are solvable by the methods proposed in this paper ; we also give solutions to the riccati equations . the first example is of an uncontrollable system with uncontrollable modes satisfying the unmixing assumption , i.e. no two of the uncontrollable poles add to zero . however , the hamiltonian matrix has eigenvalues on the imaginary axis . consider the behavior @xmath26 whose input / state / output representation is given by the following @xmath320 and @xmath308 matrices @xmath321 with @xmath322 . here @xmath323 which satisfies the unmixing assumption . an equivalent kernel representation of the behavior is given by @xmath324 in this case @xmath325 and hence @xmath326 , it can be checked that the controllable part @xmath327 is @xmath72-dissipative . and @xmath328 . thus from theorem [ main ] , @xmath26 is @xmath72-dissipative . the following real symmetric matrix induces a storage function that satisfies the dissipation inequality @xmath329 the 2-dimensional , @xmath159-invariant , @xmath106-neutral subspace which gives the solution is @xmath330 the next example is an rlc circuit shown in figure [ rlc.exa.fig ] . consider the rlc circuit system whose input @xmath331 is the current flowing into the circuit , and output @xmath332 is the current through the inductor . a state space representation of the system is found using the following definition of the states . the state variables are @xmath333 , voltage across the capacitor , and @xmath334 , the current through the inductor . assume @xmath335 . @xmath336 the system becomes uncontrollable when @xmath337 . let @xmath338 , @xmath339 and @xmath340 . the state representation of the system is @xmath341 poles of the system are @xmath342 and one of them is uncontrollable . the controllable part is dissipative and the corresponding hamiltonian matrix has eigenvalues on the imaginary axis . a solution to the are is given by the following symmetric matrix @xmath343 which induces the storage function @xmath344 . in this section we consider the case when all the states of the system are uncontrollable , in other words , when the only controllable part is static / memoryless . in a state space realization , this means that the matrix @xmath165 is zero . the case of autonomous behaviors is clearly a special case : @xmath308 also is zero and the assumptions in the lemma are satisfied . consider a behavior @xmath26 with static controllable part . let the behavior have a state representation , @xmath345 with the pair @xmath96 observable . assume @xmath138 and @xmath346 . then there does not exist a symmetric solution to the corresponding are . * proof : * the hamiltonian matrix takes the form , @xmath347 here @xmath348 , @xmath349 . next we use the following proposition to say that partial multiplicities of purely imaginary eigenvalues of the hamiltonian matrix are even . consider the matrix @xmath350 then for every purely imaginary @xmath351 such that @xmath352 , the partial multiplicities of such @xmath353 are even . in fact , they are twice the partial multiplicities of @xmath353 as an eigenvalue of @xmath103 . [ part ] consider , @xmath354 as the pair @xmath355 is observable , the pair @xmath356 is controllable . from proposition [ part ] , the partial multiplicities of purely imaginary eigenvalues are twice the partial multiplicities of purely imaginary eigenvalues of @xmath103 . in this section we investigate the requirement of unobservable variables in the definition of the storage function . as has been studied / shown so far , for controllable dissipative systems ( @xcite ) , the storage function need not depend on unobservable variables . it was later shown in @xcite that for the case of strict dissipativity of uncontrollable systems , observable storage functions are enough under unmixing and maximum input cardinality conditions . theorem [ main ] shows that this is true for a more general scenario , i.e. , dissipativity ( including non - strict dissipativity ) for all input cardinality conditions under unmixing assumption . on the other hand , when relaxing the unmixing assumption elsewhere except on the imaginary axis , under certain conditions solutions to the ari exists though are does not have a solution ( see @xcite ) . in this section , we investigate the need for unobservable storage functions for uncontrollable systems whose uncontrollable poles lie entirely on the imaginary axis . we discuss the case for autonomous behaviors below . consider an autonomous behavior @xmath357 @xmath358 with @xmath359 . let the supply rate be @xmath360 . then the following is true . if there exists a storage function @xmath304 satisfying the inequality @xmath361 then any @xmath362 is c - unobservable . * proof : * suppose if there exists a storage function which is a state function @xmath260 satisfying the dissipation lmi , then the dissipation lmi is equivalent to the lyapunov inequality @xmath363 now for every eigenvector of a corresponding to eigenvalue @xmath364 , we have @xmath365 which gives @xmath366 or @xmath367 but , as @xmath368 , we have @xmath369 for every eigenvector of @xmath103 corresponding to @xmath362 . this implies that the any @xmath362 is c - unobservable . this observation tells that for dissipativity of autonomous systems having eigenvalues on the imaginary axis , it is necessary to allow storage functions to depend on unobservable variables also . in this section we investigate the property of orthogonality of two behaviors in the absence of controllability . we propose a definition that is intuitively expected and show that by relating this definition to lossless uncontrollable behaviors , we encounter a situation that suggests an exploration whether dissipativity should be defined for behaviors for which the input - cardinality condition is not satisfied . we first review a result about orthogonality of controllable behaviors . [ prop : lossless : orthogonal ] let @xmath70 be nonsingular , and suppose @xmath370 . the following are equivalent . 1 . @xmath371 for all @xmath372 and for all @xmath373 . 2 . @xmath374 is lossless with respect to @xmath375 . 3 . there exists a bilinear differential form @xmath376 , induced by @xmath377}$ ] such that @xmath378 . statement 1 above is taken as the definition of orthogonality between two controllable behaviors @xmath379 and @xmath380 in @xcite . keeping in line with definition [ defn : dissip ] for dissipativity , we could take statement 3 above as the definition of orthogonality for behaviors not necessarily controllable . the drawback of this approach is elaborated later below in this section . we pursue a different direction as follows . notice that if @xmath379 and @xmath380 satisfy the integral condition in statement 1 , then this integral condition is satisfied for every respective sub - behaviors @xmath381 and @xmath382 also . of course , restricting to compactly supported trajectories in the integration implies only controllable parts of respectively @xmath381 and @xmath382 satisfy orthogonality . the following definition builds on this property . consider a nonsingular @xmath70 and let @xmath379 and @xmath383 . behaviors @xmath379 and @xmath380 are said to be @xmath72-orthogonal ( and denoted by @xmath384 ) if there exist @xmath385 and @xmath386 such that * @xmath387 , * @xmath388 , and * @xmath371 for all @xmath389 and for all @xmath390 . while this definition is not existential in the storage function , it is existential in @xmath385 and @xmath391 , raising questions about how to check orthogonality . note that if @xmath26 is an uncontrollable behavior , then any controllable @xmath392 such that @xmath393 satisfies @xmath394 , and @xmath395 . the question arises as to how much larger a controllable @xmath396 would have to be for it to contain @xmath26 . this problem is addressed in the following subsection . due to its significance for determining whether two uncontrollable behaviors are orthogonal , in this subsection we study the following problem : [ prob : smallestb ] let @xmath397 . find @xmath398 such that 1 . @xmath399 2 . @xmath400 i.e. , a controllable behavior 3 . @xmath380 is a behavior with the smallest input cardinality satisfying properties 1 and 2 . the following theorem answers this question . [ thm : superbehavior : controllable ] for @xmath115 , the following statements are true . 1 . there exists @xmath400 satisfying the requirements in problem [ prob : smallestb ] . the behavior @xmath380 is unique if and only if @xmath379 is controllable , and in that case @xmath401 . assume @xmath379 is uncontrollable . the input cardinality of @xmath380 , @xmath402 satisfies @xmath403 . more precisely , @xmath404 where @xmath405 where @xmath406 is a kernel representation of @xmath379 . * ( 1 ) : * this is shown by constructing such a @xmath380 . let @xmath407 and @xmath408 be the kernel representations of @xmath379 and @xmath380 respectively where @xmath409 , @xmath410 and there exists an @xmath411 such that @xmath412 . let @xmath55 be the set of uncontrollable modes of @xmath379 . without loss of generality , we assume @xmath406 to be @xmath413\ ] ] where @xmath64 is of the form @xmath414 $ ] such that @xmath415 is identity matrix and @xmath308 @xmath416 $ ] is a diagonal matrix with @xmath417 along its diagonal and satisfying the divisibility property : @xmath418 , @xmath419 , @xmath420 and @xmath421 , with degree of @xmath422 at least one , and with @xmath423 partitioning @xmath424 $ ] conforming to the row partition of @xmath406 , we have @xmath425 \left [ \begin{array}{ccc } i & 0 & 0 \\ 0 & d & 0 \end{array } \right]\ ] ] this simplifies to @xmath426\ ] ] let @xmath380 be the behaviour defined by the kernel representation of @xmath427 . for @xmath380 to be controllable , @xmath428 needs to have full row rank for all @xmath38 . as @xmath429 loses rank for @xmath430 , @xmath431 should have full row rank for every @xmath46 so that @xmath380 is controllable . this proves the existence of @xmath380 satisfying properties 1 and 2 of [ prob : smallestb ] . in order to satisfy property 3 in problem [ prob : smallestb ] , i.e. @xmath432 has to be the least , we have to choose a unimodular @xmath431 and free @xmath433 such that @xmath380 satisfies the three conditions . @xmath433 can be freely chosen because the choice of @xmath433 does not affect the input cardinality of @xmath380 . * ( 2 ) : * let @xmath379 be controllable . then , in the above , @xmath434 and @xmath308 does not exist . this means that @xmath433 does not exist . thus the kernel representation matrix of @xmath380 would be given by @xmath435 $ ] where @xmath431 is unimodular . therefore , @xmath380 is unique . let @xmath379 be uncontrollable . then @xmath308 exists and as shown above , free @xmath433 can be chosen . this makes the behaviour @xmath380 non - unique . * ( 3 ) : * the input cardinality of @xmath380 is given by @xmath436 here , @xmath437 which is determined by the size of the @xmath308 matrix . we saw above in this section that orthogonality of two uncontrollable behaviors is defined by requiring these uncontrollable behaviors to be sub - behaviors of two orthogonal controllable behaviors . using the result on existence of super - behaviors that are controllable , and their non - uniqueness even if the are the smallest controllable superbehavior , we formulate the question of whether two uncontrollable behaviors are orthogonal as a question of finding a pair of smallest controllable superbehaviors that are mutually orthogonal . the requirement of them being smallest is motivated by the fact that orthogonality of two controllable behaviors imposes an upper bound on their input cardinalities : this is reviewed below . for a behavior @xmath115 and a nonsingular matrix @xmath72 , the set @xmath438 is defined as follows @xmath439 it is straightforward that @xmath440 is also a behavior , its controllability is equivalent to that of @xmath26 , and the input cardinalities are equal . let @xmath379 and @xmath441 and suppose @xmath70 is nonsingular . then , the following are true . 1 . @xmath384 @xmath442 @xmath443 . 2 . @xmath444 @xmath445 @xmath446 . 3 . @xmath384 @xmath445 @xmath446 . due to the above inequality constraint on the input - cardinalities of orthogonal controllable behaviors , the uncontrollable behaviors too have a necessary condition to satisfy for mutual orthogonality . [ lem : uncontrol : ortho ] suppose @xmath379 and @xmath398 with at least one of them uncontrollable and let @xmath70 be nonsingular . . then @xmath447 . consider the pair of ` seemingly ' orthogonal behaviors studied in @xcite . define @xmath379 and @xmath398 by @xmath448 and the supply rate @xmath449 . thus @xmath379 is autonomous . consider the following non - observable latent variable representation for @xmath380 : @xmath450 . it can be checked that the ` storage function ' @xmath451 satisfies @xmath452 . existence of such a storage function is , in fact , reasonable for a ( different ) definition of orthogonality of the two behaviors @xmath379 and @xmath380 . in fact , any autonomous behavior @xmath453 is then ` orthogonal ' to @xmath454 ! however , the ` embeddability ' definition we have used above rules out this example for an orthogonal pair of behaviors since the necessary condition of the above lemma is not satisfied . in other words , there does nt exist a controllable behavior @xmath385 such that @xmath385 contains @xmath379 and @xmath455 . thus @xmath379 and @xmath380 are not @xmath456-orthogonal . in this section we look into the input cardinality condition for dissipative behaviors . recall that a behavior @xmath115 which is dissipative with respect to the supply rate @xmath72 ( constant , symmetric , nonsingular matrix ) satisfies the condition @xmath457 . we now look into the possibility of embedding a behavior @xmath26 in a controllable superbehavior that is @xmath72-dissipative , and into the drawback of using this as the definition of dissipativity , along the lines of orthogonality defined in the previous section . [ prob : plusminus ] given a nonsingular , symmetric and indefinite @xmath70 , find conditions for existence of a behavior @xmath71 such that * there exist @xmath458 and @xmath459 with @xmath460 . * @xmath458 is strictly @xmath72 dissipative * @xmath461 is strictly @xmath462 dissipative . the significance of the above problem is that if a nonzero behavior @xmath26 satisfying above conditions exists , then clearly such a behavior would be both strictly @xmath72 and strictly @xmath462 dissipative , raising concerns about whether embeddability in a dissipative controllable superbehavior is a reasonable definition of dissipativity ( when dealing with uncontrollable behaviors ) . the following theorem states that nonzero autonomous behaviors can indeed exist satisfying above condition . [ thm : strict : embeddable : autonomou ] let @xmath70 be nonsingular , symmetric and indefinite . then there exists a nonzero @xmath71 such that requirements in problem [ prob : plusminus ] are satisfied . further , any such @xmath26 satisfies @xmath463 , i.e. @xmath26 is autonomous . we first show that given any nonsingular , symmetric and indefinite @xmath70 , a behavior @xmath26 satisfying above properties exists . without loss of generality , let @xmath464 with sizes of the identity matrices @xmath465 and @xmath466 equal to @xmath467 and @xmath468 . due to indefiniteness of @xmath72 , @xmath469 and , due to the nonsingularity , they sum up to @xmath13 . choose any satisfying the stated conditions . ] nonzero polynomial matrix @xmath470 $ ] with @xmath471 for all @xmath82 for some @xmath472 . similarly , choose @xmath473 $ ] such that @xmath474 for all @xmath82 for some @xmath475 . now define @xmath458 and @xmath461 by image representations @xmath476 and @xmath477 respectively . define @xmath478 , thus proving and @xmath479 were nonconstant polynomial matrices , then @xmath26 is not the zero behavior . an example following the proof makes this easier to see . ] existence of @xmath26 as stated in the theorem . we now show that @xmath26 is autonomous . let @xmath480 and @xmath481 be minimal kernel representation matrices of @xmath458 and @xmath461 respectively . then @xmath460 is described by the kernel representation matrix @xmath482 $ ] with @xmath483 $ ] . clearly , @xmath484 . . then , there exists @xmath486 $ ] and @xmath487 such that @xmath488 and @xmath489 . this implies @xmath490 and @xmath491 . taking @xmath492 with @xmath493 and @xmath494 , it follows that @xmath495 . further , @xmath496 , hence @xmath497 because @xmath42 is non - zero and of compact support . further , we have @xmath498 such that @xmath499 both the above conditions can not be satisfied simultaneously for @xmath497 . thus , @xmath500 gives a contradiction . this proves @xmath501 and hence autonomy of @xmath26 . we illustrate the above theorem using an example . let @xmath502 . define @xmath458 and @xmath461 by image representations @xmath476 and @xmath477 respectively , with @xmath503 strict dissipativities is easily verified . calculating the kernel representations , we get a kernel representation for @xmath478 as @xmath504 @xmath505 clearly , @xmath24 is nonsingular and hence @xmath26 is autonomous . for non - strict dissipativity case , we have the following problem and theorem . [ prob : plusminus2 ] given a nonsingular , symmetric and indefinite @xmath70 , find conditions for existence of a behavior @xmath71 such that * there exist @xmath458 and @xmath459 such that @xmath460 . * @xmath458 is @xmath72 dissipative * @xmath461 is -@xmath72 dissipative . let @xmath70 be nonsingular , symmetric and indefinite . then there exists @xmath71 such that requirements in problem [ prob : plusminus2 ] are satisfied . any such @xmath26 satisfies @xmath506 . in case @xmath26 is uncontrollable , @xmath507 + if @xmath508 , then neither @xmath458 nor @xmath461 can be strictly dissipative . the proof proceeds in the same way as the proof for the previous theorem , except for the strictness of the dissipativities . construct @xmath458 and @xmath461 as in the previous proof , but with @xmath509 and @xmath510 equal to zero . we have @xmath511 the above two equations imply @xmath512 for all @xmath513 . since @xmath514 , the behavior @xmath26 is dissipative with respect to both @xmath72 and @xmath462 . dissipativity with respect to @xmath72 implies @xmath515 . similarly , dissipativity with respect to @xmath462 implies @xmath516 . this implies @xmath517 if @xmath26 is uncontrollable , then from theorem [ thm : superbehavior : controllable ] , then the two inequalities leading to the inequality are both strict . hence , the input cardinality of @xmath26 has to be strictly less than that of @xmath458 and @xmath461 . this implies @xmath518 if @xmath508 , then from theorem [ thm : strict : embeddable : autonomou ] , @xmath458 and @xmath461 can not be strictly dissipative with respect to @xmath72 and @xmath462 respectively . this completes the proof . as one of the consequences of the above theorem , if the input cardinality condition is satisfied for an uncontrollable behavior , i.e. @xmath519 or @xmath520 , then such a behavior can not be embedded into both a @xmath150-dissipative controllable behavior and a @xmath521-dissipative controllable behavior . however , an observable storage function for such a situation exists when the controllable part is strictly dissipative at @xmath522 and when the uncontrollable modes satisfy the unmixing condition ( see [ main ] and @xcite for this situation in presence of more assumptions ) , a situation when @xmath519 is very familiar : we deal with rlc circuits in the next section . we use this method of defining orthogonality of two behaviors to explore further the definition of dissipativity of a behavior . here we bring out a fundamental significance of the so - called input - cardinality condition : the condition that the number of inputs to the system is equal to the positive signature of the matrix that induces the power supply . we show that when this condition is not satisfied , then a behavior could be both supplying and absorbing net power , and is still not lossless . in this brief section we revisit a classical result : a rational transfer function matrix being positive real is a necessary and sufficient condition for that transfer matrix to be realizable using only resistors , capacitors and inductors ( see @xcite and also @xcite for the case with transformers ) . note that the transfer matrix captures only the controllable part of the behavior and positive realness of the transfer matrix is nothing but dissipativity with respect to the supply rate @xmath523 . this is made precise below . consider an @xmath158-port electrical network ( with each port having two terminals ) and the variable @xmath524 , where @xmath287 is the vector of voltages across the @xmath158-ports and @xmath525 is the vector of currents through these ports , with the convention that @xmath523 is the power flowing _ into _ the network . behaviors that are dissipative with respect to the supply rate @xmath523 are also called ` passive ' . given a controllable behavior @xmath26 whose transfer matrix @xmath526 with respect to a specific input / output partition , say current @xmath525 is the input and voltage @xmath287 is the output , is positive real , one can check that this behavior is passive . @xmath526 in this case is the impedance matrix and is square , i.e. the number of inputs is equal to the number of outputs . one can introduce additional laws that the variables need to satisfy , thus resulting in a sub - behavior @xmath527 which has a lesser number of inputs ; consider for example these additional laws as putting certain currents equal to zero : due to opening of certain ports . however , the transfer matrix for @xmath528 with respect to the input / output partition : input as currents through the non - open ports and output as the voltages across _ all _ the ports , is clearly not square , and in fact , tall , i.e. has strictly more rows than columns . let @xmath529 denote this transfer function . since @xmath530 , the behavior @xmath528 is also passive . of course , @xmath528 need not be controllable , even if @xmath26 is assumed to be controllable . as an extreme case , suppose all the currents are equal to zero , and we obtain an autonomous @xmath528 . while rlc realization of such transfer matrices which are tall , and further of autonomous behaviors obtained by , for example , opening all ports , has received hardly any attention , we remark here one very well - studied sub - behavior of every passive behavior : the zero behavior . consider the single port network with @xmath531 and @xmath532 . this port which is called a nullator ( also nullor , in some literature ) behaves as both the open circuit and shorted circuit : see @xcite and @xcite . the significant fact about a nullator is that a nullator can not be realized using only passive elements , and moreover any realization ( necessarily active ) leads to the realization of both a nullator and its companion , the norator ; a norator is a two - terminal port that allows both the voltage across it and current through it to be arbitrary . we briefly review the main results in this paper . we first used the existence of an observable storage function as the definition of a system s dissipativity and proved that the dissipativities of a behavior and its controllable part are equivalent assuming the uncontrollable poles are unmixed and the dissipativity at infinity frequency is strict . this result s proof involved new results in the solvability of are and used indefinite linear algebra results . we showed that for lossless autonomous dissipative systems , the storage function can not be observable , thus motivating the need for unobservable storage functions . we then studied orthogonality / lossless behaviors in the context of using the definition of existence of a controllable dissipative superbehavior as a definition of dissipativity . in addition to results about the smallest controllable superbehavior , we showed necessary conditions on the number of inputs for embeddability of lossless / orthogonal behaviors in larger controllable such behaviors . in the context of embeddability as a definition of dissipativity , we showed that one can always find behaviors that can be embedded in both a strictly dissipative behavior and a strictly ` anti - dissipative ' behavior , thus raising a question on the embeddability definition . we related this question to the well - known result that the nullator one - port circuit is not realizable using only rlc elements . j. c. willems , dissipative dynamical systems - part i : general theory , part ii : linear systems with quadratic supply rates , _ archive for rational mechanics and analysis _ , 45 , pages 321351 , 352393 , 1972 ,

a self - consistent theory involving maxwell equations and a density - matrix linear - response theory is solved for an electromagnetically - coupled doped graphene micro - ribbon array and a quantum - well electron gas sitting at an interface between a half - space of air and another half - space of a doped semiconductor substrate which supports a surface - plasmon mode in our system . the coupling between a spatially - modulated total electromagnetic field and the electron dynamics in a dirac - cone of a graphene ribbon , as well as the coupling of the far - field specular and near - field higher - order diffraction modes , are included in the derived electron optical - response function . full analytical expressions are obtained with non - locality for the optical - response functions of a two - dimensional electron gas and a graphene layer with an induced bandgap , and are employed in our numerical calculations beyond the long - wavelength limit ( drude model ) . both the near - field transmissivity and reflectivity spectra , as well as their dependence on different configurations of our system and on the array period , ribbon width , graphene chemical potential of quantum - well electron gas and bandgap in graphene , are studied . moreover , the transmitted e - field intensity distribution is calculated to demonstrate its connection to the mixing of specular and diffraction modes of the total electromagnetic field . an externally - tunable electromagnetic coupling among the surface , conventional electron - gas and massless graphene intraband plasmon excitations is discovered and explained . furthermore , a comparison is made between the dependence of the graphene - plasmon energy on the ribbon width and chemical potential in this paper and the recent experimental observation given by ju , _ et al . _ , [ nature nanotechnology , * 6 * , 630 ( 2011 ) ] for a graphene micro - ribbon array in the terahertz - frequency range .

1212.5104

there has been a high - level of research and device interests on both the electronic and optical properties of two - dimensional ( 2d ) graphene material@xcite since the report on the first successful isolation of single graphene layers as well as the related transport and raman experiments performed for such layers@xcite . the most distinctive difference between a graphene sheet and a conventional 2d electron gas ( eg ) in a semiconductor quantum well ( qw ) is the electronic band structure , where the energy dispersions of electrons and holes in the former are linear in momentum space but they become quadratic for the latter . as a result , electrons and holes in graphene behave like massless dirac fermions and show a lot of unexpected physical phenomena in electron transport and optical response , including anomalous quantum hall effect@xcite , bare and dressed state klein tunneling@xcite and plasmon excitation@xcite , a universal absorption constant@xcite , tunable intraband@xcite and interband@xcite optical transitions , broadband @xmath0polarization effect@xcite , photoexcited hot - carrier thermalization@xcite and transport@xcite , electrically and magnetically tunable band structure for ballistic transport@xcite , field - enhanced mobility in a doped graphene nanoribbon@xcite and electron - energy loss in gapped graphene layers@xcite . it has been found that most of the unusual electronic properties in graphene can be very well explained by single - particle excitation of electrons . for diffusion - limited electron transport in doped graphene , the kubo linear - response theory@xcite , as well as the hartree - fock theory in combination with the self - consistent born approximation@xcite , were used . moreover , the semiclassical boltzman theory was applied in both the linear@xcite and nonlinear@xcite response regimes . for the collective excitation of electrons in graphene , previous studies found it has an important role in the dynamical screening of the electron - electron interaction@xcite . however , the em response of graphene materials has received relatively less attention , especially for low - energy intraband optical transions@xcite . for any charged - particle systems , there always exists an electronic coupling based on the intrinsic electron - electron interaction in addition to the system response to an external electromagnetic field or an external electron ( ion ) beam . this electronic coupling can be either a classical direct coulomb interaction between any two charged particles with no requirement for an overlap of their wave functions , or quantum - mechanical tunneling and exchange coulomb interaction with a tail overlap of their wave functions@xcite . for the same charged - particle systems , there exists another electromagnetic coupling based on the light - electron interaction@xcite . this electromagnetic coupling results from an induced optical - polarization field in the system due to dipole coupling to an incident light wave . the total electromagnetic field , including the induced optical - polarization field , should satisfy the macroscopic maxwell equations@xcite . however , the induced optical - polarization field in the system is determined by the microscopic density - matrix equations@xcite . therefore , a self - consistent theory is required for studying the electromagnetic response of such a charged system to incident light . physically , a statistical ensemble average of dipole moments in an electronic system with respect to all quantum states will give rise to an induced optical - polarization field@xcite . this provides a connection between the microscopic density - matrix equations and the macroscopic maxwell equations@xcite . the former describes the full electron dynamics excited by a total electromagnetic field , including optical pumping , energy relaxation , optical coherence and its dephasing , and carrier scattering with impurities , lattice and other charged carriers@xcite . when the incident field is weak , a self - consistent linear - response theory to the total electromagnetic field ( not just the incident light ) can be applied to an electronic system as a leading - order approximation . experimentally , the inelastic light scattering@xcite and electron - energy - loss spectroscopy@xcite were widely used for probing the bulk and surface electronic coupling , respectively , in a charged - particle system . on the other hand , the elastic light scattering@xcite and the optical absorption spectroscopy@xcite were extensively applied for studying the surface and thin - film electromagnetic coupling in either a conductive or a dielectric system . for a smooth conductive surface , only electronic excitations in the long - wavelength limit can be coupled to an incident light wave , where a drude - type model is usually adopted in spectroscopy analysis . for a patterned conductive surface in the deep sub - wavelength regime , on the other hand , surface light scattering can be used to explore surface electronic excitations in the system beyond the long - wavelength limit , where a wavenumber - dependent ( nonlocal ) optical response of electrons , which is crucial for the near - field distribution and the landau damping of plasmons@xcite , must be employed to analyze the light reflectivity and transmissivity spectra . in this paper , we consider a model in which a two - dimensional electron gas ( 2deg ) and a doped linear graphene micro - ribbon array ( electronically isolated from each other ) are put on the surface of a doped semi - infinite semiconductor substrate . therefore , we expect to see the excitations of the localized surface plasmon and the conventional 2deg plasmon , as well as the 2d dirac - fermion plasmon excitation , in such a system in the presence of an incident electromagnetic wave . more interestingly , we expect there to exist strong electromagnetic couplings among these three collective excitations in the deep sub - wavelength regime , i.e. with an array period much less than the incident wavelength , and these couplings can be externally tuned by the doping densities in a graphene layer or in a qw , as well as by the driven 2deg in the qw or by the induced bandgap in the graphene . the existence of an bandgap in the graphene not only modifies the landau damping of plasmons@xcite , but also introduces the interplay between dirac - like and schrdinger - like electrons in the electromagnetic response . to demonstrate the predicted strong and tunable electromagnetic coupling among three plasmon excitations in our model system , numerical calculations and their comparisons are presented for both the transmissivity and reflectivity spectra with different configurations of our model system and various array periods , ribbon widths , chemical potentials of graphene , and induced bandgaps in graphene . the rest of the paper is organized as follows . in sec.[sect2 ] , we derive the self - consistent maxwell equations and the density - matrix linear - response theory within the random - phase approximation for our model system which consists of a combined 2deg in an inas qw and a linear graphene micro - ribbon array doped by electrons on the surface of a semi - infinite @xmath1-doped gaas substrate . this self - consistent theory , along with the detailed expressions for a linear matrix equation in appendix a , is then applied to study the transmissivity and reflectivity spectra of an incident plane wave for various structural - geometry and material parameters . in sec.[sect3 ] , numerical results , including the @xmath0polarized far- and near - field transmissivity spectra as well as the transmitted @xmath0polarized e - field intensity distribution , are presented to demonstrate and explain the existence of an externally - tunable electromagnetic coupling among the localized surface plasmon , 2deg plasmon and massless graphene plasmon excitations in our system . the conclusions drawn from these results are briefly summarized in sec.[sect4 ] . it is shown in fig.[f1 ] that the half - space @xmath2 is air with the refractive index @xmath3 whereas the half - space @xmath4 is occupied by a doped semi - infinite gaas substrate with a dynamical dielectric function @xmath5 which supports a localized surface - plasmon ( sp ) mode . at the interface @xmath6 , there are a periodic electron - doped graphene micro - ribbon array ( gmra ) and an inas quantum well ( qw ) for a 2deg . optically , the gmra and qw can be considered to be in one plane . the reason for this is that the sheet thickness is much smaller than the wavelength and the decay length of the impinging light . they can , however , still be electrically separated from each other by an energy barrier@xcite . a plane - wave electromagnetic ( em ) field incident from @xmath2 is diffracted by the gmra into both the reflection ( @xmath2 ) and transmission ( @xmath4 ) regions . the diffraction modes of the em field are modulated by the induced optical polarization at the interface with the gmra , which will be electromagnetically coupled to sp and 2deg plasmons . we assume that the gmra is periodic in the @xmath7 direction , and seek the solutions based on a plane - wave expansion for a single frequency . setting @xmath8 for simplicity and denoting the transverse electric component of the em field below the interface by @xmath9 and above by @xmath10 , we obtain@xcite @xmath11 \nonumber\\ & + & \sum\limits_{n=-\infty}^{\infty}\,e^{iq_nx - i\beta_{1,n}z } \left[\begin{array}{c } \left.b_x ( q_n\right|\omega)\\ \left.b_y ( q_n\right|\omega)\\ \left.(q_n/\beta_{1,\,n})\,b_x(q_n\right|\omega ) \end{array}\right]\ , \ \ \ \ \mbox{for $ z<0$}\ , \label{e1}\end{aligned}\ ] ] @xmath12\ , \ \ \ \ \mbox{for $ z>0$}\ . \label{e2}\end{aligned}\ ] ] here , @xmath13 label all reciprocal lattice vectors @xmath14 , @xmath15 is the period of the gmra , @xmath16 , @xmath17 , @xmath18 , @xmath19 is the refractive index of air , @xmath20 is the angle of incidence , and @xmath21 the angular frequency of the incident em field . in addition , @xmath22 and @xmath23 in eqs.([e1 ] ) and ( [ e2 ] ) are determined by the following dispersion relations for transverse em fields @xmath24=\frac{\omega^2}{c^2}\left [ \begin{array}{c } n_a^2\\ \epsilon_s(q_n,\beta_{2,n},\,\omega ) \end{array}\right]-q_n^2\ , \label{e3}\ ] ] where @xmath25 is the speed of light in vacuum , @xmath26 is the dielectric function of the doped semiconductor substrate , and we require that @xmath27 so as to ensure a finite value for the reflected or transmitted em field at @xmath28 . formally , if we keep only the term with @xmath29 in the summations in eqs.([e1 ] ) and ( [ e2 ] ) , the gmra in our model will simply change to a graphene sheet . for this situation , after removing the doping in the gaas substrate and the 2deg in the inas qw , the formalism in this paper will reduce to the previous ones@xcite for a transverse em response from a single graphene sheet . in the long - wavelength limit , the transverse dielectric function @xmath30 introduced in eq.([e3 ] ) for the doped semiconductor substrate is found to be@xcite @xmath31\ , \label{difunc}\ ] ] where @xmath32 is the bulk plasma frequency , @xmath33 is the doping concentration in gaas , @xmath34 is the homogeneous level broadening , @xmath35 is the effective mass of electrons in the substrate and @xmath36 is the relative dielectric constant of the semiconductor substrate . once the electric field @xmath37 is obtained by using the maxwell equations , the magnetic field @xmath38 for non - magnetic materials can be obtained from @xmath39 where @xmath40 is the vacuum permeability . for @xmath8 and @xmath2 , equation ( [ e4 ] ) explicitly leads to @xmath41\ , \label{e5}\ ] ] @xmath42\ , \label{e6}\ ] ] @xmath43\ . \label{e7}\ ] ] in a similar way , the magnetic field @xmath44 in the region of @xmath4 can also be obtained from @xmath45 @xmath46 @xmath47 electronically , the gmra and 2deg can be separated by an undoped wide - gap semiconductor barrier layer . optically , however , the gmra and 2deg can both be regarded as two - dimensional sheets since their thickness are much less than the em - field wavelength considered . therefore , the four independent boundary conditions for transverse em fields at the interface @xmath6 can be written as@xcite @xmath48 where @xmath49 is the induced sheet optical - polarization field ( related to the polarization current ) while @xmath50 is the induced sheet conduction - current density . furthermore , we can write @xmath51 as@xcite @xmath52 where @xmath53=\chi_1(q_n,\,\omega)\left[\begin{array}{c } \left.c_x(q_{n}\right|\omega)\\ \left.c_y(q_{n}\right|\omega ) \end{array}\right]\ ] ] @xmath54 \left[\begin{array}{c } \left.c_x(q_{n^\prime}\right|\omega)\\ \left.c_y(q_{n^\prime}\right|\omega ) \end{array}\right]\ . \label{e16}\ ] ] in eq.([e16 ] ) , @xmath55 , @xmath56 is the width of a graphene micro - ribbon , @xmath57 , and @xmath58 and @xmath59 are the 2deg and the graphene - ribbon polarizabilities , respectively . the multi - ribbon effect represented by the summation over @xmath60 , as well as the mode - mixing effect with @xmath61 , are both taken into account by the second term of eq.([e16 ] ) . moreover , the coupling between a spatially - dependent total em field and the electron dynamics in an energy band , due to the non - locality in an optical response of electrons , is also included in eq.([e16 ] ) . similarly , using ohm s law we can write the sheet conduction - current density as@xcite @xmath62 where @xmath63=\sigma_1(q_{n},\,\omega)\left[\begin{array}{c } \left.c_x(q_{n}\right|\omega)\\ \left.c_y(q_{n}\right|\omega ) \end{array}\right]\ ] ] @xmath64 \left[\begin{array}{c } \left.c_x(q_{n^\prime}\right|\omega)\\ \left.c_y(q_{n^\prime}\right|\omega ) \end{array}\right]\ , \label{e18}\ ] ] and @xmath65 and @xmath66 are the real 2deg and graphene - ribbon optical conductivities , respectively . for an incoming @xmath0polarized em field , we get @xmath67 and @xmath68 , where @xmath69 is the magnetic component and @xmath70 is the permittivity of free space . combining the sheet optical polarization in eq.([e15 ] ) and the sheet - current density in eq.([e17 ] ) , we can write the optical - response function for @xmath71 in a compact form given by @xmath72 where @xmath73 . we notice that there exists an anisotropy in the polarizability of a graphene nanoribbon along the longitudinal and transverse directions@xcite , respectively , and it can be attributed to a subband quantization due to finite - size effect . however , this anisotropy can be neglected for the width of ribbons in the micrometer range . in other words , we adopt in this paper the isotropy of the micro - ribbon array instead of anisotropy of the nanoribbon array . within the random - phase approximation , for graphene ribbons in the micrometer range we calculate the nonlocal ( @xmath74 ) optical - response function @xmath75 as@xcite @xmath76 where the non - retarded limit with @xmath77 has been used and @xmath78 is defined as @xmath79 @xmath80 is the surface area of a graphene micro - ribbon , @xmath81 is the fermi - dirac function for thermal - equilibrium electrons , @xmath82 is the kinetic energy of electrons , and @xmath83 represent the valence ( @xmath84 ) and conduction ( @xmath85 ) bands of dirac cones in a graphene layer . the dirac - cone model only holds for crystal momentum that is not too large . in fact , the energy band structure ceases to be a cone above about 2.5 ev . for our medium - high - density sample used in this paper , the fermi momentum is much less than the momentum separation between valleys at @xmath86 and @xmath87 . the coulomb interaction between electrons in different valleys will be significantly suppressed due to this large momentum separation . additionally , the electron - phonon mediated inter - valley scattering also requires a temperature higher than room temperature . since we assume very low temperature for electrons in our model , this inter - valley scattering may be justifiably neglected . the 2d dielectric function @xmath88 for graphene micro - ribbons can be obtained by@xcite @xmath89 in the non - retarded limit with @xmath90 being the dielectric constant of graphene materials . in addition , the optical conductivity @xmath91 of graphene ribbons can be calculated from eq.([e19 ] ) within the linear - response theory , i.e. @xmath92 $ ] . in a similar way , @xmath93 can be calculated from eq.([e128 ] ) below . note that @xmath94 along with eq.([e20 ] ) correspond to 2d rather than 3d fourier transform of the coulomb interaction . after a lengthy calculation@xcite , from eq.([e21 ] ) we get an analytical expression for a gapped graphene layer at @xmath95k as follows : @xmath96{\cal q}_{1_<}(x_{2,-})+\left[g_<(x_{1,-})+ig_>(x_{1,+})\right]{\cal q}_{2_<}(x_{2,-},\,x_{2,+})\right . \nonumber\\ & + & \left[g_<(x_{1,+})+g_<(x_{1,-})\right]{\cal q}_{3_<}(x_{2,-})+\left[g_<(x_{1,-})-g_<(x_{1,+})\right]{\cal q}_{4_<}(x_{2,+ } ) \nonumber\\ & + & \left[g_>(x_{1,+})-g_>(x_{1,-})\right]{\cal q}_{1_>}(x_{2,-},\,x_3)+\left[g_>(x_{1,+})+ig_<(x_{1,-})\right]{\cal q}_{2_>}(x_{2,-},\,x_{2,+ } ) \nonumber\\ & + & \left[g_>(x_{1,+})-g_>(-x_{1.-})-i\pi[2-x_0 ^ 2]\right]{\cal q}_{3_>}(x_{2,+ } ) \nonumber\\ & + & \left[g_>(-x_{1,-})+g_>(x_{1,+})-i\pi[2-x_0 ^ 2]\right]{\cal q}_{4_>}(x_{2,-},\,x_3 ) \nonumber\\ & + & \left.\left[g_0(x_{1,+})-g_0(x_{1,-})\right]{\cal q}_{5_>}(x_3)\right\}\ , \label{e22}\end{aligned}\ ] ] where @xmath97 is the chemical potential of electrons in a graphene layer with respect to zero energy at @xmath98 , @xmath99 is the fermi velocity of graphene , the kinetic energy of electrons in valence and conduction bands are @xmath100 and @xmath101 is the induced bandgap . from eq.([level ] ) we know that a finite effective mass @xmath102 of electrons will be created for a gapped graphene layer close to the edge of a gap with @xmath103 . the coulomb interaction between electrons will induce a self - energy , leading to an energy - dependent fermi velocity beyond the random - phase approximation@xcite . however , this energy - dependent behavior becomes significant only around the dirac point@xcite . for our samples in this paper with a very high doping concentration , we expect this effect is minimized . therefore , we have neglected the coulomb renormalization to the graphene fermi velocity in eq.([level ] ) . correction to the self - energy comes from correlation effects . however , this correction becomes significant only at low densities , as it occur , for example , when there is wigner crystallization . however , the sample considered in this paper is of medium - high - density concentration , where correlation effects may be neglected . since we recognize that the limitations of the model is a poor argument when it comes to the experimental data , we included in eq.(24 ) an effective mass @xmath104 term into dirac fermion picture . we note that the induced bandgap may be attributed to either photon dressed - state effects by circularly polarized light or to a substrate . there exist two main sources of disorder in graphene ribbons , i.e. , edge roughness and charged impurities . since wide micro - ribbons are considered in this paper , we can neglect the edge roughness . this leaves the charged impurities as the dominant scattering mechanism . the existence of charged impurities is also able to induce a self - energy , leading to a deformed density of states around the dirac point of graphene@xcite . however , for our medium - high - doping samples , the screening length of impurity potential is small , leading to a negligent effect on the density of states in graphene . moreover , with the added gap in eq.([level ] ) , the charged impurities create localized states only in the range between @xmath105 and @xmath101@xcite . since our transmission is not substantially affected up to @xmath106ev ( see fig.11 in sec.iii ) and the induced states by localized impurities are well above the terahertz regime of interest in this paper , we can neglect this effect here . for impurities , the interference or localization effect on the conductivity becomes important only for very high impurity concentrations . since the sample discussed in this paper is only a medium - high - density one , localization is not expected to be a dominant effect in comparison with other effects studied in this work . the detailed discussion related to the screened disordered effect on the electron polarizability of graphene beyond the random - phase approximation is , however , outside the scope of the current paper . three functions introduced in eq.([e22 ] ) are defined as @xmath107 moreover , nine region functions used in eq.([e22 ] ) are defined by @xmath108 where @xmath109 is the fermi wave number and @xmath110 is the unit step function . finally , we have defined six variables @xmath111 and @xmath112 through @xmath113 following a similar approach , within the random - phase approximation we get the nonlocal optical response function for the 2deg inside an inas qw@xcite @xmath114\right . \nonumber\\ & + & \left.i\left[d_-\sqrt{1-(z - u)^2}-d_+\sqrt{1-(z+u)^2}\,\right]\right\ } \ , \label{e128}\end{aligned}\ ] ] where @xmath115 is the 2deg density , @xmath116 and @xmath117 are the fermi wave number and the effective mass of 2deg . additionally , we have defined the notations in eq.([e128 ] ) for a driven 2deg : @xmath118 , @xmath119 , @xmath120 and @xmath121 ( @xmath122 and @xmath123 ) for @xmath124 ( @xmath125 ) , and @xmath126 and @xmath127 ( @xmath128 and @xmath129 ) for @xmath130 ( @xmath131 ) . for the specular mode with @xmath29 , from eqs.([e11])([e14 ] ) the boundary conditions for the em fields at the interface yield @xmath132 @xmath133 @xmath134 @xmath135 @xmath136 @xmath137 @xmath138 the terms with @xmath139 in eqs.([e33 ] ) and ( [ e34 ] ) reflect the coupling between the specular mode and higher - order diffraction modes . these generated higher - order diffracted em fields can be treated as source terms in addition to the incident em field in a perturbation picture . if we exclude all the terms with @xmath140 in eqs.([e33 ] ) and ( [ e34 ] ) , the solution of eqs.([e31])([e34 ] ) gives rise to a transverse em response for a single graphene sheet by taking @xmath141 , @xmath142 and @xmath143@xcite . for the diffraction modes with @xmath144 , on the other hand , from eqs.([e11])([e14 ] ) the boundary conditions for the em fields lead to @xmath145 @xmath146 @xmath147 @xmath148\,\bar{\chi}_2(q_{n^\prime},\,\omega)\left.b_y(q_{n^\prime}\right|\omega ) -\bar{\chi}_1(q_n,\,\omega)\left.b_y(q_n\right|\omega)=0\ , \label{e37}\ ] ] @xmath149 @xmath148\,\bar{\chi}_2(q_{n^\prime},\,\omega)\left.b_x(q_{n^\prime}\right|\omega ) -\bar{\chi}_1(q_n,\,\omega)\left.b_x(q_n\right|\omega)=0\ , \label{e38}\ ] ] where @xmath150 and @xmath151 correspond to a gmra only or a 2deg plus a gmra , respectively . the terms with @xmath152 in eqs.([e37 ] ) and ( [ e38 ] ) reflect the coupling between different diffraction modes with the specular mode . here , the terms with @xmath152 , excited directly by the incident em field , can be regarded as source terms for the generated higher - order diffracted em fields with @xmath144 in a perturbation picture . once the solution of the matrix equation derived from eqs.([e31])([e38 ] ) ( see appendix a ) is obtained , for @xmath0 polarization we can express the intensities of em field at the interface as @xmath153 ^ 2+\sum\limits_{n=-\infty}^{\infty}\left.|b_x(q_n\right|\omega)|^2 + \left.2a_x(q_x\right|\omega)\,{\rm re}\left.\left[b_x(q_x\right|\omega)\right ] \nonumber\\ & + & \sum\limits_{n,\,n^{\prime}}^{(n\neq n^{\prime})}\left.b_x(q_n\right|\omega ) \left.[b_x(q_{n^{\prime}}\right|\omega)]^{\ast}\,e^{i(n - n^{\prime})gx } \nonumber\\ & + & \left.2a_x(q_x\right|\omega)\,{\rm re}\,\left(\sum_{n\neq 0}\left.b_x(q_n\right|\omega)\,e^{ingx}\right)\ , \label{e39}\end{aligned}\ ] ] @xmath154 ^ 2+\sum\limits_{n=-\infty}^{\infty}\left.|b_y(q_n\right|\omega)|^2 + \left.2a_y(q_x\right|\omega)\,{\rm re}\left.\left[b_y(q_x\right|\omega)\right ] \nonumber\\ & + & \sum\limits_{n,\,n^{\prime}}^{(n\neq n^{\prime})}\left.b_y(q_n\right|\omega)\left.[b_y(q_{n^{\prime}}\right|\omega)]^{\ast}\,e^{i(n - n^{\prime})gx } \nonumber\\ & + & \left.2a_y(q_x\right|\omega)\,{\rm re}\,\left(\sum\limits_{n\neq 0}\left.b_y(q_n\right|\omega)\,e^{ingx}\right)\ , \label{e40}\end{aligned}\ ] ] @xmath155 ^ 2+\sum\limits_{n=-\infty}^{\infty}\ , \frac{q_n^2}{|\beta_{1,n}|^2}\left.|b_x(q_n\right|\omega)|^2\ ] ] @xmath156 + \sum\limits_{n,\,n^{\prime}}^{(n\neq n^{\prime})}\,\frac{q_nq_{n^{\prime } } } { \beta_{1,n}(\beta_{1,n^{\prime}})^{\ast } } \left.b_x(q_n\right|\omega)\left.[b_x(q_{n^{\prime}}\right|\omega)]^{\ast}\ ] ] @xmath157 where the interferences taking place in the @xmath7 direction at the interface between the reflected em field ( @xmath61 ) and that between the incident and reflected em fields ( @xmath158 ) are all included . using eqs.([e39])([e41 ] ) , we get the @xmath7-dependent square of the ratio @xmath159 of the reflected to the incident @xmath37-field amplitude at the interface , given by @xmath160 where @xmath161 contains both the near ( @xmath162 ) and far ( @xmath163 ) field contributions , and @xmath164^{\ast}\,e^{i(n - n^{\prime})gx}\ , \label{e43}\ ] ] @xmath165^{\ast}\,e^{i(n - n^{\prime})gx}\ , \label{e44}\ ] ] @xmath166^{\ast}\,e^{i(n - n^{\prime})gx}\ . \label{e45}\ ] ] following the same approach , we get the @xmath7-dependent square of the ratio @xmath167 of the transmitted to the incident @xmath37-field amplitude at the interface , given by @xmath168 where @xmath169 also includes both the near ( @xmath162 ) and far ( @xmath163 ) field contributions , and @xmath170^{\ast}\,e^{i(n - n^{\prime})gx}\ , \label{e47}\ ] ] @xmath171^{\ast}\,e^{i(n - n^{\prime})gx}\ , \label{e48}\ ] ] @xmath172^{\ast}\,e^{i(n - n^{\prime})gx}\ . \label{e49}\ ] ] after integrating eqs.([e42 ] ) and ( [ e46 ] ) with respect to @xmath7 , we obtain the far - field reflectivity and transmissivity spectra of the sample as @xmath173 @xmath174 where @xmath175 , the interference terms between different diffraction modes are canceled out and @xmath176 @xmath177 in our numerical calculations , we assume @xmath0 polarization for the incident em - field and set the parameters in numerical calculations as follows : @xmath178@xmath179 , @xmath180 with @xmath181 being the free electron mass , @xmath182 , @xmath183ev , @xmath184m / sec , @xmath185 , @xmath186mev , @xmath187mev , @xmath188 , @xmath189 m ( deep sub - wavelength regime ) , @xmath190 and @xmath17 with @xmath3 . in this paper , we choose the color scale from blue ( darker , minimum ) to red ( lighter , maximum ) . the change of these parameters will be directly indicated in the figure captions . in this paper , we only consider the low - energy intraband plasmon excitation in the terahertz - frequency range . to uncover the physical mechanism behind the reported graphene - plasmon peak shift with gmra period@xcite , we show a comparison in fig.[f2 ] for two calculated far - field transmissivity spectra @xmath191 , defined by eq.([e51 ] ) , as a function of photon energy @xmath192 with @xmath193 m ( red solid curve ) and @xmath189 m ( blue dashed curve ) . here , only the graphene micro - ribbon array is included in our system . as @xmath15 increases , the graphene plasmon peak at @xmath194mev for @xmath193 m shifts down to @xmath195mev for @xmath189 m and this peak shift accurately satisfies the @xmath196 scaling relation experimentally observed@xcite . a series of kinks below the plasmon peak in fig.[f2 ] corresponds to the single - particle excitations at @xmath197 ( for @xmath198 ) with @xmath199 for different values of integer @xmath1 . on the other hand , the shoulder appearing above the dominant peak comes from another severely - landau - damped graphene - plasmon peak , which becomes more visible for @xmath193 m since the peak separation is increased . to see the many - body effect on the hybridized gmra - qw plasmon excitation , fig.[f3 ] presents a comparison for @xmath191 as a function of @xmath192 with two values for the graphene chemical potential , i.e. , @xmath183ev ( red solid curve ) and @xmath200ev ( blue dashed curve ) . here , the chemical potential @xmath201 of graphene ribbons can be tuned by a gate voltage@xcite . as expected , one of the two graphene - like plasmon peaks at @xmath202mev for @xmath183ev moves up to @xmath203mev for @xmath200ev . interestingly , from this hybridized gmra - qw plasmon peak shift we find that the plasmon energy is not scaled as @xmath204 , which is different from the observation reported in ref . for the ribbon - only case . however , our calculated graphene plasmon peak shift ( from @xmath205mev to @xmath206mev ) for the ribbon - only case ( not shown here ) fully agrees with their observation@xcite . the @xmath201-insensitive plasmon peak at @xmath207mev is associated with the sp excitation at the photon energy @xmath208mev and will be addressed in fig.[f6 ] below . additionally , when @xmath200ev , we see a new strong graphene - like plasmon peak show up at @xmath209mev because one relatively - flattened portion@xcite on the plasmon dispersion curve at a relative large @xmath210 value becomes free of landau - damping . to further elucidate the importance of the em coupling between the graphene plasmon in gmra , as shown in fig.[f2 ] , and the 2deg plasmon in a qw , we displayed in fig.[f4 ] comparisons for @xmath191 [ in ( a ) ] and for the far - field reflectivity spectra @xmath211 , defined by eq.([e50 ] ) , [ in ( b ) ] as functions of @xmath192 by choosing four different array periods : @xmath212 m ( blue dash - dot - dotted curves ) , @xmath193 m ( black dashed curves ) , @xmath189 m ( red solid curves ) and @xmath213 m ( green dash - dotted curves ) . here , we fixed the ratio of the ribbon width @xmath56 to the array period @xmath15 by @xmath214 . as indicated by two downward arrows in fig.[f4](a ) for the shift of one of the two graphene - like plasmon peaks from @xmath202mev for @xmath189 m to @xmath215mev for @xmath193 m , we determine that the hybridized gmra - qw plasmon energy is approximately proportional to @xmath216 for @xmath188 . when @xmath213 m , only two weakly - split plasmon peaks are visible at @xmath217mev and @xmath218mev , respectively . as @xmath15 decreases from @xmath219 m to @xmath220 m , the peak at @xmath221mev for @xmath193 m is pushed up to @xmath222mev . these peak shifts with gmra period are also reflected in fig.[f4](b ) , as indicated by the two upward arrows for example , and the steep rise of @xmath211 around @xmath223mev is associated with the excitation of sp mode in the system . in order to clarify the fact that the graphene - like plasmon energy is actually proportional to @xmath224 , due to the non - locality in the optical response of interacting electrons , instead of @xmath225 in single - particle modes , we show @xmath191 in fig.[f5 ] for two cases with the same value for @xmath56 but different values for @xmath15 . for the case with @xmath226 and @xmath213 m , we find a weakly - split peak ( indicated by two upward arrows ) , which is aligned with a strong non - split peak ( indicated by a downward arrow ) at @xmath188 and @xmath189 m . this clearly demonstrates that the graphene - like plasmon peak energy is proportional to @xmath224 instead of @xmath225 , which is again in agreement with the observation by ju _ et al_@xcite . the finite width of a ribbon introduces a characteristic wavenumber ( proportional to @xmath227 ) , which enforces a cut - off to the short - range coulomb interaction between electrons in a micro - ribbon . a smaller @xmath228 value implies a weaker coupling between the 2deg and graphene plasmon excitations . the splitting of the plasmon peak for @xmath226 is attributed to the contribution of even - integer diffraction modes , while only odd - integer diffraction modes can contribute for @xmath188 , as can be verified by the @xmath229terms appearing in eqs.([e33 ] ) and ( [ e34 ] ) . figure [ f6 ] presents in the left panel the individual and combined effects of sp , 2deg and gmra as well as the strong em coupling among these plasmon excitations . the periodicity of gmra opens a minigap between any two adjacent branches ( bloch modes ) in a folded hybridized - plasmon dispersion curve . these minigaps can be opened either at the brillouin - zone center or at the brillouin - zone boundary . in the deep sub - wavelength regime , only the minigap at the brillouin - zone center leads to a peak in the far - field transmissivity spectrum@xcite . when the doped gaas substrate is changed into an undoped one ( no sp , blue dash - dot - dotted curve ) , only 2deg and graphene plasmons can exist in this system . in this case , there exists strong em coupling between the graphene and 2deg plasmon excitations . as labeled by the circled numbers , peak1 and peak4 are related to the 2deg - like plasmons , while peak2 and peak3 are connected to the graphene - like plasmon excitations . two small kinks on the outer shoulders of peak2 and peak3 are a result of two induced anti - crossing gaps between coupled 2deg - like and graphene - like plasmons . if the 2deg in an inas qw is removed from our system ( no qw , black dash - dotted curve ) , we observe that the sp peak is strengthened in comparison with the full system ( red solid curve ) and shifted towards the left ( indicated by two solid - line arrows ) due to turning - off the 2deg plasmon and its coupling to the sp in the system . furthermore , the graphene - like plasmon peaks superposed on the shoulder of the sp peak become greatly weakened due to the loss of the strong coupling between the 2deg and graphene plasmons . here , a simple drude - type optical - conductivity model@xcite can not be applied to our system due to the large wavenumber involved in the deep sub - wavelength regime for a higher - order diffracted em field . when only the gmra exists in the system ( ribbon only , green dashed curve ) , a downward - shifted dominant plasmon peak at @xmath195mev , along with several kinks ( a shoulder ) below ( above ) the peak , show up in this panel , as explained in fig.[f2 ] . from this figure we know that the graphene - ribbon effect can be best observed from hybridized gmra - qw plasmon modes in the absence of the sp . in the lower panel of fig.[f6 ] , we display the calculated transmitted @xmath0polarized e - field intensity @xmath230 at @xmath207mev for the full system . in this case , a resonant sp effect on the intensity distribution is expected around the qw - mediated air / gaas interface . indeed , the intensity in the gap between two neighboring graphene micro - ribbons , where the qw - mediated air / gaas interface is hosted , decreases greatly due to strong optical absorption by the sp ( @xmath187mev ) . at the same time , the intensity is built up at two edges of a micro - ribbon and spreads out to the regions covered by micro - ribbons . a slight asymmetry in the intensity distribution with respect to the left and right edges of the gap region is attributed to a finite incident angle ( @xmath231 ) . in the presence of an sp , we compare the transmitted @xmath0polarized e - field intensities @xmath230 in fig.[f7 ] for the cases of with a qw at @xmath232mev ( left panel ) or without a qw at @xmath233mev ( right panel ) . from the right panel of this figure , we find in the near - field regime that a gmra tends to build up very strong e - field intensities just at two edges of a micro - ribbon . on the other hand , the qw would like to spread the e - field intensity across the gap region between two neighboring micro - ribbons although two close - magnitude maxima in intensity distribution can still be seen between the gap center and edges . the competition of these two opposite effects constitutes a strong em coupling between the 2deg and graphene plasmon excitations , as shown in the upper panel of fig.[f6 ] . in addition , we notice that the e - field intensity for both panels is a little higher at the left edge of the gap region for the lower of the two graphene - like plasmon peaks in fig.[f6 ] . to find the strongly - diffracted near - field effect in the deep sub - wavelength regime , as shown in fig.[f7 ] , on the far - field transmissivity spectrum , we compare in fig.[f8 ] the calculated partial near - field transmissivity spectra @xmath234 for @xmath0 polarization , defined by eq.([e53 ] ) , with @xmath235 [ in ( a ) ] and @xmath236 [ in ( b ) ] . it is clear from figs.[f8](a ) and ( b ) that the graphene - like plasmon peak at @xmath232mev mainly comes from the contributions of the diffraction modes with @xmath235 , while the plasmon peak at @xmath202mev is produced jointly by the diffraction modes with both @xmath235 and @xmath236 . in general , the peak in the near - field transmissivity spectrum ( black solid and red dash - dotted curves ) can be much stronger than that in the far - field spectrum ( blue dashed curves ) due to the very strong near - field intensity as displayed in fig.[f7 ] . the non - locality in the optical response of electrons , as given by eqs.([e16 ] ) and ( [ e18 ] ) , facilitates the mixing between the em - field specular and diffraction modes , which enables transferring the peak weight from a near field to a far field . we also note that the peak strengths for the diffraction modes associated with @xmath237 are close in magnitude in this figure . surprisingly , for an inverted structure ( qw - ribbon , blue dashed curve ) , where the graphene ribbons become a graphene sheet while the inas qw sheet becomes qw ribbons , we find from fig.[f9 ] that the sp effect is greatly suppressed , compared to the original system ( g - ribbon , red solid curve ) , in addition to an overall reduction of peak strength in the transmissivity . however , the em - field reflectivity is found to be enhanced ( not shown here ) in this inverted structure for the range of photon energies shown in the figure . in this case , only one very weak peak at @xmath238mev is visible for the inverted structure , which is attributed to an order - of - magnitude lower electron density in the qw compared to that in a graphene micro - ribbon . at the same time , the sp - related peak shifts from @xmath207mev to @xmath239mev due to switching from the qw - sp coupling to the graphene - sheet - sp coupling . we display in the upper panel of fig.[f10 ] the effect of bandgap @xmath101 on @xmath191 with @xmath182 ( red solid curve ) , @xmath240ev ( black dash - dotted curve ) and @xmath241ev ( blue dashed curve ) . since a finite @xmath101 leads to an effective mass ( @xmath102 ) for electrons close to the band edge , its effect on @xmath191 is expected to be substantial ( see fig.[f9 ] ) as @xmath242 . indeed , we find from the upper panel that the strengths of the two graphene - like plasmon peaks are significantly reduced when @xmath241ev ( for @xmath243ev ) . correspondingly , from the lower panel of this figure , we see a decrease of near - field intensity as well as an enhanced em coupling between the 2deg and graphene plasmons , i.e. the e - field intensity spreads inward from the two edges to the center of the gap region . in conclusion , we have derived in this paper a self - consistent theory involving maxwell equations to determine a total em field and the density - matrix linear - response theory to determine an induced optical - polarization field . we have applied this self - consistent theory to study a model system composed of an electronically - isolated two - dimensional electron gas and an electron - doped graphene micro - ribbon array on the surface of an @xmath1-doped semi - infinite semiconductor substrate . the numerical results for the transmissivity and reflectivity spectra in the presence of an incident plane wave have been compared and physically explained in the deep sub - wavelength regime for various linear - array period , micro - ribbon width , doped graphene chemical potential and different configurations of our model system . our calculations have demonstrated the existence of a tunable em coupling among the localized surface , conventional electron - gas and massless graphene intraband collective excitations in the deep sub - wavelength regimes for terahertz frequencies . @xmath246\left.a_y(q_x\right|\omega)\\ \left[(\omega / c)\,\bar{\chi}_1(q_x,\,\omega)+(\omega / c)\,\zeta\,\bar{\chi}_2(q_x,\,\omega)-(i\omega / c\eta)\right]\left.a_x(q_x\right|\omega)\\ 0\\ \vdots\\ 0 \end{array}\right]\ . \label{a2}\ ] ] @xmath254 & \mbox{for $ j^\prime=3n+2 $ } \nonumber\\ -i\left(c\beta_{2,0}/\omega\right ) & \mbox{for $ j^\prime=7n+4 $ } \nonumber\\ -\left(\omega / c\right)\zeta\,{\rm sinc}(m^\prime\pi\zeta)\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[j_1,\,j_2]\ , \ [ j_2 + 2,\,j_3]$ } \nonumber\\ 0 & \mbox{for all other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] @xmath259 & \mbox{for $ j^\prime = n+1 $ } \nonumber\\ -i\left(\omega / c\beta_{2,0}\right)\epsilon_s(q_0,\beta_{2,0},\,\omega ) & \mbox{for $ j^\prime=5n+3 $ } \nonumber\\ -\left(\omega / c\right)\,\zeta\,{\rm sinc}(m^\prime\pi\zeta)\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[1,\,n]\ , \ [ j_1,\,j_2]$ } \nonumber\\ 0 & \mbox{for all other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] @xmath264 $ or $ j^\prime = j-3\in[n+2,2n+1]$ } \nonumber\\ -1\ \ \ \ & \mbox{for $ j^\prime = j_1\in[4n+3,\,5n+2]$ or $ j^\prime = j_1 + 1\in[5n+4,6n+3]$ } \nonumber\\ 0\ \ \ \ & \mbox{for all other $ j^\prime$ } \end{array}\right.\end{aligned}\ ] ] @xmath267 $ or $ j^\prime = j-2\in[3n+3,4n+2]$ } \nonumber\\ -1\ \ \ \ & \mbox{for $ j^\prime = j_1\in[6n+4,\,7n+3]$ or $ j^\prime = j_1 + 1\in[7n+5,8n+4]$ } \nonumber\\ 0\ \ \ \ & \mbox{for all other $ j^\prime$ } \end{array}\right.\end{aligned}\ ] ] @xmath270 & \mbox{for $ j^\prime = j-(j_1 + 1)$ } \nonumber\\ -i\left(c\beta_{2,m}/\omega\right ) & \mbox{for $ j^\prime = j+(j_1 - 3)$ } \nonumber\\ -\left(\omega / c\right)\,\zeta\,{\rm sinc}[(m^\prime - m)\pi\zeta]\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[j_1,\,j_2]$ but $ j^\prime\neq j_3 $ } \nonumber\\ 0 & \mbox{for all other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] @xmath270 & \mbox{for $ j^\prime = j - j_1 $ } \nonumber\\ -i\left(c\beta_{2,m}/\omega\right ) & \mbox{for $ j^\prime = j+j_1 - 2 $ } \nonumber\\ -\left(\omega / c\right)\,\zeta\,{\rm sinc}[(m^\prime - m)\pi\zeta]\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[j_1,\,j_2]$ but $ j^\prime\neq j_3 $ } \nonumber\\ 0 & \mbox{for all other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] @xmath278 & \mbox{for $ j^\prime = j - j_2 $ } \nonumber\\ -i\left(\omega / c\beta_{2,m}\right)\epsilon_s(q_m,\beta_{2,m},\,\omega ) & \mbox{for $ j^\prime = j-(j_1 + 1)$ } \nonumber\\ -\left(\omega / c\right)\,\zeta\,{\rm sinc}[(m^\prime - m)\pi\zeta]\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[1,\,j_1]$ but $ j^\prime\neq j_3 $ } \nonumber\\ 0 & \mbox{for all\ other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] @xmath278 & \mbox{for $ j^\prime = j - j_2 $ } \nonumber\\ -i\left(\omega / c\beta_{2,m}\right)\epsilon_s(q_m,\beta_{2,m},\,\omega ) & \mbox{for $ j^\prime = j - j_1 $ } \nonumber\\ -\left(\omega / c\right)\,\zeta\,{\rm sinc}[(m^\prime - m)\pi\zeta]\,\bar{\chi}_2(q_{m^\prime},\,\omega ) & \mbox{for $ j^\prime\in[1,\,j_1]$ but $ j^\prime\neq j_3 $ } \nonumber\\ 0 & \mbox{for all other $ j^\prime$ } \end{array}\right.\ , \end{aligned}\ ] ] k. s. novoselov , a. k. geim , s. v. morozov , d. jiang , m. i. katsnelson , i. v. grigorieva , s. v. dubonos , and a. a. firsov , `` electric field effect in atomically thin carbon films '' , sci . * 306 * , 666 - 669 ( 2004 ) . k. s. novoselov , a. k. geim , s. v. morozov , d. jiang , m. i. katsnelson , i. v. grigorieva , s. v. dubonos , and a. a. firsov , `` two - dimensional gas of massless dirac fermions in graphene '' , nat . * 438 * , 197 - 200 ( 2005 ) . r. r. nair , p. blake , a. n. grigorenko , k. s. novoselov , t. j. booth , t. stauber , n. m. r. peres , and a. k. geim , `` fine structure constant defines visual transparency of graphene '' , sci . * 320 * , 1308 ( 2008 ) . l. ju , b. geng , j. horng , c. girit , m. martin , z. hao , h. a. bechtel , x. liang , a. zettl , y. r. shen and f. wang , `` graphene plasmonics for tunable terahertz metamaterials '' , nat . * 6 * , 630 - 634 ( 2011 ) . j. c. w. song , m. s. rudner , c. m. marcus , and l. s. levitov , `` very slow cooling dynamics of photoexcited carriers in graphene observed by optical - pump terahertz - probe spectroscopy '' , nano lett . * 11 * , 4902 - 4906 ( 2011 ) . d. h. huang , c. rhodes , p. m. alsing , and d. a. cardimona , `` effects of longitudinal field on transmitted near field in doped semi - infinite semiconductors with a surface conducting sheet '' , j. appl . * 100 * , 113711 ( 2006 ) . d. h. huang , g. gumbs , p. m. alsing , and d. a. cardimona , `` nonlocal mode mixing and surface - plasmon - polariton - mediated enhancement of diffracted terahertz fields by a conductive grating '' , phys . b * 77 * , 165404 ( 2008 ) . d. h. huang , g. gumbs , and s .- y . lin , `` self - consistent theory for near - field distribution and spectrum with quantum wires and a conductive grating in terahertz regime '' , j. appl . phys . * 105 * , 093715 ( 2009 ) . d. h. huang , t. apostolova , p. m. alsing , and d. a. cardimona , `` high - field transport of electrons and radiative effects using coupled force - balance and fokker - planck equations beyond relaxation - time approximation '' , phys . b * 69 * , 075214 ( 2004 ) . d. h. huang , p. m. alsing , t. apostolova , and d. a. cardimona , `` effect of photon - assisted absorption on the thermodynamics of hot electrons interacting with an intense optical field in bulk gaas '' , phys . b * 71 * , 045204 ( 2005 ) . g. gumbs , d. h. huang , and d. n. talwar , `` doublet structure in the absorption coefficient for tunneling - split intersubband transitions in double quantum wells '' , phys . b * 53 * , 15436 - 15439 ( 1996 ) . d. c. elias , r. v. gorbachev , a. s. mayorov , s. v. morozov , a. a. zhukov , p. blake , l. a. ponomarenko , i. v. grigorieva , k. s. novoselov , f. guinea and a. k. geim , `` dirac cones reshaped by interaction effects in suspended graphene '' , nat . phys . * 7 * , 701 - 704 ( 2011 ) . and ribbon width @xmath288 and an inas quantum well ( red ) . the lower half - space ( @xmath2 ) is filled with air with refractive index @xmath3 , whereas the upper half - space ( @xmath4 ) is filled with a doped semi - infinite gaas bulk having a complex dielectric function @xmath5 or a complex refractive index @xmath289 . both the graphene micro - ribbon array [ with an optical - response function @xmath290 and the inas quantum well [ with an optical - response function @xmath291 sit on the surface ( @xmath6 ) of the semi - infinite gaas bulk . a plane - wave electromagnetic field is incident from the @xmath2 side with an incident angle @xmath20.,width=384 ] in the absence of both sp and qw for @xmath0 polarization with two given linear - array periods : @xmath193 m ( red solid curve ) and @xmath189 m ( blue dashed curve ) . two arrows indicate the shift of two corresponding peaks with @xmath15 . other parameters in calculations are given in the text.,width=288 ] for @xmath0 polarization with @xmath183ev ( red solid curve ) and @xmath200ev ( blue dashed curve ) . two downward solid - line arrows indicate the shift of two corresponding peaks with @xmath201 , while one upward dashed - line arrow indicates a new peak . other parameters in calculations are given in the text.,width=288 ] [ in ( a ) ] and the far - field reflectivity spectra @xmath211 [ in ( b ) ] for @xmath0 polarization with different linear - array periods : @xmath212 m ( blue dash - dot - dotted curves ) , @xmath193 m ( black dashed curves ) , @xmath189 m ( red solid curves ) and @xmath213 m ( green dash - dotted curves ) . two arrows indicate the shift of two corresponding peaks with @xmath15 . other parameters in calculations are given in the text.,width=288 ] [ in ( a ) ] and the far - field reflectivity spectra @xmath211 [ in ( b ) ] for @xmath0 polarization with different linear - array periods : @xmath212 m ( blue dash - dot - dotted curves ) , @xmath193 m ( black dashed curves ) , @xmath189 m ( red solid curves ) and @xmath213 m ( green dash - dotted curves ) . two arrows indicate the shift of two corresponding peaks with @xmath15 . other parameters in calculations are given in the text.,width=288 ] for @xmath0 polarization with @xmath189 m and @xmath188 ( red solid curve ) as well as with @xmath213 m and @xmath226 ( blue dashed curve ) . the peak indicated by a downward red arrow splits into two indicated by two upward blue arrows . other parameters in calculations are given in the text.,width=288 ] is presented in the upper panel for @xmath0 polarization and four different configurations of the system , including : ( i ) with gmra , qw and sp ( full , red solid curve ) ; ( ii ) @xmath292 ( no sp , blue dash - dot - dotted curve ) ; ( iii ) @xmath141 ( no qw , black dash - dotted curve ) ; ( iv ) @xmath293 ( ribbon only , green dashed curve ) . two solid - line arrows indicate the peak associated with the sp . the circled numbers label four peaks in the figure for the case of no sp . for the lower panel , the transmitted @xmath0polarized e - field intensity @xmath294 is shown at @xmath207mev for the full system , where the color scale is indicated . other parameters in calculations are given in the text.,width=288 ] is presented in the upper panel for @xmath0 polarization and four different configurations of the system , including : ( i ) with gmra , qw and sp ( full , red solid curve ) ; ( ii ) @xmath292 ( no sp , blue dash - dot - dotted curve ) ; ( iii ) @xmath141 ( no qw , black dash - dotted curve ) ; ( iv ) @xmath293 ( ribbon only , green dashed curve ) . two solid - line arrows indicate the peak associated with the sp . the circled numbers label four peaks in the figure for the case of no sp . for the lower panel , the transmitted @xmath0polarized e - field intensity @xmath294 is shown at @xmath207mev for the full system , where the color scale is indicated . other parameters in calculations are given in the text.,width=288 ] polarized e - field intensities @xmath294 in the presence of an sp with ( left ) or without ( right ) a qw at two indicated resonant photon energies , where two color scales are given in the left and right panels , respectively . other parameters in calculations are given in the text.,width=576 ] for @xmath0 polarization . in ( a ) , we take @xmath295 ( black solid curve ) , @xmath29 ( blue dashed curve ) and @xmath296 ( red dash - dotted curve ) , while in ( b ) we choose @xmath297 ( black solid curve ) , @xmath29 ( blue dashed curve ) and @xmath298 ( red dash - dotted curve ) . other parameters in calculations are given in the text.,width=288 ] for @xmath0 polarization . in ( a ) , we take @xmath295 ( black solid curve ) , @xmath29 ( blue dashed curve ) and @xmath296 ( red dash - dotted curve ) , while in ( b ) we choose @xmath297 ( black solid curve ) , @xmath29 ( blue dashed curve ) and @xmath298 ( red dash - dotted curve ) . other parameters in calculations are given in the text.,width=288 ] polarized transmissivity spectra @xmath191 for two compensated structures , including a graphene micro - ribbon array plus an inas quantum - well sheet ( g - ribbon , red solid curve ) and an inas quantum - well ribbon array plus a graphene sheet ( qw - ribbon , blue dashed curve ) . a blue upward arrow indicates a weak peak for the qw - ribbon array . other parameters in calculations are given in the text.,width=288 ] is made in the upper panel for @xmath0 polarization with three different bandgaps @xmath182 ( @xmath299mev , red solid curve ) , @xmath240ev ( @xmath300mev , black dash - dotted curve ) and @xmath241ev ( @xmath301mev , blue dashed curve ) . two pairs of arrows indicate the shift of a pair of corresponding peaks with @xmath101 . in the lower panel , a comparison of the calculated transmitted @xmath0polarized e - field intensities @xmath294 is displayed for @xmath182 ( left ) and @xmath241ev ( right ) at @xmath202mev and @xmath302mev , respectively . here , we keep the electron areal density ( @xmath303 ) in a graphene micro - ribbon unchanged for different values of @xmath101 , and two color scales are indicated in the lower - left and lower - right panels . other parameters in calculations are given in the text.,width=288 ] is made in the upper panel for @xmath0 polarization with three different bandgaps @xmath182 ( @xmath299mev , red solid curve ) , @xmath240ev ( @xmath300mev , black dash - dotted curve ) and @xmath241ev ( @xmath301mev , blue dashed curve ) . two pairs of arrows indicate the shift of a pair of corresponding peaks with @xmath101 . in the lower panel , a comparison of the calculated transmitted @xmath0polarized e - field intensities @xmath294 is displayed for @xmath182 ( left ) and @xmath241ev ( right ) at @xmath202mev and @xmath302mev , respectively . here , we keep the electron areal density ( @xmath303 ) in a graphene micro - ribbon unchanged for different values of @xmath101 , and two color scales are indicated in the lower - left and lower - right panels . other parameters in calculations are given in the text.,width=576 ]

this paper develops a discontinuous galerkin ( dg ) finite element differential calculus theory for approximating weak derivatives of sobolev functions and piecewise sobolev functions . by introducing numerical one - sided derivatives as building blocks , various first and second order numerical operators such as the gradient , divergence , hessian , and laplacian operator are defined , and their corresponding calculus rules are established . among the calculus rules are product and chain rules , integration by parts formulas and the divergence theorem . approximation properties and the relationship between the proposed dg finite element numerical derivatives and some well - known finite difference numerical derivative formulas on cartesian grids are also established . efficient implementation of the dg finite element numerical differential operators is also proposed . besides independent interest in numerical differentiation , the primary motivation and goal of developing the dg finite element differential calculus is to solve partial differential equations . it is shown that several existing finite element , finite difference and dg methods can be rewritten compactly using the proposed dg finite element differential calculus framework . moreover , new dg methods for linear and nonlinear pdes are also obtained from the framework .

1302.6984

numerical differentiation is an old but basic topic in numerical mathematics . compared to the large amount of literature on numerical integration , numerical differentiation is a much less studied topic . given a differentiable function , the available numerical methods for computing its derivatives are indeed very limited . there are essentially only two such methods ( cf.@xcite ) . one method is to approximate derivatives by difference quotients . the other is to first approximate the given function ( or its values at a set of points ) by a more simple function ( e.g. , polynomial , rational function and piecewise polynomial ) and then to use the derivative of the approximate function as an approximation to the sought - after derivative . the two types of classical methods work well if the given function is sufficiently smooth . however , the two classical methods produce large errors or divergent approximations if the given function is rough , which is often the case when the function is a solution of a linear or nonlinear partial differential equation ( pde ) . for boundary value and initial - boundary value problems , classical solutions often do not exist . consequently , one has to deal with generalized or weak solutions , which are defined using a variational setting for linear and quasilinear pdes . although numerical methods for pdes implicitly give rise to methods for approximating weak derivatives ( in fact , combinations of weak derivatives ) of the solution functions ( cf . @xcite ) , to the best of our knowledge , there is no systematic study and theory in the literature on how to approximate weak derivatives of a given ( not - so - smooth ) function . moreover , for linear second order pdes of non - divergence form and fully nonlinear pdes , it is not possible to derive variational weak formulations using integration by parts . as a result , weak solution concepts for those types of pdes are different . the best known and most successful one is the _ viscosity solution _ concept ( cf.@xcite and the references therein ) . to directly approximate viscosity solutions , which in general are only continuous functions , one must approximate their derivatives in some appropriately defined sense offline ( cf.@xcite ) , and then substitute the numerical derivatives for the ( formal ) derivatives appearing in the pdes . clearly , to make such an intuitive approach work , the key is to construct correct " numerical derivatives and to use them judiciously to build numerical schemes . this paper addresses the above two fundamental issues . the specific goals of this paper are twofold . _ first _ , we systematically develop a computational framework for approximating weak derivatives and a new discontinuous galerkin ( dg ) finite element differential calculus theory . keeping in mind the approximation of fully nonlinear pdes , we introduce locally defined , one - sided numerical derivatives for piecewise weakly differentiable functions . using the newly defined one - sided numerical derivatives as building blocks , we then define a host of first and second order sided numerical differential operators including the gradient , divergence , curl , hessian , and laplace operators . to ensure the usefulness and consistency of these numerical operators , we establish basic calculus rules for them . among the rules are the product and the chain rule , integration by parts formulas and the divergence theorem . we establish some approximation properties of the proposed dg finite element numerical derivatives and show that they coincide with well - known finite difference derivative formulas on cartesian grids . consequently , our dg finite element numerical derivatives are natural generalizations of well - known finite difference numerical derivatives on general meshes . these results are of independent interest in numerical differentiation . _ second _ , we present some applications of the proposed dg finite element differential calculus to build numerical methods for linear and nonlinear partial differential equations . this is done based on a very simple idea ; that is , we replace the ( formal ) differential operators in the given pde by their corresponding dg finite element numerical operators and project ( in the @xmath0 sense ) the resulting equation onto the dg finite element space @xmath1 . we show that the resulting numerical methods not only recover several existing finite difference , finite element and dg methods , but also give rise to some new numerical schemes for both linear and nonlinear pde problems . the remainder of this paper is organized as follows . in section [ sec-2 ] we introduce the mesh and space notation used throughout the paper . in section [ sec-3 ] we give the definitions of our dg finite element numerical derivatives and various first and second order numerical differential operators . in section [ sec-4 ] we establish an approximation property and various calculus rules for the dg finite element numerical derivatives and operators . in section [ sec-5 ] we discuss the implementation aspects of the numerical derivatives and operators . finally , in section [ sec-6 ] we present several applications of the proposed dg finite element differential calculus to numerical solutions of prototypical linear and nonlinear pdes including the poisson equation , the biharmonic equation , the @xmath2-laplace equation , second order linear elliptic pdes in non - divergence form , first order fully nonlinear hamilton - jacobi equations , and second order fully nonlinear monge - ampre equations . let @xmath3 be a positive integer , @xmath4 be a bounded open domain , and @xmath5 denote a locally quasi - uniform and shape - regular partition of @xmath6 @xcite . let @xmath7 denote the set of all interior faces / edges of @xmath8 , @xmath9 denote the set of all boundary faces / edges of @xmath8 , and @xmath10 . let @xmath11 $ ] and @xmath12 be an integer . define the following piecewise @xmath13 and piecewise @xmath14 spaces with respect to the mesh @xmath15 : @xmath16 when @xmath17 , we set @xmath18 . we also define the analogous piecewise vector - valued spaces as @xmath19^d$ ] , @xmath20^d$ ] , @xmath21^d$ ] , and the matrix - valued spaces @xmath22^{d\times d}$ ] , @xmath23^{d\times d}$ ] , and @xmath24^{d\times d}$ ] . the piecewise @xmath0-inner product over the mesh @xmath15 is given by @xmath25 and for a set @xmath26 , the piecewise @xmath0-inner product over @xmath27 is given by @xmath28 angled brackets without subscripts @xmath29 represent the dual pairing between some banach space and its dual . for a fixed integer @xmath30 , we define the standard discontinuous galerkin ( dg ) finite element space @xmath31 by @xmath32 where @xmath33 denotes the set of all polynomials on @xmath34 with degree not exceeding @xmath35 . the analogous vector - valued and matrix valued dg spaces are given by @xmath36^d$ ] and @xmath37^{d\times d}$ ] . in addition , we define @xmath38 @xmath39^d$ ] , and @xmath40^{d\times d}.$ ] we note that @xmath41 . we denote by @xmath42 the @xmath0 projection operator onto @xmath43 , @xmath44^d \to \bm{v}_r^h$ ] the @xmath0 projection operator onto @xmath45 and @xmath46^{d \times d } \to \widetilde{\bm{v}}_r^h$ ] the @xmath0 projection operator onto @xmath47 . let @xmath48 and @xmath49 . without loss of generality , we assume that the global labeling number of @xmath34 is smaller than that of @xmath50 . we then introduce the following standard jump and average notations across the face / edge @xmath51 : @xmath52 & : = v|_k - v|_{k ' } \quad & & \mbox{on } e\in { \mathcal{e}_h}^i,\qquad & & [ v ] : = v\quad & & \mbox{on } e\in { \mathcal{e}_h}^b,\\ \{v\ } & : = \frac12\bigl ( v|_k + v|_{k ' } \bigr ) \quad & & \mbox{on } e\in { \mathcal{e}_h}^i,\qquad & & \{v\}:=v\quad & & \mbox{on } e\in { \mathcal{e}_h}^b\end{aligned}\ ] ] for @xmath53 . we also define @xmath54 as the unit normal on @xmath51 . let @xmath53 . for @xmath55 , that is , @xmath56 for some @xmath57 , we write @xmath58 to be the unit normal of @xmath51 . we then define the following three trace operators on @xmath51 in the direction @xmath59 : @xmath60 for any @xmath61 and @xmath62 . we note that @xmath63 and @xmath64 can be regarded respectively as the left " and right " limit of @xmath65 at @xmath61 in the direction of @xmath59 . if @xmath66 , we simply let @xmath67 [ qidentityremark ] on an interior edge @xmath68 , we may alternatively write @xmath69,\quad \text{where}\quad { \rm sgn}(n_e^{(i ) } ) = \left\ { \begin{array}{cc } 1 & \text { if } n_e^{(i ) } \ge 0,\\ -1 & \text { if } n_e^{(i ) } < 0 . \end{array } \right.\end{aligned}\ ] ] with the help of the trace operators @xmath70 and @xmath71 we are ready to introduce our discrete partial derivative operators @xmath72 , @xmath73 . [ def1 ] for any @xmath53 , we define the discrete partial derivatives @xmath74 , @xmath75 , @xmath76 by @xmath77 \bigr\rangle_{{\mathcal{e}_h } } -\bigl(v , \partial_{x_i}\varphi_h \bigr)_{{\mathcal{t}}_h } \label{e2.5 } \\ & \hskip 1.4 in + \bigl\langle \gamma^\pm_i [ v],[\varphi_h ] \bigr\rangle_{{\mathcal{e}_h}^i } \qquad\forall \varphi_h\in v^h_r , \nonumber \\ \partial_{h , x_i } v & : = \frac12\bigl(\partial_{h , x_i}^- v + \partial_{h , x_i}^+ v\bigr ) , \label{e2.7 } \end{aligned}\ ] ] for @xmath78 . here , @xmath79 denotes the usual ( weak ) partial derivative operator in the direction @xmath59 , @xmath80 is the piecewise constant function satisfying @xmath81 and @xmath82 and @xmath83 are piecewise constants with respect to the set of interior edges . in addition , we define the discrete partial derivatives when boundary data is provided . [ def1b ] let @xmath84 be given . then for any @xmath53 , we define the discrete partial derivatives @xmath85 by @xmath86 [ rem1 ] 1 . since every function @xmath53 has a well - defined trace in @xmath87 and every function @xmath88 has a well - defined trace in @xmath89 for all @xmath90 , the last term on the right - hand side of is well defined . 2 . since @xmath1 is a totally discontinuous piecewise polynomial space , the discrete derivatives @xmath91 can also be written in their equivalent _ local _ versions : @xmath92,[\varphi_h ] \bigr\rangle_e \qquad \forall \varphi_h\in { \mathbb{p}}_r(k ) \end{aligned}\ ] ] for @xmath78 and @xmath90 . here , @xmath93 . 3 . the discrete derivatives @xmath94 and @xmath95 can be regarded , respectively , as left " , right " , and central " discrete partial derivatives of @xmath65 with respect to @xmath59 . the definitions are analogous to the weak derivative definition . we note that the discrete one - sided partial derivatives are defined for functions in the space @xmath96 , in particular , for functions in the dg finite element space @xmath41 . 5 . by the identity , we have @xmath97{\big\rangle}_{{\mathcal{e}_h } } + { \big\langle}\big(\gamma^\pm_i \pm \frac12 |n^{(i)}|\big ) [ v ] , [ \varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i } -\big ( v,{\partial}_{x_i } \varphi_h\big)_{{\mathcal{t}_h}}.\end{aligned}\ ] ] therefore , @xmath98 provided @xmath99 for all @xmath68 . [ def2 ] we define the following first order discrete operators : @xmath100 for any @xmath53 and @xmath101 . in addition , we define the discrete curl operators for @xmath102 and @xmath53 : @xmath103 when @xmath104 , and @xmath105 when @xmath106 . we also set @xmath107 when boundary data @xmath84 is provided , the analogous operators are given by [ firstordergdef ] @xmath108 in addition , for given @xmath109 , we set @xmath110 [ def3 ] denote by @xmath111 and @xmath112 the transposes of the operators @xmath113 and @xmath114 , respectively ; that is , @xmath115 and @xmath116 . we then define the following second order discrete operators : [ secondorderdef ] @xmath117 for any @xmath53 . here , @xmath118 denotes the matrix trace operator . when boundary data @xmath84 is given , we define @xmath119 [ rem2 ] 1 . the matrix - valued functions @xmath120 and @xmath121 define four copies of discrete @xmath122 hessian matrices of the function @xmath65 . similarly , @xmath123 and @xmath124 are four copies of discrete laplacians of @xmath65 . 2 . since @xmath125 , all the above second order discrete operators are well - defined . it is simple to see that the definitions of the discrete laplacians given in definition [ def3 ] are equivalent to @xmath126 , @xmath127 and @xmath128 . in this section we shall establish a number of properties for the dg finite element ( fe ) discrete derivatives defined in definitions [ def1 ] and [ def2 ] . we start with some characterization results for the dg fe discrete derivatives . we then establish an approximation property followed by several basic calculus rules such as product and chain rules , the integration by parts formula , the divergence theorem and the relationship of the dg fe discrete derivatives with standard finite difference derivative formulas . for any @xmath129 , we have @xmath130 and @xmath131=0 $ ] on @xmath55 . hence , @xmath132{\big\rangle}_{{\mathcal{e}_h } } -\bigl(v , \partial_{x_i}\varphi_h \bigr)_{{\mathcal{t}}_h } \qquad\forall \varphi_h\in v^h_r.\ ] ] integration by parts element - wise immediately yields @xmath133 hence , we have the following proposition . [ prop0a ] for any @xmath134 , @xmath91 and @xmath95 coincide with the @xmath0-projection of @xmath135 onto @xmath1 . we write @xmath136 , where @xmath137 denotes the @xmath0 projection onto @xmath43 . from the above proposition , we easily derive the following corollary . [ cor4.1 ] let @xmath138 with @xmath139 . then @xmath140 , where @xmath141 , the set with only the zero function . for an arbitrary piecewise function @xmath53 , the above characterization results are not expected to hold in general because @xmath135 may not exist . below we shall derive a similar characterization to for @xmath91 and @xmath95 when @xmath65 is an arbitrary function in @xmath96 . for any @xmath53 , let @xmath142 denote the distributional derivative of @xmath65 with respect to @xmath59 ( which does exist ) . let @xmath143 be a @xmath144-dimensional continuous and bounded surface . we define the _ delta function _ @xmath145 of variable strength supported on @xmath146 by ( cf . @xcite ) @xmath147 where @xmath148 . we also extend the above definition to test functions from @xmath1 as follows @xmath149 using @xmath145 we give the following characterization for @xmath142 . [ lem4.0c ] for every @xmath53 there holds the following representation formula : @xmath150 , x ) \qquad \mbox{for a.e . } x\in \omega,\end{aligned}\ ] ] where @xmath151 denotes the characteristic function supported on @xmath34 . by the definition of @xmath142 and integration by parts on each @xmath152 , we have for any @xmath153 @xmath154 , \varphi n_e^{(i ) } \bigr\rangle_e + \bigl ( { \partial}_{x_i } v , \varphi \bigr)_{{\mathcal{t}}_h } \\ & = -\bigl\langle \sum_{e\in{\mathcal{e}_h}^i } n_e^{(i)}\delta(e , [ v ] , \cdot ) , \varphi\bigr\rangle + \bigl\langle \sum_{k\in { \mathcal{t}}_h } { \partial}_{x_i } v\chi_k , \varphi \bigr\rangle . \end{aligned}\ ] ] here we have used the fact that @xmath155 for every @xmath90 . clearly the above identity infers . the proof is complete . [ prop0b ] set @xmath156 in . for any @xmath53 , @xmath95 coincides with the @xmath0-projection of @xmath142 onto @xmath1 in the sense that @xmath157 where the right hand - side is understood according to . we write @xmath158 . by the definition of @xmath159 and using the integration by parts formula for piecewise functions , we get , for @xmath78 , @xmath160 \bigr\rangle_{{\mathcal{e}_h } } -\bigl(v , \partial_{x_i}\varphi_h \bigr)_{{\mathcal{t}}_h } \\ & = -\bigl\langle [ v ] n^{(i ) } , \{\varphi_h\ } \bigr\rangle_{{\mathcal{e}_h}^i } + \bigl(\partial_{x_i } v , \varphi_h \bigr)_{{\mathcal{t}}_h } \nonumber\\ & = - \bigl\langle \sum_{e\in { \mathcal{e}_h}^i } n^{(i ) } \delta(e , [ v ] , \cdot ) , \varphi_h \bigr\rangle + \bigl\langle \sum_{k\in{\mathcal{t}}_h } { \partial}_{x_i } v\chi_k , \varphi_h \bigr\rangle \nonumber \\ & = \bigl\langle \mathcal{d}_{x_i } v , \varphi_h \bigr\rangle \qquad\forall \varphi_h\in v^h_r . \nonumber\end{aligned}\ ] ] here we have used and to get the final two equalities . the proof is complete . define the lifting operator " @xmath161 by @xmath162 , \cdot ) , \varphi_h \bigr\rangle = \bigl\langle [ v ] n^{(i ) } , \{\varphi_h\ } \bigr\rangle_{{\mathcal{e}_h}^i}.\ ] ] then we have @xmath163 finally , combining propositions [ prop0a ] and [ prop0b ] and the well - known limiting characterization theorem of distributional derivatives ( cf . * theorem 6.32 ) ) , we obtain another characterization for our dg fe derivatives . [ prop0c ] set @xmath156 in . then for any @xmath53 , there exists a sequence of functions @xmath164 such that , for @xmath165 , 1 . @xmath166 as @xmath167 in @xmath168 . 2 . @xmath169 as @xmath167 weakly in @xmath1 in the sense that @xmath170 let @xmath171 denote a symmetric mollifier ( or approximate identity ) with compact support . for any @xmath172 , set @xmath173 , the convolution of @xmath65 and @xmath174 . then it is easy to check that the sequence @xmath175 fulfills the properties ( i ) and ( ii ) . the proof is complete . by proposition [ prop0a ] we readily have the following approximation properties for the dg fe derivatives . [ lem4.1 ] for any @xmath176 there holds the following inequalities : @xmath177 where @xmath178 can take @xmath179 and empty value . notice that can be rewritten as @xmath180 hence , for any @xmath181 , there holds @xmath182 the identity follows from the above inequality because @xmath183 . the second inequality is obtained by similar arguments . it is also possible to derive some error estimates for @xmath184 with general @xmath53 in some weaker ( than @xmath0 ) norm . however , since the derivation is lengthy and technical , we leave it to the interested reader to exploit . the product rule and the chain rule are two very basic properties of classical derivatives and weak derivatives ( cf . * chapter 7 ) ) . the goal of this subsection is to establish both of the rules for our dg fe derivatives . as expected , these discrete rules take different forms from their continuous counterparts . first , we consider the case when functions are from the space @xmath185 . in this case , we have the following product rule and chain rule . [ product_rule1 ] let @xmath186 . for @xmath187 , there holds , for @xmath165 , @xmath188 notice that @xmath189 for any @xmath190 . by proposition [ prop0a ] and the product rule for weak derivatives ( cf . * chapter 7 ) ) we get @xmath191 hence , holds by the definition of @xmath192 . similarly , since @xmath193 for @xmath194 we have @xmath195 which then infers . the proof is complete . an immediate consequence of theorem [ product_rule1 ] and corollary [ cor4.1 ] is the following corollary . let @xmath186 . for @xmath196 with @xmath197 , there holds , for @xmath165 , @xmath198 next , we consider the general case when functions are from the space @xmath96 . as expected , the discrete product and chain rules appear in more complicated forms . [ product_rule2 ] let @xmath199 and @xmath200 . for @xmath201 , there holds for @xmath165 , @xmath202 where @xmath142 is given by and @xmath203 represents a modification of @xmath65 which assumes the same value in each element @xmath90 and takes the average value of @xmath65 on each interior edge @xmath55 . thus , @xmath204 , x ) \qquad \mbox{for a.e . } x\in \omega.\end{aligned}\ ] ] without loss of the generality , we assume @xmath205 . for @xmath201 , let @xmath206 be defined in proposition [ prop0c ] . then , by theorem [ product_rule1 ] , we have @xmath207 it follows from and that for any @xmath208 , @xmath209 on the other hand , we have @xmath210 then , follows from combining , and , and follows from combining , and . the proof is complete . the @xmath211 smoothness assumption on @xmath212 in theorem [ product_rule2 ] may be weakened to @xmath213 by following the techniques used in @xcite . we leave the generalization to the interested reader to exploit . in this subsection , we derive integration by parts formulas for the discrete differential operators that resemble the standard integration by parts formula in the continuous setting . these results will play a crucial role in developing dg methods for a variety of pde problems as well as in relating these methods to other existing dg methods in the literature . suppose that @xmath214 , that is , @xmath215 for all @xmath55 . then the integration by parts formulas @xmath216 hold for all @xmath217 . it suffices to show , since trivially follows from . by and integration by parts , we obtain @xmath218n^{(i)}{\big\rangle}_{{\mathcal{e}_h } } \\ & -{\big\langle}[v],\{\varphi_h\}n^{(i)}-\gamma^\pm_i [ \varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i}.\end{aligned}\ ] ] using the identity , we have @xmath219 , [ \varphi_h ] { \rm sgn}(n^{(i ) } ) n^{(i)}{\big\rangle}_{{\mathcal{e}_h}^i } \\ & \hskip 1 in -{\big\langle}[v],\{\varphi_h\}n^{(i)}-\gamma^\pm_i [ \varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i } \nonumber \\ & = \nonumber \big ( { \partial}_{x_i } v , \varphi_h\big)_{{\mathcal{t}_h } } + { \big\langle}[v],(\gamma^\pm_i \pm \frac12 |n^{(i)}| ) [ \varphi_h ] -\{\varphi_h\}n^{(i)}{\big\rangle}_{{\mathcal{e}_h}^i}.\end{aligned}\ ] ] now let @xmath220 . then by , and rearranging terms , we have @xmath221n^{(i)}{\big\rangle}_{{\mathcal{e}_h } } + { \big\langle}\gamma^\mp_i [ \varphi_h],[v_h]{\big\rangle}_{{\mathcal{e}_h}^i}\\ & \label{ibpline2 } = -\big ( { \partial}_{h , x_i}^\mp \varphi_h , v_h\big)_{{\mathcal{t}_h } } + { \big\langle}\{\varphi_h\},[v_h]n^{(i)}{\big\rangle}_{{\mathcal{e}_h}^i}\\ & \nonumber\hskip 0.5 in + { \big\langle}(\gamma^\mp_i \mp \frac12 |n^{(i)}| ) [ \varphi_h],[v_h]{\big\rangle}_{{\mathcal{e}_h}^i } + { \big\langle}\varphi_h , v_h n^{(i)}{\big\rangle}_{{\mathcal{e}_h}^b}.\end{aligned}\ ] ] using the identity in equation , we obtain @xmath222,[v_h]{\big\rangle}_{{\mathcal{e}_h}^i}. \nonumber\end{aligned}\ ] ] the identity then easily follows . [ firstcorollary]suppose that @xmath214 for all @xmath223 . then the integration by parts formulas @xmath224 hold for all @xmath225 and @xmath226 . if @xmath227 , then by , and , we have @xmath228 formula is obtained similarly . [ secondcorollary ] the formal adjoint of the operator @xmath229 ( resp . , @xmath230 ) is @xmath231 ( resp . , @xmath232 ) with respect to the inner product @xmath233 provided @xmath234 for all @xmath62 ; that is , @xmath235 for all @xmath236 in addition , if @xmath237 , then the formal adjoint of the operator @xmath238 ( reps . , @xmath239 ) is @xmath240 ( resp . , @xmath241 ) . this result immediately follows from theorem [ firstcorollary ] and the identities @xmath242 for all @xmath243 and @xmath226 . [ thirdcorollary ] suppose that @xmath244 . we then have @xmath245 for all @xmath53 and @xmath246 in addition , under the same hypotheses on @xmath247 , we have @xmath248 these formulas follow from theorems [ firstcorollary][secondcorollary ] with @xmath249 and @xmath250 ( cf . remark [ rem2]c ) . we now show that when @xmath251 the operators @xmath252 and @xmath253 are natural extensions ( on general grids ) of the backward / forward and central difference operators defined on cartesian grids . suppose @xmath15 is a rectangular mesh over @xmath6 which is aligned with the underlying cartesian coordinate system . let @xmath254 denote the mesh size of @xmath15 in the direction of @xmath59 . notice that when @xmath255 , the function @xmath91 is a piecewise constant function over the mesh @xmath15 . setting @xmath256 in we get @xmath257 we only consider the two dimensional case . then @xmath34 has four edges with @xmath258 . let @xmath65 be a grid function over @xmath15 and @xmath259 denote the value of @xmath65 on the @xmath260 cell / element . by we obtain @xmath261 consequently , we have @xmath262 hence , @xmath263 and @xmath264 coincide respectively with backward and forward difference operators in the direction @xmath59 , while @xmath253 results in the central difference operator in the direction @xmath59 . using the operators @xmath252 as building blocks , we can also build dg fe operators that are natural extensions of the standard finite difference approximations for second order partial derivatives . from and , we have @xmath265 thus , for @xmath266 , the discrete differential operator @xmath267 coincides with the standard second order 3-point central difference operator for nonmixed second order derivatives , @xmath268 coincides with the standard second order 3-point central difference operator with mesh size @xmath269 , @xmath270 coincides with the second order @xmath271-point central difference operator for mixed second order derivatives , and @xmath272 coincides with the standard second order central difference operator for mixed second order derivatives . in this section we explain how the discrete partial derivatives are computed . denote by @xmath273 the reference simplex , and let @xmath274 denote a basis of @xmath275 . here for example , we could take the basis to be the space of monomials of degree less than or equal to @xmath35 : @xmath277 . for @xmath278 , let @xmath279 be the affine mapping from @xmath280 onto @xmath34 , and and let @xmath281 be defined by @xmath282 , where @xmath283 . it is then easy to see that @xmath284 is a basis of @xmath285 . we then define the mass matrix @xmath286 associated with @xmath34 as @xmath287 by a change of variables we easily find @xmath288 , where @xmath289 is the mass matrix associated with @xmath280 , @xmath290 is the jacobian of the mapping @xmath291 and @xmath292 is the @xmath3-dimensional volume of the simplex @xmath34 . next , given @xmath53 , write @xmath293 with @xmath294 ( @xmath295 . we then define the vector @xmath296 by @xmath297,[\varphi_h^{(j , k)}]{\big\rangle}_e , \qquad j=1,2,\ldots , n_r.\end{aligned}\ ] ] then by , the coefficients @xmath298 are uniquely determined by the linear equation @xmath299 , where @xmath300 . equivalently , we have @xmath301 , where @xmath302 is the inverse matrix of the reference mass matrix which can be computed offline . now if @xmath303 , then we may write @xmath304 for some constants we then define the matrix @xmath306 by @xmath307,[\varphi_h^{(m , k)}]{\big\rangle}_e\end{aligned}\ ] ] for @xmath308 . again , writing @xmath309 , we find that the coefficients satisfy @xmath310 , where @xmath311 . in this section , we apply our dg fe differential calculus framework to constructing numerical methods for several different types of pdes . essentially , we replace the ( continuous ) differential operator by its discrete counterpart . in addition , we add a stability term which ensures that the discrete problem is well - posed and also enforces the given boundary conditions weakly within the variational formulation . throughout the section , we assume the penalty parameters @xmath312 ( which appear in the definition of the discrete differential operators ) are zero in the discussion below . where @xmath314 and @xmath315 are two given functions . we then consider the following discrete version of : find @xmath316 such that @xmath317 here , @xmath318 is the unique operator satisfying @xmath319,[\varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i } + { \big\langle}\eta_1 ( v - g),\varphi_h{\big\rangle}_{{\mathcal{e}_h}^b},\end{aligned}\ ] ] and @xmath320 is a penalty parameter that is piecewise constant with respect to the set of edges . problem has several interpretations . on the one hand , by the definition of the discrete laplace operator , the problem is the twofold saddle point problem @xmath321 with @xmath322 and @xmath323 . namely , problem is equivalent to finding @xmath324 , @xmath325 and @xmath326 such that [ dualmixed ] @xmath327\cdot n{\big\rangle}_{{\mathcal{e}_h}^i } - ( u_h,{{\rm div}}{{\bm \tau}}_h)_{{\mathcal{t}_h } } + { \big\langle}g,{{\bm \tau}}_h\cdot n{\big\rangle}_{{\mathcal{e}_h}^b}\quad & & \forall { { \bm \tau}}_h\in { \bm v}_r^h,\\ \label{dualmixed2 } & \big({\tilde}{{\bm r}}_h,{{\tilde}{\bm \mu}}_h\big)_{{\mathcal{t}_h } } = { \big\langle}\{{\bm q}_h\},[\tilde{\bm \mu}_h]{\big\rangle}_{{\mathcal{e}_h } } - ( { \bm q}_h,{{\rm div}}\tilde{\bm \mu}_h)_{{\mathcal{t}_h } } \quad & & \forall \tilde{\bm \mu}_h\in \widetilde{\bm v}_h^r,\\ \label{dualmixed3 } & -\big({\rm tr}(\tilde{\bm r}_h),\varphi_h\big)_{{\mathcal{t}_h } } + \big(j_{h , g}(u_h),\varphi_h\big)_{{\mathcal{t}_h } } = ( f , v)_{{\mathcal{t}_h}}\quad & & \forall \varphi_h\in v_r^h . \ ] ] on the other hand , by theorem [ firstcorollary ] , we can write problem in its primal form : find @xmath328 satisfying @xmath329 for all @xmath330 . alternatively , by theorem [ secondcorollary ] , we may write @xmath331 finally , in the case @xmath332 , problem can be viewed as finding a minimizer of the functional @xmath333\big|^2\ , ds -\int_{\omega}fv_h\ , dx\end{aligned}\ ] ] over all @xmath334 . we now discuss the well - posedness of problem as well as relate the discretization to other dg schemes . let @xmath335 , that is , @xmath322 satisfies .then by , we have @xmath336 for all @xmath88 . by the definition of the discrete gradient and integration by parts , we have @xmath337\cdot n,\{\varphi_h\ } { \big\rangle}_{{\mathcal{e}_h } } -({{\rm div}}{\bm q}_h,\varphi_h,)_{{\mathcal{t}_h } } \\ & = ( { \bm q}_h,{\nabla}\varphi_h)_{{\mathcal{t}_h } } -{\big\langle}\{{\bm q}_h\}\cdot n,[\varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i}.\end{aligned}\ ] ] using this identity in , we find @xmath338{\big\rangle}_{{\mathcal{e}_h } } + \big(j_{h , g}(u_h),\varphi_h\big)_{{\mathcal{t}_h } } & = ( f,\varphi_h)_{{\mathcal{t}_h}}\quad \forall \varphi\in v_r^h.\end{aligned}\ ] ] a similar methodology can be used to construct dg schemes for the neumann problem @xmath345 in this case , the dg method is to find @xmath328 satisfying @xmath346 where @xmath347 is the operator satisfying @xmath348,[\varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i}$ ] for all @xmath226 , and @xmath349 . we again recover the ldg method for the neumann problem ( cf.@xcite ) . in this subsection we formulate a new dg method for the poisson problem that inherits better stability than the ldg method described above . the new scheme , called the symmetric dual - wind discontinuous galerkin ( dwdg ) method , is simply given by @xmath350 where @xmath351 is defined by . by theorem [ thirdcorollary ] , problem is equivalent to finding @xmath328 such that @xmath352 for all @xmath334 . equivalently , in the case @xmath332 , problem asks to find the unique minimizer of the functional @xmath353\big|^2\ , ds -\int_{\omega}fv_h\ , dx\end{aligned}\ ] ] over all @xmath334 ( compare to ) . a complete convergence analysis of the symmetric dwdg method for the poisson problem is presented in @xcite . here , we summarize the main results , namely , well - posedness and optimal rates of convergence . [ uniquenesstheorem ] set @xmath354 . suppose that there exists at least one simplex @xmath355 with exactly one boundary edge / face . then there exists a unique solution to provided @xmath356 . furthermore , if the triangulation is quasi - uniform , and if each simplex in the triangulation has at most one boundary face / edge , then there exists a constant @xmath357 independent of @xmath358 and @xmath320 such that problem has a unique solution provided @xmath359 . [ l2estimatetheorem ] let @xmath360 be the solution to , @xmath361 be the solution to and @xmath362 . then @xmath360 satisfies the following estimate provided @xmath363 : @xmath364 and if the triangulation is quasi - uniform and @xmath365 , then there holds @xmath366 where @xmath367 denotes a generic positive constant independent of @xmath358 , and @xmath368 is the positive constant from theorem [ uniquenesstheorem ] . in light of and , we can see that when approximating with piecewise constant basis functions on cartesian grids , the dwdg method coincides with the standard finite difference method for poisson s equation while the ldg method coincides with a staggered finite difference method for poisson s equation that uses coarser second derivative approximations . in this subsection we show how to use the discrete differential calculus to develop dg methods for fourth order linear pdes . for simplicity , we focus our derivation to the model biharmonic problem with clamped boundary conditions : to develop dg methods using the discrete differential machinery , we first need to define some additional discrete operators corresponding to the biharmonic operator . given the two functions @xmath371 and @xmath372 defined on the boundary , we set @xmath373 in addition , we define the operator @xmath374 by @xmath375,[{\partial}\varphi_h/{\partial}n]{\big\rangle}_{{\mathcal{e}_h}^i } + { \big\langle}\eta_2 ( { \partial}v/{\partial}n - q),{\partial}\varphi_h/{\partial}n{\big\rangle}_{{\mathcal{e}_h}^b},\end{aligned}\ ] ] which we will use to enforce the neumann boundary condition weakly in the dg formulation . the dg method for is then defined as seeking @xmath328 such that @xmath376 similar to the discussion in section [ sec-5 - 1 ] , we may write the dg method in various mixed forms ( with up to six unknown variables ) . instead , we focus mainly on the primal formulation . the dg method closely resembles the local continuous discontinuous galerkin ( lcdg ) method proposed in @xcite ( also see @xcite ) . here , the authors consider a mixed formulation of the biharmonic problem with the hessian of @xmath379 as an additional unknown . the derivation of the lcdg method closely resembles the derivation of the ldg method for the poisson problem ; the main difference is that , as the name suggests , the lcdg method uses continuous finite element spaces . to this end , we set @xmath383 , @xmath384 , and @xmath385 so that @xmath386 . to ease notation , we define the bilinear forms @xmath387\cdot n,\{\varphi_h\}{\big\rangle}_{{\mathcal{e}_h}},\\ b_i({\bm \mu}_h,\varphi_h ) & : = -\big({{\rm div}}{\bm \mu}_h,\varphi_h\big)_{{\mathcal{t}_h } } + { \big\langle}[{\bm \mu}_h]\cdot n,\{\varphi_h\}{\big\rangle}_{{\mathcal{e}_h}^i},\\ c(\psi_h,\varphi_h ) & : = \big(j_{h,0}(\psi_h),\varphi_h\big)_{{\mathcal{t}_h}}+\big(r_{h,0}(\psi_h),\varphi_h\big)_{{\mathcal{t}_h}}.\end{aligned}\ ] ] it is then easy to verify that @xmath388 and @xmath389 for all @xmath390 and @xmath391 . furthermore , by theorem [ secondcorollary ] , we have @xmath392 and @xmath393 . it then follows that we may write in the following mixed - form : setting @xmath395 in and @xmath396 in , we have @xmath397 therefore , subtracting the two equations we get @xmath398 . next , we set @xmath399 in and @xmath400 in to obtain @xmath401 adding the two equations yields @xmath402 . therefore , @xmath403 and @xmath404 . in particular , @xmath360 vanishes on all of the boundary edges . since @xmath405 , we also have @xmath406 by . setting @xmath407 in , and @xmath408 in , we have @xmath409 since @xmath360 vanishes on the boundary edges , @xmath410 . it then easily follows that @xmath411 . finally , we have @xmath412 for all @xmath390 . this in turn implies that @xmath413{\big\rangle}_{{\mathcal{e}_h } } = ( { \bm \mu}_h,{\nabla}u_h)_{{\mathcal{t}_h}}\quad \forall { \bm \mu}_h\in { \bm v}_r^h.\end{aligned}\ ] ] therefore , @xmath414 on all @xmath278 . since @xmath415|_e = 0 $ ] across all edges , we conclude that @xmath416 . 1 . to obtain optimal order error estimates , we expect that the penalty parameters must scale like @xmath417 and @xmath418 . the construction of dg schemes with other types of boundary conditions can easily be constructed by specifying the boundary data to different discrete differential operators . for example , if simply supported plate boundary conditions @xmath419 and @xmath420 are provided , then the corresponding discrete biharmonic operator is @xmath421 . on the other hand , if cahn - hilliard - type boundary conditions @xmath422 and @xmath423 are given , then the discrete biharmonic operator is @xmath424 , where @xmath425 and @xmath426 . we now extend the discrete differential framework to some non - linear elliptic problems . although we can formulate the method for a very general class of quasi - linear pdes , we shall focus our attention to a prototypical example , the @xmath2-laplace equation @xmath427 : similar to the poisson problem , the dg method for is obtained by simply replacing the _ grad _ and _ div _ operators by their respective discrete versions and adding a stability term to ensure that the resulting bilinear form is coercive over @xmath1 . in this case , the discrete problem reads : find @xmath328 satisfying @xmath429 where @xmath430 is defined by @xmath431\big|^{p-2 } [ v],[\varphi_h]{\big\rangle}_{{\mathcal{e}_h}^i } + { \big\langle}\eta_1 |v|^{p-2 } ( v - g),\varphi_h{\big\rangle}_{{\mathcal{e}_h}^b } \ ] ] for all @xmath88 . here , @xmath340 is a penalization parameter . similar to the discrete poisson problem , the discretization has several interpretations . by the definition of the discrete divergence and gradient operators , we can write in the mixed formulation @xmath432{\big\rangle}_{{\mathcal{e}_h } } + \big(j_{h , g}^{(p)}(u_h),\varphi_h\big)_{{\mathcal{t}_h } } \\ & \qquad \nonumber = ( f,\varphi_h)_{{\mathcal{t}_h } } \qquad\forall \varphi_h\in v^h_r , \end{aligned}\ ] ] where @xmath433 satisfies . we emphasize that the gradient appearing in the left - hand side of equation is the piecewise gradient . in @xcite , burman and ern proposed and analyzed the following ldg method for the @xmath2-laplace equation ( with @xmath434 and @xmath332 ) : @xmath435\big|^{p-2}[u_h],[\varphi_h]{\big\rangle}_{{\mathcal{e}_h } } = ( f,\varphi_h)_{{\mathcal{t}_h } } \ ] ] for all @xmath88 . here , the authors showed the existence and uniqueness of the dg method provided @xmath436 . in addition , burman and ern showed that the approximate solutions converge to @xmath379 strongly in @xmath437 , and @xmath438 converges to @xmath439 strongly in @xmath440 . furthermore , in the one dimensional setting , numerical experiments indicate a convergence rate of at least @xmath441 for @xmath442 and smooth exact solution . clearly , method has a similar structure to , but they are different methods when @xmath443 . indeed , since @xmath444 , we can not use theorems [ firstcorollary][secondcorollary ] and simply write @xmath445 in the following section , we show by way of numerical experiments that the dg method converges with optimal order provided the exact solution is sufficiently smooth . in this subsection we perform some numerical experiments to gauge the effectiveness of the dg method . we take the domain to be the unit square @xmath446 and choose the data @xmath447 such that the exact solution is @xmath448 and @xmath449 . in all numerical experiments , we take the penalty parameter to be @xmath450 . the resulting errors in the cases @xmath434 and @xmath451 are recorded in table [ plaplacetable ] . the table clearly suggests that the following rates of convergence hold for smooth test problems : @xmath452 .[plaplacetable]the errors of the computed solution and rates of convergence of the dg method with solution @xmath453 on the unit square with @xmath449 and @xmath454 . [ cols="^,^,^,^,^,^",options="header " , ] [ nonlinear1 ] @xmath457,\\ \label{nonlinear1_2 } u & = g\qquad & & \text{on } \gamma \subset { \partial}{\omega}\times ( 0,t ] , \\ \label{nonlinear1_3 } u & = u_0 \qquad & & \text{on } \omega \times \ { 0\},\end{aligned}\ ] ] where the operator @xmath458 is a continuous and possibly nonlinear function and @xmath459 is a function class in which the viscosity solution @xmath379 is unique . we note that the following scheme can also be adapted for @xmath458 a function of @xmath379 , @xmath460 , and @xmath461 . let @xmath462 , @xmath463 , @xmath464 and @xmath465 denote , respectively , the spaces of bounded , bounded uniformly continuous , upper semi - continuous and lower semicontinuous functions on @xmath466 . we now recall the well - known existence and uniqueness theorem for the corresponding cauchy problem which was first proved in @xcite . [ cl84 ] let @xmath467 , @xmath468 . then there is exactly one function @xmath469)$ ] for all @xmath470 such that @xmath471 , and for every @xmath472 and @xmath470 , if @xmath473 is a local maximum ( resp . local minimum ) point of @xmath474 on @xmath475 $ ] , then @xmath476 ( resp . @xmath477 ) . recently , a nonstandard ldg method was proposed by yan and osher in @xcite for approximating the viscosity solution of the hamilton - jacobi problem . the main idea of @xcite is to approximate the left " and right " side derivatives of the viscosity solution and to judiciously combine them through a monotone and consistent numerical hamiltonian ( cf . @xcite ) such as the lax - friedrichs numerical hamiltonian @xmath478 where @xmath479 is an undetermined nonnegative vector chosen to enforce the monotonicity property of @xmath480 , or the godunov numerical hamiltonian @xmath481 where @xmath482 for @xmath483 $ ] . for @xmath255 and @xmath484 on a uniform rectangular grid , the second term in is equivalent to a second order finite difference approximation for the negative laplacian operator scaled by @xmath358 . thus , the second term in is called a _ numerical viscosity _ , and the method is a direct realization of the vanishing viscosity method from pde theory . however , for high order elements and variable coefficient vector @xmath485 , while we do not exactly recover a scaled laplacian operator , we do recover some of the stabilizing properties from adding a second - order - like perturbation . with the correct choice of discrete derivatives , we can rewrite the nonstandard ldg method of yan and osher @xcite as follows : for each time step @xmath486 with @xmath487 , find @xmath488 using the recursive relation @xmath489 in addition to the above euler time - stepping method with @xmath480 given by the lax - friedrichs numerical hamiltonian defined in , in @xcite yan and osher also implemented the explicit third - order tvd runge - kutta time - stepping method given in @xcite . tests include one and two dimensional problems for @xmath30 . where the operator @xmath212 can be nonlinear in all arguments and @xmath459 is a function class in which the viscosity solution is unique . throughout this subsection , we assume that @xmath492 $ ] is elliptic for all @xmath493 , @xmath212 satisfies a comparison principle and problem has a unique viscosity solution @xmath494 . the definitions for the above terms are given below . we again use the same function and space notations from section [ hj ] . we also note that the following scheme can also be adapted for @xmath212 a function of @xmath379 and @xmath495 . for ease of presentation , we write as @xmath496 where we have used the convention of writing the boundary condition as a discontinuity of the pde ( cf . @xcite ) . also for ease of presentation , we assume that @xmath212 is a continuous function and refer the reader to @xcite for the case when @xmath212 is only a bounded function . equation is said to be elliptic if for all @xmath497 there holds @xmath498 where @xmath499 means that @xmath500 is a nonnegative definite matrix , and @xmath501 denotes the set of real symmetric @xmath122 matrices . [ viscosity_def ] a function @xmath503 is called a viscosity subsolution ( resp . supersolution ) of if , for all @xmath504 , if @xmath505 ( resp . @xmath505 ) has a local maximum ( resp . minimum ) at @xmath506 , then we have @xmath507 ( resp . @xmath508 ) . the function @xmath379 is said to be a viscosity solution of if it is simultaneously a viscosity subsolution and a viscosity supersolution of . [ def3.2 ] problem is said to satisfy a _ comparison principle _ if the following statement holds . for any upper semi - continuous function @xmath379 and lower semi - continuous function @xmath65 on @xmath509 , if @xmath379 is a viscosity subsolution and @xmath65 is a viscosity supersolution of , then @xmath510 on @xmath509 . inspired by the work of yan and osher @xcite , the first and second authors of this paper recently proposed in @xcite a nonstandard ldg method for approximating the viscosity solution of the fully nonlinear second order problem in one - dimension . the main idea of @xcite is to use all four of the various sided " approximations for the second order derivative , , of the viscosity solution and to judiciously combine them through a g - monotone ( generalized monotone ) and consistent numerical operator such as the following lax - friedrichs - like numerical operator that has been adopted from @xcite for an arbitrary dimensional problem : @xmath511 where @xmath512 is a nonnegative constant matrix that is chosen to enforce the g - monotonicity property of @xmath513 . the consistency of @xmath513 is defined by fulfilling the following property : @xmath514 and the g - monotonicity requires that @xmath513 is monotone increasing in its first and fourth arguments ( i.e. , @xmath515 , @xmath516 ) and monotone decreasing in its second and third arguments ( i.e. , @xmath517 , @xmath518 ) . below we give a reformulation of the nonstandard ldg method of @xcite using our dg finite element differential calculus machinery . to this end , we simply replace the continuous differential operators by multiple copies of discrete ones . due to the lack of integration by parts caused by the nonlinearity , we have to project the equation onto the dg finite element space . this then leads to the following scheme of finding @xmath519 such that @xmath520 where @xmath521 , @xmath522 , @xmath523 , @xmath524 are the four sided numerical hessians ( with the prescribed boundary data @xmath371 ) defined in section [ sec-3 ] . it can be shown that is indeed equivalent to the ldg scheme of @xcite in one - dimension . 1 . when @xmath255 and @xmath525 , scheme reduces to the fd method given in @xcite , which was proved to be convergent . 2 . for @xmath255 and @xmath526 on a rectangular grid , the second term on the right hand side in is equivalent to a second order finite difference approximation for the biharmonic operator scaled by @xmath527 . thus , the second term in is called a _ numerical moment _ , and the method is a direct realization of the vanishing moment method proposed in @xcite . however , for high order elements and variable coefficient matrices @xmath528 , while we do not exactly recover a scaled biharmonic operator , we do recover some of the stabilizing properties from adding a fourth - order - like perturbation . 3 . numerical tests of @xcite show the above discretization can eliminate many , and in some cases all , of the numerical artifacts that plague standard discretizations for fully nonlinear second order pdes ( cf . @xcite and the references therein ) . fully discrete schemes of @xcite for parabolic fully nonlinear second order pdes can also be recast using the dg finite element differential calculus machinery . to solve the algebraic system , a nonlinear solver must be used . numerical tests of @xcite show that when the initial guess for @xmath360 is not too close to a non - viscosity solution of the pde problem , a newton - based solver performs well as long as the solution is not on the boundary of the admissible set @xmath459 . however , in the degenerate case , a split solver based on the dwdg discretization for the poisson equation from section [ dwdg ] and the lax - friedrichs - like operators in appear to be better suited . this new solver for is given below in algorithm [ split_dwdg ] . we note that this solver appears to work well even for some cases when the initial guess for @xmath360 is not in @xmath459 . thus , the solver uses key tools from the discretization to address the issue of conditional uniqueness of viscosity solutions . pick @xmath529 . let @xmath530 and @xmath531 denote the diagonal matrices formed by the diagonals of @xmath532 and @xmath533 , respectively , and @xmath534 and @xmath535 denote the diagonal matrices formed by the diagonals of @xmath536 and @xmath537 , respectively , for all @xmath538 . let @xmath539 denote the diagonal matrix formed by the vector @xmath540 . + for @xmath541 , 1 . for @xmath78 , set @xmath542_i^{(n ) } & : = f\bigl ( \bigl((d_{h , g}^{- + } -\lambda_{h , g}^{- + } ) /2 + ( d_{h , g}^{+ -}-\lambda_{h , g}^{+ -})/2 \bigr ) u_h^{(n-1 ) } + \widetilde{\boldsymbol{\lambda}^{(n)}\mathbf{i}},x\bigr)\\ & \qquad + a \bigl[\lambda_{h , g}^{- - } u_h^{(n-1 ) } + \lambda_{h , g}^{+ + } u_h^{(n-1 ) } - 2 \ , \widetilde{\boldsymbol{\lambda}^{(n)}\mathbf{i } } \bigr]_{ii}\end{aligned}\ ] ] for a fixed constant @xmath543 such that @xmath544 is monotone decreasing with respect to @xmath545 . 2 . solve for @xmath546 such that @xmath547_i^{(n ) } , \phi_i \bigr)_{\mathcal{t}_h } = 0\ ] ] for all @xmath548 , @xmath62 . 3 . solve for @xmath549 such that @xmath550 for all @xmath551 . the above solver is a fixed point method for the diagonal of the hessian approximation formed by @xmath553 . we can see that the nonlinear equation in step 2 is entirely monotone and local in its nonlinear components when @xmath554 is sufficiently large . step 3 is well - defined due to the well - posedness of the dwdg method for poisson s equation . lastly , as written , we can see that the numerical moment inspired by the vanishing moment methodology serves as the motivation for using the diagonal of the hessian approximation as the fixed - point parameter . numerical tests indicate that the solver destabilizes numerical artifacts even when they exist ( cf . @xcite ) . thus , the solver directly addresses the issue of conditional uniqueness by enforcing the preservation of monotonicity in the hessian approximation at each iteration . , _ convergent monotone finite difference methods for one - dimensional fully nonlinear second order partial differential equations _ , j. comput . math . , doi:10.1016/j.cam.2013.02.001 ( also downloadable at arxiv:1212.0249 [ math.na ] ) . , _ high order numerical methods for time dependent hamilton - jacobi equations _ , math . and comp . in imaging science and info . 11 , lect . notes on ser . world sci . hackensack , nj , 2007 : 47 - 91 .

we present the first direct measurement of the mean halo occupation distribution ( hod ) of x - ray selected agn in the cosmos field at , based on the association of 41 xmm and 17 c - cosmos agn with member galaxies of 189 x - ray detected galaxy groups from xmm and _ chandra _ data . we model the mean agn occupation in the halo mass range = 13- 14.5 $ ] with a rolling - off power - law with the best fit index and normalization parameter . we find the mean hod of agn among central galaxies to be modelled by a softened step function at loglog while for the satellite agn hod we find a preference for an increasing agn fraction with suggesting that the average number of agn in satellite galaxies grows slower ( ) than the linear proportion ( ) observed for the satellite hod of samples of galaxies . we present an estimate of the projected auto correlation function ( acf ) of galaxy groups over the range of 0.1 - 40 at . we use the large - scale clustering signal to verify the agreement between the group bias estimated by using the observed galaxy groups acf and the value derived from the group mass estimates . we perform a measurement of the projected agn - galaxy group cross - correlation function , excluding from the analysis agn that are within galaxy groups and we model the 2-halo term of the clustering signal with the mean agn hod based on our results .

1209.2420

in the past decade it has become clear that the majority of galaxies have likely had one or more brief active periods , shifting our view of astrophysical black holes ( bhs ) from the role of exotic phenomena to fundamental ingredients of cosmic structures . the question of when and under which physical conditions the active galactic nuclei ( agn ) activate is important for understanding not only the origin and evolution of bhs but also the origin and evolution of galaxies . the clustering analysis can powerfully test theoretical model predictions and address which physical processes trigger agn activity . in the framework of the cold dark matter structure formation ( cdm ) , the spatial distribution of agn in the universe ( described by the correlation function , cf ) can be related to how agn are biased with respect to the underlying matter distribution and to the typical mass of dark matter halos ( dmhs ) in which they reside . there have been several studies of the bias evolution of optical quasars with redshift , based on survey such as 2qz and sdss @xcite . all the previous studies infer the picture that the quasar bias evolves with redshift following a constant mass evolution in the range log@xmath11 [ ] @xmath12 , i.e. halo masses similar to group scales , where the combination of low velocity dispersion and moderate galaxy space density yields to the highest probability of a close encounter @xcite . models of major mergers between gas - rich galaxies appear to naturally produce many observed properties of quasars , such as the shape and the evolution of the quasar luminosity function and the large - scale quasar clustering as a function of luminosity and redshift @xcite , supporting the scenario in which major mergers dominate the bright quasar populations . on the other hand , the majority of the results on the clustering of x - ray selected agn , suggest a picture where moderate - luminosity agn live in massive dmhs ( @xmath13 [ ] @xmath14 ) up to @xmath15 @xcite , i.e. x - ray selected agn samples appear to cluster more strongly than luminous quasars ( see * ? ? ? * for a review on the argument ) . the reason for this is not completely clear , but several studies argued that this large bias and dmh masses could suggest a different agn triggering mechanism with respect to bright quasars characterized by galaxy merger - induced fueling . several studies on the morphology of the agn host galaxies have demonstrated that major mergers of galaxies are not likely to be the single dominant mechanism responsible for triggering agn activity at low ( @xmath16 ) @xcite and high redshift ( @xmath17 2 ) @xcite . the theoretical understanding of galaxy clustering and bias factor has been greatly enhanced through the halo occupation distribution ( hod ) framework . in this framework , the virialised dmh with typical overdensities of @xmath18 ( defined w.r.t . the mean density ) are described in terms of the probability @xmath19 of a halo of given mass @xmath20 of having @xmath21 galaxies . a simple way to model the complicated shape of @xmath22 is by assuming the existence of two separate galaxy populations within halos , central and satellite galaxies . this method has been used extensively to interpret galaxy cfs @xcite to constrain how various galaxy samples are distributed among dmh as well as whether these galaxies occupy the centers of the dmhs or are satellite galaxies @xcite . these two populations can be modeled with an hod described by a step function above a halo mass limit for central galaxies , and a power - law for satellite galaxies . similarly , the problem of discussing the abundance and spatial distribution of agn can be reduced to studying how they populate their host halos . in fact the observed departure of the agn cf from a power law on small scales ( 1 - 2 ) can be physically interpreted in the language of the halo model , as the transition between two scales - from small scales lying within the dmhs ( 1-halo term ) to those larger than the halo ( 2-halo term ) . the 1-halo term constrains the hod of satellite agn and gives us the average profile of pairs of agn in groups and clusters of galaxies . the 2-halo term reflects the large - scale agn bias driven by the typical mass of the hosting halos . due to the low number density of agn , there have been few results in the literature studying the agn correlation function using hod modeling . previous works of @xcite at @xmath23 and @xcite at @xmath24 on qso using the hod modeling found that @xmath2525% and @xmath2610% of their qsos , respectively , are satellites . @xcite described for the first time the shape of the hod of x - ray selected agn . by using the cross correlation function of rosat - rass agn with sdss galaxies , they modelled the mean agn occupation of dmhs suggesting that the satellite agn fraction increases slow ( or may even decrease ) with @xmath6 , in contrast with the satellite hod of luminosity - limited samples of galaxies . cosmological hydrodynamic simulations have been performed in @xcite to study the mean occupation function of low - luminosity agn as function of redshift and luminosity . they used a softened step function for the central component plus a rolling - off power - law for the satellite component with @xmath27 depending on the redshift and agn luminosity . their results suggest a strong evolution of the agn occupancy in the redshift range @xmath28 estimated at three different luminosities @xmath29 s@xmath30erg . @xcite modelled the hod of sdss quasars at @xmath31 following this parametrization . they found that the satellite occupation becomes significant at mass @xmath32 , i.e. only the most massive halos host multiple quasars at this redshift and only a small satellite fraction ( @xmath33 ) of sdss quasars is required to fit the clustering signal at small scales . moreover , they measured that the quasars hod steepens considerably going from z=1.4 to 3.2 over halo mass scales @xmath34 and that the characteristic halo mass increases with z for central quasars . despite the diverse methods for studying the hod , counting the number of agn within galaxy groups can constrain quite directly the average agn number within a halo as a function of halo mass . the total mass of galaxy groups can be estimated via gravitational lensing and the distribution of agn within halos can be investigated in groups by means of the distribution of the agn host galaxies . separating the contribution to the occupation of halos from agn in satellite or central galaxies can advance our understanding of the co - evolution agn / galaxy and is related to the mechanism of agn activation . on the other hand , the cross - correlation ( ccf ) of agn with galaxy groups provides additional information about how galaxies and bh co - evolve in dense environments . in fact the physical processes that drive galaxy evolution , such as the available cold gas to fuel star formation and the bh growth , are substantially different in groups and clusters compared to the field . many studies over the past decade have presented an evidence that agn at @xmath35 are more frequently found in groups compared to galaxies ( georgakakis et al . 2008 , arnold et al . x - ray observations reveal that a significant fraction of high - z clusters show overdensities of agns in their outskirts ( henry et al . 1991 , cappi et al . 2001 , ruderman et al . 2005 , cappelluti et al . 2005 ) . in this paper , we perform the first direct measurement of the mean halo occupation of x - ray agn and of the projected cross - correlation function of agn with galaxy groups , using a sample of x - ray selected agn and galaxy groups in the cosmos field at @xmath0 , from xmm and _ chandra _ data . we use a @xmath36cdm cosmology with @xmath37 , @xmath38 , @xmath39 , @xmath40 . for comparison with previous measurements we refer to correlation lengths and distances in units of @xmath41 mpc comoving , where @xmath42 h km s@xmath30 mpc@xmath30 . agn luminosities and galaxy groups masses are calculated using @xmath43 = 0.72 . the _ cosmic evolution survey _ ( cosmos ) is a multiwavelength survey over @xmath44 deg@xmath45 of equatorial sky observed by the most advanced astronomical facilities hst @xcite , subaru @xcite , spitzer @xcite , galex @xcite , _ xmm - newton _ @xcite , _ chandra _ @xcite , designed to thoroughly probe the evolution of galaxies , agn , and dark matter in the context of their cosmic environment ( lss ) . xmm-_newton _ surveyed @xmath46 deg@xmath45 of this field in the 0.5 - 10 kev energy band for a total of @xmath47 1.55 ms providing a large sample of 1822 point - like x - ray sources @xcite , down to limiting fluxes of @xmath48 , @xmath49 and @xmath50 erg @xmath51s@xmath30 in the soft , hard and ultra hard band , respectively . the inner part of the cosmos field ( @xmath52 deg@xmath45 ) has been imaged for a total of 1.8 ms by _ down to limiting fluxes 3 - 4 times deeper than xmm - cosmos . of the 1822 xmm sources , 945 have been observed by _ chandra _ and 875 of them are presented in the c - cosmos point - like source catalog @xcite . @xcite presented the xmm - cosmos multiwavelength catalog of x - ray sources with optical / near - infrared identification , multiwavelength properties and redshift information . starting from this catalog , we restricted the analysis to a sample of x - ray agn ( we removed normal galaxies and ambiguous sources ) detected in the soft band which guarantees the largest sample of x - ray agn in the cosmos field , compared to agn observed in the hard or ultra - hard band . moreover , the selection in only one band allows a more simple treatment of the agn x - ray luminosity function used to correct the agn hod ( [ sec : hod ] ) and of the xmm - and c - cosmos sensitivity maps used to generate the agn random catalog ( [ sec : ccf ] ) . specifically , we selected a sample of 280 and 83 soft xmm - cosmos agn with spectroscopic and photometric redshift @xcite @xmath0 , respectively . the use of photometric redshifts for group membership assignment has been successfully demonstrated in @xcite . note that 184/363 sources are also _ chandra _ detected agn . in order to test if the agn halo occupation significantly changes including sources from the c - cosmos catalog that are only chandra detected ( hereafter c - cosmos agn ) ( see [ sec : hod ] ) , we included in the analysis a sample of 107 and 61 agn , detected in the soft band , with known spectroscopic or photometric redshifts @xmath0 , respectively . the rest - frame soft x - ray luminosity as a function of redshift is shown in fig . [ fig : histoz ] ( _ left panel _ ) for xmm - cosmos agn ( circles ) and c - cosmos agn ( triangles ) . we used a catalog with 270 galaxy groups , detected in the co - added _ xmm - newton _ and _ chandra _ data . the general data reduction process is described in @xcite and details regarding improvements and modification to the initial catalog are given in @xcite and @xcite . the identification of the groups has been done using the red sequence technique , and spectroscopic identification of groups has been achieved through zcosmos - bright program @xcite , targeted follow - up using imacs / magellan and fors2/vlt @xcite , gemini gmos - s @xcite as well as through secondary targets on keck runs by cosmos collaboration . at @xmath53 the spectroscopic completeness has achieved 90% ( a. finoguenov 2012 , in preparation ) . as described in @xcite , the total x - ray fluxes have been obtained from the measured fluxes by assuming a beta profile and by removing the flux that is due to embedded agn point sources . some of the faint chandra agn could not be removed from the group fluxes , with their contribution to the total flux being @xmath54 . the rest - frame luminosities have been computed in @xcite and @xcite from the total flux following @xmath55 , where @xmath56 is the k - correction and @xmath57 is an iterative correction factor , while to estimate the temperature of each group they used the @xmath58 relation of @xcite . a quality flag ( hereafter xflag ) is assigned to the reliability of the optical counterpart , with flags 1 and 2 indicating a secure association , and higher flags indicating potential problems due to projections with other sources or bad photometry due to bright stars in the foreground . in detail , xflag=1,2 are assigned to groups with a confident spectroscopic association , while systems with only the red sequence identification have _ xflag=3 _ ( see * ? ? ? * for more details ) . the line - of - sight position of the group is assigned to be the centroid of the x - ray emission ( the accuracy of the determination of the x - ray center is higher for _ xflag=1 _ ) . if the x - ray centroid is not precise enough to be used directly , the most massive central galaxy ( mmcg ) located near the peak of the x - ray emission has been used to trace the center of the dm halos of groups ( see * ? ? ? * ; * ? ? ? * for more details ) . group masses @xmath59 are assigned from an empirical mass - luminosity relation , described in @xcite , @xmath60 \nonumber\end{aligned}\ ] ] where @xmath59 is the mass within the radius containing the density of matter 200 times the critical density , in units of @xmath61 . @xmath62 are the fitting parameters , @xmath63 are the calibration parameters and @xmath64 is the correction for redshift evolution of scaling relations , which has been shown in @xcite to reproduce well the @xmath65 relation of cosmos groups . in order to be consistent in comparing these mass values with the ones obtained studying the clustering properties of groups , we accounted for the difference between the mass defined with respect to 200 times the critical density and with respect to 200 times the mean density ( hereafter @xmath59 refers to masses obtained using the definition with respect to the mean density ) . in fact , the absolute bias which we are going to derive from the dm correlation function is based on the shape of the dm mass function defined with respect to the mean density . starting from the navarro - frenk - white ( nfw ) profile with a concentration parameter @xmath66 , we derived the relation between the two mass definitions , @xmath67 . in this work , we make use of galaxy groups with @xmath0 and identification flag _ @xmath68 _ which removes problematic identification cases , obtaining a catalog of 189 x - ray galaxy groups over 1.64 deg@xmath45 with a rest - frame 0.1 - 2.4 kev luminosity range of @xmath69erg]@xmath70 , and mass range of @xmath71 < 14.5 $ ] ( see fig . [ fig : histoz ] , _ right panel _ ) . we define group members as agn located within @xmath72 and @xmath73 from the group centers , where @xmath74 is the group line - of - sight velocity dispersion and @xmath75 is the virial radius of a group within which the mean density is 200 times the mean density of the universe at the group redshift . in this analysis we used the sample of 363 xmm - cosmos agn with @xmath0 described in [ subsec : agn ] and we found 41 sources ( 35/41 are also _ chandra _ detected ) in galaxy groups with median @xmath76 and median @xmath77 @xmath78erg . when we include in the analysis the sample of 168 c - cosmos agn ( [ subsec : agn ] ) , we found 17 additional agn in galaxy groups , with known spectroscopic or photometric @xmath0 . as expected c - cosmos agn have lower soft fluxes respect to xmm - cosmos agn at any redshift , with a median @xmath79 @xmath78erg . in particular we found 2 galaxy groups at @xmath80 and 2 groups at @xmath81 with 2 agn per halo and 1 group with 3 agn at @xmath82 . all the properties of agn and galaxy group samples are summarized in table [ tbl-1 ] , while table [ tbl-2 ] shows the catalog of 58 xmm + c - cosmos agn in galaxy groups . for the sources with known spectroscopic redshifts ( 47/58 ) we know the classification in bl or non - bl agn as described in @xcite and @xcite for xmm and c - cosmos agn , respectively . in detail , we found 43 non - bl and 4 bl in the xmm + c - cosmos agn sample . [ fig : histoz ] ( _ left panel _ ) shows the rest - frame soft x - ray luminosity as a function of redshift for the subsamples of xmm ( green squares ) and c - cosmos selected agn ( magenta hexagon ) within @xmath75 . we cross - matched the sample of agn in groups with a galaxy membership catalog @xcite and we verified that agn classified as group members based on our method , have host galaxies associated with the same galaxy groups . we divided the sample of 58 agn in groups in two subsets , according to their association with bcgs . in detail we found that 22/58 ( 16/41 ) agn are in central galaxies , while 36/58 ( 25/41 ) are in satellites . llll + xmm - agn & 363 & 0.66 & @xmath83 + c - cosmos agn & 168 & 0.56 & @xmath84 + xmm - agn in @xmath75 & 41 & 0.55 & @xmath85 + c - cosmos agn in @xmath75 & 17 & 0.53 & @xmath86 + xmm - agn in the field & 253 & 0.67 & @xmath83 + xmm+c - cosmos agn in @xmath75 & 58 & 0.55 & @xmath87 + - satellites & 36 & 0.56 & @xmath88 + - centrals & 22 & 0.50 & @xmath89 + + & & & @xmath90 + & & & + galaxy groups & 189 & 0.56 & @xmath91 + xmm+c - cosmos agn host groups & 52 & 0.55 & @xmath92 + [ fig : massfunc ] ( _ left panel _ ) shows the mass function of all x - ray galaxy groups and those marked by agn presence , showing separately the contributions of groups hosting an agn in central or satellite galaxies . we calculated the mass function by using the standard @xmath93 method ( schmidt 1968 ) and we counted twice galaxy groups with 2 agn . hence , in the @xmath94 mass bin , the comoving space density ( @xmath95 ) and its corresponding error ( @xmath96 ) are computed by ( see bondi et al . 2008 ) : @xmath97 in estimating the average number of agn occupying a halo of mass @xmath59 , some major effects need to be taken into consideration . the sample of agn in @xmath75 is a flux - limited sample and brighter agn are detected at higher redshift . similarly , we observe galaxy groups with small halo mass only at low redshift . however the relatively small number of cosmos agn in galaxy groups does not allow us to select a volume complete subsample by using a cut in luminosity at log@xmath98[]=42.4 . we have therefore made a correction for the effect of changes in the agn density as a function of redshift and limiting luminosity , by using the agn x - ray luminosity function ( xlf ) . it has been shown in different works that the luminosity dependent density evolution ( ldde ) model for the xlf provides the best framework that describes the evolutionary properties of agn , both in the soft @xcite and hard x - rays @xcite . in this paper we modelled the agn soft xlf with the ldde xlf described in @xcite , because they modelled the soft xlf including in the analysis both type 1 and type 2 agn , but we verified that using different best - fit parameters of the global xlf expression , the resulting mean agn occupation stays within the error bars . then for each agn redshift we defined two weights : @xmath99 to correct for the fact that we are including in the analysis agn with log@xmath100[]@xmath10142.2 and @xmath102 to correct for this effect plus the redshift evolution of the agn density : @xmath103 @xmath104 @xmath105 is the soft xlf proposed in @xcite and @xmath106 depends on the survey flux limit and is a function of redshift . in detail , following equations 9,11 and 12 and table 2 in @xcite , we described the shape of the present - day luminosity function with slopes @xmath107 and @xmath108 , log@xmath109[h@xmath110 = 43.65 @xmath111 0.05 which is the value of the luminosity where the change of slope occurs and normalization @xmath112h@xmath113mpc@xmath114 . we estimated the evolution factor assuming @xmath115 , @xmath116 , @xmath117 , log@xmath118 = 44.6h@xmath119 and @xmath120 . based on eq . [ eq : w ] and [ eq : w0 ] , the comoving space density @xmath95 of agn hosting groups corrected for the xlf is given by : @xmath121 while the @xmath95 corrected for both the xlf and the z evolution is estimated by using : @xmath122 fig . [ fig : massfunc ] ( _ right panel _ ) shows the mass function when all these effects are corrected . the ratio between the mass function of x - ray groups hosting agn within @xmath75 by that of all x - ray galaxy groups generates the mean agn hod , which describes the occupation of dm halos by agn . [ fig : hodagn ] ( _ left panel _ ) shows the observed average number of xmm - cosmos agn in a halo of given mass as a function of @xmath59 ( grey triangles ) and the average number after correcting for the xlf ( green open circles ) and for both the xlf and the z evolution ( black filled circles ) . the redshift evolution of the agn fraction is shown in fig . [ fig : massfunc ] ( _ right panel _ ) . the grey circles represent the mean agn occupation as a function of redshift for halos with @xmath123 while the green circles and the black squares show the agn fraction when we correct for the xlf only and for both the xlf and the z evolution effects , respectively . when we correct for the xlf , the mean occupation increases with z since in the agn luminosity range of our interest , the agn density increases with z , while the redshift correction removes this trend producing a constant agn fraction . we fitted the total agn hod assuming the model with a rolling - off power - law : @xmath124 where @xmath125 is the normalization , @xmath126 is the halo mass at which the number of central agn is equal to that of satellite agn , @xmath127 is a cut - off mass scale . with our data alone , we can not make meaningful constraints on @xmath126 , therefore we fixed log@xmath128 following the results of @xcite . we verified that the result does not change for log@xmath129 , due to the fact that the fraction of agn among central and satellite galaxies are comparable in this mass range . for the modelling of the rolling - off power - law , we assumed log@xmath130 which corresponds to the mass below which our data points decay exponentially . we obtained constraints in the ( @xmath131 , @xmath125 ) - space and we found as best fit parameters of the mean agn hod , @xmath132 and @xmath133 , where the @xmath134 confidence interval for a combined two parameter fit ( @xmath135 ) is given in the brackets . [ fig : hod ] ( _ left panel _ ) shows the mean occupation of xmm - cosmos agn with the best fit parameters ( solid black line ) , and 1@xmath74 confidence interval ( shaded grey region ) , compared to the mean agn hod including c - cosmos agn in the analysis . the solid red line corresponds to the best fit model for the mean occupation of xmm+c - cosmos agn , with @xmath136 and @xmath137 , while the shaded region is the @xmath138 confidence interval ( see fig . [ fig : hod ] , _ right panel _ ) . moreover by dividing the mass function of satellite ( central ) agn host groups by that of all x - ray galaxy groups , we provide the fraction of agn among satellite ( central ) galaxies as a function of halo mass ( see fig . [ fig : hodsatcen ] ) . we model the mean agn occupation function in halos by decomposing it into the central and satellite contribution @xmath139 : @xmath140 where the central agn occupation follows a softened step function and the satellite occupation a rolling - off power law ( e.g. kravtsov et a. 2004 , zheng et al . 2005 , zehavi et al . 2005 , tinker et al . 2005 , conroy et al . 2006 , chatterjee et al . 2011 , richardson et al . 2012 ) . in this formalism there are four free parameters @xmath141 , @xmath142 , @xmath143 and @xmath144 , where @xmath142 is the minimum mass where the occupation of central agn is zero . as shown in fig . [ fig : hodsatcen ] , the hod of central agn is described by a softened step function with log@xmath145=12.75(12.10,12.95)$ ] and @xmath146 where the errors are the @xmath138 confidence intervals estimated by using the @xmath147 minimization technique with 2 free parameters . on the other hand , the satellite agn hod suggests a picture in which the average number of satellite agn increases with @xmath6 ( @xmath148 and @xmath149 ) slower than the satellite hod of samples of galaxies ( @xmath150 ) . @xcite first used a sample of rosat - rass agn and sdss galaxies at @xmath151 to study the occupancy of x - ray agn in dm halos . they investigate three models : the first ( model a ) assumes that agn only reside in satellite galaxies while the others explore the effects of centrals ( model b and c ) . in particular , in model b , the hod of central agn is constant and the satellite hod has a power - law form at halo masses above @xmath142 . model c has the same form except that only less massive dm halos contain central agn . our results clearly show that only a model which includes agn in both satellite and central galaxies at any halo mass @xmath152 can reproduce the observed agn hod . in agreement with our results , they found that the upper limit of the power - law index of the satellite hod is below unity , with @xmath153 . in detail , their model b with best fit parameters @xmath154 , log@xmath142 [ ] = 12.5 and @xmath155 is in perfect agreement with our results , showing that the luminosity and redshift evolution of the mean agn hod is not strong in the luminosity and redshift ranges of our interest . in fact , we used an agn sample with @xmath156 , while the results in @xcite provide the mean agn hod at @xmath157 for more luminous agn with @xmath158 without any correction for the z evolution . then the two models suggest a similar positive @xmath144 range , but negative values of the slope are not rejected ( @xmath159 ) in their model . note that while in our case the agn fraction is a free parameter , they constrained it by normalizing the agn hod to the observed agn number density . an assumption in the corrections applied by using eq . [ eq : w ] and [ eq : w0 ] , is that the shape of the xlf remains the same between agn in groups and the general agn population at @xmath160[]@xmath161 . we still do not have sufficient observational basis to estimate the effects of possible difference in the xlf shapes between group environment and the field . at higher luminosities , krumpe et al . ( 2010 ; 2012 ) found that agn at log@xmath162 [ ] @xmath163 have higher bias values than those at @xmath164 , corresponding to the mean host halo mass of @xmath165 [ ] = 13.0 and 13.3 respectively @xcite . viewing this trend from the @xmath166 dependence of the xlf shape , the xlf in high @xmath166 environments have xlf more biased towards higher @xmath167 . if the positive correlations between @xmath167 and the host halo mass @xmath166 extends to luminosities in the range @xmath168 [ ] @xmath161 , i.e. the range of @xmath169 ( see fig . 1 left ) , the correction factors using the overall xlf in eq . 3 and 4 are overestimated at higher @xmath167 and underestimated at lower @xmath167 . a quantitative assessment of this effect needs an estimate of the bivariate x - ray luminosity - host dmh mass function , which is far from being available we comment that , if the higher @xmath100 agn preferentially occupy @xmath170[]@xmath171 dmhs than the field , our estimate of @xmath172 should be corrected to a lower value . the two - point auto - correlation function ( acf ) describes the excess probability over random of finding a pair with an object in the volume @xmath173 and another in the volume @xmath174 , separated by a distance @xmath175 so that @xmath176dv_1 dv_2 $ ] , where @xmath177 is the mean space density . a known effect when measuring pair separations is that the peculiar velocities combined with the hubble flow may cause a biased estimate of the distance when using the spectroscopic redshift . to avoid this effect it is usually computed the projected acf @xcite : @xmath178 where @xmath179 is the distance component perpendicular to the line of sight and @xmath180 parallel to the line of sight @xcite . the acf is estimated by using the minimum variance estimator described by @xcite : @xmath181 where dd , dr and rr are the normalized number of data - data , data - random , and random - random source pairs , respectively . [ eq : landy ] indicates that an accurate estimate of the distribution function of the random samples is crucial in order to obtain a reliable estimate of @xmath182 . several observational biases must be taken into account when generating a random sample of agn in a x - ray flux limited survey . in particular , in order to reproduce the selection function of the survey , one has to carefully reproduce the space and flux distributions of agn , since the sensitivity in x - ray surveys is not homogeneous on the detector and therefore on the sky . an other important choice for obtaining a reliable estimate of @xmath183 , is to set @xmath184 in the calculation of the integral above . one should avoid values of @xmath184 too large since they would add noise to the estimate of @xmath185 . if , instead , @xmath184 is too small one could not recover all the signal . we created an agn random sample where each simulated source is placed at a random position in the sky , with flux randomly extracted from the catalog of real source fluxes . the simulated source is kept in the random sample if its flux is above the sensitivity map value at that position @xcite . the corresponding redshift for a random object is assigned based on the smoothed redshift distribution of the agn sample . similarly , an unclustered catalog of galaxy groups mimicking the selection function of the survey must be employed to quantify the degree to which the groups preferentially locate themselves in one another s neighborhood . we generate a random group catalog by calculating at each area of a given sensitivity , the probability of observing a group of a given mass and redshift . we then used monte - carlo simulation of the group positional assignment , finally producing a catalog of hundred thousand objects . also the x - ray surface brightness sensitivity map is non - uniform in depth and consequently the probability of detecting groups of a particular mass is variable with redshift ; in particular the minimum mass below which a group will be detected is an increasing function of @xmath186 . we measured the projected acf of galaxy groups in the range @xmath187 by using eq . [ eq : wrp ] , with @xmath188 ( see fig.[fig : acfgroups ] ) . the errors have been estimated using bootstrap resampling of the data , which consists of computing the variance of @xmath185 in @xmath189 bootstrap realizations of the sample . each realization is obtained by randomly selecting a subset of groups from the data sample allowing for repetitions . in the halo model approach , the clustering signal can be modelled as the sum of two contributions of pairs from the same dm halo ( 1-halo term ) and those from different dm halos ( 2-halo term ) . in fourier space , the 2-halo term can be explicitly written as ( seljak 2000 , cooray & sheth 2002 ) : @xmath190 where @xmath191 is the linear power spectrum and @xmath192 is the bias factor of the sample . then the galaxy groups two - point correlation function at large scales is given by : @xmath193 where @xmath194 is the zeroth - order bessel function of the first kind . following this model , the galaxy group bias defines the relation between the 2-halo term of dm and groups clustering signal : @xmath195 according to this equation , we estimated an average bias factor in the range @xmath196 equal to @xmath197 where the error corresponds to @xmath198 using a @xmath147 minimization technique with 1 free parameter . following the bias mass relation @xmath199 described in @xcite and @xcite , this observed bias , with respect to the dm distribution at @xmath200 corresponds to a typical mass log@xmath201 [ ] = 13.65@xmath202 . on the other hand the average bias of galaxy groups can be estimated starting from the known masses @xmath59 . since usually the bias factor is related to the average dm halo mass expressed in units of @xmath203 , hereafter we refer to @xmath6 to indicate the galaxy groups masses in these units . since we know the mass estimates , we can predict the bias factor of this sample of galaxy groups . by accounting for the fact that the linear regime of the structure formation is verified only at large scales , we estimated the average bias of the sample , including only pairs which contribute to the clustering signal at @xmath196 mpc @xmath41 . as described in @xcite , we measured the average bias factor of the sample as : @xmath204 where @xmath205 is the bias factor of the group pair @xmath206 and @xmath207 is the total number of group pairs in the range @xmath196 mpc @xmath41 . the @xmath208 factor is defined by @xmath209 , where @xmath210 is the growth function ( see eq . ( 10 ) in @xcite and references therein ) and takes into account that the amplitude of the dm 2-halo term decreases with increasing redshift . + by using this approach we obtained @xmath211 where the errors have been estimated assuming a @xmath212 error on the galaxy groups masses . this value is in perfect agreement with the bias obtained from the acf , with @xmath213 . measurements of the cross - correlation function between agn and groups use a version of the estimator proposed by @xcite : @xmath214 where each data sample , with pair counts @xmath215 has an associated random catalog , with pair counts @xmath216 normalized by its number density . we estimated the cross - correlation function of 189 galaxy groups and a subset of 253 xmm - cosmos agn with known spectroscopic redshift @xmath217 , obtained excluding those within @xmath75 ( see fig . [ fig : ccf ] ) . following eq . [ eq : b1 ] , the linear bias factor of the projected ccf of agn with galaxy groups can be approximated by using the 2-halo term : @xmath218 where @xmath219 and @xmath220 are the bias factor of agn and galaxy groups , respectively and @xmath221 is the projected dark matter cf . fig . [ fig : ccf ] ( _ lower panel _ ) shows the linear bias @xmath222 as a function of @xmath179 over the scales @xmath223 . we fitted the data points with a constant by using the @xmath147 minimization technique and we found @xmath224 . the shaded regions show the bias values for which @xmath225=1 and 4 ( 68% and 99% confidence levels for one parameter ) . based on the halo model approach , the agn bias factor depends on the mean agn hod @xmath226 : @xmath227 where @xmath142 is the minimum mass below which the agn hod is zero , @xmath228 and @xmath229 are the halo mass function and the halo bias given by @xcite . note that the large - scale bias factor does not depend on the normalization log@xmath230 = 10^{13.8}$ ] and on @xmath125 . since we are estimating the ccf excluding from the xmm - cosmos sample 41 agn in galaxy groups , the clustering signal at large scale is due to agn that live in halos with @xmath231 or with masses that we can not observe at a given redshift @xmath186 . this implies that the agn bias can be written as : @xmath232 where @xmath233 is the minimum mass we can observe for a group at redshift @xmath186 ( at @xmath16 we only detect luminous and then massive groups ) . note that we are assuming the separability of the mass and redshift dependence of the @xmath234 , i.e. @xmath235 . following this method , the bias factor @xmath236 is defined by : @xmath237 where the sum is over the pairs @xmath238 contributing to the clustering signal at large scale and @xmath239 is the agn bias ( eq . [ eq : bagn ] ) assuming a rolling - off power - law hod based on our results . @xmath240 is the bias associated to the galaxy group mass and redshift following @xcite and @xmath207 is the total number of agn - group pairs in the range @xmath196 . we found @xmath241 , which is in perfect agreement with the observed bias factor defined in eq . [ eq : biasccf ] . the errors are due to the @xmath138 errors on the power - law index @xmath131 , @xmath142 and on @xmath220 . similarly we can define the average halo mass @xmath242 corresponding to the observed ccf signal , i.e. : @xmath243 where @xmath244 is the average agn host halo mass at redshift @xmath245 : @xmath246 while @xmath247 is the galaxy group mass , both in units of . fixing the values of @xmath131 and @xmath142 , we found the average mass of xmm - cosmos agn to be @xmath248 [ ] = 13.20(13.10;13.25 ) , where the errors correspond to the 68% confidence region in the log@xmath249 space . using a sample of x - ray selected agn and galaxy groups in the cosmos field at @xmath0 , we performed the first direct measurement of the agn hod in the mass range @xmath250 , based on the mass function of galaxy groups hosting agn . in contrast to previous works using the clustering signal of the sample , we directly counted the number of agn within galaxy groups and we found 58 agn in groups , associated to 22 central and 36 satellite galaxies . this allowed us to put constrains on both the mean occupation of agn among satellite and central galaxies as function of the halo mass , which provides information on the agn triggering mechanism . @xcite studied the halo occupation properties of agn detected by the chandra x - ray observatory in the bootes field over a redshift interval from z=0.17 to z=3 , showing that x - ray agns are predominantly located at the centers of dmhs with @xmath251 , with an upper limit of the satellite fraction of 0.1 ( @xmath252 ) . the central locations of the quasar host galaxies are expected in major merger models because mergers of equally sized galaxies preferentially occur at the centers of dmhs ( hopkins et al . 2008 ) . on the contrary @xcite observed the presence of the one - halo term in the cross - correlation function of optically selected @xmath253 quasars and luminous red galaxies , and use this to conclude that a large fraction of the agns is hosted by satellite galaxies . our results show that the average number of agn in satellite galaxies in the halo mass range @xmath254=13 - 14.5 $ ] , and agn luminosity @xmath255 [ ] = 42.3 might be comparable or even larger that the average number of agn in central galaxies , i.e. x - ray agn do not avoid satellite galaxies . a high fraction of agn in satellite galaxies is expected in a picture where other phenomena like secular processes , might become dominant in the agn activation . @xcite , @xcite and @xcite showed that low - luminosity agn could be triggered in more common nonmerger events , like stochastic encounters of the black holes and molecular clouds , tidal disruption or disk instability . this leads to the expectation of a characteristic transition to merger - induced fueling around the traditional quasar - seyfert luminosity division . moreover we found the power - law slope , which defines the evolution of the mean satellite hod with halo mass , to be @xmath256 , suggesting a picture in which the average number of satellite agn per halo increases with the halo mass . on the other hand , @xcite obtained @xmath257 , but negative values of the slope are not rejected ( @xmath159 ) . it is interesting to compare this result with hod analyses of galaxies . previous hod analyses of galaxies found @xmath258 for a wide range of absolute magnitudes and redshifts at least up to @xmath259 @xcite , implying a simple proportionality between halo mass and satellite number , @xmath260 . our results suggest that the mean hod of satellite agn might increase slower ( @xmath261 ) with the halo mass respect to the linear proportion ( @xmath262 ) in the satellite galaxy hod , i.e. the agn is not only triggered by the halo mass . on the contrary , a decreasing agn fraction with the halo mass is consistent with previous observations that the agn fraction is smaller in clusters than in groups in the nearby universe . in order to fully understand the growth history of smbhs as well as the physical processes responsible for the agn activity we need to explore the agn hod at different redshifts , luminosities , and agn types . the luminosity distribution of agn that reside in halos of a given mass provides a tool to examine the distribution of halo mass for a given luminosity and study luminosity dependent clustering . while this formalism has been widely used in modelling galaxy clustering , it is still not applicable to agn . in fact , it is also important to have larger numbers of agn in galaxy groups which will enable stronger constraints on the shape of the satellite and central agn hod , hopefully as function of agn luminosity and redshift . we have performed the first direct measurement of the mean halo occupation distribution of x - ray agn as function of halo mass , by directly counting the number of agn within x - ray galaxy groups with masses @xmath1 = 13- 14.5 $ ] , in the cosmos field at @xmath0 . our findings are summarized as follows . 1 . we identified 41 xmm - cosmos agn within galaxy groups , defined as agn located within 3 times the group line - of - sight velocity dispersion and within @xmath75 and 17 additional sources including in the analysis c - cosmos only selected agn . we measured the mean agn occupancy of galaxy groups as function of halo mass in the range log@xmath263 = 13- 14.5 $ ] and we modelled the data points with a rolling - off power - law with the best fit index @xmath264 and normalization parameter @xmath265 . 3 . using a galaxy membership catalog , we associated 22/58 and 36/58 agn to central and satellites galaxies , respectively . we constrained that the mean agn occupation function among central galaxies is described by a softened step function above log@xmath5 while the satellite agn hod increases with the halo mass ( @xmath266 ) slower than the satellite hod of sample of galaxies ( @xmath267 ) . we presented an estimate of the projected acf of galaxy groups over the range @xmath9 0.1 - 40 at @xmath10 . we verified that the bias factor and the corresponding typical halo mass estimated from the observed galaxy group acf , are in perfect agreement with the values @xmath220 and @xmath268 obtained by using the galaxy group mass estimates . in particular we found @xmath211 and log @xmath268 [ ] = 13.61@xmath269 , llllllllllll + 11 & 150.1898 & 1.65725 & 0.22 & 14.28 & -99 & 150.19864 & 1.671924 & 12390 & 0.228@xmath270 & 41.27 & 0 + 17 & 149.96413 & 1.68033 & 0.372 & 14.02 & 30744 & 149.96396 & 1.6805983 & -99 & 0.372 & 43.28 & 1 + 19 & 150.37282 & 1.60944 & 0.103 & 12.98 & 2021 & 150.37256 & 1.6093965 & 1678 & 0.104 & 41.22 & 1 + 20 & 150.32494 & 1.60313 & 0.22 & 13.38 & 2186 & 150.33601 & 1.6012201 & 1671 & 0.234@xmath270 & 41.88 & 0 + 35 & 150.2066 & 1.82327 & 0.53 & 13.70 & -99 & 150.20702 & 1.823398 & 1292 & 0.529 & 41.73 & 1 + 39 & 149.82381 & 1.8252701 & 0.531 & 13.63 & 5502 & 149.81248 & 1.8238187 & 219 & 0.529 & 42.37 & 0 + 52 & 150.44704 & 1.8828501 & 0.671 & 13.64 & 30681 & 150.44731 & 1.8832886 & 690 & 0.670@xmath270 & 42.75 & 1 + 69 & 150.42012 & 1.9708 & 0.862 & 13.49 & 5519 & 150.42473 & 1.9693011 & 1185 & 0.854@xmath270 & 42.77 & 0 + 78 & 150.27461 & 1.98884 & 0.838 & 13.61 & -99 & 150.27823 & 1.990302 & 685 & 0.838 & 42.35 & 1 + 78 & 150.27461 & 1.9888 & 0.838 & 13.61 & 206 & 150.27434 & 1.9884306 & 1100 & 0.673 & 42.57 & 0 + 79 & 150.44728 & 2.05392 & 0.323 & 13.77 & -99 & 150.44765 & 2.053961 & 2605 & 0.323 & 41.62 & 1 + 87 & 150.51109 & 2.0269899 & 0.899 & 13.60 & 2387 & 150.51036 & 2.0293686 & 496 & 0.899 & 43.15 & 1 + 93 & 149.6692 & 2.0740 & 0.338 & 13.59 & 417 & 149.6692 & 2.073962 & 168 & 0.34 & 42.76 & 0 + 110 & 150.17979 & 2.1103699 & 0.361 & 13.43 & 6 & 150.17978 & 2.1101542 & 42 & 0.360 & 43.19 & 1 + 118 & 149.63463 & 2.1357 & 0.962 & 13.70 & 411 & 149.63828 & 2.1494889 & 323 & 0.952 & 43.25 & 0 + 124 & 150.05656 & 2.2085 & 0.186 & 13.22 & 302 & 150.0571 & 2.2063098 & 1297 & 0.186 & 41.25 & 0 + 124 & 150.05656 & 2.20854 & 0.186 & 13.22 & -99 & 150.09554 & 2.2202671 & 1221 & 0.186 & 40.86 & 0 + 128 & 150.58435 & 2.1811 & 0.556 & 13.58 & 2855 & 150.57634 & 2.1812408 & 1634 & 0.554 & 42.41 & 0 + 127 & 150.44104 & 2.15873 & 0.377 & 13.32 & -99 & 150.44202 & 2.1551671 & 2313 & 0.376 & 41.24 & 1 + 130 & 150.02382 & 2.2032 & 0.942 & 13.84 & -99 & 150.02303 & 2.206567 & 2413 & 0.94 & 42.45 & 0 + 136 & 150.17493 & 2.2170 & 0.676 & 13.56 & -99 & 150.17592 & 2.2156539 & 877 & 0.667@xmath270 & 42.08 & 0 + 137 & 149.96271 & 2.2102399 & 0.425 & 13.32 & -99 & 149.96216 & 2.2102211 & 12011 & 0.425 & 41.44 & 1 + 138 & 149.50874 & 2.2614 & 0.943 & 13.97 & 5539 & 149.50914 & 2.261276 & -99 & 0.962 & 43.19 & 1 + 142 & 150.28798 & 2.2769 & 0.122 & 13.09 & 317 & 150.25331 & 2.2779887 & 3718 & 0.165 & 41.08 & 0 + 143 & 150.21454 & 2.2801 & 0.880 & 13.77 & 116 & 150.20604 & 2.2857771 & 22 & 0.874 & 43.33 & 0 + 149 & 150.41566 & 2.4302 & 0.124 & 13.85 & 45 & 150.33598 & 2.4335926 & 658 & 0.121 & 41.31 & 0 + 149 & 150.41566 & 2.4302001 & 0.124 & 13.85 & 73 & 150.416 & 2.4299896 & 634 & 0.124 & 42.00 & 1 + 161 & 149.95262 & 2.3418 & 0.941 & 13.74 & 430 & 149.95949 & 2.3560882 & 497 & 0.889 & 42.85 & 0 + 171 & 149.66328 & 2.2677 & 0.676 & 13.56 & 5411 & 149.66243 & 2.2693584 & 1646 & 0.676 & 42.76 & 0 + 173 & 150.05804 & 2.38045 & 0.347 & 13.48 & 241 & 150.05794 & 2.3805208 & 898 & 0.347 & 42.14 & 1 + 174 & 149.63988 & 2.3491 & 0.950 & 13.97 & 53937 & 149.64459 & 2.3587193 & -99 & 0.962@xmath270 & 42.88 & 0 + 175 & 150.24123 & 2.34835 & 0.723 & 13.55 & -99 & 150.24109 & 2.3483839 & 977 & 0.703 & 41.72 & 0 + 186 & 150.21748 & 2.4003 & 0.905 & 13.77 & 53303 & 150.21141 & 2.4022212 & 975 & 0.894@xmath270 & 42.62 & 0 + 194 & 149.69957 & 2.4028 & 0.354 & 13.45 & 135 & 149.70084 & 2.4025679 & 417 & 0.376 & 43.00 & 0 + 196 & 150.27898 & 2.4192 & 0.123 & 13.11 & -99 & 150.27975 & 2.4222209 & 1315 & 0.122 & 40.25 & 0 + 216 & 150.06664 & 2.6474 & 0.696 & 13.64 & 5113 & 150.06633 & 2.642817 & 143 & 0.693 & 42.45 & 0 + 217 & 150.00713 & 2.4534299 & 0.731 & 13.54 & -99 & 150.00188 & 2.4606521 & 1243 & 0.732 & 42.39 & 1 + 219 & 150.27148 & 2.5134399 & 0.704 & 13.54 & 158 & 150.27411 & 2.5117824 & 138 & 0.703 & 42.51 & 1 + 220 & 149.92343 & 2.5249 & 0.729 & 14.39 & 486 & 149.92007 & 2.5143571 & 562 & 0.698 & 42.52 & 0 + 220 & 149.92343 & 2.5249901 & 0.729 & 14.39 & -99 & 149.91588 & 2.52139 & 1310 & 0.729 & 42.38 & 0 + 231 & 150.05421 & 2.58885 & 0.675 & 13.73 & 8 & 150.05383 & 2.5896702 & 142 & 0.699 & 44.03 & 1 + 234 & 150.15816 & 2.6082 & 0.893 & 13.66 & -99 & 150.15796 & 2.6113441 & 727 & 0.863@xmath270 & 42.35 & 0 + 237 & 150.11774 & 2.6842501 & 0.349 & 14.00 & 5118 & 150.11783 & 2.6840661 & -99 & 0.349 & 42.18 & 1 + 259 & 149.65717 & 2.8195 & 0.703 & 13.73 & 60270 & 149.65953 & 2.8270493 & -99 & 0.708@xmath270 & 42.33 & 0 + 262 & 149.60007 & 2.8211 & 0.344 & 14.03 & 5320 & 149.63142 & 2.8174951 & -99 & 0.34 & 42.93 & 0 + 275 & 149.83878 & 2.6750801 & 0.259 & 13.45 & 5112 & 149.83847 & 2.6750875 & 1608 & 0.259 & 43.08 & 1 + 277 & 150.00462 & 2.63275 & 0.677 & 13.45 & 5091 & 150.00452 & 2.6328416 & 616 & 0.678 & 42.73 & 1 + 289 & 150.11256 & 2.5560 & 0.501 & 13.57 & 142 & 150.10342 & 2.5504889 & 148 & 0.498 & 42.62 & 0 + 292 & 150.03307 & 2.5524 & 0.747 & 13.48 & 398 & 150.02638 & 2.5620575 & 991 & 0.745 & 42.82 & 0 + 296 & 149.55516 & 2.0020 & 0.894 & 13.90 & 10732 & 149.56131 & 2.0087938 & 3549 & 0.930@xmath270 & 42.85 & 0 + 298 & 149.78191 & 2.1390 & 0.354 & 13.59 & 63 & 149.78223 & 2.1387713 & 313 & 0.355 & 42.57 & 0 + 298 & 149.78191 & 2.13906 & 0.354 & 13.59 & 392 & 149.79369 & 2.1256437 & 679 & 0.353 & 41.98 & 1 + 298 & 149.78191 & 2.13906 & 0.354 & 13.59 & -99 & 149.77357 & 2.141633 & 1498 & 0.354 & 41.34 & 0 + 300 & 149.72893 & 2.2373 & 0.381 & 13.47 & -99 & 149.74792 & 2.253087 & 2876 & 0.356 & 41.06 & 0 + 303 & 149.99364 & 2.2585399 & 0.660 & 13.53 & 19 & 149.99367 & 2.2585886 & 450 & 0.659 & 43.38 & 1 + 322 & 150.2254 & 2.26872&0.677 & 13.51 & -99 & 150.228 & 2.2698331 & 24 & 0.678 & 42.46 & 0 + 324 & 150.02414 & 2.36050&0.726 & 13.39 & 254 & 150.03079 & 2.358371 & 533 & 0.786 & 42.89 & 0 + 333 & 150.0423 & 2.6949 & 0.219 & 13.31 & 5075 & 150.04155 & 2.6945302 & 623 & 0.221 & 41.46 & 1 + allevato v. , et al . 2011 , , 736 , 99 avni , y. & bahcall , j. n. , 1980 , , 235 , 694 balogh , m. l. , et al . 2011 , , 412 , 2303 berlind a. a. , et al . 2003 , , 593 , 1 bierly , r. m. , et al . 2010 , a&a , 523 , 66 bonoli , s. , marulli , f. , springel , v. , et al . 2009 , , 396 , 423 brusa , m. , civano , f. , comastri , a. , et al . 2010 , , 716 , 348 cappelluti , n. , et al . 2007 , , 172 , 341 cappelluti , n. , brusa m. , hasinger g. , et al . 2009 , a&a , 497 , 635 cappelluti , n. , et al . 2010 , , 716 , 209 cappelluti , n. , allevato , v. , finoguenov , a. , advances in astronomy , vol . 2012 , i d . 853701 chatterjee s. , et al . 2012 , , 419 , 2657 cisterans , m. , et al . 2011 , , 726 , 57 civano , f. , et al . 2012 , , 752 , 49 , a. l. , georgakakis , a. , newman , j. a. , et al . 2009 , 701 1484 cooray , a. , sheth , r. , 2002 , phr , 372 , 1 croom , scott m. , boyle , b. j. , shanks , t. , smith , r. j. , et al . 2005 , , 356 , 415 davis , m. , peebles , p. j. e. , 1983 , , 267 , 465 dunlop j. s. , mclure r. j. , kukula , m. j. , et al . 2003 , , 340 , 1095 dunlop j. s. , mclure r. j. , kukula , m. j. , et al . 2003 , , 340 , 1095 elvis , m. , chandra - cosmos team , 2007 , in bullettin of the american astronomical society , vol . 39 , p.899 elvis m. , et al . 2009 , , 184 , 158 eisenstein , daniel j. , hu , wayne . , 1999 , , 511 , 5 finoguenov , a. , et al . 2007 , , 172 , 182 finoguenov , a. , et al . 2010 , , 403 , 2063 fisher , k. b. , davis , m. , strauss , m. a. , yahil , a. & huchra , j. , 1994 , , 266 , 50 gabor , j. m. , et al . 2009 , , 691 , 705 georgakakis , a. , nandra , k. , laird , e. s. , et al . 2007 , , 660 , 15 georgakakis , a. , coil , a. l. , laird , e. s. , et al . 2009 , , 397 , 623 george , matthew r. , et al . 2011 , , 742 , 125 george , matthew r. , et al . 2012 , submitted gilli , r. , daddi , e. , zamorani , g. , et al . 2005 , a&a , 430 , 811 gilli , r. , et al . 2009 , a&a , 494 , 33 grogin , n. a. , et al . 2005 , , 627 , 97 hamana , t. , ouchi , m. , shimasaku , k. , kayo , i. , & suto , y. 2004 , , 347 , 813 hasinger , g. , hasinger , g. , schmidt , m. , 2005 , a&a , 441 .. 417h hasinger , g. , et al . 2007 , , 172 , 29 hickox , r. c. , jones , c. , forman , w. r. , 2009 , , 696 , 891 hogg , d.w , 1999 , arxiv : astro - ph/9905116 hopkins , p.f . , et al . 2007 , , 662 , 110 hopkins , p.f . , hernquist , l. , cox , t.j . , & keres , d. 2008 , , 175 , 356 hopkins , p.f . , hernquist , l. , 2006 , , 166 , 1 hopkins , p.f . , hernquist , l. , 2009 , , 694 , 599 kauffmann , g. , nusser , a. & steinmetz , m. , 1997 , 286 , 795 kocevski , d. d. , et al 2011 , arxiv1109.2588k kravtsov a. v. , et al . 2004 , , 609 , 35 krumpe , m. , miyaji , t. , coil , a. l. 2010 , , 713 , 558 la franca , f. , et al . 2005 , , 635 , 864 landy , s. d. , & szalay a. s. , 1993 , , 412 , 64 leauthaud , a. , finoguenov , a. , kneib , j. p. , et al 2010 , , 709 , 97 leauthaud , a. , et al . 2007 , , 172 , 219 lilly , s. j. , le fvre , o , renzini , a , et al . 2007 , , 172 , 70 lilly , s. j. , le brun , v. , mayer , c. , et al . 2009 , , 184 , 218 markevitch , m. , 1998 , , 504 , 27 mcintosh , d. h. , guo , y. , mo , h. j. , van den bosch , f. , & yang , 2009 , bulletin of the american astronomical society , 41 , 423.09 miyaji , t. , hasinger , g. , schmidt , m. , 2000 , a&a , 353 , 25 miyaji , t. , zamorani , g. , cappelluti , n. , et al . , 2007 , , 172 , 396 miyaji , t. , krumpe , m. , coil , a. l. , & aceves , h. 2011 , , 726 , 83 milosavljevic , m. , merritt , d. , & ho , l. c. 2006 , , 652 , 120 mullis , c. r. , henry , j. p. , gioia , i. m. , et al . , 2004 , , 617 , 192 padmanabhan , n. , white , m. , norberg , p. , & porciani , c. 2009 , , 397 , 1862 pierce , c. m. , et al . 2007 , , 660 , 19 peacock , j. a. , smith , r. e. 2000 , 318 , 1144 phleps s. , peacock j. a. , meisenheimer k. , wolf c. 2006 , a&a , 457 , 145 peebles p. j. e. , 1980 , the large scale structure of the universe ( princeton : princeton univ . press ) porciani , c. , norberg , p. , 2006 , , 371 , 1824 puccetti , s. , et al . 2009 , , 185 , 586 reichard , t. a. , heckmas , t. m. , rudnick , g. , et al . 2009 , , 691 , 1005 rosario , d. j. , mcgurk , r. c. , max , c. e. , et al . 2011 , 2011arxiv1102.1733r richardson , j. , zheng , z. , chatterjee , s. , nagai , d. , and shen , y. , 2012 , arxiv:1104.3550v2 ross , n. p. , shen , y. , strauss , m. a. , et al . 2009 , , 697 , 1634 salvato , m. , et al . 2011 , , 742 , 61 sanders , d. , salvato , m. , aussel , h. , et al . 2007 , , 172 , 86 schawinski , k , , treister , e.,urry , c. m. , et al . 2011 , , 727 , 31 scoville , n. , abraham , r. g. , aussel , h. , et al . 2007 , , 172 , 38 shankar , f. , weinberg d. h. , et al . 2009 , , 690 , 20 shankar f. , et al . , 2010 , , 718 , 231 shankar f. , 2010 , iaus , 267 , 248 shen y. , strauss , m. a. , ross , n. p. , hall , p. b. , et al . 2009 , 697 , 1656 shen y. , 2009 , , 704 , 89 shen , y. , et al . 2010 , , 719 , 1693 sheth r. k. , mo h. j. , tormen g. 2001 , , 323 , 1 silverman , j. d. ; kovac , k. , knobel , c. , , 695 , 171 silverman , j. d. et al . 2011 , , 743 , 2 starikova , s. et al . , 2011 , , 741 , 15 tal , t. , van dokkum p. g. , nelan , j. , et al . 2009 , , 138 , 1417 taniguchi , y. , scoville , n. z. , murayama , t. , et al . 2007 , , 172 , 9 tinker , j. l. , weinberg , d. h. , zheng , z. , zehavi , i. 2005 , , 631 , 41 ueda , y. , akiyama , m. , ohta , k. , & miyaji , t. , 2003 , , 598 , 886 yang , y. , mushotzky , r. f. , barger , a. j. , & cowie , l. l. 2006 , , 645 , 68 van den bosch , f. c. 2002 , , 331 , 98 zamojski , m. a. , schiminovich , d. , rich , r. m. , et al . 2007 , , 172 , 468 zehavi , i. , et al . 2005 , , 630 , 1 zehavi , i. , et al . 2010 , arxiv:1005.2413 zheng , z. , et al . 2005 , , 633 , 791 zheng , z. , coil , a. l. , & zehavi , i. 2007 , , 667 , 760 zheng , z. , zehavi , i. , eisenstein , d. j. , weinberg , d. h. , & jing , y. p. 2009 , , 707 , 554